Coda Liquid Production and Perception in Puerto Rican Spanish
Dissertation
Presented in Partial Fulfillment of the Requirements for the Degree Doctor of Philosophy in the Graduate School of The Ohio State University
By
Mary Elizabeth Beaton
Graduate Program in Spanish and Portuguese
The Ohio State University
2015
Dissertation Committee:
Dr. Rebeka Campos-Astorkiza, Co-Advisor
Dr. Scott A. Schwenter, Co-Advisor
Dr. Terrell A. Morgan
Copyright by
Mary Elizabeth Beaton
2015
ABSTRACT Dialects of Spanish in the Caribbean and southern Spain are described as
"switching" liquids in syllable-final position, resulting in the neutralization of the two sounds. This dissertation considers liquid variation in San Juan Spanish (SJS), which is frequently cited as neutralizing /r/ to /l/ such that arma ('weapon') and alma ('soul') are both pronounced [al.ma]. In light of recent work suggesting that neutralization is often incomplete, i.e. small but significant differences exist between two sounds previous considered to be merged, this study examines the formant structure (F1, F2, F3, F4) and duration of rhotics and laterals in SJS to determine the neutralization status of /r/ and /l/.
In addition, this dissertation features a perception experiment which tests how well SJS listeners are able to hear acoustic differences between the liquids.
Using twenty-four sociolinguistic interviews with SJS speakers, I extracted 2,212 vowel+/r/ and 728 vowel+/l/ sequences. The conditionings of word position, stress, vowel, preceding and following consonants, gender, and age are considered for two separate data analyses. The first analysis considers the conditioning on the manner of articulation of the liquid. Then, approximant liquids, which are the site of potential neutralization, were analyzed for formant structure and duration. In order to develop an understanding of the dynamic formant trajectories, seven equidistant points were sampled for all four formants. These measurements were submitted to both linear regression analyses and Smoothing Spline ANOVAs (SS ANOVAs), which are relatively new to the
ii field. To test liquid perception, an online survey with vowel+liquid audio clips with varying formant structure was presented to both SJS and northern Spain Castilian Spanish
(CS) listeners.
The results for the production of approximant liquids indicate that rhotics are far more variant in this dialect than laterals and their realization depends on linguistic and social factors. Therefore, I propose viewing this dialect as possessing a liquid continuum, rather than as switching /r/ for /l/. I find that rhotics have similar formant structure to the vowel that precedes them, thus becoming more /l/-like after vowels with high or fronted articulation. While rhotics assimilate to vowels, they dissimilate from surrounding consonants. I assert that liquid variation in SJS is motivated by low coarticulatory resistance of rhotics to vowels combined with a sensitivity to sonority sequencing principles such that the liquid segments contrast maximally in sonority from surrounding consonants. In terms of social variables, gender plays a role in rhotic manner of articulation; women produce more tap and fricative rhotics, men produce more deletions, and gender plays no role in approximant productions. Age, on the other hand, is significant for the production of F3 in approximant liquids. Younger speakers have more tongue bunching for rhotics than laterals, whereas older speakers neutralize the sounds.
The tendency for younger speakers to neutralize less may be due to the negative social evaluation attributed to this phenomenon. In the perception experiment, SJS listeners, unlike CS listeners, were able to hear differences in /r/ and /l/ in the context of less neutralizing vowel environments. This finding suggests that SJS listeners are able to hear liquid differences in their own dialect, whereas listeners with more distinct sounds are unable to utilize these small articulatory differences.
iii
A la memoria de Inés Fraile Martín, quien sirvió como madre tanto en la adquisición del español como en mi desarrollo personal.
iv
ACKNOWLEDGEMENTS
Many people have supported me along my path. First and foremost, I'd like to thank my advisors, Dr. Rebeka Campos-Astorkiza and Dr. Scott Schwenter. Rebeka has met with me weekly since the beginning of the candidacy process through the completion of this dissertation. I benefited not only from Rebeka's keen phonetic insights, but also from her ability to map out otherwise overwhelming tasks into manageable pieces. The awards that she has won over the past few years are a testament that I am not alone in my awe of her accomplishments as a researcher and as a mentor. To Scott, I owe the origins of my passions for sociolinguistic theory and quantitative approaches to language. It was through his classes and guidance that I found my footing as a researcher and developed my first scholarly contributions. His informal coffee shop meetings for his advisees helped create an invaluable sense of community. I'd also like to thank Dr. Terrell
Morgan, who served as a committee member for this dissertation and who was the catalyst for my career. It was Terrell's enthusiasm for Spanish dialectology that first enticed me to undertake graduate study in the field.
The fieldwork for this dissertation was made possible by the LoAnn Crane
Hispanic Studies Award, the Josaphat Kubayanda Graduate Student Scholarship, and the
Alumni Grant for Graduate Research and Scholarship. I am grateful to all of the individuals who financed these grants as well as to Melody McGrothers, Tiffany Garner,
v and Rosemarie Thornton who organized the dispersal of this funding. The data collection process would not have been possible without the generosity of many Puerto Ricans who allowed me to record them or spent the time to complete my perception survey. Special thanks goes to Marisol Gutiérrez and all of the librarians at the Biblioteca de Educación at the University of Puerto Rico Río Piedras for their support. I am grateful to my honorary borícua colleague, Ashlee Dauphinais Civitello, who helped me navigate San
Juan and took it upon herself to introduce me to many beautiful places on the island. I also want to thank Jonathan Washington and Hannah Robertson for technical data analysis help. Finally, thank you to Jeff Gory and the Ohio State Statistical Consulting
Service for their assistance.
My colleagues have made the six years of graduate study fly by. I cannot imagine a better work environment or better friends. I'm especially grateful to Hannah
Washington and Christy García, who started the MA with me. Hannah has been my co- author and travel companion. Of all of the wonderful things that have come about from graduate school, her friendship is the most precious. She and her partner Lizzie Gordon have fed me more meals than I can count and have been there for me since the first year of the program. Christy has been my source of sociophonetic genius, teaching ideas, and hugs. Her help with the set-up for the perception experiment and document formatting were immensely helpful for this project. I'd also like to thank Becca Mason and Meghan
Dabkowski for their encouragement and humor.
Outside of the academic realm, my friends, teachers, and family have enriched my life. My yoga community has given me upekshanam, a wide perspective, that has been extremely helpful in this process. I'd especially like to thank my teachers Linda Oshins
vi and Marcia Miller and friends Gina Derry and Alicia Rieske for their guidance and support. Danny Ash has been a confidant, my musical soul mate, and the savior of my computer on numerous occasions - thank you. The members of my "chosen extended family" - Amanda Metskas, August Brunsman, and Teal Larsen - also deserve my gratitude for their love, advice, food, and knitting counsel. Chloe Page, the newest addition, has my deepest respect and affection.
My grandmother, Dorothy Sheffer, who braves life with the ideal blend of tenderness and sarcasm, has always been my role model. James and Sharon Beaton, my parents, instilled in me the value of education. They have worked tirelessly to provide me with the resources to pursue my dreams and their unfaltering belief in my abilities has made this degree possible. My big sister, Katy Sweet, has always held a place for me.
Despite our different lives, we share a very special bond.
Finally I want to thank Gary Morkassel and Michael Riggs. Gary's calm, steady, sweet companionship has balanced out the stress of this project. His DIY mastermind - ranging from fixing my car to building our future home - rocks my world. Mike has shown me the usefulness of viewing reality through multiple paradigms, changing the way I see myself and my thought processes. His creative drive and brilliant understanding of logic and philosophy have served as inspiration for this dissertation and beyond. I cannot begin to express the boundless joy that both of you bring to my life. Thank you.
vii
VITA
2007 ...... B.A. Spanish and Portuguese Literatures and Cultures, The Ohio State University
2011 ...... M.A. Hispanic Linguistics, The Ohio State University
Publications
Beaton, Mary Elizabeth & Hannah B. Washington. Lexical Pejoration and Reappropriation in the Indexical Field: The Case of Favelado. Language Sciences. December 2014.
Fields of Study
Major Field: Spanish and Portuguese
viii
TABLE OF CONTENTS ABSTRACT ...... ii
ACKNOWLEDGEMENTS ...... v
VITA ...... viii
TABLE OF CONTENTS ...... ix
LIST OF TABLES ...... xv
LIST OF FIGURES ...... xx
CHAPTER 1: INTRODUCTION ...... xx
1.1 Description of the phenomenon and focus of the study ...... 1
1.2 Research Questions ...... 7
CHAPTER 2: LITERATURE REVIEW ...... 9
2.1 Introduction ...... 9
2.2.1 Laterals ...... 12
2.2.2 Rhotics ...... 13
2.3 Liquids in Spanish...... 16
2.3.1 Laterals in Spanish ...... 16
2.3.2 Rhotics in Spanish...... 18
2.4 Coda Liquids in Puerto Rican Spanish ...... 24
ix 2.4.1 Impressionistic Studies ...... 24
2.4.1.1 Dialectological Survey of the Island ...... 24
2.4.1.2 San Juan ...... 26
2.4.1.2 Caguas ...... 28
2.4.1.3 Lares ...... 29
2.4.1.4 Mainland United States ...... 30
2.4.2 Acoustic Studies...... 32
2.4.2.1 Ponce ...... 32
2.4.2.2 San Juan and Mayagüez ...... 33
2.5 Incomplete Neutralization in Production and Perception ...... 38
CHAPTER 3: CODA LIQUID PRODUCTION IN PUERTO RICAN SPANISH ...... 48
3.1 Introduction ...... 48
3.2 Methodology ...... 48
3.2.1 Participants ...... 48
3.2.2 Tasks and Recording Procedures ...... 50
3.2.3 Data Analysis ...... 52
3.2.4 Independent Variables ...... 58
3.2.4.1 Liquid Type ...... 58
3.2.4.2 Vowel ...... 59
3.2.4.3 Stress ...... 60 x 3.2.4.4 Word Position ...... 61
3.2.4.5 Preceding Sound ...... 62
3.2.4.6 Following Sound ...... 63
3.2.4.7 Gender ...... 64
3.2.4.8 Age ...... 65
3.2.4.9 Summary of Independent Variables...... 65
3.2.5 Statistics ...... 66
3.2.5.1 Statistical Analysis for Manner of Articulation Data ...... 67
3.2.5.2 Statistics for the Acoustic Features of Approximant Liquids ...... 69
3.2.5.2.1 Linear Regression ...... 69
3.2.5.2.2 Smoothing Splines ...... 71
3.3 Hypotheses and Goals ...... 77
3.4 Results ...... 79
3.4.1 Liquid Manner of Articulation Analyses ...... 80
3.4.1.1 Distribution of the Data ...... 80
3.4.1.2 Rhotic Manners of Articulation ...... 82
3.4.1.2.1 Approximant Productions of /r/ ...... 82
3.4.1.2.2 Tap Productions of /r/ ...... 84
3.4.1.2.3 Deleted Productions of /r/ ...... 86
3.4.1.2.4 Fricative Productions of /r/ ...... 88 xi 3.4.2 Approximant Liquids Analysis of Acoustic Characteristics ...... 90
3.4.2.1 Duration ...... 90
3.4.2.2 First Formant (F1) ...... 92
3.4.2.2.1 Linear Regression Models for F1 at Seven Time Points ...... 92
3.4.2.2.2 Smoothing Splines Analysis for F1 ...... 98
3.4.2.3 The Second Formant (F2) ...... 105
3.4.2.3.1 Linear Regression Models for F2 at Seven Time Points ...... 105
3.4.2.3.2 Smoothing Splines Analysis for F2 ...... 111
3.4.2.4 The Third Formant (F3) ...... 116
3.4.2.4.1 Linear Regression Models for F3 at Seven Time Points ...... 116
3.4.2.4.2 Smoothing Splines Analysis for F3 ...... 121
3.4.2.5 The Fourth Formant (F4) ...... 125
3.4.2.5.1 Linear Regression Models for F4 at Seven Time Points ...... 126
3.4.2.5.2 Smoothing Splines Analysis for F4 ...... 129
CHAPTER 4: CODA LIQUID PERCEPTION IN PUERTO RICAN AND CASTILIAN SPANISH ...... 135
4.1 Introduction ...... 135
4.2 Methodology ...... 136
4.2.1 Experimental Design ...... 136
4.2.1.1 Selection and Preparation of Stimuli ...... 136
4.2.1.1 Survey Design ...... 146 xii 4.2.2 Participants ...... 149
4.2.3 Data Analysis ...... 150
4.3 Research Questions and Hypotheses ...... 152
4.4 Results ...... 153
CHAPTER 5: DISCUSSION ...... 169
5.1 Predictors of Coda Liquid Production ...... 169
5.1.1 Summary of Manner of Articulation Results ...... 169
5.1.2 Summary of Duration Results ...... 170
5.1.3 Summary of Results by Formant ...... 172
5.1.4 Linguistic Conditioning Factors ...... 175
5.1.4.1 Word Position ...... 176
5.1.4.2 Vowel ...... 179
5.1.4.3 Preceding Sound ...... 184
5.1.4.4 Following Sound ...... 187
5.1.5 Extra-linguistic factors and the social status of liquid neutralization ...... 190
5.2 Main Findings from the Perception Experiment ...... 194
5.3 Implications...... 198
5.3.1 Incomplete neutralization of coda liquids ...... 198
5.3.2 Methodological considerations ...... 203
5.4 Understanding the liquid switching phenomenon...... 207 xiii CHAPTER 6: CONCLUSION ...... 216
6.1 Conclusions and contributions ...... 216
6.2 Future research ...... 221
REFERENCES ...... 225
APPENDIX A: INTERVIEW PROTOCOL ...... 235
APPENDIX B: LINEAR REGRESSION MODELS FOR FOUR FORMANTS AT SEVEN TIME POINTS
AND FOR DURATION ...... 237
xiv
LIST OF TABLES
Table 1. Allophonic distribution of rhotics in Spanish ...... 20
Table 2. Age, gender, and education for the speakers in the corpus used for the
production analyses ...... 49
Table 3. Vowel+liquid sequences excluded from the acoustic analysis of approximant
liquids ...... 57
Table 4. Examples from the coding of liquid type ...... 59
Table 5. Examples from the coding of vowel ...... 60
Table 6. Examples from the coding of stress ...... 61
Table 7. Examples from the coding for word position ...... 61
Table 8. Examples from the coding for preceding sound ...... 63
Table 9. Examples from the coding for following sound ...... 64
Table 10. Independent variables and factor levels summary ...... 66
Table 11. Interaction terms considered in the linear regression models ...... 70
Table 12. Factors in the Time and Liquid Type SS ANOVAs...... 75
Table 13. Factors in the Time, Liquid Type, and Word Position SS ANOVAs...... 76
Table 14. Manner of articulation for orthographic coda /r/ tokens in the corpus ...... 81
Table 15. Manner of articulation for orthographic coda /l/ tokens in the corpus ...... 81
Table 16. Logistic regression results for approximant vs. other (tap, deletion, fricative)
pronunciations of /r/ ...... 83
xv Table 17. Duration values for overall data set and for significant interactions ...... 91
Table 18. Overall F1 predicted values of vowel+liquid sequences at each time point ..... 93
Table 19. Values for F1 for vowel+/r/ and vowel+/l/ overall and in particular contexts . 96
Table 20. KL Projection Model Comparison for F1 Time and Liquid Type...... 99
Table 21. KL Projection model comparisons for F1 Time, Liquid Type, and Word
Position and all of the two-way interactions ...... 104
Table 22. Overall F2 predicted values of vowel+liquid sequences at each time point ... 106
Table 23. Values for F2 for vowel+/r/ and vowel+/l/ overall and in particular contexts 108
Table 24. KL Projection model comparison for F2 Time and Liquid type ...... 112
Table 25. KL Projection model comparisons for F2 Time, Liquid Type, and Word
Position and all of the two-way interactions ...... 115
Table 26. Overall F3 predicted values of vowel+liquid sequences at each time point ... 117
Table 27. Values for F3 for vowel+/r/ and vowel+/l/ overall and in particular contexts 119
Table 28. KL Projection model comparison for F3 Time and Liquid Type ...... 122
Table 29. KL Projection model comparisons for F3 Time, Liquid Type, and Word
Position and all of the two-way interactions ...... 125
Table 30. Overall F4 predicted values of vowel+liquid sequences across seven time
points ...... 127
Table 31.Values for F4 for vowel+/r/ and vowel+/l/ overall and in particular contexts 129
Table 32. KL projection model comparison for F4 Time and Liquid Type ...... 131
Table 33. KL Projection model comparisons for F4 Time, Liquid type, and Word Position
and all of the two-way interactions ...... 133
xvi Table 34. Token counts for each informant from the production data for /a/, /e/, and /o/
with coda /r/ and /l/ ...... 139
Table 35. Ideal categories for perception stimuli formant structure ...... 140
Table 36. Schema for coding low, average, and high formant trajectories across the seven
time points ...... 141
Table 37. Stimuli for perception experiment ...... 145
Table 38. Perception according to orthography for vowel+/r/ and vowel+/l/ stimuli taken
together ...... 153
Table 39. Responses to orthographic /r/ approximant stimuli from SJS and CS listeners
...... 154
Table 40. Responses to orthographic /l/ stimuli for SJS and CS listeners...... 154
Table 41. SJS listeners' performance on orthographic /r/ vs. /l/ ...... 155
Table 42. CS listeners' performance on orthographic /r/ vs. /l/ ...... 155
Table 43. Responses to orthographic /r/ tap stimuli from SJS and CS listeners ...... 156
Table 44. SJS listeners' performance on tap vs. approximant orthographic /r/ stimuli .. 156
Table 45. CS listeners' performance on tap vs. approximant orthographic /r/ stimuli ... 157
Table 46. Logistic regression for SJS with orthographic liquid, vowel, and interaction
between the two ...... 158
Table 47. Logistic regression for CS with orthographic liquid, vowel, and interaction
between the two ...... 159
Table 48. Logistic regression for SJS for time point 1 ...... 161
Table 49. Logistic regression for CS for time point 1 ...... 161
Table 50. Logistic regression for SJS for time point 2 ...... 161
xvii Table 51. Logistic regression for CS for time point 2 ...... 161
Table 52. Logistic regression for SJS for time point 3 ...... 162
Table 53. Logistic regression for CS for time point 3 ...... 162
Table 54. Logistic regression for SJS for time point 4 ...... 163
Table 55. Logistic regression for CS for time point 4 ...... 163
Table 56. Logistic regression for SJS for time point 5 ...... 163
Table 57. Logistic regression for CS for time point 5 ...... 164
Table 58. Logistic regression for SJS for time point 6 ...... 164
Table 59. Logistic regression for CS for time point 6 ...... 164
Table 60. Logistic regression for SJS for time point 7 ...... 165
Table 61. Logistic regression for CS for time point 7 ...... 165
Table 62. Findings for which liquid is perceived with higher values of F1 across seven
time points ...... 166
Table 63. Findings for which liquid is perceived with higher values of F2 across seven
time points ...... 166
Table 64. Findings for which liquid is perceived with higher values of F3 across seven
time points ...... 166
Table 65. Findings for which liquid is perceived with higher values of F4 across seven
time points ...... 166
Table 66. Word position effects for the time points at which vowel+/r/ and vowel+/l/ are
neutralized in one position and distinct in the other...... 177
Table 67.Word position effects for the time points at which vowel+/r/ and vowel+/l/
neutralized in both positions, but have opposite directionality ...... 178
xviii Table 68. Direction of effect of F1 according to vowel height...... 181
Table 69. Direction of effect of F2 according to vowel frontness ...... 183
Table 70. Direction of effect of F3 by vowel...... 184
Table 71. Effects of preceding sound on formant structure ...... 186
Table 72. Effects of following sound on formant structure ...... 188
xix
LIST OF FIGURES
Figure 1. Map of Puerto Rico by municipality ...... 6
Figure 2. Fricative /r/ in entrar ‘to enter’ ...... 53
Figure 3. Deleted /r/ in decir que ‘to say that’ ...... 53
Figure 4. Tap /r/ in conocer y 'to know and' ...... 54
Figure 5. Approximant /r/ in fuerza ‘strength’ ...... 54
Figure 6. Conditional inference tree for taps...... 85
Figure 7. Conditional inference tree for deletions...... 87
Figure 8. Conditional inference tree for fricatives ...... 89
Figure 9. Overall F1 predicted values of vowel+liquid sequences across seven time points
...... 94
Figure 10. Smoothing spline ANOVA of F1 values over time for vowel+/r/ and vowel+/l/
...... 98
Figure 11. Smoothing spline ANOVA of F1 values with word position for vowel+/r/ and
vowel+/l/ ...... 100
Figure 12. Smoothing spline ANOVA of F1 values for word-final liquids ...... 100
Figure 13. Smoothing spline ANOVA of F1 values for word-medial liquids ...... 101
Figure 14. Smoothing spline of F1 values for vowel+/l/ sequences in both word positions
...... 102
xx Figure 15. Smoothing spline of F1 values for vowel+/r/ sequences in both word positions
...... 102
Figure 16. Overall F2 predicted values of vowel+liquid sequences across seven time
points ...... 108
Figure 17. Smoothing Spline of F2 values over time for vowel+/r/ and vowel+/l/ ...... 111
Figure 18. Smoothing spline ANOVA of F2 values with word position for vowel+/r/ and
vowel+/l/ ...... 113
Figure 19. Overall F3 predicted values of vowel+liquid sequences across seven time
points ...... 118
Figure 20. Smoothing spline ANOVAs of F3 values over time for vowel+/r/ and
vowel+/l/ ...... 121
Figure 21. Smoothing spline ANOVA of F3 values with word position for vowel+/r/ and
vowel+/l/ ...... 123
Figure 22. Smoothing spline ANOVA of F3 values for vowel+/r/ and vowel+/l/ for word-
medial position only ...... 123
Figure 23. Overall F4 predicted values of vowel+liquid sequences across seven time
points ...... 128
Figure 24. Smoothing spline ANOVAs of F4 values over time for vowel+/r/ and
vowel+/l/ ...... 130
Figure 25. Smoothing spline ANOVA of F4 values with word position for vowel+/r/ and
vowel+/l/ ...... 131
Figure 26. Example of perception data coding ...... 142
xxi Figure 27. Sample page from perception survey with an audio stimulus and two possible
choices...... 148
Figure 28. The vowel inventory of Spanish...... 181
Figure 29. Smoothing spline ANOVA of F3 values for vowel+/r/ and vowel+/l/ for older
speakers ...... 192
Figure 30. Smoothing spline ANOVA of F3 values for vowel+/r/ and vowel+/l/ for
younger speakers ...... 192
Figure 31. Parker's (2002:240) universal sonority hierarchy. Note that /r/ represents a
retroflex approximant, like that of English...... 213
xxii
CHAPTER 1: INTRODUCTION
1.1 DESCRIPTION OF THE PHENOMENON AND FOCUS OF THE STUDY
The pronunciation of syllable-final liquids /r/ and /l/ is one of the most salient characteristics of Puerto Rican Spanish. Speakers in the capital city of San Juan describe their dialect by saying "confundimos la /r/ por la /l/ al hablar" [we confuse /r/ for /l/ when we speak]. Impressionistic linguistic accounts transcribe [l] for orthographic /r/ when it appears in coda position, such that mar and mal, which are distinct in most dialects, are neutralized in San Juan Spanish (henceforth SJS), as shown in example (1).
(1) alma 'soul' [al.ma] arma 'sea' [al.ma]
This phenomenon has come to be known in the Hispanic linguistics literature as trueque de líquidas 'liquid switching'. This switch can go in either direction, depending on the dialect. That is, some dialects, such as SJS, lateralize /r/, like in (1), whereas other dialects, such as that spoken in Seville, Spain, rhotacize /l/, as in (2).
(2) alma 'soul' [aɾ.ma] arma 'sea' [aɾ.ma]
Lateralization and rhotacization of coda liquids occurs in parts of Cuba, the
Dominican Republic, southern Spain, and Puerto Rico. Other coda liquid phenomena, such as deletion, deletion with gemination of the following consonant, and vocalization are also present in certain parts of these regions (Lipski 1994). Similar processes with
1 coda liquids also exist in other Romance languages, such as rhotacization in certain sociolects of Brazilian Portuguese (Bortoni Ricardo 1985). The phonetic processes that underlie these dialectal features and the linguistic and social conditionings on liquid variation is only beginning to be understood. The present study illuminates the phonetic complexity of coda liquids in SJS and contributes to a broader debate on liquids cross- linguistically.
This dissertation is the first sociophonetic approach to liquid switching in San
Juan Spanish. All of the previous sociolinguistic literature on this topic analyzes orthographic /r/ in coda impressionistically. This methodology is problematic because classifying sounds by ear means developing a limited number of categories, and the researcher has to decide whether orthographic /r/ is pronounced, for example, as [ɹ] or [l].
In a dialectological account in 1948, Navarro Tomás remarked that he heard an intermediate sound between /r/ and /l/ on the island. López Morales (1983a, 1983b), whose work constitutes the largest sociolinguistic account of liquid switching in SJS previous to the present study, reports that he also heard some occurrences of orthographic
/r/ as "mixed" liquid sounds, but said that the "elemento lateral es sobresaliente" [the lateral element stands out] and thus he classifies these sounds along with occurrences of
/r/ that he heard as completely neutralized to /l/. Thus, the potential phonetic gradience that exists between more /r/-like and more /l/-like sounds is lost in these studies. Another problem with impressionistic work is the inherent subjectivity. What sounds like an /r/ to one observer may sound more like an /l/ to another. Subjectivity in sound classification is always a problem in impressionistic study of language, but recent evidence from a cross- dialectal perception study suggests that this sort of analysis of liquid switching in SJS
2 may be even more fraught than previously assumed. Paz (2005) had a lateralizing SJS speaker read minimal pairs with /r/ and /l/ such as mal/mar and alma/arma and discovered that SJS listeners correctly heard the word being pronounced over 80% of the time, while Argentinian listeners, who do not have liquid switching as a feature of their dialect, performed at about chance. Based on Paz's findings, it seems that in the case of impressionistic studies, whether a liquid is categorized as sounding like an /r/ or an /l/ can be highly dependent on the native dialect of the researcher.
A couple of recent studies have circumvented the categorization problem by performing acoustic analysis on liquid switching in Puerto Rico. As part of a broader project on the Spanish spoken in Ponce, a city on the southern coast of the island, Luna
(2010) finds that /r/ and /l/ produced as approximants are significantly different in terms of F3 and F4. Simonet, Rohena-Madrazo, & Paz (2008) explore acoustic dimensions of these sounds in the speech of three SJS speakers and one speaker from Mayagüez, a city on the western coast. The results from their analysis of F1, F2, F3, and duration for orthographic /r/ and /l/ revealed that each of the four speakers used at least one of the acoustic dimensions to differentiate the two liquid sounds when they are produced as approximants.
Simonet et al.'s study (2008) was the first to empirically demonstrate that Puerto
Ricans do not completely neutralize orthographic /r/ and /l/ in coda position. This finding is relevant not only to our understanding of SJS, but also contributes to the growing body of work on neutralization phenomena. Recent advances in phonetic analysis have led researchers to re-examine sounds that were previously assumed to be contextually neutralized, or merged into a single pronunciation. By using fine-grained spectrographic
3 analysis, linguists have discovered that in many cases where neutralization was assumed, sounds are actually incompletely neutralized. That is, there are small differences in the articulation of these sounds such that they are in fact not merged, even though impressionistic descriptions report them as such. Most of the work on incomplete neutralization has been on coda stop voicing, and prior to Simonet et al.'s investigation, no liquid neutralization phenomena had been studied acoustically to determine the nature of the "intermediate" or "mixed" sounds, as reported by Navarro Tomás (1948) and
López Morales (1983a, 1983b). Chapter 2 of this dissertation further discusses these studies and others on liquid switching in SJS and also provides a general literature review on liquid sounds and on incomplete neutralization studies.
Unlike previous studies, the present study combines the advantages of both sociolinguistic and acoustic analysis in an effort to make use of the methodological benefits that each approach offers. Sociophonetic research was rare to non-existent before the 1990s, but has since become a productive field of inquiry (Thomas 2011).
Sociolinguists aim to capture the most natural speech possible, through conversational interviews with participants, rather than the reading tasks usually employed in laboratory phonetics. Also, since sociolinguists aim to understand how language is stratified in a particular community, extralinguistic independent variables such as gender and age are usually included in their research. Phoneticians, on the other hand, tend to be more concerned with linguistic variables and often do not consider social factors. Due to this, phoneticians often draw their data from a smaller pool of speakers than do sociolinguists.
Phonetic research, however, generally provides a more in-depth understanding about acoustic and articulatory properties of pronunciation phenomena, which allows for a
4 better understanding of cross-linguistic patterns that are sometimes overlooked in sociolinguistic research due to the focus on social stratification. A sociophonetic approach combines the data collection methods from sociolinguistics with the detailed phonetic analysis from laboratory approaches. By using acoustic analysis, the present study improves upon the past sociolinguistic studies by doing away with impressionistic categorization.
Furthermore, this dissertation considers both the production and the perception of rhotics and laterals in syllable-final position in the Spanish spoken in San Juan, Puerto
Rico. Analyzing both production and perception provides a more complete picture of the status of incomplete neutralization in this dialect; the production analysis determines whether or not there are acoustic differences between orthographic /r/ and /l/ and the perception analysis examines whether or not these differences can be used by SJS listeners to correctly identify the sounds under study. The data for the production study comes from 24 sociolinguistic interviews that I conducted in San Juan near the University of San Juan, Río Piedras campus in June-July 2013. The speakers in the corpus ranged from ages 18 to 62 and were equally divided for gender. All of the participants have spent most of their lives in the San Juan area and worked or studied in the capital city at the time of the recording. The speakers all live in San Juan proper or in the surrounding municipalities of Carolina, Trujillo Alto, Caguas, Guaynabo, or Bayamón. Figure 1 is a map of Puerto Rico divided by municipalities; the red oval indicates the area from which the sample was taken.
5
Figure 1. Map of Puerto Rico by municipality
To analyze the production data for SJS, which is the subject of Chapter 3, I first categorize the liquid productions in terms of their manner of articulation and provide the distribution of these realizations. Then, I extract vowel+liquid sequences from words with orthographic /r/ and /l/ that are realized as approximants. Following Simonet et al.'s
(2008) methodology, I take formant measurements at seven equidistant points within each vowel+liquid sequence to get a picture of the shape of the formant trajectories in order to observe the formant changes and transitions over the duration of the sequences.
The dependent variables for the production analysis are the formant measurements for
F1, F2, F3, and F4 as well as the duration of the vowel+liquid sequences. The effects of the preceding and following sounds, stress, word position, age of the informant, and gender of the informant are considered as independent variables for the production analyses. Two statistical tools, linear regression and smoothing spline ANOVAs, are used together for the analysis of the production data in order to understand the complex conditioning of the predictors on the variation and the dynamic trajectories of liquid sounds.
6 For the perception experiment, presented in Chapter 4, I extracted vowel+liquid sequences from a speaker in the production study who exhibits considerable variation in his liquid production. These stimuli were presented to listeners using an online survey format in which participants chose between two options, such as /ar/ and /al/, after hearing an audio stimulus. In order to test whether the perception of liquid switching by
SJS speakers was different from that of speakers from non-liquid switching dialects, a control group of listeners from central and northern Spain also completed the experiment.
The logistic regression analyses for the experiment compare how successfully the two groups of listeners identify the difference between orthographic /r/ and /l/ as well as the effect of the vowel that is paired with the liquids. In addition, the conditioning of the formant measurements for F1, F2, F3, and F4 on listeners' selection of /r/ or /l/ is considered.
1.2 RESEARCH QUESTIONS
The following questions represent the overarching lines of investigation in the present study.
What are the different manners of articulation in the production of liquids in coda
position in SJS?
In SJS, do approximant realizations of vowel+/r/ differ from vowel+/l/
productions in terms of formant structure (F1, F2, F3, F4) and duration? If so,
how are they different?
Are approximant realizations of vowel+/r/ and vowel+/l/ completely or
incompletely neutralized? 7 What social and linguistic factors influence the variation of the formant structures
and durations of vowel+/r/ and vowel+/l/ sequences?
Do SJS listeners hear liquid sounds differently than Spaniards?
What role do the formants (F1, F2, F3, and F4) play in listeners' ability to
distinguish vowel+/r/ from vowel+/l/?
How can we understand or model the behavior of coda liquids in SJS based on the
present study’s results?
Chapter 2 recounts relevant past research on liquid sounds cross-linguistically, in the Spanish language in general, and finally, the studies that have been conducted on coda liquid switching in Puerto Rico. The chapter concludes with a discussion of previous research on incomplete neutralization phenomena in a variety of languages.
Chapter 3 presents the production study of coda liquids in SJS based on my interview data, including the methodology and results. The following chapter, Chapter 4, relates the perception experiment, detailing its methodology, as well as the results. Chapter 5 discusses the production and perception results from the previous two chapters in light of relevant phonetic and sociolinguistic theories and presents the implications of these results for our understanding of liquid switching. Finally, Chapter 6 presents contributions, conclusions, and avenues for further research.
8
CHAPTER 2: LITERATURE REVIEW
2.1 INTRODUCTION
This chapter provides a survey of the literature on liquid sounds and studies on incomplete neutralization that are pertinent to the understanding of coda liquids in San
Juan Spanish. First, an overview of liquid behavior across languages is presented (section
2.2), which builds a general understanding of the articulation, acoustics and phonological patterning of these sounds. Section 2.3 covers specific studies on rhotics and laterals in
Spanish across dialects, and then section 2.4 considers work on liquids in Puerto Rican dialects with a focus on coda segments. The last section, 2.5, discusses the experimental literature on neutralization phenomena across languages, which will serve as a basis to inform my contribution to theories of incomplete neutralization.
2.2 LIQUID SOUNDS CROSSLINGUISTICALLY
Quilis (1981) describes liquids as intermediate sounds between vowels and consonants. The author notes that [ɾ] and [l] tend to have similar formant structures to vowels but these liquids have a lower fundamental frequency and are less intense than vowels due to the partial obstruction of the oral cavity by the tongue. Ladefoged &
Maddieson (1996) point out that the vowel-like nature of rhotics and laterals grants them status as the most sonorous consonants. Due to their vowel-like qualities, liquids are 9 more frequently permitted as the second member of complex onsets and they are more likely to appear in coda than other consonants across languages. Based on this observation that liquids pattern close to vowels and have different phonotactics than other consonants, they occupy a position after vowels and before other consonants in sonority scales. These scales are built around cross-linguistic observations about syllable structure whereby the most sonorous sounds are found in the center (nucleus) of the syllable whereas the margins of the syllable feature the least sonorous sounds. (1) below illustrates the sonority scale proposed by Clements (1990).
(1) Sonority Scale (from Clements 1990) vowels > glides > liquids > nasals > obstruents
As well as having similar phonological patterning, acoustically, /r/ and /l/ tend to have characteristics in common.1 Delattre (1958) describes a few phonetic cross- linguistic tendencies of liquids: 1) during the stable state, both /r/ and /l/ have an F1 of about 400 Hz for male voices, 2) all of the formant values are greater than those for nasals, 3) F2 and higher formants are lower than those for vowels and 4) transitions from vowels are more gradual for liquids than for stops.
The phonotactics, acoustic properties, and articulatory characteristics of /r/ and /l/ sounds discussed thus far are used in the literature as evidence to consider these sounds as one class, i.e. the liquid class of consonants. Despite their similarities, the idea that rhotics and laterals should be considered together as a class has been widely debated in the literature. Many phoneticians argue that including all laterals and rhotics together as a
1 The use of forward slash notation in this dissertation, i.e. /r/ and /l/, is meant to represent the classes of rhotic and lateral sounds represented by the graphemes
Steele (2005) give evidence from Spanish and French both for and against considering rhotics as part of the liquid class. They conclude that rhotics are "unstable" and are more likely to leave the class than are laterals. Since much of the evidence against rhotic membership in the liquid class is phonetic in nature, the authors posit that it is possible to consider liquids as a phonological class but not a phonetic class. Lindau (1985) furthers this idea that rhotics should be considered part of a phonological class with laterals. She says that the extensive variety of rhotic sounds should be thought of as related to each other by historical and articulatory associations and therefore should rightfully be called liquids.
An argument for considering /r/ and /l/ together as a class that is particularly relevant for the present study is that these sounds tend to pattern together in coda.
Throughout historical changes from Latin to modern Romance, /r/ and /l/ have participated in diachronic and synchronic processes whereby a word with two non- contiguous /r/ or /l/ sounds undergo dissimilation so that the same liquid sound does not appear twice in a lexical item. An example of such a process is Latin 'arbor' to Spanish
'árbol.' These changes of /r/ to /l/ or /l/ to /r/, known as lateralization and rhotacization respectively, are attested in the development of many modern Romance dialects of
Spanish, Portuguese, and French (Fontanella de Weinberg 1984, Straka 1979). The dissimilation pattern is one of many instantiations of "switching" in these languages. The coda liquids under study for this dissertation, whether or not they are in stable variation
11 or represent a change in progress for SJS, can provide evidence for or against considering
/r/ and /l/ sounds together as a liquid class.
2.2.1 Laterals
Laterals are less cross-linguistically and cross-dialectally variable than rhotics in terms of manner of articulation, and are more likely to be sonorant and continuant, as is characteristic for members of the liquid class (Colantoni & Steele 2005). The most common type of lateral across languages is the voiced approximant [l]; voiceless approximants [ɬ], lateral flaps [ɺ], fricatives [ɮ] and affricates [dɮ] are rare in the world's languages (Ladefoged & Maddieson 1996).2 The first formant (F1) in the steady state of the voiced approximant [l] is low in frequency, usually below 400 Hz for males. Since F1 has an inverse relationship with the height of the tongue, this low value represents a high tongue position. The second formant (F2) varies, depending on where the occlusion of the tongue is made along the palate - generally, the further forward the constriction, the higher the second formant. (Ladefoged & Maddieson 1996).
In order to show the range of F2 values across laterals with more front and more back pronunciations, Ladefoged and Maddieson (1996) take measurements of F2 values in Eastern Arrernte, Kaititj and Alyawarra, which are Australian languages that have phonemic lateral sounds that differ by place of articulation. The F2 values in these languages range from 1350 to 2324 Hz, with the lower values corresponding to more back articulations and the higher values representing more front articulations. The general
2 Spanish only has approximant laterals, so the acoustic characteristics of these more rare laterals will not be discussed here. 12 shape of the tongue and the surrounding vowel sounds are also important in determining the F2 value (Di Paolo & Yaeger-Dror 2011). F3 in lateral sounds has a "relatively strong amplitude and high frequency" and higher formants, i.e. F4 and above, are typically close to F3 (Ladefoged & Maddieson 1996:193). An important signature feature of laterals is the presence of an anti-formant between F2 and F3 due to the absorption of sound waves by the chamber that the tongue creates in the center of the mouth (Johnson 2003). Apical lateral articulations display the quickest transitions to and from surrounding sounds, whereas laminal and dorsal productions have more gradual transitions (Ladefoged &
Maddieson 1996).
Much of the work on laterals has been done on English, which most commonly has two allophonic laterals (see Gick 2002, Giles & Moll 1975, Horvath & Horvath 2001,
Newton 1996). The onset lateral is alveolar, [l], and is frequently referred to as 'clear /l/' in the literature. The lateral found in coda and often in consonant clusters in English, [ɫ], has both the alveolar articulation and a secondary velar tongue dorsum gesture and is referred to as a 'dark /l/'. Acoustically, there is a difference in F2 values for the two variants - the alveolar variant has a higher F2 than the velarized version (Di Paolo &
Yaeger-Dror 2011). Some dialects of English have a vocalized coda /l/, which is harder to distinguish acoustically, but generally can be identified by a low F3 (Baranowski 2013).
2.2.2 Rhotics
Rhotics - for the purposes of this study, sounds written with
(1985). Taps are also perceptually similar to trills since many trills have only one or two occlusions (Lindau 1985).
Despite their range of productions, rhotics tend to pattern similarly phonologically across languages (see Ladefoged & Maddieson 1996:216 for specifics) and this general observation as well as the aforementioned historical arguments have been used to justify rhotics as a group of sounds that can be considered together and alongside laterals as liquids. Researchers have tried to pinpoint specific acoustic correlates for rhotics as a
14 class. The most commonly agreed upon feature of rhoticity is a lower F3 than that found for laterals, which corresponds to the more bunched, retracted articulation of rhotics
(Fujimura & Erickson 1997, Johnson 2003, Ladefoged & Maddieson 1996, among others). But, counter-examples, such as uvular trills with very high F3 values, are present in many languages (Lindau 1985).
Given the debate surrounding the acoustic correlates for rhoticity, and specifically the role of F3 in distinguishing rhotics from laterals, a number of perception studies have been conducted to shed light on this issue. O’Connor et al. (1957) use computer- generated formant structures to determine the acoustic differences necessary in the perception of liquids (/r/ and /l/) and glides (/w/ and /j/) in onset position in American
English.3 The authors find that the most important distinction in perception between /r/ and /l/ onsets in English lies in F3 – "the third-formant starting frequency for /l/ is close to that of the vowel third formant, whereas the third-formant onset of /r/ needs to be lower in frequency, fairly close to the second-formant onset" (34). Despite the importance of F3, the authors also find that modifications of F2 can change the sound perceived. Specifically, for liquids, a high F2, like that of front vowels can lead to a rhotic percept while a low F2 sounds more like /l/. The authors also point out that percepts of /r/ and /l/ vary greatly depending on the vowel that follows the segment. In addition, in a perception study on coda rhotic percepts for a relatively /r/-less dialect of
British English, Heselwood (2009) finds that removing F3 from coda rhotics entirely often leads listeners to hear more rhoticity. This result leads the researcher to propose that rhoticity could actually lie in F2 values.
3 The '/r/' here does not refer to a trill, as explained in footnote 1, but rather refers to rhotic sounds in general. The rhotic sound in English is usually transcribed as [ɹ]. 15 In sum, the cross-linguistic literature on liquids highlights F3 as the most telling formant to distinguish rhoticity from laterality, but also cites F2 as potentially different in rhotic vs. lateral sounds, although the influence from the vowel on F2 makes the role of this formant more difficult to ascertain. Due to the importance of formant structure and the influence of vowel transitions, the present study considers the vowel and the liquid as a single vowel+liquid trajectory or sequence, as discussed in Chapter 3.
2.3 LIQUIDS IN SPANISH
2.3.1 Laterals in Spanish
According to the literature, Spanish laterals, unlike English, are articulated with only a tongue tip gesture and do not have a secondary velar articulation [ɫ]. In other words, lateral articulations in Spanish favor more fronted, 'clear' articulations, generally with the tongue tip or blade (Navarro Tomás 1970). The occlusion for the tongue tip in
Spanish is much like [t], but unlike this stop, for /l/ the air escapes through one or both sides of the tongue. Also, laterals tend to assimilate in place of articulation to following consonants produced with the tongue tip or blade, such that the closure can be produced as a dental, alveolar, or palatal in pre-consonantal positions (Hualde 2005).4
Palatographic studies show that there is a lot of variation for /l/ in onset in terms of the place of articulation and the amount of tongue contact (Martínez Celdrán & Fernández
Planas 2007). Midsagittal x-rays taken in the middle of the lateral gesture for a single speaker of Peninsular Spanish show that the back of the tongue has a low position in
4 Historically, and today in a limited number of conservative dialects, Spanish has a palatal lateral phoneme [ʎ] that corresponds with
(2009:98).
Most of the acoustic studies for /l/ in Spanish have been performed on onsets. In an acoustic study of Argentine Spanish, Massone (1988) shows that onset /l/ has an F1 value that is consistently about 400Hz and an F2 that is higher for front vowels and lower for back vowels (1500 Hz with [u]; 1700 Hz with [o] and [a]; 1800 with [e]; 2000 with
[i]). The value of F3 for onset /l/ was around 2700 Hz for all vowels. Quilis (1981) claims that onset laterals in Spanish (presumably Spanish in Madrid) are fairly uninfluenced by the vowels that follow them - F1 is usually lower than the following vowel, F2 descends during the duration of the lateral from a frequency similar to that of /i/ to that of /u/. It should be noted that this observation for F2 runs counter to Massone's study in which the following vowel did influence the lateral. This may be due to dialect differences or simply different sampling techniques; Quilis does not make it clear how he derived his measurements. In their study of three Argentinian male speakers, Guirao & García Jurado
(1991) find that onset laterals in the syllable /la/ have an F1 of about 350 Hz (about half the frequency of the [a]) while F2 is much higher than the vowel at 1600 Hz to start with, before swooping down towards the vowel. F3 is about 2300 Hz in their study.
17 Quilis (1981) claims that stress affects formant structure for /l/ in Spanish in onset position. According to Quilis, /l/ in the onset of stressed syllables features a more open articulation, measured by a higher F1, and a longer duration than in unstressed syllables.
F2 for /l/ is higher for stressed than for unstressed syllables, due probably to a more anterior articulation (Delattre 1951). According to Quilis (1981:276), the duration of /l/ in onset or coda position averages at 60.3 ms.5 The segment is longer in stressed than unstressed syllables; also it is longest before a pause and shortest between vowels.
Laterals in Spanish vary depending on syllable position. Spanish coda laterals tend to have higher F1 values and lower F2 values than laterals in word onset or intervocalic positions; these acoustic features correspond to a greater degree of closure and a more fronted tongue position in coda than in onset (Quilis 1981). Guirao & García
Jurado (1991) point out that transitions are more abrupt from the lateral to a vowel (CV) than from a vowel to the lateral (VC). In their study, F1 lowered gradually from the vowel to the lateral and F2 raised to about 1800 Hz and F3 to 2650 Hz.
2.3.2 Rhotics in Spanish
The literature on rhotic sounds in Spanish claims that there are two rhotics in
'standard' dialects - tap [ɾ] and trill [r] (Quilis 1993), which are phonemic in intervocalic position. A great deal of dialectal and social variation exists for trilled [r], which can be realized in a number of ways, as will be discussed in more detail below. The tap and trill
5 Quilis does not give the differences in duration for onset vs. coda position but his phrasing suggests that they are very similar: "La duración media de /l/ en posición silábica prenuclear o postnuclear, y sin formar parte de una secuencia consonántica tautosilábica es de 6,03 cs" (Quilis 1981:276). 18 (or other sound in its place) are allophones in most onsets and in codas. In word-internal intervocalic onsets, unlike in other contexts, the tap and trill are phonemic. The allophonic distribution for Spanish rhotics is illustrated in Table 1. It is important to note that word-final consonants in Spanish tend to resyllabify to the following word if the following word begins with a vowel so that the consonant ends up as an onset. Syllable- final rhotics are variable between taps and trills, as shown in (e), (f) and (g). Hualde
(2005) claims that the tap is more common than the trill and that the trill is only common of emphatic speech.
19 [r] trill
(a) #___ (Word Initial) [ro.ka] 'roca'
(b) C.___ (after a heterosyllabic consonant) [al.re.ðe.ðoɾ] 'alrededor'
[ɾ] tap
(c) σ[C___ (in onset clusters) [bɾo.ma] 'broma'
(d) V___#V (word-final before a vowel) [se.ɾa.mi.gos] 'ser amigos'
Variable Rhotic (usually [ɾ], sometimes [r] in 'emphatic' speech)
(e) V___.C (before a consonant) [paɾ.te] ~ [par.te] 'parte'
(f) V___#C (word-final before a consonant) [seɾ.po.e.ta] ~ [ser.po.e.ta] 'ser poeta'
(g) V___## (word-final before pause) [se.ɾo.no.seɾ] ~ [se.ɾo.no.ser] 'ser o no ser'
Table 1. Allophonic distribution of rhotics in Spanish (adapted from Hualde 2005)
Due to the very specific aerodynamic conditions necessary for the realization of trills (cf. Solé 2002), even dialects that have trill productions usually present 'failed' realizations such as fricatives. Related to this idea, many dialects of Spanish frequently utilize a completely different sound in place of the normative trill. An assibilated rhotic
[ř] is common in onset in parts of Central America, the Andes region, Paraguay and northern Argentina (Hualde 2005). In Cochabamba, Bolivia, Sessarego (2011) finds that assibiliated rhotics are the most common production of /r/ across phonological contexts.
Retroflex [ɹ] is less common, particularly in onset, but appears in this position in some speakers of Costa Rican Spanish as well as in some varieties of Spanish in the United
States (Hualde 2005). In the Dominican Republic, speakers produce the canonical trill with pre-breathy voice followed by either a tap [ɦɾ] or trill [ɦr] in lieu of the standard trill
20 most of the time (Willis 2007, Willis & Bradley 2008). These productions are also present in Cuba and Puerto Rico, but are less frequent than in the Dominican Republic
(Hualde 2005). According to Lipski (1994), the Puerto Rican canonical trill is realized as dorsal rather than apical in some speakers; voiced or voiceless velar fricatives [x,ɣ] or uvular trills [ʀ] are possible variants. These posterior productions coexist in Puerto Rico with the apicoalveolar trilled prestige variant; dorsal rhotics are particularly common in
Mayagüez, Hormigueros, Cabo Rojo, Maricao, San Germán, Sabana Grande and rural areas (Megenney 1978). In an acoustic study on Equatorial Guinean Spanish rhotics,
O'Brien (2013) finds that the trill-tap distinction is neutralized intervocalically.
In complex onsets, rhotics are more variant than laterals. The tap production that
Hualde (2005) claims for this position (see (c) in Table 1) occurs in many dialects, but closer phonetic studies reveal that a very brief vowel segment is inserted before the tap.
This short svarabhakti vowel tends to have similar formant structure to the following vowel (Quilis 1993). In /Cl/ structures, intrusive vowels are exceedingly rare (Colantoni
& Steele 2005). Furthermore, in some dialects, the rhotic in consonant clusters is coarticulated with the previous consonant sound (Bradley 2006), resulting in an assibilated /r/ sound due to gestural overlap. In some dialects, this coarticulation is generalized, while in others, such as Highland Ecuadorian Spanish, the rhotic is only coarticulated if the preceding consonant is a coronal stop (Argüello 1978).
Taps, while they are stable across dialects in intervocalic position, present a great deal of variation in coda. According to Martínez Celdrán & Fernández Planas (2007), coda rhotics in most dialects of Peninsular Spanish are variably pronounced as taps, trills, and fricatives. Taps across dialects of Spanish are frequently produced as approximated
21 taps when the tongue tip gesture does not reach the alveolar ridge (e.g. Almeida & Dorta
1993, Massone 1988). Rhotics in syllable final position pattern with laterals in much of the Caribbean and southern Spain. Assimilation of both /r/ and /l/ to following consonants with a coronal closure is common in Cuba, such that carne is pronounced as [kan.ne]
(Alfarez 2007). Hualde (2005) describes this assimilation alongside /r/ aspiration and the trueque de líquidas as coda liquid weakening. In the Cibao region in the Dominican
Republic, /r/ and /l/ neutralize to a glide [j] in coda position, such that arma and alma are realized as [aj.ma] (Alba 1990). Andalusian Spanish, like Caribbean dialects, features a trueque de líquidas, usually realized as rhotacization of /l/ (Ruiz-Peña 2013). Deletion of coda liquids is fairly common across dialects of Spanish, especially in word-final position
(Hualde 2005).
Acoustic characteristics for taps are different for onsets than for codas. Guirao &
García Jurado (1991) describe the principal characteristic of the tap both in production and perception as a short interval of silence that separates it from the following vowel in onset or from the preceding vowel in coda. Before the start of the vowel when the tap is in onset, there is a brief open phase with a vocalic segment similar to an [e] with reduced amplitude right after the release of the tap. When the tap is in coda, the transition from the vowel is slower than the transition to the vowel after onset. Coda taps, like those in onset, also have an open phase with a vocalic segment similar to [e] upon release. The contrast in amplitude between the surrounding vowel and the tap is lower in coda than onset. The average closure of the tap is 20 ms (22 ms in stressed syllables, 18.6 ms in unstressed syllables), which is considerably shorter than the duration of /l/ (Quilis 1981).
Almeida & Dorta (1993) give measures of Spanish tap approximants - those without full
22 closure - as 45 ms in tonic syllables and 33 ms for unstressed syllables. The approximant realization lengths that these authors give are considerably longer than the lengths of taps cited above.
In an unpublished perception experiment mentioned in her production work,
Massone (1988) finds that stimuli with taps in onset (tap+vowel stimuli) are easily identified by participants as rhotics, whereas taps in coda (p+vowel+tap) are frequently confused with /l/. She attributes this finding to the shorter durations and frequent approximant realizations of taps in coda. After establishing the acoustic correlates for the production of [ɾ] and [l] in Argentine Spanish, Guirao & García Jurado (1991) perform a perception experiment to see how long the liquid+vowel durations need to be in order for listeners to perceive these liquids in different syllabic positions. The duration of [ɾa] is usually about 430ms in production, but listeners are capable of identifying the sequence with a total duration of 108ms. If the total duration is reduced to 50ms, listeners hear stops with the vowel [e]. With /r/ in coda ([aɾ]), the average duration in production is
285ms - the first 230ms belong to the vowel, 15ms of closure and 45ms of the release vowel similar to [e]. The total duration can be reduced to 80ms (25ms for [a], 15ms for the occlusion, 40ms for the release vowel) and still be perceived correctly. The average duration of /la/ in the production experiment is 380ms, but listeners can deduce what they are hearing with as little as 80ms. The sequence /al/ is longer in duration, averaging at
410ms. The vowel occupies the first 180ms of the sequence; when this part is taken out, the last 230ms is not identifiable. The whole sequence can be reduced to 100ms and still be recognizable. The authors conclude that [ɾ] and [l] do not have independent acoustic profiles and cannot be separated from vowels in either onset or coda position. These two
2 3 perception studies suggest that coda liquids require longer durations to be recognized than do onsets, and coda liquids are more likely to be misidentified than are onsets. This finding is not surprising, given that consonants in postvocalic contexts across languages are frequently cited as being more difficult to perceive than those in onset (see Steriade
1997).
2.4 CODA LIQUIDS IN PUERTO RICAN SPANISH
This section reviews the findings from previous studies on coda liquids in dialects of Puerto Rican Spanish, including the dialect spoken in San Juan, which is the focus of the current investigation. Section 2.4.1 covers studies that used impressionistic categorization of coda liquids in various locations throughout the island. Section 2.4.2 reviews the investigations of coda liquids that were carried out using acoustic measures in Ponce, Mayagüez, and San Juan.
2.4.1 Impressionistic Studies
2.4.1.1 Dialectological Survey of the Island
Navarro Tomás (1948) completed one of the first and most comprehensive dialect surveys of the island. Following the dialectological paradigm, Navarro Tomás had one or two informants from each place he visited in his travels around the island. He presented them with a “cuestionario de 445 preguntas” [445-question questionaire] (10) which encompassed not only liquid switching, but also other phonological variables, such as /s/ aspiration, as well as local differences in vocabulary and syntactic variables. His study of
24 coda liquids consists of 7 words with orthographic /r/ and 7 words with orthographic /l/: barba, carbón, mazorca, muerte, puerta, verdad, tarde, caldo, calvo, dulce, espalda, palmillo, pulga, and soldado. As I mentioned in Chapter 1, Navarro Tomás, despite his impressionistic methods, made the observation that an intermediate sound between /l/ and
/r/ existed on the island.6 He thus categorizes coda liquids as /r/, /l/, or "mixed." When all of the surveyed areas on the island are taken together for his 14 studied words, he finds
52.5% /r/, 41.0% /l/, and 6.5% mixed sounds. The mixed sounds that Navarro Tomás reports are both for orthographic /r/ and /l/; in the case of the San Juan area, orthographic
/r/ resulted in complete neutralization with /l/ and in mixed sounds, whereas /l/ was more stable. He noted that rhotacization of /l/ occurred primarily on the western side of the island, while lateralization of /r/ was more frequent on the northeast side, which encompasses the San Juan area. Despite these overall patterns, Navarro Tomás emphasized that the same informant often varied in their production of different words on the list, such that a single speaker might lateralize or rhotacize some of the time but pronounce other words on the list "correctly." This irregularity was the rule for most of the island; only the southeast and northeast extremes of the island uniformly switched their liquid sounds. The areas with the most standard pronunciation of these sounds, that is the places that have the least liquid switching, were located in the interior western portion of the island. In this interior region, he also noted that elision of rhotics and laterals occurs word-finally, but claims that this does not happen in other parts of the island. The mixed variants of /r/ were most common in San Juan and the surrounding
6 Navarro Tomás does not make it clear if /r/ is produced as a tap or approximant. Presumably, there are instances of both, with a predominately approximant production, as has been found in later studies, including this one. 25 areas. He states that “even well-educated people” in San Juan have the /r/-/l/ trueque de líquidas (83). Obviously, Navarro Tomás's work on liquids is impressionistic and his elicitation techniques and number of speakers per dialect area are lacking by modern standards, but his observation about the intermediate sound between /r/ and /l/ is important; his intuition underlines what later research would show about phonetic gradience in lateralization and rhotacization.
2.4.1.2 San Juan
López Morales (1983a) looked at coda /r/ realizations and found [ɾ], fricatives, approximants with a rhotic percept [ɹ], approximants with a lateral percept [l], and deleted segments. He notes that he had to decide whether to classify "unos pocos sonidos mixtos en los que el elemento lateral es sobresaliente" [some mixed sounds in which the lateral sound is the outstanding one] as laterals. Thus, unlike Navarro Tomás (1948),
López Morales did not create a category for "mixed" sounds, but rather categorized them as [l]. Of a total of 12,146 tokens of orthographic /r/, 45.6% are fricatives and approximants with a rhotic percept, 38.9% are lateralized approximants, 14% are taps and the remainder are deleted (5.6%). As for the most influential linguistic factors, he found that 1) word-final position is a more common site of lateralization than word-internal position, and that 2) lateralization is most common in pre-pausal position, then before an obstruent. Although participants of all classes in López Morales’s study lateralized, lower class speakers and men had higher rates than higher classes and women. Also, speech style (read vs. interview) matters – López Morales found more lateralization in the interview data. 26 In a detailed study of several phonological variables including coda liquids, the same author, López Morales (1983b), analyzed 9,802 tokens of coda /r/ from speakers in
San Juan and found the same five variants with the same ordering in regards to rhotic token frequency. He also analyzed the effect of the following linguistic factors: phonological context (pre-consonantal vs. pre-vocalic vs. pre-pausal), and morphemic status. He observed that lateralization was more frequent before another lateral, an obstruent or a pause and was less common in infinitives than other words. Prevocalic position disfavored lateralization and favored a fricative realization, as did pre-nasal contexts. In terms of social factors, López Morales again found that men favored lateralization while women disfavored it. While age was not significant in his previous study, in this study he found that younger people disfavor lateralization more than older people. He also discovered that only the highest of four social classes disfavored lateralization overall - the middle and lower classes all had more lateralized tokens than rhotic variants. López Morales's work (1983a, 1983b), unlike previous dialectological accounts, provides a detailed account of SJS based off of a large data set in which different ages, genders, and social classes are represented. However, as discussed in
Chapter 1, his work relies on impressionistic judgments, which is problematic in terms of the categorization of sounds. Besides the lack of nuance in his categorical analysis, López
Morales is not a native speaker of SJS, which may affect his ability to hear mixed sounds in the same way as they are perceived by native SJS listeners, as will be shown in
Chapter 4 of the present study.
27 2.4.1.2 Caguas
Medina-Rivera (1999) considered both onset and coda realizations of
(9.3% of his data). In terms of syllable-final segments, the author reported to find mostly taps and fricatives (63.9%), which he refers to as "standard variants." The author also finds a substantial number of "non-standard" variants: [l], [h] and deleted segments.
Lateralized segments are the most common in this group, accounting for 92.4% of these non-standard variants. Stressed syllables preferred non-standard productions of coda /r/; lateralized, aspirated and deleted segments considered together have a factor weight of
.57. Medina-Rivera found that infinitives (with a factor weight of .56) were more likely to undergo lateralization than other words. He also found that the word ‘porque’ favored lateralization (.53). Words with one or two syllables favored lateralization, while longer words disfavored it. The social variables of gender and age were not significant in
Medina-Rivera’s study; he does not look at class – all of his speakers were university- educated. What sets Medina-Rivera's study apart from others is his inclusion of different oral speech styles: individual conversation, group conversation, and oral presentation. In conversation groups, the informants use 37.2% "standard" /r/ vs. 47.1% in interviews vs.
93.4% in presentations. The type of speech also matters – "standard" forms were used
41.6% of the time in dialogue, 56.7% of the time in narrative and 62.5% of the time in description, argumentation and exposition. Medina-Rivera also controlled for how well the speakers knew him; his subjects were divided into two groups – those that knew him personally (family members/friends) and strangers. The speakers that knew the researcher 28 used the non-standard variants of /r/ in coda significantly more than participants who did not know him personally; the factor weights were .58 vs. .39 respectively for production of the non-standard variant. Overall, the most important contribution of Medina-Rivera's study is the understanding it lends to liquid switching across speech styles, which had previously not been explored in depth. However, like the other studies mentioned thus far, the impressionistic classification of sounds is problematic - though less so for this researcher than Navarro Tomás (1948) or López Morales (1983a, 1983b) given that he is a native speaker of the dialect he studies.
2.4.1.3 Lares
In his studies of Castañer, a town with 2,000 people in west-central Puerto Rico in the municipality of Lares, Holmquist (2003, 2005) studied the effect of social variables on vowel raising, velarization of onset /r/, and maintenance of coda /r/ and /l/ - all of which are hallmark features of the Lares dialect. Holmquist does not give specifics for the manner of articulation for coda /r/ - he reported 'normative' /r/ maintenance, which probably refers to taps and approximant tap realizations. He also mentioned that deletion occurs with /r/ segments in coda, which aligns in part with Navarro Tomás's (1948) description of this region as deleting liquids in word-final positions.
In the first Castañer study, Holmquist (2003) examined the speech of men in the community and only includes extralinguistic variables. The author analyzed both linguistic and social variables in his second (2005) study, which considered the women in the community. For the coda liquids, he considered the following context (vowel, consonant or pause) and word position (internal vs. final). Overall, men and women 29 lateralized or deleted coda /r/ 11% of the time. Holmquist considered time out of Lares as a variable and found that greater rates of lateralization correlated with having spent time outside of Lares, specifically for participants who had lived in San Juan. Speakers who have not left Lares were more likely to rhotacize
2.4.1.4 Mainland United States
In a study on a Puerto Rican community in Springfield, Massachusetts, Shouse de
Vivas (1978) impressionistically categorized coda /r/ as [ɾ], [l], an intermediate liquid, or elided. She does not mention the fricative realization that López Morales described. From her data, Shouse de Vivas forms what she calls an implicational scale for the realization of coda /r/ according to the following phonological context, which is shown in (3) below.
In her scale, the realization of coda /r/ as [l] is least common before fricatives and most common before other laterals:
(3) Shouse de Vivas's implicational scale for following phonological context (1978:626)
fricatives > nasals > stops > laterals least lateralization ------most lateralization
30 In terms place of articulation, Shouse de Vivas claims that lateralization is more common before alveolars and dentals than before other consonants. Also, the trueque was more common in her data in infinitives than in words where
Ramos-Pellicia (2007) examined coda /r/ variation in the read and spoken Spanish of a Puerto Rican community in Lorain, Ohio. The main focus of her study is the use of the retroflex American English [ɹ] in the Spanish of three generations of Lorain Puerto
Ricans. Ten percent of the /r/ tokens in the third (youngest) generation are [ɹ]. The second and third generation have much less [ɹ] with 2% and 3% of the total /r/ realizations. As far as lateralization is concerned, the author finds that more speakers use [ɾ] in the oldest generation, the second generation uses [ɾ] and [l] about equally and the third generation prefers [l]. Ramos-Pellicia attributes this ordering to the level of social awareness that the speakers have about the stigma of the trueque de líquidas on the island – the oldest speakers grew up in Puerto Rico and therefore are more careful about monitoring their use of lateralized /r/, since it is the variant that is most likely to be stigmatized. Lorain speakers favor lateralization in word-final context and disprefer it in word-internal environments. Ramos Pellicia assumes that the retroflex [ɹ] is a product of contact with
American English and uses the generational differences to support her claim. Although the use of the retroflex pronunciation may be due to contact with English, I see the preference for a bunched approximant similar to that of American English as potentially due to an internal motivation in Puerto Rican Spanish towards approximant liquids.
31 Shouse de Vivas (1978) and Ramos Pellicia (2007) provide insight into the linguistic systems of Puerto Rican speakers living away from the island. Furthermore, Shouse de
Vivas's findings for differences in conditioning for following environment and Ramos
Pellicia's conclusions about generational differences provide possibilities for instrumental research to test their impressionistic findings.
2.4.2 Acoustic Studies
2.4.2.1 Ponce
Luna (2010) considered coda /r/ and /l/ as part of a broader acoustic study on phonetic and intonational features of Ponce, a city on the southern coast of the island.
The author's formant measurements came from recorded interviews with 9 speakers.
Luna extracted a total of 90 tokens of coda
- 10 of each sound from each of the 9 speakers - and used Praat7 tools to measure the duration of the liquid, the formant values from one point in the middle of the segment, and the average value for each formant over the entire segment. The author reported that all of the liquids in coda position were approximant (and therefore did not mention other manners of articulation) and that all of the coda /r/ segments were perceptually lateralized. Luna found no significant differences for any of the formants (F1, F2, F3, F4) at the midpoint measurements, but did find that F3 and F4 measurements of orthographic
/r/ and /l/ were significantly different when the values were averaged across the durations
7 Boersma, Paul & Weenink, David (2015). Praat: doing phonetics by computer [Computer program]. Version 5.4.08, retrieved 24 March 2015 from http://www.praat.org/.
32 of the segments. The average values for F3 were 500-1000Hz lower for coda /r/ than for
/l/ and F4 was about 120 Hz lower. The author describes the coda /r/ segments in his data as sounding similar to the retroflex [ɹ] in American English and thus proposes [l] as a symbol for the lateralized rhotic sound. Duration was not significantly different between
Also, unlike other authors, Luna claims that all tokens of syllable-final /r/ are approximants and are lateralized, which lends some doubt to the rigor of his analysis, since other authors find many variants for coda /r/.
2.4.2.2 San Juan and Mayagüez
Except for Luna’s (2010) dissertation, Simonet, Rohena-Madrazo & Paz (2008) is the only other published acoustic study of coda liquids in Puerto Rican Spanish. The study was inspired by an unpublished production and perception study by the third author, Paz (2005), in which she found different acoustic correlates to the lateralized rhotic and the lateral. The production portion consisted of a word list read by a native male speaker of San Juan Spanish who was instructed to read in a casual manner. Paz 33 reported that the F3 values were lower for /r/ than for /l/. For the perception experiment,
Paz used the audio files of minimal pairs (such as alma and arma) from the production experiment and had participants identify whether the words they heard had an orthographic /r/ or /l/. Thirty Argentinians averaged at 48% correct (about chance) and thirty Puerto Ricans chose the correct segment 81% of the time, which is significantly above chance. Paz concluded that lateralized rhotics were incompletely neutralized, which confirms Navarro Tomás’s (1948) intuition about the existence of an intermediate sound between /r/ and /l/. Paz (2005) was the first to give acoustic and perceptual evidence that Puerto Rican speakers do not completely neutralize /r/ to /l/ when they lateralize. This finding has important implications for phonological theory since the phenomenon of the trueque de líquidas can result in phonetic gradience, something that not all approaches to phonology consider as part of their formal representations. The implication that speakers can hear categories in their own dialect that outsiders cannot is crucial to our understanding of dialectal differences and how phoneme boundaries are organized cross-dialectally.
Simonet et al. (2008) decided to further investigate Paz's (2005) claims about incomplete neutralization by considering the production of coda liquids of four speakers - three from San Juan and one from Mayagüez - in both a map task and a reading task. A total of 331 tokens with orthographic /r/ and 298 tokens with orthographic /l/ were extracted. The second author, who is Puerto Rican, impressionistically classified coda liquids as shown in (4).
34 (4) Liquid classification used by Simonet et al. (2008:77)
a) Deleted or elided liquid
b) Tap or trill realization
c) Inconclusive tap/approximant percept
d) Approximant realization. (This is much like what Paz (2005) called a
“retroflex or bunched rhotic”, and was heard as /l/ by Argentinean listeners.)
e) Inconclusive approximant/lateral percept
f) Lateral realization: clear /l/ percept
Variants (a), (d) and (f) were used in the acoustic analysis. That is, tokens spelled with /l/ perceived as /l/ (f) and approximant tokens spelled with /r/8 but not perceived as completely neutralized to /l/ (d) were compared. Deleted tokens (a) were also used to compare durations. Unlike Luna (2010), Simonet et al. analyzed the liquid and the preceding vocalic segment together due to the difficulty of separating the two and the importance of the vowel-to-liquid transitions. They controlled for preceding vowel (/a/ or
/i/) and following segment (/p/, /t/, /k/) and ran separate tests for each speaker individually. In order to look at the formant structure of F1, F2 and F3, the researchers took seven points along the trajectory of the vowel+liquid sequence. The authors then used these measurements for two different types of statistical tests. First, they submitted the measurements to smoothing spline ANOVAs, a statistical technique that compares curves, to see if the trajectories were different between the lateral /l/ and the approximant
8 Most of these tokens are orthographic /r/ and have undergone lateralization. However, there are nine tokens that are rhotacized orthographic /l/. 285 out of 298 tokens of coda orthographic /l/ are pronounced as laterals. Of the approximant rhotic tokens, 112 are orthographic /r/.
35 /r/.9 They then averaged the values for the first three time points (T1) and the next three time points (T2) and performed linear regressions for the two averaged time points (T1 and T2) for each formant for each speaker. The durations of vowel+/r/, vowel+/l/, and vowels without coda segments were compared using a three-factor ANOVA for each of the four speakers.
Simonet et al.'s (2008) results for duration for the three San Juan speakers show that vowel segments without coda liquids are the shortest and vowel+/l/ sequences are the longest, with vowel+/r/ in between the two. The results for the Mayagüez speaker for duration show vowel+/l/ and bare vowels are different, but vowel+/r/ is not significantly different from either of the other two contexts. The authors conclude that vowel+/l/ sequences are longer than deleted or vowel+/r/ sequences because laterals require a greater degree of tongue closure and therefore take longer to be produced. In their regression models, the results from the production of vowel+/r/ had significantly higher
F1 values than vowel+/l/ for three of the four speakers. The values for F3 for vowel+/r/ were significantly lower than those for vowel+/l/ for three of the four speakers.10 The second formant (F2) only displayed significant results for the speaker from Mayagüez and the authors do not state whether vowel+/r/ or vowel+/l/ has a higher F2 value. The authors conclude that the finding for F2 could just be due to chance and do not give it further attention. The spline analyses are fairly comparable with the regression analyses,
9 Time points over the duration of the segment are as follows: p0= 25%; p1= 37.5%; p2= 50%; p3= 62.5%; p4= 75%; p5= 87.5%; p6= 100%. These same points are used for the present study. I adapt the smoothing spline ANOVA for this dissertation - this statistical modeling technique is explained in depth in Chapter 3 in section 3.2.5.2.2 and the advantages and disadvantages of smoothing splines are discussed in Chapter 5 in section 5.3.2.
10 The speaker that does not have a significant effect for F1 is not the same speaker who does not have a difference for F3. Both of these speakers are from San Juan. 36 but since the splines are presented separately by vowel, more detail can be gained. The speaker whose F3 was not significant for the regression analysis has overlapping confidence intervals in the splines for both /a/ and /i/, which shows that there are not significant differences for either vowel. The authors did not include all of the formants in their splines graphics, so results for F1 are not visible for three of the speakers. The spline results for the second formant from the Mayagüez speaker show that results for F2 are different towards the end of the liquid trajectories with the vowel /a/, but not for /i/.
The authors point out that "acoustic differences between coda approximant /r/ and
/l/ are robust, at least for speakers from the San Juan metropolitan area" (85).11 It is likely then, that the Puerto Rican listeners in Paz’s (2005) study were able to use these formant and durational cues to distinguish approximant /r/ from /l/, which were unavailable for
Argentinian listeners. Simonet et al. (2008) conclude that neutralization is incomplete for their three San Juan speakers because all three of them produce orthographic /r/ and /l/ segments differently on at least one acoustic dimension.
Simonet et al.'s (2008) study provides a fascinating point of departure for the current study. Due to the small sample size, the authors were not able to find consistent patterns. That is, each of the three San Juan speakers have different formant patterns in their lateralizations. Also, the researchers only acoustically analyze the tokens that the second author (who is Puerto Rican) hears as intermediate segments. In order to fully understand the neutralization situation, it is important to look at all approximant realizations of orthographic /r/ to see if some of the lateralized tokens are indeed fully
11 The Mayagüez speaker also shows differences between vowel+/r/ and vowel+/l/ in formant structure, even though she does not have differences in duration. The authors seem to imply that the results for Mayagüez are not substantial, perhaps due to only having one speaker. 37 neutralized. Simonet et al. do not acoustically examine sounds spelled with /r/ that were deemed as fully neutralized (produced as [l]) in the initial impressionistic categorization, so their claim about incomplete neutralization deserves further investigation. The present study acoustically analyzes all approximant realizations of both orthographic liquids in order to pinpoint the linguistic and social conditionings of liquid realization. In other words, this study aims to investigate not only if incomplete neutralization exists, but also in what contexts complete neutralization occurs.
2.5 INCOMPLETE NEUTRALIZATION IN PRODUCTION AND PERCEPTION
Neutralization occurs when two sounds that are contrastive in some contexts do not contrast in others. For example, /t/ and /d/ are phonemic for German speakers in onset position, but they are neutralized in syllable-final position. The words Rad ('wheel') and
Rat ('advice') are homophones pronounced as [ʁat]. Until about thirty years ago, it was assumed that this neutralization was complete in that there were no phonetic differences between words like Rad and Rat, i.e. they were produced exactly the same (Kleber et al.
2010). However, this view of neutralization has been called into question by recent studies that have shown that small phonetic differences exist in contexts that were before assumed to be neutralizing. In the case of German voicing neutralization, researchers have found that vowel duration before /d/ is slightly longer than before /t/ (c.f. Charles-
Luce 1984). The observation that neutralization is often not complete is referred to in the literature as 'incomplete neutralization.'
Many researchers (e.g. Fourakis & Iverson 1984, Jassem & Richter 1989, Warner et al. 2006) claim that incomplete neutralization is an artifact of orthography and from 38 this, a great deal of debate has arisen over the existence of the phonetic differences found in recent studies. In careful speech styles, speakers tend to increase the differentiation between word pairs that are otherwise less distinctive or totally neutralized, which is used as a point of argument for researchers who believe that neutralization is complete when speakers are not monitoring their speech. A number of studies address this disagreement about whether or not spelling is serving as a model of the underlying form for speakers; representative studies are discussed below.
Dinnson & Charles-Luce (1984) point out that considering incomplete neutralization as resulting from orthography is a circular argument. Speakers could be making coda distinctions that otherwise would be completely neutralized because they know that the segments they see 'should' be different given that they are distinct in onsets.
On the other hand, the orthography that was implemented in the first place could be seen as a result of real differences in pronunciation. The authors use Catalan to explore incomplete neutralization in the context of a language that does not have spelling differences for singular forms but has distinctive voicing for plurals. They employ five minimal pairs like /kap/ ('toward') and /kab/ ('head'), both spelled
While they do not have significant results for their five speakers considered together, speakers vary a great deal in the phonetic cues under investigation. Two speakers show evidence of incomplete neutralization - one in closure duration and the other for vowel length. Although their results suggest that neutralization is sometimes complete or at least
39 highly variable, the authors are hesitant to discard incomplete neutralization of these segments as a possibility. They point out that "in order to establish a case of neutralization, it would be necessary to show that all phonetic parameters that could possibly be influenced by the underlying distinction have been identified and have failed to yield any phonetic differences" (56). In other words, proving complete neutralization requires an exhaustive look at phonetic detail.
Another study that attempts to get around the issue of orthographical influence on neutralization is Röttger et al.'s (2014) work on the production of coda stop voicing in
German. Rather than a reading task where participants are likely to reference the spelling, the authors use an audio prompt with the plural form and have participants produce the singular form after listening to the plural. Due to resyllabification with the plural morpheme [ɐ], full voicing contrast is maintained in the plural forms. For example, the
/d/ in Rad ('bike') is resyllabified in the plural form, Räder, [ʁæ.dɐ]. Unlike Catalan, where spelling of singular forms is the same despite plural voicing contrasts, German singular forms have stop voicing contrasts encoded in the spelling. Since Röttger et al. find that the main cue for voicing is a vowel that is, on average, 16% longer than for the voiceless stop, vowel length in the carrier phrase is modified to be: 1) twice as long for voiced sounds 2) normal (16% longer for voiced) 3) exactly the same length as voiceless
4) 16% shorter for voiced than for voiceless (inverse condition). The results indicate that speakers are sensitive to the vowel length in the input plural; the longer the vowel in the plural form, the more likely speakers are to voice their productions of the codas of singular forms. In other words, speakers are tuned in to the cues that they hear and their production is influenced by these cues. The authors conclude that incomplete
40 neutralization, at least in terms of production, does exist, independently of orthography.
In a production study on Dutch, Warner et al. (2004) find significant differences in vowel length before supposedly completely neutralized singleton vs. geminate stops, suggesting incomplete neutralization. The authors use minimal pairs like heten (‘to be called’) and heetten (‘were called’), which are spelled differently due to morphological differences but can both be broadly transcribed as /heːtən/. Since these spelling differences are due to a morphemic difference, the authors attempt to disambiguate the influences of spelling and morphology on neutralization in a later study by using minimal pairs that have the same morphological contrast without spelling differences. The pairs used in this study are first and third person singular forms, which differ only in the pronoun employed. For example, /pɪt/ is paired with ik ('I') and with hij ('he'). The participants were instructed to read the pronoun silently and the verb form out loud.
Warner et al. (2006) measure vowel duration, final stop closure duration and total duration of final consonant. They find a null effect, in other words, neutralization is complete for forms with underlyingly different morphemic structure. That is, there is no difference in vowel length for the first and third persons of 'to be',
The studies discussed up to this point dealt with the influence of orthography in the production of (in)completely neutralized sounds. Next, I discuss two perception
41 studies of these phenomena, which show that production differences in neutralizing environments can be used by listeners in perception. In Eastern Andalusian Spanish
(EAS), both /s/ and voiceless stops are aspirated in coda position. For example, 'hasta',
'apta,' and 'acta' are all pronounced as apparent homophones [aht:a]. Gerfen & Hall
(2001) observe durational differences in the production of aspirated segments in EAS whereby words spelled with /s/ have significantly longer aspiration durations than words in which /p/ or /k/ is aspirated. They also find that the length of the stop is longer for words spelled with /p/ than words with /s/. Bishop (2007) follows up Gerfen & Hall's study with a perception experiment on EAS, in which he finds convincing evidence for incomplete neutralization. To test whether or not listeners can perceive differences in aspiration duration and as a result can differentiate between lexical items that are apparent homophones, the word 'hasta' is recorded by a native EAS speaker and an aspirated token is modified in duration of aspiration to 30ms, 55ms and 80ms. The stop segment /t/ is modified for length of stop closure in 50ms intervals from 60-210ms. These continua are used because they mirror the durations from Gerfen & Hall's (2001) production data. The stimuli are presented in a carrier sentence (Oye, dime _____, tío) and 23 listeners choose the word they hear from multiple choice questions. Length of aspiration is not significant in Bishop's results, but length of stop closure is – a longer closure duration causes listeners to hear /apta/ rather than /asta/, which leads the author to conclude that the contrast between coda /p/ and /s/ is not completely neutralized and is used in the perception of these sounds. Bishop does not consider combining aspiration and stop closure, which may be in reality what listeners use to perceive differences and which could therefore produce even more convincing results for incomplete
42 neutralization.
Kleber et al. (2010) examine stop neutralization in German to answer questions about the effects of linguistic environment, language-specific frequency effects, and phonotactics on listeners' ability to discriminate between voiced and voiceless forms. In
German, voiceless consonants typically follow lax vowels, so the researchers expect listeners to more readily hear voiceless stops after lax than tense vowels. The authors invent city names to form minimal pairs with vowel differences - Widdlinn— Wittlinn
(/VlaxCalv/), Bigglinn—Bicklinn (/VlaxCvel/), Niedlinn—Nietlinn (/VtnsCalv/) and
Mieglinn—Mieklinn (/VtnsCvel/). By forming continua of the names with modification of the vowel lengths in proportion with the stop lengths, the authors are able to test whether participants hear voicing in the different environments. As expected, listeners label more stimuli as voiceless after lax vowels (36.6%) than tense vowels (10.9%), which shows that the lexical frequency of the lax vowel with devoicing influences listeners when they are presented with new lexical items. Since tense vowels before alveolar stops have more variation than tense vowels before velar stops, the researchers expected listeners to be more sensitive to voicing contrasts before alveolar stops. This is not the case, so Kleber et al. propose that this is possibly due to the ability of velar stops in coda to resyllabify to onset and the impossibility of alveolar stops resyllabifying
(German phonotactics permit /kl/ and /gl/ but do not permit /tl/ and /dl/ in onsets, like
English). A second experiment confirms that the ability to resyllabify helps listeners hear voicing. Kleber et al. also show that voicing of /t/-/d/ and /k/-/g/ is easier for listeners to perceive before /l/, which is a cue-rich context, than before the fricative /ʃ/. The authors' contribution demonstrates the importance of considering lexical type frequency and 43 phonotactic restraints specific to the language under study as well as universals such as the richness of the cue environment in which the variable is situated. The authors find that listeners can use small acoustic differences to differentiate between voiced and voiceless sounds, suggesting that incompletely neutralized sounds can be perceived as different. Moreover, their perception is affected by type frequency and language-specific phonotactics.
Other studies consider the effect of semantic and pragmatic processing on incomplete neutralization. In situations where the context obviates which word of a minimal pair is being referred to, less careful articulation may be employed and neutralization may be complete due to the lack of importance afforded to the differentiation of the pair (Pollack & Pickett 1964). Grammaticality of utterances also is expected to have an influence on how carefully speakers produce forms - it has been shown that ungrammatical sentences tend to cause speakers to produce longer vowel durations (Charles-Luce & Walker 1981). The premise of Charles-Luce's (1993) production study on the neutralization of voicing in final stops in Catalan is that having semantic information (i.e. semantic bias) would lead speakers to neutralize more than if they were unsure of the meaning of a lexical item. The author hypothesizes that the degree of neutralization is “inversely related to the degree of biasing information present"
(30). Vowel duration, voicing during closure and closure duration are measured. Vowel duration before the voiced and voiceless final stops is significantly different for the neutral context but not the biasing context. The author uses this finding to argue against scholars like Fourakis & Iverson (1984) who claim complete neutralization occurs for final stops in German. Charles-Luce states that research contexts give participants the
44 assumption that researchers know the words being studied and therefore may completely neutralize because the gradient differences that might otherwise be present are unnecessary for the context. She argues that more careful articulation leading to incomplete neutralization is not an artifact of certain kinds of data, but rather a learned habit that aids people in everyday communication. According to Charles-Luce, if the listener (and the speaker) is not primed, articulation will be more careful as an accommodation strategy.
Port & Crawford (1989) also seek to determine how context, i.e. the level of pragmatic information available to the participant, affects the degree of neutralization.
Using 3 minimal pairs of German words with the typically studied t/d codas (bunt, Bund;
Rat, Rad; seit, seid), the researchers designed different tasks ranging from neutral sentences (e.g. 'She rode her bike (Rad) to the store.') vs. contrastive sentences (e.g. 'I said 'bike' (Rad), not 'advice' (Rat)'). The contrastive sentences were used a second time and one of the researchers had the participant read them so that he could write down the word in question to make the task more communicatively purposeful. Finally, the minimal pairs were included on a word list for a traditional reading task. Port & Crawford measured the duration of the vowel, the duration of the stop closure, the duration of the burst, the nasal closure, where applicable, and the voicing present during the stop closure.
They found that voicing differences were not statistically significant within any of the contexts and the contexts were not significantly different from each other. Results for the first two conditions (different types of sentences) were not significantly different, but the participants used a greater deal of contrast when the researcher was writing the words
(78%) than when the participants were simply reading the words (63%). Port & Crawford
45 conclude that incomplete neutralization is not due to unnatural speech production since all conditions showed distinctions between the voiced/voiceless words in the minimal pairs above chance. Furthermore, the authors argue that speakers can modify the level of contrast in order to navigate different communicative situations. The authors then performed a perception task to test how well listeners heard the productions of two speakers from the dictation and word list sections of the first experiment. The motivation for this second part was to see if the subtle differences in production would be different enough for perceptual discrimination. Altogether, the five listeners perform 20% better than chance (70% correct responses), the less neutralizing speaker of the two was better understood overall and perceptual performance on the dictation section was better than the word list reading. The researchers claim that the congruency between the two experiments makes a strong argument for incomplete neutralization.
The studies discussed above conclude that neutralization is complete or incomplete in their data based on differences in the production of underlying
(orthographic) representations or the perception of these differences. In other words, studies on neutralization phenomena often decide if segments are incompletely neutralized by testing if there are acoustic differences in production. Four of the seven studies mentioned in this section look at production exclusively. Perceptual tests examine whether the differences between cues in production are accessible to listeners. By considering both production and perception, the researcher gains more information about how neutralization phenomena function. In his study on the perception of /s/ and stop aspiration and stop closure length in EAS, Bishop (2007) points out that not all acoustic differences in the production of segments will necessarily be used by listeners. His study
46 is a case in point - listeners that are not able to utilize the aspiration duration did distinguish underlying /s/ from underlying /p/ based on the duration of the stop closure.
The speech signal is said to be redundant - that is, it is often the case that more acoustic information is provided than is employed in perception (c.f. Lisker 1983). It is also possible that certain segments are incompletely neutralized by speakers in production even though native listeners cannot separate and therefore cannot categorize the sounds.
Some researchers call these sounds 'near mergers.'12
The acoustic analysis of the production data for SJS in Chapter 3 of the present study will give further evidence for either complete or incomplete neutralization of coda liquids based on whether or not there are differences between orthographic vowel+/r/ and vowel+/l/ in terms of F1, F2, F3, F4 and duration. The perception experiment in this dissertation (Chapter 4) will contribute to a more complete picture of the neutralization situation in SJS codas by investigating whether or not Puerto Ricans can utilize small phonetic differences in the production of liquids in their perception.
12 According to Röttger (2012), the difference of the terms 'incomplete neutralization' and 'near mergers' really has to do with the history of the research on different phenomena. Segments that are considered incompletely neutralized are the same segments that were thought to have complete neutralization whereas the discovery of near mergers came from sound change in English vowels where speakers pronounced two different sounds but only heard one category. 47
CHAPTER 3: CODA LIQUID PRODUCTION IN PUERTO RICAN SPANISH
3.1 INTRODUCTION
The goal of this chapter is to understand the characteristics of syllable-final /r/ and
/l/ in the Spanish spoken in San Juan, Puerto Rico in order to determine whether or not these sounds are neutralized in this dialect. The previous studies discussed in Chapter 2 do not provide a clear picture of coda liquid variation in SJS. This chapter reports results from the analysis of acoustic data using a variety of statistical methods to quantify the conditioning of social and linguistic variables on the production of liquid sounds. Section
3.2 describes the methodology developed to collect and analyze the data for the production study presented in this chapter. Section 3.3 presents the hypotheses and goals of the study and section 3.4 details the results. A synthesis and discussion of the results will be presented in Chapter 5.
3.2 METHODOLOGY
3.2.1 Participants
The corpus for this project comes from interviews with twenty-four SJS speakers that I recorded in San Juan in the summer of 2013. The sample is evenly distributed in terms of gender, with twelve women and twelve men. Due to my point of entry near the
48 University of Puerto Rico Río Piedras campus, all but three of the participants have at least a bachelor's degree or are currently university students. Speaker age ranges from eighteen to sixty-two with a gap in the thirties range. That is, speakers fall into two age groups - a younger group from ages 18-28 and an older group from ages 42-62. Table 2 shows the age, gender, and education for each of the speakers included in this study. All of the speakers have spent the majority of their lives in the municipality of San Juan or the surrounding municipalities of Carolina, Trujillo Alto, Caguas, Guaynabo, or
Bayamón which all fall within a 20-minute radius of the city center by car. More specifically, all twenty-four speakers spent their childhood and adolescence in the San
Juan area from birth or the age of two in the case of one speaker. Three speakers spent a number of years in New Jersey, New York, and Florida respectively and most of the participants have at least an intermediate level of proficiency in English.13
Informant Number Age Gender Education
1 51 Male BA
2 50 Female Associate
3 22 Female BA
4 60 Male High School
Continued Table 2. Age, gender, and education for the speakers in the corpus used for the production analyses
13 According to the U.S. Census of 2000, 71.9% of Puerto Ricans speak English less than "very well." This number is an overall estimate for the island. The percentage of proficient English speakers in San Juan, especially among educated speakers, is probably considerably higher. Therefore, the level of bilingualism in the corpus is not surprising. 49 Table 2: Continued Informant Number Age Gender Education 5 45 Female MA 6 19 Female BA 7 27 Male MA 8 18 Female BA in progress 9 18 Female BA in progress 10 28 Female High School 11 24 Female MA in progress 12 25 Male BA 13 60 Female MA 14 56 Male BA 15 22 Male BA in progress 16 18 Female BA in progress 17 23 Female BA 18 55 Male BA 19 62 Male PhD 20 22 Male BA 21 42 Male BA 22 22 Female BA 23 20 Male BA in progress 24 21 Male BA in progress
3.2.2 Tasks and Recording Procedures
The data for this project comes from sociolinguistic interviews based on a flexible protocol, included in Appendix A. The protocol contains questions about family, work, education, hobbies, and life in San Juan and was designed to elicit spontaneous speech in accordance with first wave variationist analysis (Labov 1966, 1984). Most of the interviews adhered closely to the questions on the protocol, but were tailored to a certain
50 extent to each individual's interests to keep conversations as natural as possible. For example, if an individual seemed more excited to talk about her profession than her family, more follow-up questions were asked about her profession and the questions about family were not emphasized. The interviews lasted between forty minutes and an hour and fifteen minutes. Participants also performed a reading task after completing the interview, but this task will not be considered for this study.14 I conducted the interviews myself, and it should be noted that although I speak Spanish fluently, I am not a native speaker. My speech has influences from a number of different dialects of Spanish and my accent is notably different from that of my interviewees. While some level of speech accommodation is inevitable in this context, the participants seemed comfortable speaking to me and did not speak more slowly or carefully with me than they did with other Puerto Rican interlocutors.
The interviews were recorded with a Zoom H2 audio handheld recorder using a sampling rate of 44.1 kHz. A preprocessing step was required to reduce the presence of background air-conditioning noise in the audio data. This white noise removal was performed by a spectral noise gaiting filter in Audacity.15
14 The reading task consists of 102 sentences on index cards. These sentences contain real and invented words with coda liquids, as well as words with the same phonological context but no coda segment for potential use as comparisons for duration. The target words are inserted in carrier sentences such that each word appears both phrase-medially and phrase-finally to control for any effects of prosody. Word-final liquids sometimes appear before words that start with a vowel in order to examine resyllabification effects. As stated, the reading task will not be used for this project but will be useful for future studies on the influence of speech style on liquid production. 15 http://audacity.sourceforge.net/ 51
3.2.3 Data Analysis
The data discussed in this section pertains to two separate analyses: 1) an analysis of the manner of articulation for orthographic coda /r/ and /l/ and 2) an analysis of the duration and formant structure for approximant articulations of vowel+orthographic /r/ sequences compared to vowel+orthographic /l/ sequences. The details of the two analyses will be discussed in depth in terms of statistical procedures in section 3.2.5 and the results are presented in section 3.4. Here, it is simply important to note that data was extracted simultaneously for both analyses, but that certain tokens were included for one analysis but fell outside the envelope of variation for the other.
The first five minutes of each interview were excluded from the analyses in order to avoid the more formal speech that participants often use before they have settled into the interview situation. After the first five minutes, I marked sequences of vowel+orthographic liquid and identified the manner of articulation using evidence from the spectrogram and waveform in Praat. Virtually all /l/ productions were realized as approximants. Coda rhotics presented more variation and were classified as fricatives, deletions, taps, approximants trills, or vocalizations. Productions with any evidence of frication were classified as fricatives, even if part of the /r/ segment did seem approximant in nature; see Figure 2 for an example of a fricative token. If there was no acoustic evidence of the presence of a sound corresponding to the orthographic rhotic, this was coded as deletion; see the example in Figure 3. Since full closures for taps were extremely rare, taps were characterized by a brief interruption in the formant structure with an extreme reduction of intensity. See Figure 4 for an example. Productions that
52 have continuous formant structure from the previous vowels but with some decrease in intensity were coded as approximants, as in example Figure 5. Only ten trills were found in coda position and these sounds often occurred in metalinguistic discussion, so they will be left of the manner of articulation quantitative analysis. Vocalizations will also not be included in the analysis due to a fairly low token count of 25 and the fact that the majority of these occurrences are found in a single lexical item, porque 'because'.
Figure 2. Fricative /r/ in entrar ‘to enter’
Figure 3. Deleted /r/ in decir que ‘to say that’
53
Figure 4. Tap /r/ in conocer y 'to know and'
Figure 5. Approximant /r/ in fuerza ‘strength’
Only productions classified as approximants were subject to the acoustic analysis, where the values of F1, F2, F3 and F4 and the duration of the sequences were extracted.
In this analysis, the vowel and the liquid were considered together, as a unit. The vowel was included as part of the sequence in the acoustic analysis for two reasons. First, given the acoustic similarities of the vowel and an approximant liquid, it is very difficult to determine where the vowel ends and the liquid begins. Second, and more importantly, previous studies on coda liquids have found that these sounds have coarticulatory effects with the preceding vowels which are crucial in the perception of these consonant sounds.
In fact, it has been shown that the exclusion of the vowel from the sequence can render the liquid unidentifiable (cf. Guirao & García Jurado 1991 - see section 2.3.1). A study 54 on coda liquid production, therefore, must take the vowel and the transition from vowel to liquid into account in order to accurately and completely describe these segments.
With the few exceptions that I will outline below, I coded all occurrences of orthographic vowel+/r/ and vowel+/l/ sequences from five minutes into each interview until I had found a minimum of 75 occurrences of vowel+approximant /r/ and 30 occurrences of vowel+approximant /l/ per speaker. In other words, since the questions most central to this study pertain to the approximant liquid sounds given that these are the site for possible neutralization, the goal was to obtain a relatively even sample of these types of sounds from each of the speakers. Since words with coda /r/ are much more frequent in the Spanish language than words with coda /l/, it was not possible to extract a similar number of tokens with /r/ and /l/. Generally, about an hour of interview time yielded the target 75 vowel+approximant /r/ and 30 vowel+approximant /l/ sequences.
Some speakers were less talkative or used fewer words containing coda liquids, which meant that a few speakers fell slightly below these target numbers. In the coding process,
I recorded all tokens containing these coda sounds until I reached either the end of the interview or the appropriate number of tokens for each sound. If the target number of one sound had been reached but not the other, numbers exceeding the goal were sometimes extracted. For example, I sometimes obtained 75 vowel+approximant /r/ tokens while only having found 20 vowel+approximant /l/, meaning that I would continue extracting both sounds until 30 /l/ segments were located. Therefore, some speakers have slightly higher counts of one of the two sounds. While coding for approximant sounds, all other productions (taps, deletions, etc.) of rhotics and laterals were also extracted in order to get a full picture of coda liquid production in this dialect. The vowel+approximant liquid
55 sequences were manually marked for the onset of the vowel and the offset of the liquid, as described below. The exact onset and offset of non-approximant liquids were not marked since these sounds were not analyzed for formant structure and duration.
Exclusions from the envelope of variation for both the analysis of the manner of articulation and the acoustic analysis include tokens of vowel+liquid that are not analyzable due to background noise or low volume on the part of the speaker. Liquids that appear in English words incorporated by interviewees either due to lexical borrowing or code switching were also left out of both analyses. Occurrences of the word verdad
'truth'/'right' in which the /d/ following the /r/ was elided, thus making a lateralized /r/ into an onset, were also left out of both analyses.16 Vowel+liquid sequences in the phonological contexts listed in Table 3 below were excluded from the acoustic analysis of approximant liquids due to segmentation difficulties. These sequences were included, however, for the analysis of manner of articulation, since this analysis did not require segmentation. It is important to note that word-final consonants in Spanish resyllabify with the following word if the word begins with a vowel. Since dialectological accounts of SJS describe lateralization even in the context of resyllabification, sequences of vowel+liquid before a word beginning with a vowel are included in the analysis. See section 2.3.2 for a discussion of resyllabification in Spanish.
16 The product of this /d/ elision sounds something like [be.ˈla] to my ears. This is a high frequency discourse item, making reduction unsurprising. However, this particular reduction is not, to my knowledge, especially common among dialects of Spanish and merits a closer look. 56 Excluded Context Example vowel+liquid followed by /l/ adoptar los 'to adopt the' vowel+liquid preceded by another vowel que al 'that to the'
Table 3. Vowel+liquid sequences excluded from the acoustic analysis of approximant liquids
Moving on to the acoustic analysis of the approximant tokens, the vowel+liquid sequences were first manually segmented in Praat. The onset of the vowel was identified using the beginning or increase of energy in F2 in the spectrogram and the periodicity in the waveform that is characteristic of vowel production. The offset of the liquid was determined using evidence of the beginning of the following sound. In the case of following stops and fricatives, the onset of the closure or frication and the sudden drop of
F2 energy marks the end of the liquid. If the liquid is word-final and the following word begins with a vowel, the end of the approximant liquid was determined by the increase of
F2 intensity of the vowel. The end of pre-pausal liquid segments were marked where the acoustic energy dropped off into silence.
For the approximant liquids analysis, a Praat script was used to extract the durations of all the segmented vowel+liquid sequences as well as formant values for F1,
F2, F3, and F4. Following the methodology developed by Simonet et al. (2008), formant value measurements were taken at seven evenly-spaced points throughout the duration of each sequence starting at 25% into the duration of each vowel+liquid sequence until the offset of the liquid. The script therefore took measurements of F1, F2, F3, and F4 at 25%,
37.5%, 50%, 62.5%, 75%, 87.5%, and 100% of each vowel+liquid sequence. Taking measurements at the seven points rather than just one or two places along the duration of 57 the sequence helps develop an understanding of how these sequences change through time.
3.2.4 Independent Variables
In this section, I describe the independent variables that are explored in both analyses, i.e. the manner of articulation analysis and the analysis of the acoustic features of approximant liquids. The linguistic, or internal, variables are those that are part of the structure of the language itself. Six linguistic independent variables are considered: liquid type, vowel, stress, word position, previous sound, and following sound. Each of these variables are explained below in their corresponding sections. The social variables of gender and age have been investigated in past studies and have been shown to have significant conditioning on the realization of liquids in SJS (see section 2.4.1 for details), and for this reason they were considered in both of the analyses presented here. Due to the limited amount of data from speakers with less than a college degree, education will not be examined in this study. All of the independent variables apply to both analyses, except for liquid type, which only applies to the approximant liquids analysis since sequences with /l/ are not examined in the other analysis.
3.2.4.1 Liquid Type
This variable refers to the orthographic liquid present in each vowel+liquid sequence under analysis. In other words, it codes whether a word is spelled with /r/ or /l/.
This is the only independent variable that is considered exclusively in the approximant
58 liquids analysis. Since the rhotic manner of articulation analysis only considers occurrences of orthographic /r/ and not those of /l/, this variable is unnecessary. As will be explained in more detail in section 3.2.5.2.1, the influence of liquid type on the dependent variables will be considered on its own and through interactions with the rest of the independent variables. See Table 4 for an example of each level of liquid type.
Example Token Liquid Type Level parte 'part' /r/ cultura 'culture' /l/
Table 4. Examples from the coding of liquid type
3.2.4.2 Vowel
The vowel refers to the vocalic element in the sequence of vowel+liquid. Besides the five simple vowels in the Spanish inventory - /a/, /e/, /i/, /o/, and /u/ - a number of diphthongs appear before liquids. In the corpus for this project, /ja/, /je/, /jo/, /wa/, /we/, and /wi/ occurred with coda liquids. The diphthongs /jo/ and /wi/ were excluded from the analysis due to the low token counts of ten and one token respectively. The remaining diphthongs were then grouped according to the main vowel such that /ja/ and /wa/ form a factor level called 'glide+/a/' and /je/ and /we/ compose the level 'glide+/e/'. Thus, the five simple vowels and two diphthong categories create seven factor levels for the vowel variable.
59 Example Token Vowel Level estar encontrado 'to be found' /a/ disertaciones 'dissertations' /e/ fácil 'easy' /i/ por un 'for a' /o/ cursos 'courses' /u/ estudiar primero 'to study first' glide +/a/ vuelve 'returns' glide +/e/
Table 5. Examples from the coding of vowel
3.2.4.3 Stress
The independent variable of stress reflects whether or not the syllable containing the liquid sound is stressed or unstressed. When coding for stress, I created special categories for word-final liquids before vowels to take into account the fact that syllable- final consonants in Spanish resyllabify to the following syllable when followed by a vowel. These categories specified whether the syllable with the liquid was stressed or unstressed and also whether the syllable of the next word was stressed or unstressed, which produced four categories for word-final liquids before vowels. However, these categories were collapsed according to whether or not the vowel+liquid sequence was stressed or unstressed such that resyllabification is not taken into account. This reduction of the six groups listed in the 'Initial Coding' column of Table 6 to the two groups in the
'Final Coding' column was necessary due to the small numbers of tokens in the groups with following vowels. While the smaller number of factor levels introduced to the model 60 reduces noise and empty categories, a certain amount of nuance is lost. The effect of stress on liquid production of following vowels merits future study.
Example Tokens Initial Coding Final Coding carta 'letter' stressed ser este 'to be this' stressed before stressed stressed vowel
hacer el 'to make the' stressed before unstressed
vowel por fotografía 'by unstressed photography' por algo 'for something' unstressed before stressed vowel unstressed el estudiante 'the student' unstressed before unstressed vowel Table 6. Examples from the coding of stress
3.2.4.4 Word Position Word position is a binary variable that indicates whether the vowel+liquid sequence is located in the middle or end of a word. Table 7 gives an example of each word position.
Example Token Word Position Level porque 'because' word-medial ayudar 'to help' word-final
Table 7. Examples from the coding for word position
61 3.2.4.5 Preceding Sound
This variable captures the manner of articulation of the sound that precedes the vowel+liquid sequence. Table 8 shows examples of all of the preceding sounds found in the corpus. The orthographic preceding sound is shown in bold in the first column, where the vowel+liquid sequence is underlined, and represented by its corresponding IPA symbol in the second column. During the coding process, the specific sound, shown in the second column of Table 8, was recorded. If the vowel+liquid token came after a pause, either the beginning of a sentence or after the speaker hesitated before saying something, 'pause' was recorded as the previous sound. The preceding sounds in the second column were then collapsed according to manner of articulation to form seven groups: pause, approximant, fricative, lateral, nasal, rhotic, and stop. Complex onsets such as /br/ and /bl/ were collapsed according to the second element of the sequence - rhotic or lateral, respectively.
62 Example Tokens Corresponding Preceding Sound Pronunciation Level (pause) algo 'something' (N/A) pause (pause) hermosa 'pretty' universidad 'university' [β] poder económico 'economic [ð] power' approximant18 lugar que 'place that' [ɣ] mayor 'older' [ʝ]17 forma 'shape' [f] mejor por 'better for' [x] ciertos 'certain' [s] fricative es algo 'it's something' [h] escuchar y 'to listen and' [tʃ] calor pues 'heat well' [l] lateral hablar 'to speak' tomar muchas 'to take a lot of' [m] final de 'end of' [n] nasal español y 'Spanish and' [ɲ] correr 'to run' [r] general 'general' [ɾ] rhotic abril 'April' porque 'because' [p] tal nota 'such a grade' [t] azúcar pero 'sugar but' [k] stop invierten 'they invest' [b] expanderse 'to expand' [d] Table 8. Examples from the coding for preceding sound
3.2.4.6 Following Sound This variable captures the manner of articulation for the sound after the vowel+liquid sequence. It was coded similarly to the preceding sound variable from section 3.2.4.5, except that following laterals were excluded from the envelope of
17 The pronunciation of orthographic
63 variation for the analysis of approximants (but included for the manner of articulation analysis), as explained above, and following rhotics and complex onsets were left out due to the very low token counts. Also, vowel is a factor level for following sound, although it was not a possible level in the preceding sounds factor. The vowel [a] is given as an example, but various vowels and diphthongs also enter into the vowel factor level.
Example Tokens Corresponding Following Sound Level Pronunciation escribir (pause) 'to write' (N/A) pause volver 'to return' [β] acuerdo 'agreement' [ð] approximant algo 'something' [ɣ] del yunque 'from the forest' [ʝ] perfecto 'perfect' [f] margen 'margin' [x] fricative persona 'person' [s] parcha 'passion fruit' [tʃ] el mayor 'the oldest' [m] nasal gobierno 'government' [n] llegar a 'to arrive at' [a] (etc. - all vowels and vowel diphthongs included) aprender pero 'to learn but' [p] faltaba 'was missing' [t] porque 'because' [k] stop mil dolares 'a thousand [d] dollars' Table 9. Examples from the coding for following sound
3.2.4.7 Gender
The participants in the sample self-identified as male or female and were coded as such. As discussed in section 3.2.1, the gender sample for this study is even with twelve men and twelve women.
64 3.2.4.8 Age
Since the most of the participants were recruited near a university campus, the participants for this study are either young students completing their undergraduate or graduate education or older university employees. Therefore, the data divides naturally into two age groups - a younger group comprised of participants aged 18-28 and an older group with speakers from ages 42-62. The statistical analysis considers age as a binary variable with 'younger' and 'older' speakers. See Table 2 in section 3.2.1 for a list of participants with their age, gender, and education information.
3.2.4.9 Summary of Independent Variables
Table 10 contains a summary of all of the independent variables used in the statistical analyses. Note that, as explained earlier, liquid type is not used for the manner of articulation analysis, since that analysis focuses on orthographic /r/ tokens exclusively.
65 Independent Variable Factor Levels
Liquid Type /r/ /l/ Vowel /a/ glide+/a/ glide+/e/ /e/ /i/ /o/ /u/ Stress stressed unstressed Word Position medial final Preceding Sound pause approximant fricative lateral nasal rhotic stop Following Sound pause approximant fricative nasal vowel stop Gender male female Age younger older Table 10. Independent variables and factor levels summary
3.2.5 Statistics
All of the statistical analyses were performed using R (R Development Core Team
2015). The following subsections review the data under analysis and describe the particular statistical tools and techniques employed for the manner of articulation analyses (3.2.5.1) and the various analyses of the acoustic measurements of the 66 approximant liquids (3.2.5.2). The analyses for manner of articulation consider categorical responses whereas formants and duration analyses of approximant liquids are examined via continuous dependent variables. The regression analyses for both manner of articulation and approximant liquid analyses were performed with sum of squares contrasts. Therefore, each factor group does not have a reference level, but rather is compared to the average of all factor levels taken together.19
3.2.5.1 Statistical Analysis for Manner of Articulation Data
Since occurrences of orthographic /l/ in the corpus are almost exclusively produced as approximants, after presenting percentages and token numbers of the different articulations of both /r/ and /l/, the statistical analysis will examine the variation of /r/ only. The goal of the manner of articulation analysis is to see how rhotic realizations are conditioned by the independent variables described in section 3.2.4 and whether or not certain productions are favored over others in certain linguistic environments or by speakers of a particular gender or age group. The four most common rhotic productions - approximants, taps, deletions, and fricatives - were fitted to four
19 The use of sum of squares contrasts is common practice in sociolinguistics, especially in the tradition of software designed for sociolinguistic research, such as Varbrul and related programs, which do not have other options for contrasts. In research involving experiments, treatment contrasts are the norm and in fact are the default setting in R. With treatment contrasts, there is a reference level for each factor such that observations are compared to a baseline of a particular factor. For example, if this study had been carried out with treatment contrasts, the preceding sounds might all be compared to preceding stops. Sum of squares contrasts lack a reference level and instead compare all levels of a factor to the average of all of the factors taken together. The sum of squares contrasts were used for the present project because this technique was more appropriate for the research questions. In other words, the questions deal with what contexts stand out as different from all of the others, rather than having some baseline upon which other observations were being tested.
67 different models in which one of the productions is compared to the other three combined. The dependent variable for each analysis, therefore, is binary. For example, the analysis for approximants involves comparing approximant realizations to 'other'
(taps, deletions, and fricatives combined) articulations. All of the linguistic and social independent variables in section 3.2.4 except liquid type are included in the analyses.
Approximant /r/ was analyzed using a mixed effects logistic regression model with speaker as a random effect, while taps, deletions, and fricatives were each examined using conditional inference trees. The latter type of analyses was employed due to the relatively low token counts for taps, deletions, and fricatives, which means that these productions do not lend themselves to regression analysis. The conditional inference trees were built in R using the packages party and partykit (Hothorn & Zeileis 2015).
Conditional inference trees show the relationship between independent variables and how they interact with the selection of the two possible dependent variables. The independent variable at the top of the tree is the most important predictor in the selection of the dependent variable. This variable branches out with the factor levels that interact with other independent variables. At the bottom of each inference tree, the branches end in nodes that show the number of tokens in that particular branch and the percentage breakdown of the dependent variables. While conditional inference trees are not a predictive tool that generalize to a population in the same way that regression analysis does, they are a useful way of visualizing the data and understanding the interactions between the variables. Conditional inference trees are a suitable way of dealing with variables that have a low number of observations and will therefore be used to explore rhotic manner of articulation in the corpus.
68 3.2.5.2 Statistics for the Acoustic Features of Approximant Liquids
The data for these analyses comprise 1,982 tokens of vowel+approximant orthographic /r/ and 703 tokens of vowel+approximant orthographic /l/ which were extracted from the speech of twenty-four informants, as discussed in section 3.2.4. The formant values taken at each time point were used as dependent variables for both the linear regression analysis in 3.2.5.2.1 and the smoothing splines analysis in 3.2.5.2.2. The linear regression analysis also considers duration as a dependent variable.
3.2.5.2.1 Linear Regression
The linear regression models serve to test the effect of the independent variables
(see section 3.2.4) on the duration and formant values of vowel+approximant liquid sequences. Since measurements were taken at seven points, a separate statistical model was built for each formant at each time point, resulting in 28 models (seven time points for four formants). An additional model considered the duration of the segments. All of the 29 analyses are generalized linear mixed-effects models built with the lmer function in R. It has become standard practice in sociolinguistic methodology to employ speaker as a random effect in order to capture the fact that the individuals in the sample are not generalizable to the population at large in the same way that other factors, like gender or age, are (Tagliamonte & Baayen 2012). Using mixed effects models also eases the violation of the assumption made by regression models that each occurrence is independent, i.e. that there is only one token per speaker. For these reasons, all the mixed-effects models used in this study include speaker as a random effect.
69 Since the goal of these analyses is to understand where and how orthographic /r/ and /l/ differ from each other, interactions between the independent variable of liquid type (/r/ vs. /l/ in the orthography, see section 3.2.4.1) and the other independent variables were also added to the model. See Table 11 below for a listing of the interaction terms considered in the duration and formant models.
Liquid Type and Vowel Liquid Type and Stress Liquid Type and Word Position Liquid Type and Previous Sound Liquid Type and Following Sound Liquid Type and Gender Liquid Type and Age Table 11. Interaction terms considered in the linear regression models
For each formant, interactions that were not selected as significant in any of the seven time-point models were left out of the seven models for that formant. If an interaction was significant at any time-point, it was included in all seven models. The results sections for F1 (3.4.2), F2 (3.4.3), F3 (3.4.4), and F4 (3.4.5) give details as to the significant and insignificant interaction terms that were included and excluded from the statistical models for formant analysis. Insignificant interactions were also removed from the final model for the duration analysis, as will be discussed in detail in section 3.4.2.1.
When an interaction term is selected as significant by the model, this means that at least one of the levels within that term has a difference between vowel+/r/ and vowel+/l/ that is significantly different from the difference between the two sounds in the overall model.
For example, if the interaction term of Liquid Type and Previous Sound is selected as significant, previous nasals may be selected as significant. If this is the case, this means
70 that the difference between vowel+/r/ and vowel+/l/ in the context of a preceding nasal is significantly different than the difference between the two liquids in the overall model.
This allows us to see if the difference between orthographic /r/ and /l/ vary according to certain factor levels of the independent variables considered in this study. This is important given that, even though the overall difference between /r/ and /l/ in a given model might not be significant, there often are significant differences in particular contexts.
In all statistical testing, it is important to make sure that the data does not violate the assumptions of the model. Regression models assume a normal distribution. The distribution of the data for the four formants for vowel+liquid sequences in the corpus was normal. However, the duration data has a rightward skew. Therefore, a log transformation was performed on the duration data in order to make the data conform to the normality assumption. See the results for duration (section 3.4.2.1) for more details on the log transformation.
3.2.5.2.2 Smoothing Splines
Smoothing Splines predict the shapes of curves and provide a visual tool for understanding dynamic movement, which is appropriate for studying the formant trajectories of the vowel+liquid sequences in this study. The Smoothing Spline ANOVAs
(henceforth SS ANOVAs) performed for this project were fitted using the ssanova function in the gss package in R (Gu 2014). The ssanova function takes data points along a curve as input and essentially connects the dots into a "best-fitting" spline. This fitting process uses a penalized likelihood method to identify the curve that best fits the data 71 points but minimally violates a lack of smoothness condition (Gu 2004).20 The use of smoothing splines in linguistic research is fairly recent. Davidson (2006) used splines to study how tongue-shape differs in the production of certain consonants in word-medial vs. word-final position using data points extracted from ultrasound imaging of sagittal views of the tongue. The curvatures of the tongue for minimal pairs such as 'jazz dancer' and 'NASDAQ' are then compared. While smoothing splines had been used in other fields that consider curves from other aspects of human behavior and in nature, Davidson's work served as an impetus for the incorporation of splines into phonetic research.
As discussed in Chapter 2, Simonet et al. (2008) utilized SS ANOVAs to compare the curves of the formant structures for vowel+/r/ and vowel+/l/ in their data from Puerto
Rican Spanish. This study builds upon their methodology of taking seven time-points from occurrences of vowel+approximant /r/ and /l/ and comparing the splines built from these time-points through SS ANOVAS. The obvious advantage of SS ANOVAs over regression models is the ability to visualize and compare the full trajectory of tongue movement over time. Since formants change continuously throughout speech, splines are a much better fit for this kind of data than are regression models that necessitate the consideration of measurements extracted from a single time point.
The ability to capture movement rather than static time points for things that move, like formants, is an exciting and welcome advancement for the field. However, SS
ANOVAs have a number of drawbacks over the traditional use of regression models in
20 The algorithm used to create the splines is stochastic in nature, meaning that the penalized likelihood function uses a random number generator. Thus, splines are actually slightly different every time they are run with the same data. These differences are extremely small and do not have consequences for the study at hand.
72 linguistics. Perhaps the most serious disadvantage of splines is that the models do not account for the uneven numbers of tokens in certain factor levels that inevitably arise from linguistic interviews due to the frequency of certain structures and sounds over others in a given language. For example, the corpus considered for this project features the vowel /a/ more frequently than other vowels, which likely reflects a higher frequency of this vowel before liquids in the Spanish language. SS ANOVAs are easily overwhelmed by the consideration of multiple factors, and the plot outputs become crowded if independent variables with several levels, like those for previous and following sound in this particular study, are included in the model. Also, to my knowledge, the random effect of speaker used to control for differences between individual speakers cannot be employed with splines.
Another important difference is that SS ANOVAs do not return p-values like regression models. It is therefore not possible to discuss output in terms of significance.
The closest measurement to satisfy the question of whether or not two curves are different from each other is the Bayesian 95% confidence interval, which calculates an area around the spline in which one can have a 95% level of certainty that the best-fitting curve is within the bounds of the intervals. In other words, each curve in the output receives a confidence interval above it and below it. If the lower boundary of one curve is above the upper boundary of another, one can be confident that the two curves are different from each other at that point in the trajectory (Wahba 1983).
Apart from confidence intervals, there are there are also a number of what Gu
(2002, 2004), a statistician who has written extensively about splines, calls "diagnostic heuristics" available with SS ANOVAs to test the relative importance of each factor in
73 the model. In this study, I consider the Kullback-Leibler (KL) projection to examine the components of each model. The KL projection compares a more complex model, i.e., a model with more independent variables, to a simpler model and gives a ratio. The higher the ratio, the more variation is explained by the more complex model. Gu (2004) explains that a ratio of 0.02 or higher suggests that the more complex model better explains the variation than the simpler model. This test is meant to help choose between models of the data, but is also a useful tool to determine how important individual variables are for the variation within a model. The splines shown in the figures for each of the formants (in sections 3.4.2.2.2, 3.4.2.3.2, 3.4.2.4.2, and 3.4.2.5.2) are built with the most complex model possible for two different model configurations, which are discussed below. The potential simplifications to the models suggested by low KL projections and the particular factors they reflect are pointed out and explored in the discussion for each formant and model configuration.
For each of the four formants, I consider two spline model configurations. The first model, shown in Table 12 below, is an extremely simplified look at the data; it includes the main effects of liquid type and time as well as the interaction between them.
This model outputs two splines, one for vowel+/r/ and another for vowel+/l/. The effect of time indicates whether vowel+liquid trajectories change over time as opposed to being a flat line. Liquid type tells us whether or not there is distance between vowel+/r/ and vowel+/l/. The interaction is perhaps the most interesting metric that splines give since it gives us a sense of how different the shape of the curves for /r/ and /l/ sequences are from each other.
74 Time Liquid Type Interaction - Time and Liquid Type Table 12. Factors in the Time and Liquid Type SS ANOVAs.
As will be seen in the results from the regression models, word position plays a very important role in determining whether or not liquid sounds are distinct or neutralized. Due to the importance of word position as a predictor of liquid production, I decided to explore the addition of word position to the time and liquid type spline models. For these models, I added in the effect for word position and all of the two-way interactions possible between the main effects of time, liquid type, and word position: time and liquid type, time and word position, and liquid type and word position. The word position models output four splines: word-medial vowel+/r/, word-medial vowel+/l/, word-final vowel+/r/, and word-final vowel+/l/. As shown in Table 13 below, the word position model adds two new interaction terms to the mix, apart from the type of liquid and time interaction from the simpler model - an interaction between time and word position and an interaction between liquid type and word position. The interaction for time and word position tells us whether or not the shape of the curve over time for a sequence at the end of a word is different from that of a sequence in the middle of a word.
The liquid type and word position interaction indicates whether vowel+/r/ and vowel+/l/ differ based on whether they are found word-medially or word-finally.
75 Time Liquid Type Word Position Interaction - Time and Liquid Type Interaction - Time and Word Position Interaction - Liquid Type and Word Position Table 13. Factors in the Time, Liquid Type, and Word Position SS ANOVAs.
The use of SS ANOVAs is intended to complement the results from the regression models and offer an exploratory view of the dynamical analysis of formants.
The models presented here are limited both by the reasons explained above and by the fact that these splines are fitted to the data of 24 speakers who do not always have the same curve shapes as each other due in part because of differences between them in their liquid production and in part due to uneven conditioning contexts. Ideally, separate SS
ANOVAs would be run for each speaker and each speaker would have the same conditioning factors as the others and the environment would be controlled. For example, each speaker, in an ideal context, would have the same number of vowel+liquid tokens after stops. In a natural speech context, it is possible, for example, for one speaker to have many preceding stops and not many preceding nasals and another speaker to have the opposite situation, which would be likely to influence the types of curves produced for individuals. Since SS ANOVAs do not take all of the possible conditioning factors and their uneven distribution into account, the regression analysis is a more reliable way to understand the data. The regression models at seven time points lend a more dynamic understanding of the liquid trajectories than previous studies which sample at one or two points in the liquid trajectory. The analysis with splines supplements the regression models and provides a way to consider entire curves instead of seven static points. The
76 use of SS ANOVAs is quickly becoming more popular in phonetic studies considering formant trajectories (Docherty et al. 2015, Nycz & De Decker 2006, Haddican et al.
2013, among others).
3.3 HYPOTHESES AND GOALS
The study presented in this chapter addresses the first three research questions introduced in chapter 1. Hypotheses and goals for each point of inquiry are reviewed below.
1) What are the different manners of articulation in the production of orthographic liquids in coda position in SJS? In SJS, do approximant realizations of vowel+/r/ differ from vowel+/l/ productions in terms of formant structure (F1, F2, F3, F4) and duration? If so, how are they different?
Based on the findings of previous studies discussed in section 2.4, I expect that the speakers in my corpus will produce /l/ with an approximant articulation but will have a variety of productions of /r/ in coda, such as taps, trills, approximants, fricatives, and deletions. This will be evaluated in section 3.4.1, where the analysis of all liquid realizations according to their manner of production will be presented. Next, I will examine the formant values and duration for vowel+orthographic /r/ sequences produced as approximants, as well as approximant vowel+orthographic /l/ sequences. Following the findings of previous studies (see Chapter 2), I expect rhotic sounds to have shorter durations than lateral sounds. As discussed in Chapter 2, F1 varies according to the laterality of the segment in question; in the case of incomplete neutralization, I expect to find slightly higher F1 values overall for vowel+orthographic /r/ sequences than for vowel+orthographic /l/ sequences. Since previous studies on F2 values for liquids 77 conclude that these values are affected more by the previous vowel than by rhoticity and laterality, I do not expect to find significant differences among F2 values for liquid segments. Past research has consistently found F3 to be an important indication of rhoticity; a lowering of the F3 trajectory during the vowel+liquid sequence is expected for vowel+orthographic /r/ sequences. Although F4 is rarely considered in studies on liquids, Luna's (2010) findings discussed in section 2.4.2.1 show convincing differences in F4 values. I expect to find similar results in that vowel+orthographic /r/ will have significantly lower F4 values than orthographic lateral sequences.
2) Are approximant realizations of vowel+/r/ and vowel+/l/ completely or incompletely neutralized?
The analysis of formant structure and duration discussed in the previous question will serve to distinguish /r/ and /l/ sequences from each other acoustically. If there are significant differences in duration or formant structure between sequences spelled with /r/ and sequences spelled with /l/, this will support the hypothesis that neutralization in this dialect is incomplete. If there are no significant differences, the null hypothesis that sequences with vowel+/r/ and vowel+/l/ are neutralized will be supported.
3) What social and linguistic factors influence the variation of the formant structures and durations of /r/ and /l/ segments?
Despite the differences in methodology discussed in Chapter 2, I do expect the findings from previous impressionistic studies to be mirrored to some extent by the acoustic analysis. Previous studies have found that men lateralize more than women, so I expect this finding to be consistent in my data. In terms of linguistic variables, the vowel that is included in the segment will be likely to greatly affect the realization of the liquid
78 segment due to potential coarticulation, as well as the overall acoustic trajectory of the sounds under study. Since Shouse de Vivas (1978) finds that the realization of /r/ by
Puerto Ricans in Springfield, Massachusetts, is affected by the manner of articulation of the following sound, it is possible that this will also be a conditioning factor in my data.
Her findings indicate that /r/ is least likely to lateralize before fricatives, slightly more likely before nasals, more likely before stops and very likely before laterals. Luna (2010) shows that resyllabification of coda liquids to the onset of a word that starts with a vowel does not block lateralization, despite the fact that orthographic onset rhotics do not lateralize. The larger sample size under investigation in the present study will help to better understand if and how resyllabification affects SJS liquid sounds. The word position of the coda liquids in SJS, that is, whether the codas are word-internal or word- final, has been found to influence liquid realization. Following the results of previous studies, I expect lateralization and thus more neutralized sequences to be more common in word-final position. Previous acoustic studies have not examined the effects of stress, but following from the results of impressionistic studies, I hypothesize that there will be more lateralization in tonic syllables than in unstressed environments.
3.4 RESULTS
This section details the results from the analyses of the production data. Section
3.4.1 reports descriptive and non-parametric statistics for all of the data in terms of the manner of articulation and the conditioning factors on the selection of manner of articulation for orthographic /r/. Section 3.4.2 presents the results of the analysis of the
79 duration and formant structure of approximant liquids through linear regression models and smoothing splines.
3.4.1 Liquid Manner of Articulation Analyses
This analysis considers all of the coda /r/ and /l/ tokens coded from the corpus in order to form a clear picture of coda liquid behavior in SJS. Section 3.4.1.1 discusses the distribution of the data in terms of percentages and token counts. Section 3.4.1.2 presents an analysis with logistic regression for approximant articulations (3.4.1.2.1) and inference trees for coda /r/ tokens produced as taps (3.4.1.2.2), deletions (3.4.1.2.3), and fricatives
(3.4.1.2.4).
3.4.1.1 Distribution of the Data
Table 14 below shows the percentage of each type of /r/ articulation found in the data. The approximant realization of /r/ is by far the most common rhotic manner of articulation. However, a substantial number of taps, deletions, and fricatives are also present in this dialect. Vocalizations and trills are very rare and will not be analyzed further in this study.
80 Realization of /r/ Percentage Occurrence Token Count
Approximant 75.5% 2,212
Tap 9.8% 288
Deletion 7.2% 210
Fricative 6.3% 185
Vocalization 0.009% 25
Trill 0.003% 10
Total 100% 2,930
Table 14. Manner of articulation for orthographic coda /r/ tokens in the corpus
Realizations of orthographic coda /l/ in the data were fairly uniform, with 96.6% approximant productions. Deletions and fricatives were very infrequent, but a few tokens of each were found in the corpus. Table 15 below shows the distribution of the data for
/l/. Due to the lack of variation of laterals articulation in this dialect, manners of articulation of /l/ will not be further explored in this analysis.
Realization of /l/ Percentage Occurrence Token Count
Approximant 96.6% 703
Deletion 2.1% 15
Fricative 1.5% 11
Total 100% 728
Table 15. Manner of articulation for orthographic coda /l/ tokens in the corpus
81 3.4.1.2 Rhotic Manners of Articulation
The following analyses explore each the four most common types of rhotic productions in the corpus - approximants, taps, deletions, and fricatives - individually.
Each analysis considers the conditioning of the manner of articulation by the linguistic and social independent variables discussed in section 3.2.4.
3.4.1.2.1 Approximant Productions of /r/
The data for this analysis combines the tap, deletion, and fricative tokens to one category of 683 tokens called 'other' and compares it to the 2,212 approximant tokens.
Rhotic production, either 'other' or 'approximant', is the dependent variable for this regression analysis. A stepwise procedure was used and an ANOVA was performed to determine which of the independent variables should be included in the model. The mixed effects model that best fits the data, shown in Table 16, includes the main effects of stress, vowel, preceding sound, and following sound as well as a random effect for speaker. The factor levels selected as significant are bolded. The levels with a positive value in the estimate column favor an approximant production of /r/, whereas the negative values prefer non-approximant (tap, deletion, fricative) articulations.
82
Estimate Standard Error zValue p-value (Intercept) 1.38608 0.18646 -7.434 1.06e-13 Stress stressed 0.34712 0.07832 -4.432 9.34e-06 unstressed -0.34712 0.07832 4.432 9.34e-06 Vowel /a/ 0.17574 0.13036 1.348 0.177596 glide+/a/ 0.27552 0.31251 0.882 0.377969 /e/ -0.05988 0.12824 -0.467 0.640547 glide+/e/ 1.16415 0.27311 4.263 2.02e-05 /i/ -0.91417 0.18945 -4.825 1.40e-06 /o/ 0.04925 0.14768 0.333 0.738775 /u/ -0.69069 0.30524 -2.263 0.023651 Preceding Sound stop -0.29463 0.15777 -1.867 0.061838 pause -0.72360 0.59332 -1.220 0.222623 approximant 0.14090 0.17349 0.812 0.416708 fricative -0.37423 0.16208 -2.309 0.020947 lateral 0.50441 0.32008 1.576 0.115048 nasal 0.80351 0.23613 3.403 0.000667 rhotic -0.05638 0.33964 -0.166 0.868163 Following Sound stop 0.41331 0.10757 3.842 0.000122 pause -0.18439 0.15566 -1.185 0.236206 approximant 1.27574 0.21131 6.037 1.57e-09 fricative -0.70318 0.13343 -5.270 1.36e-07 nasal 0.87749 0.17232 5.092 3.54e-07 vowel -1.67907 0.11441 -14.676 < 2e-16 Table 16. Logistic regression results for approximant vs. other (tap, deletion, fricative) pronunciations of /r/
As can be observed in Table 16 above, stressed syllables favor approximant realizations. In terms of the vowel before the rhotic sound, glide+/e/ favors approximant realizations while high vowels /i/ and /u/ disfavor approximant /r/. Preceding nasals favor approximant /r/ whereas preceding fricatives disfavor this articulation. Approximant
83 realizations of /r/ are disfavored when the following sound is a fricative or vowel and favored in all other contexts, except before pauses, which are not selected as significant.
3.4.1.2.2 Tap Productions of /r/
As discussed above in sections 3.4.1.2 and 3.2.5.1, the small token count (288 occurrences) for taps make regression analysis inappropriate. In order to better understand how tap articulations of /r/ are selected over other productions, the data was submitted to a conditional inference tree model. The biggest takeaway from the tree, shown in Figure 6, is that taps are most frequent when the following sound is a vowel.
Furthermore, it is /r/ in unstressed syllables followed by vowels that tend to be produced as taps, as can be seen by the far right-branching structure of the inference tree. In stressed syllables with a following vowel, taps are more common when preceded by stops, fricatives, and pauses than by other preceding sounds. There are very few tokens of taps in the data where the following sound is not a vowel, but of these tokens, women produce them more frequently than do men.
84
1 afterManner p < 0.001
S, A, F, N, P V
2 11 Gender Stress p = 0.013 p < 0.001
f m s u
3 12 ageTwoGroups beforeManner p = 0.008 p < 0.001
old young
4 beforeManner p = 0.016
85 S, A, N, R F, L S, F, P A, L, N, R
6 Stress p = 0.039
s u
Node 5 (n = 262) Node 7 (n = 59) Node 8 (n = 9) Node 9 (n = 738) Node 10 (n = 1137) Node 13 (n = 178) Node 14 (n = 163) Node 15 (n = 94)
r r r r r r r
r 1 1 1 1 1 1 1 1
e e e e e e e e
h h h h h h h h
t t t t t t t t
o o o o o o o o 0.8 0.8 0.8 0.8 0.8 0.8 0.8 0.8
0.6 0.6 0.6 0.6 0.6 0.6 0.6 0.6
0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.4
0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2
p p p p p p p p
a a a a a a a a
t t t t t t t t 0 0 0 0 0 0 0 0
Figure 6. Conditional inference tree for taps. 'afterManner' corresponds with the independent variable Following Sound. 'beforeManner' represents the Preceding Sound. Both of these categories have abbreviations for their factor levels: S (stop), A (approximant), F (fricative), N (nasal), P (pause), R (rhotic), and L (lateral). The factor levels for Stress are also abbreviated as 's' for stressed syllables and 'u' for unstressed syllables. For the Gender variable, 'f' represents females and 'm' males.
85 3.4.1.2.3 Deleted Productions of /r/
Like the analysis above for tap productions in section 3.4.1.2.2, deletions will also be examined using a conditional inference tree due to small token counts (210 occurrences of deletions). The most important variable for deletions is the vowel preceding the rhotic. As can be seen in the right-branching structure in Figure 7, the vowels /i/ and /o/ have the highest number of deletions. Upon a closer examination of the tokens with these vowels, it becomes apparent that deletions are common at the end of infinitives with /ir/ and in the word porque 'because.' In the context of the vowels /i/ and
/o/, men and women have similar rates of deletion. In other contexts, men delete more frequently than women. Since the conditional inference trees do not have random effects and individual speakers cannot be accounted for, the observation of a gender difference cannot be substantiated since this effect could be due to particular speakers in the sample.
This finding, therefore, merits further exploration in a future study.
86
1 VowelNew p < 0.001
a, da, de, e, u i, o 2 11 afterManner afterManner p < 0.001 p < 0.001
S, A, P F, N, V S, P A, F, N, V 3 8 12 19 beforeManner Gender Gender VowelNew p < 0.001 p = 0.005 p = 0.012 p = 0.003
S, A, F, L, N P, R f m 4 14 wordPosition Stress p = 0.005 p = 0.048
f m s u i o 16 wordPosition fin mid p = 0.031
87 fin mid
Node 5 (n = 405) Node 6 (n = 537) Node 7 (n = 25) Node 9 (n = 372) Node 10 (n = 394) Node 13 (n = 272) Node 15 (n = 101) Node 17 (n = 19) Node 18 (n = 175) Node 20 (n = 90) Node 21 (n = 250)
r r r r r r r r r r
r 1 1 1 1 1 1 1 1 1 1 1
e e e e e e e e e e e
h h h h h h h h h h h
t t t t t t t t t t t
o o o o o o o o o o o 0.8 0.8 0.8 0.8 0.8 0.8 0.8 0.8 0.8 0.8 0.8
0.6 0.6 0.6 0.6 0.6 0.6 0.6 0.6 0.6 0.6 0.6
0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.4
e e e e e e e e e e e
t t t t t t t t t t
t 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2
e e e e e e e e e e e
l l l l l l l l l l l
e e e e e e e e e e e
d d d d d d d d d d d 0 0 0 0 0 0 0 0 0 0 0
Figure 7. Conditional inference tree for deletions. 'afterManner' corresponds with the independent variable Following Sound. 'beforeManner' represents the Preceding Sound. Both of these categories have abbreviations for their factor levels: S (stop), A (approximant), F (fricative), N (nasal), P (pause), R (rhotic), and L (lateral). The factor levels for the variable for Stress are also abbreviated as 's' for stressed syllables and 'u' for unstressed syllables. For the Gender variable, 'f' represents females and 'm' males.
87 3.4.1.2.4 Fricative Productions of /r/
The conditional inference tree for fricative productions in Figure 8 shows that the production of /r/ as a fricative happens most frequently when the following sound is a fricative. This is particularly common in the word persona 'person' (as well as personas
'people' and personal 'personal'), which explains the large percentage of occurrences of fricative /r/ in Node 12. There are a large number of fricative productions before pauses, as shown in the right-branching Node 2. Women have a higher percentage of fricative /r/ tokens than do men in this pre-pausal position. As discussed in section 3.4.1.2.3 above, the lack of a random effect in these models makes the social variables particularly unreliable since they could pinpoint specific individuals. In fact, a close look at the data reveals that the effect for gender comes from three particular women in the data, while other women in the corpus pronounce /r/ with other articulations. There are only a few tokens of men using fricative /r/ before pauses. It is difficult, therefore to draw any definitive conclusions about gender. There is a possibility that pre-pausal fricatives are a gendered strategy since men rarely have these pronunciations. However, plenty of women also do not produce /r/ as a fricative before pauses.
88 1 afterManner p < 0.001
S, A, N, P, V F
2 9 afterManner beforeManner p < 0.001 p < 0.001
S, A, N, V P S A, F, L, N, P, R
4 10 13 Gender wordPosition VowelNew p < 0.001 p = 0.012 p = 0.009
f m
5 ageTwoGroups fin mid a, de, e, o da, i, u p < 0.001
89 old young
Node 3 (n = 2060) Node 6 (n = 36) Node 7 (n = 77) Node 8 (n = 123) Node 11 (n = 20) Node 12 (n = 167) Node 14 (n = 134) Node 15 (n = 23)
r r r r r r r
r 1 1 1 1 1 1 1 1
e e e e e e e e
h h h h h h h h
t t t t t t t t
o o o o o o o o 0.8 0.8 0.8 0.8 0.8 0.8 0.8 0.8
0.6 0.6 0.6 0.6 0.6 0.6 0.6 0.6
0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.4
0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2
c c c c c c c c
i i i i i i i i
r r r r r r r r
f f f f f f f f 0 0 0 0 0 0 0 0
Figure 8. Conditional inference tree for fricatives. 'afterManner' is the independent variable for the Following Sound. 'beforeManner' represents the Preceding Sound. Both of these categories have abbreviations for their factor levels: S (stop), A (approximant), F (fricative), N (nasal), P (pause), R (rhotic), and L (lateral). The factor levels for the variable for Stress are also abbreviated as 's' for stressed syllables and 'u' for unstressed syllables. For the Gender variable, 'f' represents females and 'm' males.
89 3.4.2 Analysis of Acoustic Characteristics of Approximant Liquids
The following analyses compare the acoustic properties of approximant vowel+/r/ tokens from section 3.4.1 to the approximant vowel+/l/ tokens from the corpus. Since the possibility for neutralization with approximant /l/ sounds lies in the approximant production of /r/, these analyses were performed in order to ascertain whether or not these liquid sounds are neutralized in SJS. Section 3.4.2.1 presents results for the differences in duration between vowel+/r/ and vowel+/l/ sequences. The following sections are devoted to formant analyses for F1 (3.4.2.2), F2 (3.4.2.3), F3 (3.4.2.4), and F4 (3.4.2.5). Each of the formant sections are divided into two parts: 1) the results from the seven linear regression time-point models, 2) the results from the smoothing spline analyses. A comparison of the results from the regression and spline models will be presented in chapter 5.
3.4.2.1 Duration
Overall, the average duration for vowel+/r/ is 132.158 ms and the average duration for vowel+/l/ is 135.639 ms. This difference of 2.57% is not significant.
However, differences in duration are significant in certain contexts. The results of the mixed-effects model shows that there are significant interactions between liquid type and with certain vowels, with preceding nasals, and with following approximants. Table 17 below gives the lengths of vowel+/r/ and vowel+/l/ in the environments where the difference between vowel+/r/ and vowel+/l/ is significantly different from the overall difference of 2.57%. The overall values are also provided in the first row for reference.
90 As discussed in section 3.2.5.2, the regression model for duration, unlike those for formant values, was run using a log transform due to the skewedness of the data.
Therefore, differences between vowel+/r/ and vowel+/l/ in the third column are reported as percentages rather than additive differences in milliseconds. The fourth column reports the direction of the difference between vowel+/r/ and vowel+/l/ sequences. In other words, the direction column shows whether vowel+/r/ is longer (>) or shorter (<) than vowel+/l/.
/r/ /l/ Difference Direction Overall 132.158 ms 135.639 ms 2.57% /r/ < /l/ (p > .05) preceding nasal 140.471 ms 128.895 ms 11.85% /r/ > /l/ following 138.934 ms 128.509 ms 10.96% /r/ > /l/ approximant following vowel 112.168 ms 129.024 ms 10.77% /r/ < /l/ glide+/a/ in 166.501 ms 141.034 ms 21.17% /r/ > /l/ sequence glide+/e/ in 118.475 ms 157.701 ms 22.89% /r/ < /l/ sequence /e/ in sequence 126.091 ms 143.309 ms 9.70% /r/ < /l/ /i/ in sequence 130.582 ms 104.376 ms 28.40% /r/ > /l/ Table 17. Duration values for overall data set and for significant interactions
It is interesting to note that the different vowels in the sequence might have opposite pulls on the direction of the difference between vowel+/r/ and vowel+/l/.
Sequences with the vowel /e/ or glide+/e/ go along with the overall trend of vowel+/r/ being shorter than vowel+/l/. However, sequences with glide+/a/ and the simple vowel /i/ have the opposite effect. That is, vowel+/r/ is longer than vowel+/l/ with these vowels.
91 3.4.2.2 First Formant (F1)
The following analyses considers F1 as the dependent variable in order to determine whether or not vowel+/r/ and vowel+/l/ are different from each other in terms of the first formant, which inversely corresponds to the height of the tongue. That is, a lower F1 value corresponds to a higher placement of the tongue in the mouth whereas a higher F1 indicates a lower tongue position. The results from the linear regression models for the seven time points are presented in section 3.4.2.2.1 and the exploratory analysis of
F1 using smoothing splines are discussed in section 3.4.2.2.2.
3.4.2.2.1 Linear Regression Models for F1 at Seven Time Points
This analysis was performed using seven mixed effects linear regression models.
The time points 1 through 7 represent measurements of F1 taken at 25%, 37.5%, 50%,
75%, 87.5%, and 100% of the duration of the vowel+liquid sequences. Each of the seven regression models takes the values of F1 at each respective time point as the dependent variable and weighs the correlation of the independent variables (see section 3.2.4 for details) and the interaction terms between liquid type and the other independent variables with the values for F1. The interaction terms for liquid type with gender, age, and stress were not selected as significant by any of the seven time point models for F1 and were therefore left out of the final models for this formant. The models reported here, therefore, have all of the independent variables discussed in section 3.2.4 as well as the interaction terms between liquid type and word position, preceding sound, following
92 sound, and vowel.21 Table 18 shows the overall predicted values for F1 in vowel+/r/ and vowel+/l/ and whether the difference was significant. These overall predicted values reflect all of the data taken together. As discussed above, each time point represents a separate mixed effects regression model.
Time Points vowel+/r/ vowel+/l/ Significant Difference? 1 518.67 523.49 no 2 513.92 518.18 no 3 503.89 493.99 no 4 481.39 471.15 no 5 452.52 440.7 no 6 442.28 427.16 no 7 445.05 426.23 no Table 18. Overall F1 predicted values of vowel+liquid sequences at each time point
Figure 9 below is a plot of the values from Table 18. Visualizing the values of vowel+/r/ and vowel+/l/ over the duration of the vowel+liquid sequences allows us to see that F1 values for sequences with /r/ are slightly lower than those for /l/ at the first two time points, but are higher from time point 3 through 7.
21 The main effects were left in the model to account for the differences in the formant values for all vowel+liquid sequences. These effects tended to be significant, so all of them were left in the models to make the models easier to compare. 93 520 r l 500
480
z H 460 440 420
1 2 3 4 5 6 7
Time Points
Figure 9. Overall F1 predicted values of vowel+liquid sequences across seven time points
As reported in Table 18, the overall differences between vowel+/r/ and vowel+/l/ are not significant. However, significant differences arise in the interactions between liquid type and word position, previous sound, following sound, and vowel. The specific contexts that were selected as significant are reported in Table 19 below. Each row reports findings for a model at a specific time point, which is indicated in the first column. The second column summarizes whether or not the overall predictions from
Table 18 above are significant. Note that /r/ and /l/ are shorthand for vowel+/r/ and vowel+/l/ sequences. The greater than (>) and less than (<) signs show the direction of the difference in F1. For example, /r/ < /l/ means that the value for F1 for vowel+/r/ is lower than the value for vowel+/l/. The third column lists the levels of interaction terms that were selected as significant by the model for the time point in each row. The fourth column displays three characteristics about the interaction context. First, the relationship between the predicted values for that context are compared to the overall values for that
94 time point. For example, for the first time point in Table 19, '/r/, /l/ > overall' means that the values for vowel+/r/ and vowel+/l/ are both higher when preceded by a nasal than the predicted values for vowel+/r/ and vowel+/l/ in the overall data set. After the comparison with the overall values, the relationship between vowel+/r/ and vowel+/l/ in that particular environment is established. Thus, in the first time point after the overall comparison for preceding nasals, '/r/ > /l/' means that in the context of preceding nasals at this time point, the F1 value for vowel+/r/ is higher than that of vowel+/l/. Finally, the difference between vowel+/r/ and vowel+/l/ in the particular context is compared to the overall difference. For example, for time point 1, 'Difference > Overall' in the row for preceding nasals means that the difference between vowel+/r/ and vowel+/l/ in the environment of a preceding nasal is greater than the difference between vowel+/r/ and vowel+/l/ for the overall data set. The factor levels where the difference is greater than the overall (Difference > Overall) are those that are of interest for this formant: since the overall values for /r/ and /l/ are not significantly different at any of the time points for F1, if the difference between the sounds in particular contexts is significantly different than the overall, it can be concluded that /r/ and /l/ are not neutralized in those contexts. These contexts are bolded in Table 19. In the contexts where the difference is significantly less than the overall difference (Difference < Overall), /r/ and /l/ are even more similar than in the overall comparison. For example, at time point 4 for F1, the difference between /l/ and /r/ in the context of the vowel glide+/e/ is even more similar than the overall values for the liquid sounds. The tables for F2, F3, and F4 will follow the same conventions and thus will not be explained in the subsequent sections.
95 Time Overall Significant Interaction Effects Point difference interactions significant? 1 no Preceding nasal: /r/, /l/ > overall; /r/ > /l/; Difference > nasal Overall. /r/ < /l/ Vowels glide +/a/: /r/, /l/ > overall; /r/ < /l/; Difference > glide+/a/, /e/ Overall. /e/: /r/ > overall; /l/ < overall; /r/ > /l/; Difference > Overall.
Following vowel: /r/, /l/ > overall; /r/ < /l/; Difference > vowel Overall. 2 no Preceding nasal: /r/, /l/ > overall; /r/ > /l/; Difference > nasal Overall. /r/ < /l/ Vowel /e/ /e/: /r/ > overall; /l/ < overall; /r/ > /l/; Difference > Overall.
Following approximant: /r/ < overall; /l/ > overall; /r/ < /l/. approximant Difference > Overall. 3 no Following approximant: /r/ < overall; /l/ > overall; /r/ < /l/; approximant, Difference > Overall. /r/ > /l/ fricative fricative: /r/ > overall; /l/ < overall; /r/ > /l/; Difference > Overall.
Vowels /e/, /e/: /r/, /l/ < overall; /r/ > /l/; Difference > Overall. /o/, /u/ /o/: /r/ > overall; /l/ < overall; /r/ > /l/; Difference > Overall. /u/: /r/, /l/ < overall; /r/ < /l/; Difference > Overall.
Word medial: /r/ > overall; /l/ < overall; /r/ > /l/; Position Difference > Overall. final: /r/ < overall; /l/ > overall; /r/ < /l/; Difference < Overall. Continued
Table 19. Values for F1 for vowel+/r/ and vowel+/l/ overall and in particular contexts
96 Table 19: Continued
Time Overall Significant Interaction Effects Point difference interactions significant? 4 no Following pause: /r/, /l/ < overall; /r/ > /l/; Difference > pause, Overall. /r/ > /l/ approximant approximant: /r/ < overall; /l/ > overall; /r/ < /l/; Difference > Overall.
Vowel glide +/e/: /r/ < overall; /l/ > overall; /r/ < /l/; glide+/e/ Difference < Overall.
Word medial: /r/ > overall; /l/ < overall; /r/ > /l/; Position Difference > Overall. final: /r/ < overall; /l/ > overall; /r/ < /l/; Difference < Overall. 5 no Following pause: /r/, /l/ < overall; /r/ > /l/; Difference > pause, Overall. /r/ > /l/ approximant approximant: /r/, /l/ < overall; /r/ < /l/; Difference > Overall.
Vowel /a/ /a/: /r/, /l/ > overall; /r/ > /l/; Difference > Overall.
Word medial: /r/ > overall; /l/ < overall; /r/ > /l/; Position Difference > Overall. final: /r/ < overall; /l/ > overall; /r/ > /l/; Difference < Overall. 6 no Following approximant: /r/, /l/ < overall; /r/ < /l/; Difference < approximant Overall. /r/ > /l/ Vowel /a/ /a/: /r/, /l/ > overall; /r/ > /l/; Difference > Overall.
Word medial: /r/ > overall; /l/ < overall; /r/ > /l/; Position Difference > Overall. final: /r/ < overall; /l/ > overall; /r/ > /l/; Difference < Overall. 7 no Word medial: /r/ > overall; /l/ < overall; /r/ > /l/; Position Difference > Overall. /r/ > /l/ final: /r/ < overall; /l/ > overall; /r/ > /l/; Difference < Overall.
97 3.4.2.2.2 Smoothing Splines Analysis for F1
This section shows the F1 results of the Smoothing Spline ANOVAs, which compare the entire trajectories of vowel+approximant /r/ and vowel+approximant /l/. See section 3.2.5.2.2 for details on the statistical modeling for SS ANOVAs.
Figure 10 below shows the resulting SS ANOVA for the first formant across time. The
model considers time, liquid type, and the interaction between the two.
0
0
6
0
5
5
0
z
0
H
5
0
5 4
r
0 l
0 4
1 2 3 4 5 6 7
Time Points
Figure 10. Smoothing spline ANOVA of F1 values over time for vowel+/r/ and vowel+/l/
The Bayesian 95% confidence intervals for the splines in Figure 10 overlap in the middle of the trajectories around time points 3 and 4, but are otherwise distinct. In order
98 to better understand the model and the results, a Kullback-Leibler (KL) projection was conducted. See the output from the projection in Table 20 below.
Model 1 Time, Liquid Type, Interaction between Time and Liquid Type Model 2 Time and Liquid Type KL Ratio for Model 1 vs. Model 2 0.02753673 Table 20. KL Projection Model Comparison for F1 Time and Liquid Type.
The KL ratio in Table 20 is above 0.02, which means that the extra term in Model
1 is useful in accounting for the variation in the data. The interaction between time and liquid, which is the extra term in Model 1, compares the shape of vowel+/r/ across time to that of vowel+/l/ across time. In other words, the interaction term considers whether or not the shape of the /r/ curve is different from the /l/ curve. The SS ANOVA model, therefore, helps us see that vowel+/r/ and vowel+/l/ are different in terms of the overall shape of their trajectories.
Splines for a slightly more complex model that adds word position to the time and liquid type model are presented in Figure 11 below. Since the model has four splines that have a lot of overlap, the word-final and word-medial contexts are displayed individually in Figures 12 and 13 for the sake of visual clarity.
99 final /r/ final /l/ middle /r/
middle /l/
0
0
6
0
5
5
z
H
0
0
5
0
5 4
1 2 3 4 5 6 7
Time Points
Figure 11. Smoothing spline ANOVA of F1 values with word position for vowel+/r/ and vowel+/l/
final /r/
final /l/
0
0
6
0
5
5
z
H
0
0
5
0
5 4
1 2 3 4 5 6 7
Time Points
Figure 12. Smoothing spline ANOVA of F1 values for word-final liquids
100 middle /r/
middle /l/
0
0
6
0
5
5
0
0
5
0
5 4
1 2 3 4 5 6 7
Figure 13. Smoothing spline ANOVA of F1 values for word-medial liquids
These figures show that liquids behave differently based on word position. Word- final vowel+/r/ and vowel+/l/ are different from each other towards the end of the vowel+liquid sequence, whereas word-medial segments differ the most in the initial part of the sequence, overlap in the middle, and become distinct again at the end. It is also interesting to note that vowel+/l/ sequences are fairly uniform despite word position, whereas vowel+/r/ sequences are different from each other in different word positions.
Figures 14 and 15 below separate out liquid type to help to see this more clearly.
101 final /l/
middle /l/
0
0
6
0
5
5
z
H
0
0
5
0
5 4
1 2 3 4 5 6 7
Time Points
Figure 14. Smoothing spline of F1 values for vowel+/l/ sequences in both word positions
final /r/
middle /r/
0
0
6
0
5
5
z
H
0
0
5
0
5 4
1 2 3 4 5 6 7
Time Points
Figure 15. Smoothing spline of F1 values for vowel+/r/ sequences in both word positions
In order to test the contribution of main effects (time, liquid type, and word position) and the interaction terms to explaining the variation in the data, various KL projections that leave out different terms from the model were compared to the SS
102 ANOVA which considers all three main effects and all of the two-way interactions (see section 3.2.5.2.2 for further explanation). Model 1 in Table 21 below represents this most complex model, whereas Models 2.1 through 2.6 represent various less complicated models. The KL ratio for Model 1 compared to each respective Model 2 is shown below the presentation of each Model 2.
103 Model 1 Time Liquid Type Word Position Interaction - Time and Liquid Type Interaction - Liquid Type and Word Position Interaction - Time and Word Position Model 2.1 - Just main effects Time Liquid Type Word Position KL Ratio for Model 1 vs. Model 2.1 0.03675748 Model 2.2 - Without Liquid Type and Time Word Position Interaction Liquid Type Word Position Interaction - Time and Liquid Type Interaction - Time and Word Position KL Ratio for Model 1 vs. Model 2.2 0.007745009 Model 2.3 - Without Time and Liquid Time Type Interaction Liquid Type Word Position Interaction - Liquid Type and Word Position Interaction - Time and Word Position KL Ratio for Model 1 vs. Model 2.3 0.02318353 Model 2.4 - Without Time and Word Time Position Interaction Liquid Type Word Position Interaction - Time and Liquid Type Interaction - Liquid Type and Word Position KL Ratio for Model 1 vs. Model 2.4 0.003004173 Model 2.5 - Without Liquid Type or Time its Interactions Word Position Interaction - Time and Word Position KL Ratio for Model 1 vs. Model 2.5 0.03575966 Model 2.6 - Without Word Position or Time its Interactions Liquid Type Interaction - Time and Liquid Type KL Ratio for Model 1 vs. Model 2.6 0.05177153 Model 2.7 - Without Time or its Liquid Type Interactions Word Position Interaction - Liquid Type and Word Position KL Ratio for Model 1 vs. Model 2.7 0.9499964 Table 21. KL Projection model comparisons for F1 Time, Liquid Type, and Word Position and all of the two-way interactions
104 The KL value above 0.02 in Model 2.1 shows that including the interaction terms in useful in explaining the variation of vowel+/r/ and vowel+/l/ values for F1. None of the models which exclude time and liquid type have low KL values, which shows that the shapes of the curves for the two liquid sounds are different, even in the context of these models with word position effects. Two of the models, 2.2 and 2.4 are below the KL threshold of 0.02. These lower numbers mean that the interaction terms left out of the models - time with word position and word position and liquid type - have less explanatory power than the other terms.
3.4.2.3 The Second Formant (F2)
The following analyses considers F2 as the dependent variable in order to determine whether or not vowel+/r/ and vowel+/l/ are different from each other in terms of the second formant, which corresponds to the frontness/backness of the tongue. That is, a higher F2 value corresponds to a more front placement of the tongue in the mouth, whereas a lower F2 corresponds to a more back position of the tongue. The results from the linear regression models for the seven time points are presented in section 3.4.2.3.1 and the exploratory analysis of F2 using smoothing splines are in section 3.4.2.3.2.
3.4.2.3.1 Linear Regression Models for F2 at Seven Time Points
Like the previous analysis for F1, this analysis was carried out using seven mixed effects linear regression models. The time points 1 through 7 represent measurements of
F2 taken at 25%, 37.5%, 50%, 75%, 87.5%, and 100% of the duration of the vowel+liquid sequences. The interaction terms for liquid type with following sound, 105 stress, gender, and age were not selected as significant by any of the seven time point models for F2 and were left out of the final models for this formant. The models reported here, therefore, have all of the independent variables discussed in section 3.2.6.1 as well as the interaction terms between liquid type and word position, preceding sound, and vowel. Table 22 shows the overall predicted values for vowel+/r/ and vowel+/l/ and whether the difference was significant. These overall predicted values reflect all of the data taken together. As discussed above, each time point represents a separate mixed effects regression model.
Time Points vowel+/r/ vowel+/l/ Significant Difference? 1 1608.19 1461.07 yes 2 1613.19 1522.49 yes 3 1620.84 1557.3 no 4 1642.29 1590.63 no 5 1623.4 1567.76 no 6 1617.9 1557.3 no 7 1627.45 1627.51 no Table 22. Overall F2 predicted values of vowel+liquid sequences at each time point
Figure 16 below is a plot of the values from Table 22. Visualizing the values of vowel+/r/ and vowel+/l/ over the duration of the vowel+liquid sequences allows us to see that F2 values for sequences with /r/ are consistently higher than those for /l/ until the last time point where the two converge. The difference between vowel+/r/ and vowel+/l/ is selected as significant for the first two time points, but is not significant throughout the rest of the trajectory. However, significant differences arise in the interactions between liquid type and word position, previous sound, and vowel at all of the time points. The
106 specific contexts that were selected as significant are reported in Table 23 below. See the explanation in section 3.4.2.2 on how to read the table. F2, unlike F1, has time points that have a significant overall difference between /r/ and /l/. When an overall difference is significant at a time point, the interpretation for significant interaction terms is slightly different. When the difference for an interaction term is significantly larger than the overall distance between vowel+/r/ and vowel+/l/ (Difference > Overall), this indicates that although the average distance between the sounds is significant, the liquids in the context of the particular interaction are even more different than in the overall measurement. In the case of interactions that are significantly lower than the overall significant difference (Difference < Overall), these particular contexts are neutralizing.
For example, at time point 1 for F2, the overall difference between vowel+/r/ and vowel+/l/ is significant, but there are exceptions in particular contexts. For example, vowel+/r/ and vowel+/l/ are neutralized in word-final position.22
22 At a time point that has a significant overall difference between vowel+/r/ and vowel+/l/, it is possible, although improbable, to have interaction terms with significantly lower differences between the liquid sounds than the overall and still maintain a significant difference between the sounds. In order to test for this, the binary independent variables were submitted to t-tests to confirm that the cases of Difference < Overall were in fact instances of an insignificant difference between vowel+/r/ and vowel+/l/. I did not perform tests for the instances of lower than overall differences for predictors with more than two factor levels. Although it is fairly safe to assume that these contexts are neutralizing, setting up a way to test these cases would be an improvement upon the current methodology. 107 1640 1620 1600 1580
1560 z
H 1540 1520 1500 1480 r 1460 l 1440
1 2 3 4 5 6 7
Time Points
Figure 16. Overall F2 predicted values of vowel+liquid sequences across seven time points
Time Overall Significant Interaction Effects Point difference interactions significant? 1 yes Word Position medial: /r/, /l/< overall; /r/ > /l/; Difference > Overall. /r/ > /l/ final: /r/, /l/ > overall; /r/ > /l/; Difference < Overall.
Preceding stop: /r/, /l/ < overall; /r/ > /l/; Difference < stop, fricative Overall. fricative: /r/ < overall; /l/ > overall; /r/ > /l/; Difference < Overall.
Vowels glide +/a/: /r/ > overall; /l/ < overall; /r/ > /l/; glide+/a/, /e/, Difference > Overall. /o/ /e/: /r/, /l/ > overall; /r/ > /l/; Difference < Overall. /o/: /r/, /l/ < overall; /r/ < /l/; Difference < Overall. Continued
Table 23. Values for F2 for vowel+/r/ and vowel+/l/ overall and in particular contexts
108 Table 23: Continued
Time Overall Significant Interaction Effects Point difference interactions significant? 2 yes Word Position medial: /r/ > overall; /l/ < overall; /r/ > /l/; Difference > Overall. /r/ > /l/ final: /r/ < overall; /l/ > overall; /r/ > /l/; Difference < Overall.
Preceding stop: /r/ < overall; /l/ > overall; /r/ < /l/; Difference stop, fricative, < Overall. nasal fricative: /r/ < overall; /l/ > overall; /r/ < /l/; Difference < Overall. nasal: /r/ < overall; /l/ > overall; /r/ < /l/; Difference < Overall.
Vowel /o/ /o/: /r/, /l/ < overall; /r/ < /l/; Difference < Overall. 3 no Word Position medial: /r/ > overall; /l/ < overall; /r/ > /l/; Difference > Overall. /r/ > /l/ final: /r/ < overall; /l/ > overall; /r/ > /l/; Difference < Overall.
Preceding stop: /r/ < overall; /l/ > overall; /r/ < /l/; Difference stop, fricative, < Overall. lateral fricative: /r/ < overall; /l/ > overall; /r/ overall; /l/ < overall; /r/ > /l/; Difference > Overall.
Vowels /i/, /o/ /i/: /r/, /l/ > overall; /r/ > /l/; Difference > Overall. /o/: /r/, /l/ < overall; /r/ < /l/; Difference < Overall.
4 no Preceding stop: /r/ < overall; /l/ > overall; /r/ < /l/; stop, fricative Difference > Overall. /r/ > /l/ fricative: /r/ < overall; /l/ > overall; /r/ < /l/; Difference < Overall.
Vowels /i/, /o/ /i/: /r/, /l/ > overall; /r/ > /l/; Difference > Overall. /o/: /r/, /l/ < overall; /r/ < /l/; Difference > Overall. Continued
109 Table 23: Continued
Time Overall Significant Interaction Effects Point difference interactions significant? 5 no Preceding fricative: /r/ < overall; /l/ > overall; /r/ < /l/; fricative, Difference > Overall. /r/ > /l/ nasal, lateral nasal: /r/ < overall; /l/ > overall; /r/ < /l/; Difference > Overall. lateral: /r/ > overall; /l/ < overall; /r/ > /l/; Difference > Overall.
Vowels /i/, /u/ /i/: /r/, /l/ > overall; /r/ > /l/; Difference > Overall. /u/: /r/ < overall; /l/ > overall; /r/ < /l/; Difference > Overall. 6 no Preceding stop: /r/ < overall; /l/ > overall; /r/ < /l/; stop, fricative, Difference > Overall. /r/ > /l/ lateral, nasal fricative: /r/ < overall; /l/ > overall; /r/ < /l/; Difference > Overall. lateral: /r/ > overall; /l/ < overall; /r/ > /l/; Difference > Overall. nasal: /r/ < overall; /l/ > overall; /r/ < /l/; Difference > Overall.
Vowels /i/, /u/ /i/: /r/, /l/ > overall; /r/ > /l/; Difference > Overall. /u/: /r/, /l/ < overall; /r/ < /l/; Difference > Overall. 7 no Word Position medial: /r/ > overall; /l/ < overall; /r/ > /l/; Difference > Overall. /r/ < /l/ final: /r/ < overall; /l/ > overall; /r/ < /l/; Difference > Overall.
Preceding pause: /r/, /l/ > overall; /r/ > /l/; Difference > pause, Overall. approximant, approximant: /r/, /l/ < overall; /r/ > /l/; nasal Difference > Overall. nasal: /r/ < overall; /l/ > overall; /r/ < /l/; Difference > Overall.
Vowels /o/, /o/: /r/, /l/ < overall; /r/ < /l/; Difference > /u/ Overall. /u/:/r/, /l/ < overall; /r/ < /l/; Difference > Overall.
110 3.4.2.3.2 Smoothing Splines Analysis for F2
This section shows the F2 results of the Smoothing Spline ANOVAs, which compare the entire trajectories of vowel+approximant /r/ and vowel+approximant /l/. See section 3.2.5.2.2 for details on the statistical modeling for SS ANOVAs.
The splines for vowel+/r/ and vowel+/l/ overall are shown below in Figure 17.
The two liquid sounds remain distinct until the very end of the trajectory, where the
confidence intervals cross.
0
5
6
1
0
0
6
1
0
z
5
5
H
1
0
0
5 1
r 0
5 l
4 1 1 2 3 4 5 6 7
Time Points
Figure 17. Smoothing Spline of F2 values over time for vowel+/r/ and vowel+/l/
111 Visually, vowel+/r/ and vowel+/l/ look very similar in terms of curvature in that the trajectory ascends over the first four time points and then remains level to the end of the sequences. It seems then, that it is the distance between the two curves that distinguishes /r/ from /l/ sequences - vowel+/l/ tends to have a higher F2 value than vowel+/r/. The small KL ratio (<0.02) for the model without the interaction term in Table
24 below shows that the simpler model more effectively accounts for the variation. This supports the assertion that the vowel+/r/ and vowel+/l/ sequences have similar F2 trajectories in terms of shape, and that it is the distance between the curves that distinguishes the liquid sounds.
Model 1 Time, Liquid Type, Interaction between Time and Liquid Type Model 2 Time and Liquid Type KL Ratio for Model 1 vs. Model 2 0.002807813 Table 24. KL Projection model comparison for F2 Time and Liquid Type
It is important to note that the spline model in Figure 17 above shows the reverse pattern of the overall regressions models for F2 (see section 3.4.2.3.1) in that the F2 for vowel+/l/ is higher than that for vowel+/r/ in the SS ANOVA, while the opposite patterns emerged from the regression results. A fine-grained analysis of the data reveals that this effect is due to the conditioning by previous sound. The results from the regression analyses in Table 23 show that preceding stops, fricatives, and nasals reverse the overall trend for the time points. In other words, the results for liquids preceded by these sounds is exactly what we see in the overall model for splines in which F2 values for vowel+/r/ are lower than those for vowel+/l/. Preceding stops, fricatives, and nasals together make
112 up for 76.3% of the vowel+/r/ and vowel+/l/ data. Since the spline model does not normalize the data over all of the possible contexts like the regression model does, the more frequent occurrence of these preceding sounds allows the direction of effect of these sounds to take precedence in the model.
The addition of word position into the spline model shows that final liquids tend to have higher and flatter trajectories than word-medial liquids, as can be seen in Figure
18. Vowel+/r/ and vowel+/l/ sequences are distinct in terms of F2 values for both word- medial and word-final contexts. The direction of effect in word-final context mirrors that of the overall model in that vowel+/r/ is lower than vowel+/l/. The word-medial context is similar to the regression models in that vowel+/r/ is higher than vowel+/l/.23
final /r/ final /l/ middle /r/
middle /l/
0
0
7
1
0
0
6
1
0
z
0
5
H
1
0
0
4
1
0
0
3 1 1 2 3 4 5 6 7
Time Points
Figure 18. Smoothing spline ANOVA of F2 values with word position for vowel+/r/ and vowel+/l/
23 This effect is surprising given the frequency of the preceding sounds, but probably has to do with an interaction between preceding sound and vowel. Further research on this would be useful for a more complete understanding of the role of F2 in liquid production. 113
In order to consider how well the components of the model shown in Figure 18 above explain the variation, this model was compared to other simpler models with KL projections. The results are in Table 25 below.
114 Model 1 Time Liquid Type Word Position Interaction - Time and Liquid Type Interaction - Liquid Type and Word Position Interaction - Time and Word Position Model 2.1 - Just main effects Time Liquid Type Word Position KL Ratio for Model 1 vs. Model 2.1 0.2972433 Model 2.2 - Without Liquid Type and Time Word Position Interaction Liquid Type Word Position Interaction - Time and Liquid Type Interaction - Time and Word Position KL Ratio for Model 1 vs. Model 2.2 0.1635685 Model 2.3 - Without Time and Liquid Time Type Interaction Liquid Type Word Position Interaction - Liquid Type and Word Position Interaction - Time and Word Position KL Ratio for Model 1 vs. Model 2.3 0.001059225 Model 2.4 - Without Time and Word Time Position Interaction Liquid Type Word Position Interaction - Time and Liquid Type Interaction - Liquid Type and Word Position KL Ratio for Model 1 vs. Model 2.4 0.1326577 Model 2.5 - Without Liquid Type or its Time Interactions Word Position Interaction - Time and Word Position KL Ratio for Model 1 vs. Model 2.5 0.183927 Model 2.6 - Without Word Position or Time its Interactions Liquid Type Interaction - Time and Liquid Type KL Ratio for Model 1 vs. Model 2.6 0.6916325 Table 25. KL Projection model comparisons for F2 time, Liquid Type, and Word Position and all of the two-way interactions
The only model comparison with a KL Ratio below 0.02 is that for Model 2.3.
This finding goes along with the finding for the simple model in Figure 17 in that the interaction between time and liquid type, and thus the shape of the vowel+/r/ spline 115 including both word positions, is not very different from the vowel+/l/ curve. in Figure
17. The fact that the rest of the models have values above 0.02 suggests that the main effects of time, liquid type, and word position, as well as the interaction terms of word position with both time and liquid type are important for explaining the variation of liquids for this formant.
3.4.2.4 The Third Formant (F3)
The following analyses considers F3 as the dependent variable in order to determine whether or not vowel+/r/ and vowel+/l/ are different from each other in terms of the third formant. As discussed in chapter 2, low F3 values are a sign of rhoticity. The results from the linear regression models for the seven time points are presented in section 3.4.2.4.1 and the exploratory analyses of F3 using smoothing splines are in section 3.4.2.4.2.
3.4.2.4.1 Linear Regression Models for F3 at Seven Time Points
Like the previous analyses for F1 and F2, this analysis was performed using seven mixed effects linear regression models. The time points 1 through 7 represent measurements of F3 taken at 25%, 37.5%, 50%, 75%, 87.5%, and 100% of the duration of the vowel+liquid sequences. Each of the seven regression models takes the values of
F3 at its respective time point as the dependent variable and weighs the correlation of the independent variables (see section 3.2.4) and the interaction terms between liquid type and the other independent variables (see section 3.2.5.2.1) with the values for F3. The interaction terms for liquid type with following sound, stress, and gender were not 116 selected as significant by any of the seven time point models for F3 and were therefore left out of the final models for this formant. The models reported here, therefore, have all of the independent variables discussed in section 3.2.4 as well as the interaction terms between liquid type and age, word position, preceding sound, and vowel. Table 26 shows the overall predicted values for vowel+/r/ and vowel+/l/ for F3. These overall predicted values reflect all of the data taken together. As discussed above, each time point represents a separate mixed effects regression model.
Time Points vowel+/r/ vowel+/l/ Significant Difference? 1 2534.82 2514.94 no 2 2496.04 2490.96 no 3 2476.88 2500.34 no 4 2500.31 2548.07 no 5 2501.53 2580.51 yes 6 2551.34 2614.26 no 7 2591.14 2596.3 no Table 26. Overall F3 predicted values of vowel+liquid sequences across seven time points
Figure 19 below is a plot of the values from Table 26. Visualizing the values of vowel+/r/ and vowel+/l/ over the duration of the vowel+liquid sequences allows us to see that F3 values for vowel+/r/ differ from those with vowel+/l/ in the middle of the sequences from time points 3 through 6 where the trajectory of vowel+/r/ curves downwards whereas the trajectory of vowel+/l/ curves upwards. Despite the different trajectories, the only significant difference between vowel+/r/ and vowel+/l/ is at time point 5. However, significant differences arise in the interactions between liquid type and age, word position, previous sound, and vowel at all of the time points. The specific
117 contexts that were selected as significant are reported in Table 27 below. See the explanation in section 3.4.2.2 on how to read the table.
2620 2600 r l 2580 2560
z 2540 H 2520 2500 2480 2460
1 2 3 4 5 6 7
Time Points
Figure 19. Overall F3 predicted values of vowel+liquid sequences across seven time points
118 Time Overall Significant Interaction Effects Point difference interactions significant? 1 no Vowels /e/, /e/: /r/ < overall; /l/ > overall; /r/ < /l/; Difference /o/ > Overall. /r/ > /l/ /o/: /r/ < overall; /l/ > overall; /r/ < /l/ Difference > Overall. 2 no Word medial: /r/, /l/ < overall; /r/ > /l/; Difference > Position Overall. /r/ > /l/ final: /r/, /l/ > overall; /r/ < /l/; Difference > Overall.
Age older: /r/ > overall; /l/ < overall; /r/ > /l/; Difference > Overall. younger: /r/ < overall; /l/ > overall; /r/ < /l/; Difference > Overall.
Vowels /e/, /e/: /r/ < overall; /l/ > overall; /r/ < /l/; Difference /o/ > Overall. /o/: /r/ < overall; /l/ > overall; /r/ < /l/; Difference > Overall. 3 no Word medial: /r/ > overall; /l/ < overall; /r/ > /l/; Position Difference < Overall. /r/ < /l/ final: /r/ < overall; /l/ > overall; /r/ < /l/; Difference > Overall.
Age older: /r/ > overall; /l/ < overall; /r/ > /l/; Difference < Overall. younger: /r/ < overall; /l/ > overall; /r/ < /l/; Difference > Overall.
Vowels /i/, /i/: /r/, /l/ > overall; /r/ > /l/; Difference > Overall. /o/ /o/: /r/ < overall; /l/ > overall; /r/ < /l/; Difference > Overall. Continued
Table 27. Values for F3 for vowel+/r/ and vowel+/l/ overall and in particular contexts
119 Table 27: Continued
Time Overall Significant Interaction Effects Point difference interactions significant? 4 no Word medial: /r/ > overall; /l/ < overall; /r/ < /l/; Position Difference < Overall. /r/ < /l/ final: /r/ < overall; /l/ > overall; /r/ < /l/; Difference > Overall.
Preceding approximant: /r/ > overall; /l/ < overall; /r/ > /l/; approximant Difference > Overall.
Vowel /o/ /o/: /r/, /l/ < overall; /r/ < /l/; Difference > Overall. 5 yes (none) (none)
/r/ < /l/ 6 no Word medial: /r/ > overall; /l/ < overall; /r/ < /l/; Position Difference < Overall. /r/ < /l/ final: /r/ < overall; /l/ > overall; /r/ < /l/; Difference > Overall;
Preceding approximant: /r/ > overall; /l/ < overall; /r/ > /l/; approximant, Difference < Overall. nasal nasal: /r/ < overall; /l/ > overall; /r/< /l/; Difference > Overall.
Vowels /i/, /i/: /r/ > overall; /l/ < overall; /r/ > /l/; Difference /u/ > Overall. /u/: /r/ < overall; /l/ > overall; /r/ < /l/; Difference > Overall. 7 no Word medial: /r/ > overall; /l/ < overall; /r/ > /l/; Position Difference > Overall. /r/ < /l/ final: /r/ < overall; /l/ > overall; /r/ < /l/; Difference > Overall.
Vowel /i/ /i/: /r/ > overall; /l/ < overall; /r/ > /l/; Difference > Overall.
120 3.4.2.4.2 Smoothing Splines Analysis for F3
This section shows the F3 results of the Smoothing Spline ANOVAs, which compare the entire trajectories of vowel+approximant /r/ and vowel+approximant /l/. See section 3.2.5.2.2 for details on the statistical modeling for SS ANOVAs. In the basic model in Figure 20 below, which shows the vowel+/r/ and vowel+/l/ trajectories for F3 over time, it can be seen that both splines have different shapes and that the confidence
intervals only overlap at the very beginning of the vowel+liquid trajectories.
0
0
6
2
0
5
5
2
0
z
0
5
H
2
0
5
4 2
r 0
0 l
4 2 1 2 3 4 5 6 7
Time Points
Figure 20. Smoothing spline ANOVAs of F3 values over time for vowel+/r/ and vowel+/l/
121 The KL projection for this model in Table 28 confirms that the shapes are indeed different from each other, since the ratio value is over 0.02.
Model 1 Time, Liquid Type, Interaction between Time and Liquid Type Model 2 Time and Liquid Type KL Ratio for Model 1 vs. Model 2 0.02446816 Table 28. KL Projection model comparison for F3 Time and Liquid Type
As is evident in Figure 21, the addition of word position to the above model for
F3 shows that final liquids differ from each other more dramatically in word-final position than in word-medial position. The confidence intervals for word-final liquids, overlap at the very beginning of the trajectory, but otherwise are distinct both in terms of height and in terms of the shapes of the curves. Word-medial liquids, on the other hand, have overlapping confidence intervals and have similar shapes. The medial liquids splines are a bit difficult to see in the figure with all four splines, so they are presented on their own in Figure 22 below. Vowel+/r/ and vowel+/l/ in word-medial position have overlapping confidence intervals throughout the duration of the liquid trajectories and therefore appear to be neutralized in this position. This finding is supported in the regression analysis in section 3.4.2.4.1 wherein word-final sounds are distinct but word- medial sounds are neutralized.
122 final /r/ final /l/ middle /r/
middle /l/
0
0
6
2
0
5
5
2
z
0
H
0
5
2
0
5
4
2
0
0
4 2 1 2 3 4 5 6 7
Time Points
Figure 21. Smoothing spline ANOVA of F3 values with word position for vowel+/r/ and vowel+/l/
middle /r/
middle /l/
0
0
6
2
0
5
5
2
0
0
5
2
0
5
4
2
0
0
4 2 1 2 3 4 5 6 7
Figure 22. Smoothing spline ANOVA of F3 values for vowel+/r/ and vowel+/l/ for word- medial position only
In order to explore the main effects and interactions in the model in Figure 21, KL projections were performed for a variety of different simpler models excluding one or
123 more terms in order to compare them to the main model (Model 1). All of the model comparisons received a KL ratio above 0.02, which tells us that all of the main effects and interactions in Model 1 are helpful in explaining the variation in F3.
124 Model 1 Time Liquid Type Word Position Interaction - Time and Liquid Type Interaction - Liquid Type and Word Position Interaction - Time and Word Position Model 2.1 - Just main effects Time Liquid Type Word Position KL Ratio for Model 1 vs. Model 2.1 0.1032441 Model 2.2 - Without Liquid Type and Time Word Position Interaction Liquid Type Word Position Interaction - Time and Liquid Type Interaction - Time and Word Position KL Ratio for Model 1 vs. Model 2.2 0.04577876 Model 2.3 - Without Time and Liquid Time Type Interaction Liquid Type Word Position Interaction - Liquid Type and Word Position Interaction - Time and Word Position KL Ratio for Model 1 vs. Model 2.3 0.02141535 Model 2.4 - Without Time and Word Time Position Interaction Liquid Type Word Position Interaction - Time and Liquid Type Interaction - Liquid Type and Word Position KL Ratio for Model 1 vs. Model 2.4 0.04346469 Model 2.5 - Without Liquid Type or Time its Interactions Word Position Interaction - Time and Word Position KL Ratio for Model 1 vs. Model 2.5 0.2229087 Model 2.6 - Without Word Position or Time its Interactions Liquid Type Interaction - Time and Liquid Type KL Ratio for Model 1 vs. Model 2.6 0.4283061 Table 29. KL Projection model comparisons for F3 Time, Liquid Type, and Word Position and all of the two-way interactions
3.4.2.5 The Fourth Formant (F4)
The following analyses considers F4 as the dependent variable in order to determine whether or not vowel+/r/ and vowel+/l/ are different from each other in terms 125 of the fourth formant, which is usually left unstudied in work for consonants and also often for vowels. F4 corresponds to voice quality from the opening and closing of the larynx (Sundberg 1995). In his work on coda rhotics and laterals in the Spanish spoken in
Ponce, Puerto Rico, Luna (2010) finds that F4 values for /l/ are significantly higher than those for /r/. This study, therefore, seeks to further investigate this finding on the role of
F4 in liquid production. The results from the linear regression models for the seven time points are presented in section 3.4.2.5.1 and the exploratory analysis of F4 using smoothing splines are in section 3.4.2.5.2.
3.4.2.5.1 Linear Regression Models for F4 at Seven Time Points
Like the previous analyses for F1, F2, and F3, this analysis was performed using seven mixed effects linear regression models. The time points 1 through 7 represent measurements of F4 taken at 25%, 37.5%, 50%, 75%, 87.5%, and 100% of the duration of the vowel+liquid sequences. Each of the seven regression models takes the values of
F4 at its respective time point as the dependent variable and weighs the correlation of the independent variables (see section 3.2.4) and the interaction terms between liquid type and the other independent variables (see section 3.2.5.2.1) with the values for F4. The interaction terms for liquid type with following sound, stress, and gender were not selected as significant by any of the seven time point models for F4 and were therefore left out of the final models for this formant. The models reported here, therefore, have all of the independent variables discussed in section 3.2.4 as well as the interaction terms between liquid type and age, word position, preceding sound, and vowel. Table 30 shows the overall predicted values for vowel+/r/ and vowel+/l/ for F4 and whether or not these 126 differences were found to be significant. These overall predicted values reflect all of the data taken together. As discussed above, each time point represents a separate mixed effects regression model.
Time Points vowel+/r/ vowel+/l/ Significant Difference? 1 3531.53 3461.33 no 2 3478 3496.14 no 3 3438.99 3486.83 no 4 3471.46 3510.06 no 5 3533.22 3502.76 no 6 3563.53 3516.51 no 7 3582.25 3601.53 no Table 30. Overall F4 predicted values of vowel+liquid sequences at each time point
Figure 23 below is a plot of the values from Table 30. Visualizing the values of vowel+/r/ and vowel+/l/ over the duration of the vowel+liquid sequences allows us to see that F4 values for vowel+/r/ have a different trajectory from those with vowel+/l/.
Despite the different trajectories, there are no significant differences between vowel+/r/ and vowel+/l/ for F4. As reported in Table 30, the overall differences between vowel+/r/ and vowel+/l/ are not significant. However, significant differences arise in the interactions between liquid type and age, word position, previous sound, and vowel at all of the time points. The specific contexts that were selected as significant are reported in
Table 31 below. See the explanation in section 3.4.2.2 on how to read the table.
127 3600 r 3580 l 3560 3540
z 3520 H 3500 3480 3460 3440
1 2 3 4 5 6 7
Time Points
Figure 23. Overall F4 predicted values of vowel+liquid sequences across seven time points
128 Time Overall Significant Interaction Effects Point difference interactions significant? 1 no none N/A
/r/ > /l/ 2 no Word medial: /r/, /l/ < overall; /r/ > /l/; Difference > Position Overall. /r/ < /l/ final: /r/ > overall; /l/ < overall; /r/ < /l/; Difference > Overall. 3 no Vowel /a/ /a/: /r/, /l/ > overall; /r/ < /l/; Difference > Overall. /r/ < /l/ 4 no Preceding rhotic: /r/ < overall; /l/ > overall; /r/ < /l/; rhotic Difference > Overall. /r/ < /l/ Vowel /o/ /o/: /r/, /l/ < overall; /r/ < /l/; Difference > Overall. 5 no Age older: /r/, /l/ > overall; /r/ > /l/; Difference > Overall. /r/ > /l/ younger: /r/, /l/ < overall; /r/ < /l/; Difference < Overall.
Preceding lateral: /r/ > overall; /l/ < overall; /r/ > /l/; lateral, rhotic Difference > Overall. rhotic: /r/ < overall; /l/ > overall; /r/ < /l/; Difference > Overall. 6 no Preceding lateral: /r/, /l/ < overall; /r/ > /l/; Difference > lateral, nasal Overall. /r/ > /l/ nasal: /r/ < overall; /l/ > overall; /r/ < /l/; Difference > Overall. 7 no Vowel glide +/a/: /r/ < overall; /l/ > overall; /r/ < /l/; glide+/a/ Difference > Overall. /r/ < /l/ Table 31.Values for F4 for vowel+/r/ and vowel+/l/ overall and in particular contexts
3.4.2.5.2 Smoothing Splines Analysis for F4
This section shows the F4 results of the Smoothing Spline ANOVAs, which compare the entire trajectories of vowel+approximant /r/ and vowel+approximant /l/. See section 3.2.5.2.2 for details on the statistical modeling for SS ANOVAs. 129 The two splines in Figure 24 differ from each other near the beginning of their trajectories until around time point 6, where the two converge. The shapes of the F4 curves do not appear to be very different from each other, and this visual observation is confirmed by the KL projection in Table 32. The KL projection is below 0.02, which suggests that the interaction term is not useful to explain the variation and that the curves
are not very different from each other.
0
0
7
3
0
0
6
3
z
H
0
0
5 3
r 0
0 l
4 3 1 2 3 4 5 6 7
Time Points
Figure 24. Smoothing spline ANOVAs of F4 values over time for vowel+/r/ and vowel+/l/
130 Model 1 Time, Liquid Type, Interaction between Time and Liquid Type Model 2 Time and Liquid Type KL Ratio for Model 1 vs. Model 2 0.01821351 Table 32. KL projection model comparison for F4 Time and Liquid Type
Figure 25 below considers word position for F4. The confidence intervals overlap for vowel+/r/ and vowel+/l/ in both word-final and word-medial positions. Word-final and word-medial liquids are more different from each other in the beginning of the trajectories, which correspond to the vocalic portion, and more similar at the end of the trajectories. The confidence intervals for vowel+/r/ and vowel+/l/ overlap for both word- final and word-medial liquids, which shows that the differences between the liquids are negligible in both word positions.
final /r/ final /l/ middle /r/
middle /l/
0
0
7
3
0
0
6
3
z
H
0
0
5
3
0
0
4 3 1 2 3 4 5 6 7
Time Points
Figure 25. Smoothing spline ANOVA of F4 values with word position for vowel+/r/ and vowel+/l/
131 In order to analyze the components of the model in terms of whether or not they are useful for explaining the variation, KL projects for different model comparisons were run. See Table 33 below.
132 Model 1 Time Liquid Type Word Position Interaction - Time and Liquid Type Interaction - Liquid Type and Word Position Interaction - Time and Word Position Model 2.1 - Just main effects Time Liquid Type Word Position KL Ratio for Model 1 vs. Model 2.1 0.09684753 Model 2.2 - Without Liquid Type and Time Word Position Interaction Liquid Type Word Position Interaction - Time and Liquid Type Interaction - Time and Word Position KL Ratio for Model 1 vs. Model 2.2 0.0006963347 Model 2.3 - Without Time and Liquid Type Time Interaction Liquid Type Word Position Interaction - Liquid Type and Word Position Interaction - Time and Word Position KL Ratio for Model 1 vs. Model 2.3 0.004447502 Model 2.4 - Without Time and Word Time Position Interaction Liquid Type Word Position Interaction - Time and Liquid Type Interaction - Liquid Type and Word Position KL Ratio for Model 1 vs. Model 2.4 0.08521984 Model 2.5 - Without Liquid Type or its Time Interactions Word Position Interaction - Time and Word Position KL Ratio for Model 1 vs. Model 2.5 0.02716173 Model 2.6 - Without Word Position or its Time Interactions Liquid Type Interaction - Time and Liquid Type KL Ratio for Model 1 vs. Model 2.6 0.3815938 Table 33. KL Projection model comparisons for F4 Time, Liquid Type, and Word Position and all of the two-way interactions
133 The model that excludes the interaction for liquid type and word position (Model
2.2) receives a low KL ratio when compared to Model 1. This suggests that vowel+/r/ and vowel+/l/ are not different from each other based upon whether they appear at the middle or end of a word. This finding is not surprising considering that the confidence intervals overlap for the two liquids in both word positions. Model 2.3, which excludes the interaction between time and liquid type, also has a low value, suggesting that this interaction is not useful for explaining the variation. This is consistent with the findings for the time and liquid type model in Figure 24. The only interaction that is helpful in the model, then, is that of time and word position. This tells us that while vowel+/r/ and vowel+/l/ are not different from each other, liquids taken together do behave differently in word-medial vs. word-final positions in terms of F4.
The results for the production study presented in this chapter will be explained and interpreted in Chapter 5. In the next part of this dissertation, Chapter 4, I present the methodology and results of a perception experiment that tests whether or not SJS listeners can use the acoustic differences between vowel+/r/ and vowel+/l/ to distinguish the liquid sounds.
134
CHAPTER 4: CODA LIQUID PERCEPTION IN PUERTO RICAN AND CASTILIAN SPANISH
4.1 INTRODUCTION
The findings for the production study presented in Chapter 3 indicate that there are significant differences between the pronunciation of vowel+/r/ and vowel+/l/ in San
Juan Spanish (SJS). Although it is clear that
135 4.2 METHODOLOGY
4.2.1 Experimental Design
4.2.1.1 Selection and Preparation of Stimuli
The purpose of this study is to see how SJS and CS speakers hear SJS vowel+liquid sequences produced in natural speech. I therefore decided to choose a speaker from the corpus used for the production study in Chapter 3 who exhibited a variety of different formant contours in his or her production of vowel+liquid sequences.
In order to decide which speaker to use and which vowel+liquid sequences in that speaker's interview to include in the perception experiment, I considered several factors.
The first consideration for the experiment was which vowels would be included in the vowel+liquid stimuli. While it would be ideal to look at all vowels and glide+vowel combinations, including enough stimuli from each vowel to consider a variety of formant contours would make the experiment too long and it would be less likely that participants would complete the task. Another pitfall of including all of the vowels in the perception study is the low frequency of certain vowels before liquids. The vowel /u/, for example, rarely appears before
136 but also serves to explore how liquids are perceived in the contexts that are the most frequent in the Spanish language.
The next consideration in the design of the experiment was whether or not to include any surrounding context from the vowel+liquid sequence or just the sequence in isolation. A variety of contextual information would be possible; participants could hear whole words, two syllable utterances, single syllables with a consonant before the vowel+liquid sequence, or simply the vowel+liquid sequence itself. The idea to use entire words was discarded since there are very few minimal pairs in the corpus and the listener would likely be influenced by the existence of one word and not another. For example, if a listener was asked to choose between having heard parte 'part' and palte (nonsense word)', the participant would likely be influenced by their lexicon. The two-syllable option would give listeners a more naturalistic context than simpler tokens, but ultimately would result in similar problems to the inclusion of entire words, since many words are two syllables long and many two-syllable combinations are more common with /r/ and others with /l/. The decision, then, came to whether or not to include previous consonant sounds before each vowel+liquid sequence. I opted to exclude previous sounds due to the difficulty of finding enough vowel+liquid occurrences with the same previous sound within the speech of a single speaker. Given the limited number of vowel+liquid tokens within each interview from the production data, the inclusion of the previous sound would mean including a broad variety of previous sounds, which would result in a limited number of tokens from each sound, making the independent variable of previous sound fairly meaningless for the analysis. Given all of these considerations, I chose to use
137 stimuli that consisted of vowel+liquid sequences without any preceding or following sounds.
In order to limit the number of independent variables in the experiment, I decided to choose stimuli from the interview of a single speaker. The selection of a speaker from the corpus whose vowel+liquid sequences would serve as the stimuli for the perception experiment was predicated on the quality of speech; the number of possible /a/+liquid,
/e/+liquid, and /o/+liquid tokens from the interview; and the amount of coda liquid variation within the speech of each single informant. Five informants (speakers 11, 14,
21, 22, and 23) were excluded from consideration due to creaky voice. The other nineteen speakers were evaluated according to the number of tokens with the vowels under consideration. Table 34 below shows the number of tokens per vowel+liquid sequence for each speaker.
138 Informant # /ar/ /er/ /or/ /al/ /el/ /ol/ 1 33 22 49 15 10 2 2 15 18 17 10 6 1 3 23 32 22 17 6 0 4 24 26 18 13 6 3 5 31 25 33 15 7 1 6 18 23 12 5 4 3 7 30 34 30 9 7 2 8 18 18 21 10 6 12 9 23 13 38 0 7 0 10 27 21 11 6 4 7 12 25 29 18 13 4 3 13 19 34 30 11 5 2 15 19 18 28 15 4 2 16 13 8 11 9 4 8 17 14 14 35 7 4 4 18 22 17 7 7 3 5 19 21 17 26 14 5 1 20 34 8 24 13 10 0 24 16 14 24 4 6 4 Table 34. Token counts for each informant from the production data for /a/, /e/, and /o/ with coda /r/ and /l/. Informants 11, 14, 21, 22, and 23 were not considered due to creaky voice.
Informants 3, 9, and 20 were excluded due to empty categories. The rest of the speakers were included in an analysis of formant variation. This analysis involved examining the formant measurements for F1, F2, F3, and F4 at the seven time points for each of the occurrences of /a/, /e/, and /o/+liquid. Originally, I hoped to find a speaker who had tokens pertaining to the categories listed in Table 35 below.
139 Variable Formant Constant Formants high F1 average F2, average F3, average F4 low F1 average F2, average F3, average F4 high F2 average F1, average F3, average F4 low F2 average F1, average F3, average F4 high F3 average F1, average F2, average F4 low F3 average F1, average F2, average F4
Table 35. Ideal categories for perception stimuli formant structure
The idea behind the categorizations in Table 35 above was that I could control for three of the formants (constant formants) while varying one formant (variable formant). I decided to find tokens with varying F1, F2, and F3, since these formants had more significant findings than did F4 in the production analysis in Chapter 3. By selecting tokens that differed in one formant at a time, I could see which of the formants contributed to a perception of /r/ or /l/ in the experiment. For example, if stimuli with low
F3 were more frequently perceived as /r/ than stimuli with a high F3, the conclusion that low F3 contributes to a rhotic percept could be drawn. In order to categorize the formants of individual tokens as having a high, average, or low trajectory, I considered each of the seven time points for each formant separately. I divided the data into twelve separate spreadsheets corresponding to each of the six vowel+liquid combinations (/ar/, /er/, /or/,
/al/, /el/, and /ol/) for both men and for women. Men and women were considered separately due to anatomically determined differences in F0 and formant values; an
140 overall formant value that was low for women might be high for men.24 Each time point for each formant was then sorted in Excel from the lowest to highest value. I divided the number of tokens by three and color-coded the first third as low (green), the second third as average (yellow), and the last third as high values (red). The color-coding scheme made it possible to visualize the amount of variation in the formant trajectories for individual tokens and then for individual speakers. Next, in order to code the formant values globally, the green, yellow, and red values for individual time points were assessed across the seven time points. Each of the four formant values for each token was then classified according to the categories in Table 36 below.
Label Color Scheme for Seven Time Points LOW all 7 cells are green low 5-6 cells are green, the other 1-2 are yellow low? 4 cells are green, 3 are yellow AV all 7 cells are yellow av 5-6 cells are green, the other 1-2 are either red or green (but not one of each) av? 4 cells are yellow, the other 3 are either red or green (but not a combination of the two) HIGH all 7 cells are red high 5-6 cells are red, the other 1-2 are yellow high? 4 cells are red, 3 are yellow Table 36. Schema for coding low, average, and high formant trajectories across the seven time points
The capitalized (LOW, AV, HIGH) represent the best tokens of their categories in terms of being high, average, or low across the seven time points. The lower case and question mark categories are less than ideal, but still constitute measurements pertaining
24 Since every speaker has a different fundamental frequency, this methodology could be improved upon in future study. The idea here was to find a speaker with a great deal of variance in their liquid production, which could be achieved through this methodology, despite the inherent differences between individuals. 141 mostly to the category desired. An example of the implementation of the color scheme and the classification of low, average, and high trajectories is given in Figure 26, which shows F1 values of /ar/ tokens for informant 1.
Figure 26. Example of perception data coding. This data comes from F1 for /ar/ tokens from a male speaker. Each row represents one occurrence of /ar/ from the production data. The seven time points are represented across the top as f1.1, f1.2, f1.3, f1.4, f1.5, f1.6, and f1.7. The right-most column shows how each trajectory was coded.
The schema for coding the seven time points, as discussed above, made it possible to classify the full duration of the vowel+liquid sound and then multiple formants could be compared. Thus, I could look at the values for F1, like in Figure 26 and then look at values for F2, F3, and F4. The goal was to find a speaker who produced tokens like those in Table 35 above. That is, I was searching for a speaker who produced high and low
142 trajectories of each formant while maintaining average trajectories for the others. I eventually had to discard this idealized notion of controlled formant trajectories - none of the speakers showed this pattern. I therefore decided to pick the speaker with the highest number of marked low, average, and high trajectories. That is, rather than controlling across the four formants, I decided to pick a speaker who exhibited a wide range of liquid variation across the formants. The generalizations about the trajectories could then be used in the statistical modeling of the perception results to help explain why participants heard certain trajectories as vowel+/r/ and others as vowel+/l/. Two of the speakers - 1 and 7 - were selected as candidates for having varying formant trajectories based on the categorization as low, average, and high. Speaker 7 was chosen because of his slower speech rate than that of speaker 1. Since the stimuli for the perception experiment were vowel+liquid syllables, which are already very short in duration, the slower speech rate would be helpful to ensure that the syllables were long enough to be distinguished out of context. Although, as described below, the durations of the speech data were modified for the experiment, starting out with the longer durations in speaker 7's speech allowed for more minimal modifications.
Next, all 112 occurrences of /ar/ (N=30), /er/ (N=34), /or/ (N=30), /al/ (N=9), /el/
(N=7), and /ol/ (N=2) that were produced as approximants in informant 7's speech were spliced from the interview and saved as individual files. Four tap productions of /r/ were also included in the experiment to see if these productions would be easier to identify as rhotics than approximant /r/ productions. I expected taps to be more easily perceived as rhotics than approximant /r/ productions by both SJS and CS listeners, since these sounds are considered to be the canonical coda rhotic production and since taps have a different
143 manner of articulation than approximant liquids. Seventy tokens were excluded due to creaky voice or other sound quality problems, or due to the lack of a liquid percept in the vowel+liquid sequence. In order to make sure that the experimental stimuli were alike in every way except the vowel+liquid formant trajectories under study, the remaining 42 approximant and tap tokens were adjusted in duration and amplitude. The original durations of the usable sequences ranged from 58.667ms to 283.738ms with one outlier of 439.523ms. In order to determine an appropriate duration for the experimental stimuli,
I modified each of the stimuli twice to both an average value of all of the stimuli
(153.124ms) and to the length of the longest stimulus other than the outlier (283.738ms).
To perform the duration modifications, I used the 'lengthen' function in Praat and multiplied the original duration of each token by the amount necessary to obtain the desired length. The stimuli durations were then evaluated by two trained phoneticians, one of whom is a native speaker of Spanish. Both phoneticians agreed that the longer duration was easier to judge in terms of the liquid sound being produced. The stimuli were thus modified to 283.738ms and then were adjusted with a Praat script to equate their amplitude so that participants would not hear some stimuli as louder than others.
Finally, 30ms of silence were added to the beginning and end of each sound file. The duration adjustment caused a few of the tokens to sound unnatural, so these tokens were left out of the analysis. The experiment includes 33 approximant vowel+liquid sequences and 4 vowel+tap /r/ productions. Table 37 shows the words from which the stimuli were drawn. Note that the participants did not hear the word, but rather simply the vowel+liquid.
144 Order Orthography Options Stimulus Liquid Presented in of Liquid in Available to Word Manner of Survey Stimulus Listener Articulation 1 r ar / al pARte approximant 3 r er / el pERder approximant 5 l ar / al cÁLculo approximant 7 r or / ol mejOR approximant 9 r ar / al quedAR approximant 11 r ar / al fARmacia approximant 13 l er / el EL nuevo approximant 15 r or / ol pOR ejemplo tap 17 r or / ol mejOR te approximant 19 r ar / al tocAR pero approximant 21 r or / ol mejOR approximant 23 r ar / al estAR en approximant 25 r er / el aprendER tap 27 r er / el intERmedio approximant 29 l ar / al mAL approximant 31 r er / el cERca approximant 33 r ar / al ARtístico approximant 35 l or / ol cOLmado approximant 37 r ar / al dAR approximant 39 r er / el vER niño approximant 41 l ar / al mentALmente approximant 43 r or / ol pORque approximant 45 r or / ol pOR eso tap 47 r er / el pERsonas approximant 49 r or / ol pORque approximant 51 r ar / al girAR approximant 53 r or / ol impORta approximant 55 r or / ol pOR eso tap 57 l ar / al elementAL approximant que 59 r or / ol pORque approximant 61 r ar / al tARjeta approximant 63 l er / el nivEL de approximant 65 r ar / al manipulAR approximant Continued Table 37. Stimuli for perception experiment. Distractor stimuli are inserted between each liquid stimulus. Listeners heard vowel+liquid only. The words from which the stimuli were extracted are provided for reference. 145 Table 37: Continued 67 r or / ol mejOR approximant 69 r er / el metER arte approximant 71 r or / ol impORta approximant 73 r ar / al entregAR approximant
Distractor stimuli were also extracted from informant 7's speech and were included in the experiment interspersed among the vowel+liquid stimuli. For the distractor items, I selected stimuli that I expected to be easily identifiable in order to avoid tiring the participants with the filler stimuli. The distractor items also served to make sure that participants understood the task, since it is expected that participants will perform better on these stimuli than the harder-to-distinguish liquid stimuli under study.
Since onsets are generally easier to perceive than codas (Steriade 1997), I chose two onset contexts with the same vowels used in the liquid stimuli (/a/, /e/, and /o/) for fillers.
For the distractor items, participants chose between m+vowel and ñ+vowel or between t+vowel and k+vowel. The same 12 stimuli (3 vowels for /m/, /ɲ/, /t/, and /k/) were repeated three times each during the experiment, in addition to three of the fillers in the training phase of the experiment (see 4.2.1.1 for more on the phases of the experiment).
The filler tokens were adjusted in duration to match the 283.738ms durations of the vowel+liquid stimuli and received the same modification for amplitude and the addition of 30ms of silence at both ends of each sound file.
4.2.1.1 Survey Design
The online platform SurveyGizmo (Vanek and McDaniel 2006) was used in order to create and administer the survey. The 37 vowel+liquid stimuli presented in Table 37
146 were presented in the order listed in the table with distractor stimuli in between them such that participants never heard two liquid stimuli in a row. Before listening to the stimuli, participants were presented with written instructions that explained that the goal of the study was to understand how speech sounds in Spanish are perceived by native speakers and that there were not 'right' or 'wrong' answers. No specific dialect details were given in order to keep any dialect-specific biases from playing into the judgments listeners made as to what they heard. After reading the instructions, participants were asked for their consent to participate and then filled out a demographic questionnaire which asked for the participant's birthplace, current town of residence, other places where he/she has lived, gender, age, occupation, and other languages spoken.
Following the demographic information, the participants read a page that told them that they would first have a short practice phase. The instructions for the practice phase informed the participants that the task was to listen to the stimulus and then choose which of two possibilities they had heard. Each stimulus was presented on a separate page and played automatically when the page opened. After hearing the stimulus, participants were asked to choose between two options, which included the vowel in the sequence and only differed in the consonant. For example, participants would choose between ar and al when hearing an /a/+liquid stimulus and would choose between mo and ño when the audio stimulus was a nasal+vowel. During the practice phase, participants heard fillers on pages one, three, and five and heard vowel+liquid stimuli on pages two and four. Having a practice phase was particularly important for this experiment due to the presence of a timer on each page which allowed participants to spend a maximum of 10 seconds listening to each stimuli and choosing a response.
147 Participants were instructed to try to choose an answer after hearing the audio one time so that they could record an answer within the ten seconds. If they had time to change their answer after listening again, they could do so. Limiting the time on each stimulus was important both to keep the survey under 15 minutes to ensure that participants would complete it and to avoid a great deal of variation between participants on the amount of time they spent listening to each stimulus. The disadvantage to the presence of the timer is that participants inevitably failed to choose an answer for some of the stimuli. After the five practice stimuli, participants were instructed that the practice phase was over and that the questions for the task would be exactly like those in the practice phase. Figure 27 is an example of a stimulus page from the survey.
Figure 27. Sample page from perception survey with an audio stimulus and two possible choices
148 After listening to all of the stimuli, participants were asked if they have a partner, close friend, or colleague who speaks another variety of Spanish and if so, where that person is from. Including this question was important in order to consider separately participants who have significant influence from other dialects and therefore may not faithfully represent the way that /r/ and /l/ are perceived in their native dialect. Unlike the other demographic information questions, this question was asked at the conclusion of the survey after listening to the stimuli so that participants would be less likely to suspect that the survey concerned any particular dialect of Spanish.
4.2.2 Participants
Participants for the perception survey were recruited through professional and personal contacts and online through social media platforms. In order to test whether or not Puerto Ricans perceive coda liquids differently than speakers of dialects with more distinct coda /r/ and /l/, data was collected from participants whose native dialect is
Puerto Rican Spanish (SJS) and from participants who speak Castilian Spanish (CS) from central and northern Spain. 25 Any variety of Spanish not described as a liquid- neutralizing dialect could serve equally well for the control group. The decision to collect data in Spain was a logistical one - I have professional and personal contacts there. Since regions of southern Spain have liquid switching, it was important to only recruit
25 The term Castilian Spanish was chosen to describe this group of participants for the experiment since it generally refers to the dialects spoken in central and northern Spain. The other common term to refer to Spanish spoken in Spain, Peninsular Spanish, generally refers to the entire peninsula, including the southern region. This nomenclature does not work in Spanish - use of the term castellano 'Castilian' in Spanish refers to the Spanish language itself and is used by many speakers throughout the Spanish-speaking world as synonymous with español 'Spanish' when naming the Spanish language. 149 participants from the central and northern regions. Forty-five Puerto Ricans and thirty- four Spaniards completed the survey. Nineteen of the 45 Puerto Ricans were excluded from the analysis because they were not from the San Juan area or did not clearly indicate which region of the island they were from or currently lived in. One of the Spaniards who took the survey was excluded due to having lived most of her life in the southern part of the country. Of the remaining twenty-six Puerto Ricans and thirty-three Spaniards, none of them had a parent or significant other from a dialect with a radically different liquid production behavior. That is, none of the Puerto Ricans considered had sustained daily contact with a dialect without liquid switching and none of the Spaniards had a close contact with a person from a liquid switching dialect. Many of the participants included in the analysis have friends from a variety of dialects, but occasional contact with other dialects is unlikely to drastically affect participants' evaluations of liquid sounds. The
Puerto Rican participants range from 18 to 65 years in age and the Spaniards range from
24 to 62. Both groups were primarily women: 19 of the Puerto Rican listeners were women, 7 were men and 22 Spanish women and 11 men participated.
4.2.3 Data Analysis
The participants' responses to the stimuli along with the demographic information that they filled out was exported from SurveyGizmo to Excel for analysis. Descriptive statistics for the number of /r/ and /l/ responses in the SJS group vs. the CS group are explored for all 33 approximant stimuli taken together. Chi-square tests were performed for the approximant liquid stimuli to compare the overall number of /r/ vs. /l/ selections by the two groups to see if the groups behaved differently from each other and to explore 150 whether or not performance was different within each dialect depending on the orthography of the stimulus. Then, the responses to orthographic /r/ tokens produced as taps from both groups are compared to see if SJS and CS listeners perceive taps differently. Responses to tap productions are then compared to responses to approximant
/r/ articulations within each dialect group to explore the differences in perception for the two articulations within each group. All of the explorations for tap productions are carried out with chi-square tests.
Finally, mixed effects logistic regression models were built in R using the lmer function of the lme4 package (Bates et al. 2011). Each model employs listeners' selection of /r/ or /l/ as the binary dependent variable and individual participants as a random effect. Separate models were built for SJS and CS listeners and the models were then compared. The regression analyses consist of two sets of models that differ only in the independent variables that they consider: the first set contains the vowel in the sequence
(/a/, /e/, or /o/), the orthography of the liquid (/r/ or /l/), and the interaction between vowel and orthography; the second set considers the four formant measurements. Since it is expected that the vowels and liquids in the spelling of the stimuli are highly correlated with certain formant structures, I chose not to include all of these independent variables together to avoid colinearity effects. The four formant measurements are considered together at each time point, resulting in seven comparisons between SJS and CS models for the seven time points.
151 4.3 RESEARCH QUESTIONS AND HYPOTHESES
1) Do Puerto Ricans hear liquid sounds differently than Spaniards?
Following Paz's (2005) findings, I expect that SJS speakers can hear the small but significant differences in the formant structures for vowel+/r/ and vowel+/l/.
Therefore, I expect SJS listeners to hear /r/ at a significantly higher rate than CS listeners when presented with stimuli with orthographic /r/.
2) What role do the formants (F1, F2, F3, and F4) play in listeners' ability to distinguish vowel+/r/ from vowel+/l/?
In the production data presented in Chapter 3, there are significant differences between vowel+/r/ and vowel+/l/ for all of the formants within specific contexts.
Therefore, it would not be surprising if all of the formants were important in distinguishing the two sounds. However, past research points to a low F3 as the strongest indicator of rhoticity (Lagefoged & Maddieson 1996, Lindau 1985) and it is this formant that Paz (2005) points out as the only differing acoustic correlate in her study. It is expected that certain stimuli in this experiment are more /r/-like or /l/-like due to formant structure differences and that more participants will hear /r/ on some stimuli than others.
It is expected that orthographic /r/ stimuli will be more reliably heard as /r/ by SJS listeners than by CS listeners. SJS listeners are also likely to be better at distinguishing orthographic /l/ stimuli from their dialect more accurately than CS lisenters. Differences between SJS listeners and CS listeners on these stimuli will help illuminate which formants are important for SJS listeners that are not attended to by CS listeners.
152 4.4 RESULTS
The first step in interpreting the results was to make sure that the participants understood the task. This was accomplished by checking the percentage of correct responses for distractor items. The nasal+vowel and stop+vowel filler stimuli were correctly identified by both groups at a rate around 90%, which is well above chance.
Therefore, it can be reasonably concluded that the participants understood the nature of the task.
When all 33 approximant liquid stimuli are considered together, both SJS and CS listeners hear the vowel+liquid stimuli as /r/ only about 37% of the time, even though
76% of the stimuli (26 of the 33) are spelled with /r/. The results for correctly identifying the liquid stimuli according to their spelling (orthographic /r/ as /r/ and orthographic /l/ as
/l/) are about at chance for both groups when all the data is considered together. See
Table 38 below for these results.
Dialect Number of Stimuli Number of Stimuli Total Number Percentage Correctly Identified Incorrectly Identified of Responses Correct Puerto 393 453 846 46.5% Rico Spain 448 599 1047 42.8% Table 38. Perception according to orthography for vowel+/r/ and vowel+/l/ stimuli taken together
Looking at the responses separately according to orthography, the results of the descriptive statistics on vowel+approximant liquids show that SJS listeners perform about 3% better than Spaniards when presented with orthographic /r/ stimuli. With stimuli with orthographic /l/, Puerto Ricans perform 9% better than Spaniards in that they 153 are less likely to choose /r/. Both dialect groups perform much better when presented with orthographic /l/ tokens - SJS listeners identify these stimuli correctly 71.3% of the time and the CS group hear orthographic /l/ as /l/ 63.3% of the time. Table 39 presents the results for orthographic /r/ and Table 40 presents results for orthographic /l/.
Dialect Total Number of Number of Percentage Number of /l/ responses /r/ responses /r/ responses Responses Puerto Rico 665 401 264 39.7% Spain 826 518 308 37.3% Table 39. Responses to orthographic /r/ approximant stimuli from SJS and CS listeners. Yates Chi-Square p=0.3681.
Dialect Total Number of Number of Percentage Number of /l/ responses /r/ responses /l/ responses Responses Puerto Rico 181 129 52 71.3% Spain 221 140 81 63.3% Table 40. Responses to orthographic /l/ stimuli for SJS and CS listeners. Yates Chi- Square p=0.0838.
As indicated by the p-values in Tables 39 and 40, the differences between the two groups for both orthographic /r/ and /l/ stimuli are not significant. However, the differences in performance on orthographic /r/ and /l/ within each group yields significant results for the SJS group, whereas the CS group does not perform differently on orthographic /r/ vs. orthographic /l/ stimuli. In other words, SJS listeners chose /r/ much more frequently for stimuli spelled with /r/ than they did for stimuli spelled with /l/, whereas there was no such effect for the CS group. Since /l/ is the most common response to the perception stimuli for both groups, the fact that /r/ is chosen for
154 orthographic /l/ less frequently than for orthographic /r/ by SJS listeners is noteworthy.
See Table 41 and Table 42 for the response counts and p-values from the chi-square tests.
Orthographic Orthographic /l/ /r/ Response /l/ 129 401 Response /r/ 52 264 Total 181 665 Percent /r/ 28.7% 39.7% Table 41. SJS listeners' performance on orthographic /r/ vs. /l/. Yates Chi-Square p < 0.01.
Orthographic Orthographic /l/ /r/ Response /l/ 140 518 Response /r/ 81 308 Total 221 826 Percent /r/ 36.7% 37.2% Table 42. CS listeners' performance on orthographic /r/ vs. /l/. Yates Chi-Square p=0.9203.
Four stimuli with /r/ with a tap, rather than an approximant, were included in the experiment. These stimuli were left out of the above analyses and considered separately.
Table 43 shows the overall results for the selection of /r/ vs. /l/ by both the SJS and the
CS groups. The SJS group recognizes tap as /r/ about 6% more frequently than does the
CS group. The difference between the two groups, however, is not statistically significant. Table 43 displays the results from the dialect comparison for taps.
155 Dialect Total Number of Number of Percentage Number of /l/ Responses /r/ /r/ responses Responses Responses Puerto Rico 102 50 52 50.9% Spain 124 69 55 44.4% Table 43. Responses to orthographic /r/ tap stimuli from SJS and CS listeners. Yates Chi- Square p=0.3904.
The selection of /r/ vs. /l/ was then compared for tap vs. approximant articulations of orthographic /r/ within each dialect to see whether or not there was a difference in how accurately each group identified /r/ depending on the two different manners of articulation. SJS listeners were significantly more likely to identify taps as /r/ than they were to correctly identify approximants as /r/. The CS group also identified taps as /r/ more frequently than approximants, but the difference between the judgments for the CS group were not statistically significant. Although further research would be necessary to explore this finding, it could be the case that tap production differs across the two dialects, making it easier for SJS listeners to identify taps in their own dialect. Table 44 shows the results for the tap vs. approximant data for SJS listeners and Table 45 displays the same information for CS listeners.
Orthographic Orthographic Tap /r/ Approximant /r/ Response /l/ 50 401 Response /r/ 52 264 Total 102 665 Percent /r/ 50.9% 39.7% Table 44. SJS listeners' performance on tap vs. approximant orthographic /r/ stimuli. Yates Chi-Square p=0.0407.
156 Orthographic Orthographic Tap /r/ Approximant /r/ Response /l/ 69 518 Response /r/ 55 308 Total 124 826 Percent /r/ 44.4% 37.2% Table 45. CS listeners' performance on tap vs. approximant orthographic /r/ stimuli. Yates Chi-Square p= 0.1583.
As explained in section 4.2.3, two sets of logistic regression models were built to explore what factors influenced listeners to choose /r/ or /l/. The first set of models consider the vowel in the sequence, the orthography of the liquid, and the interaction between orthography and vowel for SJS and CS separately. The interaction term serves to clarify which combinations of vowels and liquids behaved differently than others. The results for the step-wise comparison for SJS selected the interaction between vowel and orthography as significant. The CS model comparison did not find the interaction term to be useful in explaining the variation; this term was left in the model so that the model would be comparable to the SJS model. Table 46 shows the results for the selection of /r/ for SJS listeners and Table 47 contains the results for CS listeners.
157 Estimate Standard Error z-Value p-value (Intercept) -0.6607 0.2194 -3.012 0.00260 Orthography of Liquid /l/ -0.1067 0.1134 -0.941 0.34666 /r/ 0.1067 0.1134 0.941 0.34666 Vowel /a/ -0.2059 0.1394 -1.477 0.13979 /e/ -0.6526 0.1658 -3.937 8.25e-05 /o/ 0.8585 0.1760 4.879 1.07e-06 Interaction: Vowel with Liquid Orthography /al/ -0.3656 0.1397 -2.617 0.00886 /ar/ 0.3656 0.1397 2.617 0.00886 /el/ -0.1207 0.1648 -0.732 0.46388 /er/ 0.1207 0.1648 0.732 0.46388 /ol/ 0.4863 0.1749 2.781 0.00542 /or/ -0.4863 0.1749 -2.781 0.00542 Table 46. Logistic regression for SJS with orthographic liquid, vowel, and interaction between the two. Input value= /r/.
158 Estimate Standard Error z-Value p-value (Intercept) -0.64148 0.24099 -2.662 0.00777 Orthography of Liquid /l/ 0.09684 0.10046 0.964 0.33503 /r/ -0.09684 0.10046 -0.964 0.33503 Vowel /a/ -0.03421 0.12215 -0.280 0.77940 /e/ -0.42416 0.14269 -2.973 0.00295 /o/ 0.45838 0.16003 2.864 0.00418 Interaction: Vowel with Liquid Orthography /al/ -0.20389 0.12228 -1.667 0.09544 /ar/ 0.20389 0.12228 -1.667 0.09544 /el/ -0.02480 0.14214 -0.174 0.86149 /er/ 0.02480 0.14214 -0.174 0.86149 /ol/ 0.22869 0.15960 1.433 0.15189 /or/ -0.22869 0.15960 1.433 0.15189 Table 47. Logistic regression for CS with orthographic liquid, vowel, and interaction between the two. Input value= /r/.
The results for the logistic regression show that there were no significant results for orthography on its own, but there are significant differences when orthography is combined with vowel. The vowel+liquid combinations with /a/ and /o/ are significant for
SJS listeners, but there are no significant effects for the CS group. This means that liquids with these two vowels are distinguishable for SJS listeners, but not for CS listeners. For
SJS listeners, /al/ disprefers the selection of /r/ and /ol/ prefers the selection of /r/. The significant effects for the interaction terms for SJS listeners suggest that they are hearing the orthographic /r/ and /l/ segments differently than listeners from the CS group. To further investigate what acoustic cue SJS listeners use to identify orthographic /r/ and /l/, the selection of liquid according to formant values for the stimuli is explored below.
159 The second set of models has the same dependent variable - the selection of /r/ vs.
/l/ - as the first set but utilizes the formant measurements for the independent variables in place of the vowels and orthography. Section 4.2.3 gives an explanation for the separate consideration of the independent variables. Since formant values are continuous rather than discrete, the predictors in this model should be interpreted as having a positive or negative correlation with the response variable. In other words, a positive value for the estimate of a formant value indicates that the likelihood of the selection of /r/ increases as the formant value increases whereas a negative value tells us that the selection of /r/ is less likely as the formant value increases. In other words, a positive value means that a higher formant value makes /r/ more likely and a negative value means that a lower formant value makes /r/ more likely. For each of the time points, I conducted step-wise
ANOVA model comparisons to determine the model that best fits the data. In order to compare the two dialects, I selected the most complex model necessary in order to compare similar models. For example, if the best model for one dialect included only three formants but the optimal model for the other dialect selected the model with four formants, the model with four formants was selected for both dialects.
The output for the models for time point 1, shown in Table 48 for SJS and Table
49 for CS, show the same effect for F3. As the value of F3 increases, the likelihood of selecting /r/ decreases for listeners in both dialects. The significant effect for F2 in the
SJS data shows that SJS listeners are less likely to choose /r/ in the presence of high F2 values. There is no significant effect for F2 in the CS data.
160 Estimate Standard Error z-Value p-value (Intercept) 4.6840326 1.2093886 3.873 0.000107 F1 -0.0008775 0.0011814 -0.743 0.457604 F2 -0.0009734 0.0004752 -2.048 0.040531 F3 -0.0015948 0.0006211 -2.568 0.010235 Table 48. Logistic regression for SJS for time point 1. Input value= /r/.
Estimate Standard Error z-Value p-value (Intercept) 5.6909276 1.1596928 4.907 9.24e-07 F1 -0.0003292 0.0010949 -0.301 0.764 F2 0.0000338 0.0004457 0.076 0.940 F3 -0.0028098 0.0005983 -4.696 2.65e-06 Table 49. Logistic regression for CS for time point 1. Input value= /r/.
At the second time point, like in time point 1, both groups have significant effects for F3 values in that /r/ is disfavored as F3 increases. CS listeners have significant effects for F1 and F2 as well - as these values increase, the likelihood of hearing /r/ also increases. Tables 50 and 51 show these results.
Estimate Standard Error z-Value p-value (Intercept) 6.8094528 1.2379004 5.501 3.78e-08 F1 0.0021634 0.0014578 1.484 0.138 F2 -0.0004162 0.0005075 -0.820 0.412 F3 -0.0035471 0.0007413 -4.785 1.71e-06 Table 50. Logistic regression for SJS for time point 2. Input value= /r/.
Estimate Standard Error z-Value p-value (Intercept) 9.4633655 1.2485287 7.580 3.47e-14 F1 0.0047884 0.0013822 3.464 0.000531 F2 0.0012528 0.0004836 2.591 0.009577 F3 -0.0063644 0.0007534 -8.448 < 2e-16 Table 51. Logistic regression for CS for time point 2. Input value= /r/.
161 The same effect for F3 in both groups from the previous two time points persists for time point 3. Here, increasing F1 values significantly increase the likelihood of choosing /r/ for both groups. In time point 2, this effect was only significant for CS. At this time point, there is a significant effect for F2 among SJS listeners in that as F2 increases, the likelihood of selecting /r/ decreases. The same effect is not significant in the CS group. The results for time point 3 are shown in Tables 52 and 53.
Estimate Standard Error z-Value p-value (Intercept) 2.9304108 0.9035575 3.243 0.00118 F1 0.0034391 0.0012422 2.769 0.00563 F2 -0.0016787 0.0005767 -2.911 0.00360 F3 -0.0013419 0.0005171 -2.595 0.00946 Table 52. Logistic regression for SJS for time point 3. Input value= /r/.
Estimate Standard Error z-Value p-value (Intercept) 2.9748810 0.8744653 3.402 0.000669 F1 0.0053963 0.0011742 4.596 4.31e-06 F2 -0.0003789 0.0005410 -0.700 0.483687 F3 -0.0026208 0.0005122 -5.117 3.10e-07 Table 53. Logistic regression for CS for time point 3. Input value= /r/.
In the step-wise model comparisons for time point 4, F4 was useful for the variation in the CS data, but not in the SJS data. In order to compare SJS and CS, I chose the models including F4 for both groups in order to avoid comparing models with different terms in them. The effect for F3 persists at this time point for both groups. The likelihood for /r/ to be selected significantly more frequently than /l/ with an increase in
F1 is also present for both groups. An effect for F2 wherein the increase of values for this formant makes the selection of /r/ less likely is significant in the SJS data, but not in the
162 CS data. Finally, an inverse effect for F4 exists in the CS data, but not in the SJS data.
Tables 54 and 55 show the results for the two groups at time point 4.
Estimate Standard Error z-Value p-value (Intercept) 2.747e+00 9.188e-01 2.989 0.0028 F1 2.934e-03 1.218e-03 2.409 0.0160 F2 -2.158e-03 5.202e-04 -4.148 3.36e-05 F3 -8.236e-04 3.920e-04 -2.101 0.0356 F4 -1.506e-05 1.842e-04 -0.082 0.9348 Table 54. Logistic regression for SJS for time point 4. Input value= /r/.
Estimate Standard Error z-Value p-value (Intercept) 1.9992717 0.8651952 2.311 0.020845 F1 0.0038239 0.0011344 3.371 0.000749 F2 -0.0008318 0.0004800 -1.733 0.083069 F3 -0.0008898 0.0003670 -2.425 0.015315 F4 -0.0004409 0.0001720 -2.564 0.010355 Table 55. Logistic regression for CS for time point 4. Input value= /r/.
For time point 5, the same effects for F3 and F1 are found in both groups. The effect for F2 found in time points 3 and 4 for the SJS group persists here. The same effect for F4 for the CS group persists here from time point 4. The output from the models for time point 5 are shown below in Table 56 and Table 57.
Estimate Standard Error z-Value p-value (Intercept) 2.9096501 1.0549493 2.758 0.00581 F1 0.0032657 0.0011138 2.932 0.00337 F2 -0.0014298 0.0005761 -2.482 0.01307 F3 -0.0010056 0.0004477 -2.246 0.02469 F4 -0.0002646 0.0002076 -1.275 0.20236 Table 56. Logistic regression for SJS for time point 5. Input value= /r/.
163 Estimate Standard Error z-Value p-value (Intercept) 3.2819709 1.0142440 3.236 0.001213 F1 0.0039273 0.0010485 3.746 0.000180 F2 0.0002913 0.0005426 0.537 0.591421 F3 -0.0017736 0.0004285 -4.139 3.49e-05 F4 -0.0006597 0.0001934 -3.411 0.000647 Table 57. Logistic regression for CS for time point 5. Input value= /r/.
The same effects for F3 and F1 persist at time point 6 for both groups. The same effect for F2 for SJS also persists. At this time point, F4 is significant for both groups.
For both SJS and CS listeners, as F4 at this time point increases, so does the likelihood of choosing /r/. This is the opposite effect from the decreased likelihood of choosing /r/ as
F4 increases for the CS group in time points 4-5. Tables 58 and 59 show the results for the two groups at time point 6.
Estimate Standard Error z-Value p-value (Intercept) 3.0871635 1.0429156 2.960 0.00308 F1 0.0024130 0.0009229 2.615 0.00893 F2 -0.0015178 0.0005148 -2.949 0.00319 F3 -0.0023484 0.0004542 -5.170 2.34e-07 F4 0.0007201 0.0002224 3.238 0.00120 Table 58. Logistic regression for SJS for time point 6. Input value= /r/.
Estimate Standard Error z-Value p-value (Intercept) 2.8230240 1.0027575 2.815 0.00487 F1 0.0024095 0.0008503 2.834 0.00460 F2 -0.0001203 0.0004722 -0.255 0.79886 F3 -0.0028876 0.0004337 -6.658 2.77e-11 F4 0.0005813 0.0002050 2.836 0.00458 Table 59. Logistic regression for CS for time point 6. Input value= /r/.
At time point 7, the same effects for F3 and F1 persist for both SJS and CS listeners. The same significant effect for F2 for SJS listeners, which has been present
164 since time point 3, also persists. The effect for F4 continues in the same direction for both groups, but is slightly above the alpha-value (p > 0.05) for SJS. Tables 60 and 61 display the models for time point 7 for SJS and CS.
Estimate Standard Error z-Value p-value (Intercept) 0.0987779 1.1377790 0.087 0.93082 F1 0.0030288 0.0010176 2.976 0.00292 F2 -0.0013211 0.0004607 -2.868 0.00414 F3 -0.0008675 0.0003568 -2.431 0.01504 F4 0.0004131 0.0002113 1.955 0.05056 Table 60. Logistic regression for SJS for time point 7. Input value= /r/.
Estimate Standard Error z-Value p-value (Intercept) 0.3028769 1.0802128 0.280 0.77918 F1 0.0016500 0.0009205 1.793 0.07304 F2 -0.0007235 0.0004252 -1.701 0.08887 F3 -0.0013560 0.0003376 -4.017 5.9e-05 F4 0.0006213 0.0001981 3.136 0.00171 Table 61. Logistic regression for CS for time point 7. Input value= /r/.
In order to summarize and help visualize the results for the regression models at the seven time points, the following tables present the findings for the likelihood of liquid selection (/r/ vs. /l/) for each of the formants at each of the time points. For each time point, the direction of effect is noted for both dialects. The liquid sound (/r/ or /l/) preferred for higher values of each formant for each time point is presented in Tables 62,
63, 64, and 65 for F1, F2, F3, and F4 respectively.
165 1 2 3 4 5 6 7 Puerto not not /r/ /r/ /r/ /r/ /r/ Rico significant significant Spain not /r/ /r/ /r/ /r/ /r/ /r/ significant Table 62. Findings for which liquid is perceived with higher values of F1 across seven time points
1 2 3 4 5 6 7 Puerto /l/ not /l/ /l/ /l/ /l/ /l/ Rico significant Spain not /r/ not not not /l/ not significant significant significant significant significant Table 63. Findings for which liquid is perceived with higher values of F2 across seven time points
1 2 3 4 5 6 7 Puerto /l/ /l/ /l/ /l/ /l/ /l/ /l/ Rico Spain /l/ /l/ /l/ /l/ /l/ /l/ /l/ Table 64. Findings for which liquid is perceived with higher values of F3 across seven time points
1 2 3 4 5 6 7 Puerto not not not not not /r/ not Rico significant significant significant significant significant significant Spain not not not /l/ /l/ /r/ /r/ significant significant significant Table 65. Findings for which liquid is perceived with higher values of F4 across seven time points
The most obvious result from the formant analysis is the reliance on F3 to determine rhoticity. Both dialects use this formant in the same way throughout the duration of the vowel+liquid sequences such that a lower F3 is more likely to be perceived as /r/. Findings for F1 are also consistent for both CS and SJS listeners from
166 time points 3 through 7. Both F1 and F3 results align with the results for the production study in that F1 values for vowel+/r/ are higher than those for vowel+/l/ and for F3 vowel+/r/ values are consistently lower than F3 measurements for vowel+/l/.
The results for the perception of vowel+liquid sequences in terms of F2 are interesting in that the SJS group seems to be utilizing F2 to distinguish /r/ from /l/, whereas only a couple of the time points are significant for the CS group. This finding provides a possible explanation for the findings in the regression models conducted with vowel and orthography as the independent variables (Table 47). The interaction terms in this model for the vowel /a/ are unsurprising: SJS listeners are significantly more likely to identify stimuli spelled with as /or/. This tendency is not significant in the CS group, but the results for the SJS group are robust (p < 0.01). On the surface, it appears that SJS listeners are worse than Spaniards at identifying liquid sounds after /o/, since SJS listeners misidentify orthographic /r/ and /l/ while there are no significant findings for /o/ in either direction for the CS listeners. However, a closer look at the data suggests that this result may be explained by a confluence of SJS listeners' sensitivity to F2 values for determining rhoticity and laterality, and the atypical F2 values of the /o/ stimuli in the experiment.
167 The results from the perception of F2 for all of the vowels taken together show that SJS listeners choose /l/ significantly more than /r/ with higher F2 values. There is only one stimulus in the data with the vowel /o/ that is spelled with /l/, and the F2 values for this stimulus are lower than those of the 9 /or/ stimuli. Since the production of liquids after /o/ in SJS tend to have higher F2 values for /ol/ than for /or/, it is not surprising that the participants in the perception study behaved in the opposite direction than the orthography would suggest. In other words, the stimuli with /o/ for the particular speaker chosen for the perception experiment run counter to the general trend in SJS. In essence, the SJS listeners were accurately perceiving the general trends for the pronunciation of their dialect's orthography, which led them to select /or/ and /ol/ for the way they are usually pronounced, which runs counter to the speech used in the stimuli.
The results for F2 for SJS listeners, as will be discussed in more detail in Chapter
5, are similar to the spline models, but contrast with the regression models from the production results. That is, lower F2 values correspond to rhotic percepts for these listeners. The findings for the perception experiment for F4, like the findings for its role in liquid production in Chapter 3, are less substantial than those of the other formants.
The connections between liquid perception and production will be explored in depth in
Chapter 5.
168
CHAPTER 5: DISCUSSION
5.1 PREDICTORS OF CODA LIQUID PRODUCTION
This section discusses the results of the coda liquid production study both in terms of overall findings (sections 5.1.1-5.1.3) and the patterns according to particular linguistic and extralinguistic contexts (sections 5.1.4-5.1.5). The purpose of this discussion is to identify where speakers of San Juan Spanish (SJS) neutralize coda /r/ and /l/ and where they are kept distinct. Furthermore, the means of distinction and neutralization for approximant liquids, i.e. the acoustic correlates studied (F1, F2, F3, F4, and duration), are explored in depth.
5.1.1 Summary of Manner of Articulation Results
The majority of coda liquids in SJS are realized as approximants. However, while the articulation of orthographic /l/ segments are nearly categorically approximants
(96.6%), orthographic /r/ has a much more variable production with approximant realizations constituting 75.5% of rhotics. The likelihood for SJS speakers to display approximant articulations of /r/ over other pronunciations (taps, fricatives, deletions) given the variables of stress, word position, vowel, preceding sound, following sound, gender, and age was tested using logistic regression analysis (see section 3.4.1.2.1). The analysis revealed that stress, vowel, and preceding and following sounds are important in 169 determining whether or not orthographic /r/ is produced as an approximant. Stressed syllables favor approximant rhotics, whereas unstressed syllables disfavor this articulation. The high vowels, /i/ and /u/, disfavor approximant articulations. On the other hand, the vowel category for glide+/e/ favors approximant realizations over other rhotics.
However, this finding for glide+/e/ could be a frequency effect due to the words Puerto
Rico and puertorriqueño ('Puerto Rican'), which tend to be pronounced as approximants.
The implications of these findings for stress and vowel will be discussed in more detail in section 5.4. In terms of the following sound context, approximant /r/ is dispreferred before fricatives and vowels, which are precisely the environments where /r/ tends to be produced as a fricative or a tap, respectively. All other environments except for pre- pausal contexts favor approximants rhotics. In the context of a pause, either approximant or other articulations have a similar chance of occurring. Preceding nasals prefer approximant articulations and preceding fricatives disfavor approximant rhotics. Tap, fricative, and deleted rhotics did not constitute enough data to consider regression analyses. The inference trees for these manners of articulation suggest that a number of conditioning effects may be in play, including gender, which will be discussed in section
5.1.5. However, more data for these manners of articulation would be necessary to conduct more robust statistical tests for their conditioning.
5.1.2 Summary of Duration Results
There is not a significant effect for duration in the present study when all of the linguistic and social contexts for the approximant liquid sounds are taken together. That is, overall, vowel+/r/ sequences are slightly shorter than vowel+/l/ at 132.158 ms and 170 135.639 ms respectively, but this difference is not significant. Most descriptions of rhotic and lateral sounds (such as Lagefoged & Maddieson 1996) posit that lateral sounds tend to be significantly longer than rhotics because of the greater degree of closure associated with the lateral tongue gesture. However, these descriptions usually compare approximant laterals with tap /r/, which is more common across languages than approximant productions of /r/.
As will be discussed in the independent variable sections for vowel (5.1.4.2), preceding sound (5.1.4.3), and following sound (5.1.4.4), significant differences in duration exist between the liquid sounds for certain factor levels of these three linguistic variables. Despite the number of significant differences in formant structure across word positions, no effects for duration were found for this variable. The factor levels for the social variables of gender and age also do not differ significantly from the overall means for vowel+/r/ and vowel+/l/ sequences.
The overall findings for this study contrast with those from Simonet et al.'s (2008) report of significant durational differences between the liquid sounds in the speech of their three participants from San Juan. In their results, vowel+/l/ sequences are longer than vowel+/r/ sequences, making them incompletely neutralized in terms of duration.
The difference in results between Simonet et al.'s study and the present investigation may be due to the fact that Simonet et al. excluded tokens that the second author, a native SJS speaker, deemed to be totally neutralized. The tokens analyzed for incomplete neutralization in Simonet et al.'s (2008) study were therefore pre-screened to exclude the most neutralized instances of coda liquids and is not as representative of what is happening across all approximant liquids as is the present study. The authors also
171 mention that vowel plays a role in the duration of the vowel+liquid sequences for one of their speakers, but it appears that they did not look at the interaction between vowel and the difference between vowel+/r/ and vowel+/l/ sequences. The mean durations for /ar/ and /ir/ are shorter than those for /al/ and /il/ in their study for all three speakers. The finding for /a/ goes along with that for the present study in that /ar/ sequences are shorter than /al/ sequences, albeit insignificantly. The results for /ir/, however, are surprising, since these sequences are significantly longer than /il/ sequences in the present study.
In his study of liquid switching in Ponce, Luna (2010), did not find effects for duration.
5.1.3 Summary of Results by Formant
The current study, similarly to the two previous acoustic studies on coda liquids in
SJS (Simonet et al. 2008, Luna 2010), found significant results for the differences between orthographic /r/ and /l/ in terms of F3, whereby a lower F3 correlates with orthographic /r/. The three participants for San Juan, as well as the participant from
Mayagüez, in Simonet et al.'s (2008) study showed significant differences between vowel+/r/ and vowel+/l/ for one or both of the two vowels, /a/ and /i/, that they considered in the analysis.26 The lower value for F3 for vowel+/r/ than vowel+/l/ sequences in their study was the same as the overall effect for the twenty-four speakers in the production analysis of this study. Luna (2010) also found this effect for the Spanish of
Ponce when considering the average value of /r/ and /l/ over the duration of the liquid
26 Findings were significant for all four speakers in the regression analysis that the authors conducted by taking the average value of time points 4 through 6 for both vowels in the study, /a/ and /i/, taken together. The spline analyses, which are conducted by vowel and by considering the speakers separately, show a somewhat distinct picture. The confidence intervals overlap for one of the speakers for both vowels and for another speaker for the /i/ vowel, indicating that these sounds may not be different when the time points are considered separately. 172 segments. A robust effect for F3 was also found for the perception experiment in the present study (see section 5.2 for a summary and discussion of the perception results). As will be discussed in section 5.1.5, F3 is the only formant in this study that shows significant differences between the liquids for the two age groups.
The finding in the present study that F1 values for vowel+/r/ tend to be somewhat higher than those for vowel+/l/ is corroborated in Simonet et al.'s work in that there are significant differences between vowel+/r/ and vowel+/l/ in the speech of all four of their informants in the regression analyses.27 In the present analysis, none of the time points featured a significant distinction between the liquid sounds. However, the overall direction of the two sounds is the same as that in Simonet et al.'s study. The results of the perception experiment for listeners from both SJS and CS listeners also align with the findings that a higher F1 value distinguishes vowel+liquid sequences as being spelled with /r/.
The role of F2 in the production of coda liquids in SJS is less straightforward than the behavior of F3 and F1. In the regression models for this formant in the current study,
F2 values for vowel+/r/ are significantly higher than those of vowel+/l/ at the first two time points. Values for vowel+/r/ for the rest of the trajectory given by the regression models continue to be higher than those of vowel+/l/, though not significantly so, until the very end of the liquid segments, where the two sounds are practically identical. The results from the spline analysis for the two liquid sounds overall, without consideration of any of the independent variables, show precisely the opposite picture. In this analysis, vowel+/l/ sequences are consistently higher than vowel+/r/ until the end of the trajectory,
27 These analyses are like those for F3 in that they average across time points. In the spline analysis, the results appear to be neutralized for one of the speakers for both vowels. 173 where the two sounds converge, like in the regression model for time point 7. The spline analysis of F2 for word position shows the same overall pattern for word-final vowel+liquid sequences as the overall spline model, whereas the word-medial liquids have higher F2 values for sequences with /r/ than /l/, like the regression models. The results from the perception experiment for the present study show that SJS listeners tend to identify stimuli with low F2 values as /r/, which goes along with the splines for overall values and word-final contexts. In their production study, Simonet et al. (2008) have somewhat sporadic results for F2 in that one of the speakers shows differences between the liquids in the spline models for one of the vowels and another speaker presents differences in the regression model. The authors are dismissive of both of these findings and assert that F2 does not play a role in coda liquid differentiation in SJS.
A possible explanation for the difference in the results in the present study for F2 in the regression analysis vis-à-vis the splines and perception analyses is the effects from previous sounds and the higher frequency of certain consonants than others before vowel+liquid sequences in the Spanish language (see section 3.4.2.3.2). Stops, fricatives, and nasals make up the majority of the sounds preceding both vowel+/r/ and vowel+/l/ sequences in the data set (about 79% and 70% respectively), and by extension, probably in the language as a whole. The interactions in the regression model select these preceding sounds as instances in which the difference between /r/ and /l/ and the direction of effect is the reverse of the overall predictions for the model. That is, while the F2 values for vowel+/r/ are greater than those for vowel+/l/ for most preceding contexts
(pauses, approximants, laterals, and rhotics), F2 values for a minority of the preceding contexts (stops, fricatives, and nasals) account for the greatest number of instances of
174 vowel+liquid sequences in the data. Regression models are designed to make predictions about values based upon all of the possibilities presented by the independent variables while controlling for uneven representations of factor levels in the data set. Therefore, frequency effects can be lost in the regression model, as seems to be the case for F2 in the present study. The results from the perception experiment seem to suggest that SJS listeners are sensitive to the frequency of contexts where F2 values are lower for vowel+/r/ sequences than for vowel+/l/ sequences, since this is precisely the context in which SJS listeners favor the selection of /r/. Listeners from Spain, by contrast, do not seem to be sensitive to F2 differences in their evaluation of vowel+liquid sequences.
In the present study, there are no significant overall findings for F4 at any of the time points and there are fewer significant findings for particular contexts than there are for the other three formants. Furthermore, the results for the perception experiment are inconsistent across the time points. Taken together, the results from the perception and production of F4 do not seem to imply that this formant is particularly important in distinguishing liquid sounds in this dialect. This is counter to Luna's (2010) claim that F4 plays a role in differentiating the liquid sounds in Puerto Rican Spanish. In his study, F4 values of /r/ were significantly lower than those for /l/ when the average measurement across the formant was taken.
5.1.4 Linguistic Conditioning Factors
The following subsections review the results for the linguistic conditioning factors of word position (5.1.4.1), vowel (5.1.4.2), preceding sound (5.1.4.3), and following sound (5.1.4.4). 175 5.1.4.1 Word Position
The position within words of vowel+liquid sequences stands out as a robust predictor for whether or not sequences with orthographic /r/ and /l/ produced as approximants are distinct from each other or neutralized. Other manners of articulation of
/r/ in this study do not appear to be affected by word position. This finding is similar to that of López Morales (1983b), who finds fricatives and deletions in his data on San Juan at about the same rate in word-medial and final contexts. For approximant liquids, the interaction term between liquid type (/r/ vs. /l/) and word position (word medial vs. word final) was significant in fifteen of the twenty-eight models performed for the seven time points for each of the four formants. A significant interaction means that the difference between vowel+/r/ and vowel+/l/ for at least one word position is significantly different from the overall difference between the two liquid sounds. In eight of the fifteen significant time points, t-tests showed that vowel+/r/ and vowel+/l/ were significantly different in one word position and were not significantly different in the other word position. That is, in one of the two word positions, the liquid sounds are neutralized in terms of a particular formant, whereas in the other, they are distinct. Table 66 below shows the formants and time points where this is the case.
176 Formant Time Overall Direction Word-Medial Word-Final Points of Difference
F1 3 /r/ > /l/ /r/ > /l/ /r/ < /l/ 4 Liquids are distinct. F1 values of liquids 6 are neutralized. 7 F2 2 /r/ > /l/ /r/ > /l/ /r/ > /l/ 3 Liquids are distinct. F2 values of liquids are neutralized. F3 4 /r/ < /l/ /r/ < /l/ /r/ < /l/ 6 F3 values of liquids Liquids are distinct. are neutralized. Table 66. Word position effects for the time points at which vowel+/r/ and vowel+/l/ are neutralized in one position and distinct in the other.
In other cases, word position was selected as significant because the directionality of the liquid sounds was different depending on word-position. Table 67 shows the time points where this happens without a significant difference between /r/ and /l/ in either word position. Since the differences between /r/ and /l/ within the word positions are not significant, /r/ and /l/ can be said to be nearly-neutralized in these time points within both word positions. However, it is interesting that word position affects the directionality of
/r/ and /l/. For example, F2 values at time point 7 are higher for /r/ than /l/ word-medially, but lower word-finally. Although the actual distance between /r/ and /l/ is not significant in either of the word positions, the change in relationship between the two is significant to the overall model.
177
Formant Time Points Overall Direction of Difference Word-Medial Word-Final
F2 7 /r/ < /l/ /r/ > /l/ /r/ < /l/
F3 2 /r/ > /l/ /r/ > /l/ /r/ < /l/ 3 /r/ < /l/ 7 F4 2 /r/ < /l/ /r/ > /l/ /r/ < /l/ Table 67.Word position effects for the time points at which vowel+/r/ and vowel+/l/ neutralized in both positions, but have opposite directionality
There are two significant word position interactions that do not fit into either of the patterns in the above two tables. The results for F1, time point 5, like those in Table
67 are not significant for either word-medial or word-final position. But, the directionality is the same for both positions; F1 values for /r/ are higher in both contexts.
The value for /r/ in word-final context, however, is only about 1 Hz higher than that for
/l/, whereas in word-medial context, /r/ is about 20 Hz higher. The first time point for F2 has significant differences between /r/ and /l/ in both word-medial and word-final position. F2 values for /r/ at this time point are higher than those for /l/ for both word positions, but the difference is greater for word-medial contexts, which follows the trend for this formant.
To summarize, the word position of coda liquids in SJS strongly influences the formant values for these sounds. Values for F1 are higher for vowel+/r/ than for vowel+/l/ word-medially at time points 3 through 6 and word-medial liquids are significantly different in all except for time point 5. The difference between /r/ and /l/ in terms of F2, like for F1, is also greater for word-medial than word-final context, with the
F2 values for vowel+/r/ sequences higher than values for vowel+/l/. Values for F3 are
178 always lower for vowel+/r/ than for vowel+/l/ in word-final position and the liquids are more distinct in this position, unlike for the first two formants where word-medial liquids are more distinct. In summary, in word-medial position, vowel+/r/ and vowel+/l/ are differentiated in terms of F1 and F2, and word-finally the difference between the two lies in the F3 value.
This study is the first acoustic approach to the role of word position in the realization of coda liquids in SJS. Previous impressionistic studies claim that word-final contexts are the more frequent locus of neutralization (López Morales 1983a, Ramos
Pellicia 2007). The results from the present study contribute a more nuanced view to how liquids in this dialect behave - word-final contexts are indeed more neutralizing in terms of F1 and F2, but in terms of F3, liquids in word-medial context are more neutralized. In terms of articulation this means that in word-medial contexts, vowel+/r/ is more lowered and fronted than vowel+/l/ and word-finally, the height and frontness is similar between the liquids and the difference lies in a more retracted tongue root for vowel+/r/.
5.1.4.2 Vowel
This is the first acoustic study to examine the effect of all of the vowels in the
Spanish language on liquid switching in SJS. Liquid sounds are unique in that they possess both consonant-like and vowel-like qualities, more precisely they have some degree of constriction but have clear, continuous formant structure. Some researchers posit that liquid sounds are formed by a vowel-like articulation produced with the tongue dorsum along with a consonant-like gesture of the tongue tip (Proctor 2009, Sproat &
Fujimura 1993). In a study that utilizes MRI imaging to examine liquid-inserting dialects 179 of English, Gick et al. (2002) find that the tongue dorsum shapes for the vowels after which /r/ and /l/ are inserted (/ɔ/ and /ʌ/ respectively) are virtually identical to those of the liquids that get inserted. It seems, then, that the shape of the tongue body for the vowel could have a great deal of influence on the outcome of coda liquid production. In fact, the results for the independent variable of vowel in the present study strongly support this view. The findings for vowel in this study show that whether vowel+/r/ or vowel+/l/ sequences have higher formant values is strongly influenced by the vowel in the sequence. This section reviews the directions of effect for both the simple vowels (/a/,
/e/, /i/, /o/, and /u/) and the diphthongs (glide+/a/ and glide+/e/) for F1, F2, and F3.28 The results for F4 did not present a significant effect for any particular vowel for more than one time point. Since the results for this formant are spurious and inconclusive, they are not discussed here. For reference, the vowel inventory of Spanish, arranged by frontness and height is provided in Figure 28 below.
28 Due to small token counts, the diphthongs in this study were collapsed into the groups glide+/a/ and glide+/e/, as explained in section 3.2.4.2. It is important to note that each category includes both /j/ and /w/ as the glide part of the vowel. The diphthongs, then, really belong to two categories at a time. The beginnings of the sequences, either /j/ or /w/, are radically different on the axis of frontness and backness, and therefore are likely to pattern with the front or back vowels depending on the frequency of each within each category. The glide+/a/ category has an equal representation of /j/ and /w/, whereas the glide+/e/ factor group has a strong majority of /w/ glides, at 72.7% of the data. Therefore, the early time points for glide+/e/ are likely to behave similarly to /u/, which is indeed found for F1. 180 High Front Back
Low Figure 28. The vowel inventory of Spanish.29
The first formant has an inverse relationship with the height of the tongue such that vowels with a high tongue position have low F1 values and low vowels have high values. Table 68 shows the overall results for the directionality of the liquid sounds according to vowel height for the five simple vowels.
Vowel Height Vowel General effect on F1 of Liquids low a /r/ > /l/ mid e /r/ > /l/ o /r/ > /l/ high i small effects /r/ > /l/ until time point 6, then /r/ < /l/ u small effects /r/ < /l/ Table 68. Direction of effect of F1 according to vowel height.
Lateral segments in Spanish generally have a lower F1 than rhotics because of the degree of closure (Martínez Celdrán & Fernández Planas 2007, Quilis 1981). This expected effect is observed in the data for the present study for low and mid vowels (/a/,
/e/, and /o/); for these vowels, F1 is lower for sequences with /l/ than /r/, meaning that sequences with /r/ present less closure. In the case of the high vowels and glide+/e/ at early time points, the directionality reverses - F1 for vowel+/r/ sequences is higher than
29 Reproduced and modified from http://www.indiana.edu/~hlw/PhonUnits/vowels.html with permission. Indiana University and Michael Gasser. 2009. Edition 3.0.
181 that for vowel+/l/. The tendency to realize vowel+/r/ sequences in a more constricted way, which is usually associated with a more /l/-like production, makes sense, given that the tongue is already high when there is a preceding high vowel. In other words, the /l/- like production is a result of coarticulation with the high vowel. The high vowels are also of interest in the manner of articulation results, since these two vowels disfavor approximant realizations to tap, deleted, and fricative articulations. In essence, orthographic /r/ sequences have more approximant /r/-like articulation with low vowels and have variable manners of articulation that generally require a higher and more front tongue position in the context of high vowels. If the vowel+/r/ sequence is realized as an approximant in the context of high vowels, the result is a more /l/-like sound.
The effects of vowel for F2 values in vowel+liquid production, summarized in
Table 69 below, align with what is expected in terms of the frontness or backness of the vowels and liquids. It is generally expected that /l/ will have a more forward pronunciation than will approximant /r/ because of the coronal occlusion of the tongue tip for /l/. The vowels produced with the tongue furthest back, /o/ and /u/, correlate with lower F2 values for vowel+/r/ than those of vowel+/l/. This direction, however, is reversed in the case of more front vowels. With these vowels, it appears that the tongue position stays more fronted, making more /l/-like rhotics. The more front the vowel, the greater this effect. The effect for F2 is much like that for F1 in that the more similar the vowel is to the tongue position associated with /r/, the more /r/-like the production of sequences with orthographic /r/ is. In summary, vowel has an coarticulatory effect on orthographic /r/ segments, such that the height and frontness of the vowel in the sequence
182 determines where the rhotic will be produced and thus vowel plays a role in determining whether the rhotic will sound more /r/-like or /l/-like.
Vowel Frontness Vowel Effect on F2 front i /r/ > /l/ mid-front e /r/ > /l/, but less so than /i/ central a /r/ > /l/, but less so than /e/ back o /r/ < /l/ u /r/ < /l/ Table 69. Direction of effect of F2 according to vowel frontness
At most of the time points, F3 values for vowel+/r/ are lower than those from vowel+/l/, which is expected. The third formant is considered to be an acoustic result of tongue bunching such that more retraction of the tongue results in lower F3 values. All of the vowels except for /i/ follow this pattern of having a lower F3 value for vowel+/r/ than vowel+/l/ sequences. The overall trend is especially prominent in the mid vowels (/e/ and
/o/), in which sequences with /r/ are dramatically lower than sequences with /l/. The degree of bunching of the tongue is likely related to the position of the tongue for the vowel. Thus, the higher, more fronted position of
183
Vowel Effect on F3 i /r/ > /l/ a /r/ < /l/, but less difference than /e/ and /o/ u /r/ < /l/ e /r/ < /l/ o /r/ < /l/ Table 70. Direction of effect of F3 by vowel.
Vowel also plays a role in the results for duration for the directionality of vowel+/r/ and vowel+/l/. In the context of glide+/e/ and /e/, sequences with orthographic
/r/ are significantly longer, whereas sequences with orthographic /l/ were longer with glide+/a/ and /i/ vowels. These effects for duration do not correlate with the patterns seen for the first and second formants. That is, the liquid and vowel combinations with more or less closure or fronting do not correspond with the vowel+liquid sequences with durational differences. One possible explanation is some effect of F3 by which vowel+/r/ sequences with more bunching are longest whereas the vowel+/l/ sequences without bunching represent the longest sequences of this type.
5.1.4.3 Preceding Sound
To my knowledge, this is the first study to consider the effects of the sound preceding vowel+liquid sequences in SJS. The effects found in this study for preceding and following sounds on the production of coda liquids can be seen as opposite to those of the vowel sounds. That is, liquid sounds in SJS assimilate to the vowel in their environment, as explained in the previous section, but dissimilate from the surrounding consonantal sounds. Table 71 gives an overview of the directionality of vowel+/r/ and vowel+/l/ for F1, F2, and F3 according to previous consonant. To the right of the 184 directionality of the liquids for each formant, there is a label ("/r/-like" or "/l/-like") that indicates whether the orthographic /r/ is behaving in a more rhotic or lateral way. That is,
"/r/-like" describes an open, back, bunched production (high F1, low F2, low F3 value) whereas "/l/-like" is a more closed, front, less bunched production (low F1, high F2, high
F3). Orthographic /r/ preceded by sounds with the most occlusion - stops, fricatives, and nasals - have more open, back, bunched productions than vowel+/l/ sequences. In other words, the most /r/-like sounds in this dialect occur in the context of constrictive preceding consonants with full or critical closure. Preceding pauses, laterals, and rhotics have the same direction of effect for openness and bunching, but vowel+/r/ in these contexts is more fronted, i.e. has a higher F2, than vowel+/l/. The environment of preceding approximant is particularly interesting in that vowel+/r/ sequences are /l/-like in all three dimensions - vowel+/r/ is more closed, fronted, and unbunched than vowel+/l/ in this position.
185 Preceding F1 F2 F3 Sound nasal /r/ > /l/ /r/- /r/ < /l/ /r/-like /r/< /l/ /r/-like like stop /r/ > /l/ /r/- /r/ < /l/ /r/-like /r/< /l/ /r/-like like fricative /r/ > /l/ /r/- /r/ < /l/ /r/-like /r/< /l/ /r/-like like (pause) /r/ > /l /r/- /r/ > /l/ /l/-like /r/< /l/ /r/-like like lateral /r/ > /l/ /r/- /r/ > /l/ /l/-like /r/< /l/ /r/-like like rhotic /r/ > /l/ /r/- /r/ > /l/ /l/-like /r/< /l/ /r/-like like approximant /r/ < /l/ /l/- /r/ > /l/ /l/-like /r/ > /l/ /l/-like like Table 71. Effects of preceding sound on formant structure. The labels '/r/-like' and '/l/- like' describe the behavior of orthographic vowel+/r/ sequences as expected from descriptions of approximant liquids.
The overall pattern of liquid behavior with respect to preceding sounds is dissimilatory, but the acoustic parameters for dissimilation and the degree of difference between vowel+/r/ and vowel+/l/ vary according to the specific environment. Preceding nasal sounds have effects for duration and all of the formants. In addition, nasal sounds are interesting in that they are the only preceding sound selected in the manner of articulation results for preferring approximant realizations of /r/ over productions of taps, deletions, and fricatives. These sounds are also the only preceding sound with an effect for duration; in the case of other preceding sounds, vowel+/l/ sequences are slightly, but insignificantly, longer than vowel+/r/ sequences. With a preceding nasal, the directionality is reversed; sequences with /r/ are nearly 12% longer than sequences with
/l/. The effects of preceding nasals on F1 are fairly small, but are interesting because nasals are the only preceding sound with significant results for the first two time points
186 for this formant. For F3, vowel+/r/ sequences after nasals have a lower value than vowel+/l/ sequences, like the overall direction for this formant in most preceding contexts. The strongest effect for nasals is on F2, where the values for vowel+/r/ sequences are around 100 Hz lower than those for vowel+/l/ sequences, which correlates with a more back articulation of rhotics.
Like with preceding nasals, in the contexts of stops and fricatives F2 values for vowel+/r/ are significantly lower than those for vowel+/l/, which correlates with a more back articulation for rhotics. In the case of preceding sounds with less closure than nasals, stops, and fricatives, more anterior articulations of rhotics are found. Preceding lateral and rhotic sounds exhibit a more fronted vowel+/r/ than vowel+/l/ articulation, which is evinced by the higher F2 value for vowel+/r/ sequences. In the context of preceding approximants, vowel+/r/ and vowel+/l/ sequences are fairly similar in terms of F1 and F2.
The largest difference between the vowel+liquid sequences in this environment lies in F3.
After approximants, vowel+/r/ sounds run counter to the overall trend of a low F3 and they exhibit a higher F3 than vowel+/l/ sequences, which correlates with a lack of the tongue bunching usually associated with rhoticity.
5.1.4.4 Following Sound
The present study is the first acoustic approach to consider the effect of a full range of following sounds on the production of coda liquids in SJS. Simonet et al. (2008) limited the following contexts in their study to the voiceless stops /p/, /t/, and /k/ and found no significant differences in duration, F1, F2, or F3 for vowel+liquid sequences with the different places of articulation of the consonants. Luna (2010) does not consider 187 following sound in a rigorous way, but does comment that lateralization occurs even when the liquid is word-final and followed by a vowel. In the present study, following sounds only have significant results for approximant realizations for the first formant, meaning that the directionality of vowel+/r/ and vowel+/l/ is only influenced in the dimension of tongue height and the subsequent open or closed jaw position that aligns with the tongue dimension. Table 72 outlines the directions of effect for the six following sound contexts.
Following Sound F1 stop /r/ > /l/ /r/-like fricative /r/ > /l/ /r/-like nasal /r/ > /l/ /r/-like (pause) /r/ > /l/ /r/-like approximant /r/ < /l/ /l/-like vowel /r/ < /l/ /l/-like
Table 72. Effects of following sound on formant structure. The labels '/r/-like' and '/l/- like' describe the behavior of orthographic vowel+/r/ sequences as expected from descriptions of approximant liquids.
Following stops, fricatives, and nasals show a similar pattern to that found for these same sounds preceding the vowel+liquid sequences in that vowel+/r/ sounds are more open than vowel+/l/ sounds. In the context of following approximants and vowels, which have more open productions or realizations, coda sequences with /r/ have a higher
(more closed) tongue configuration than sequences with /l/. The pattern for following sounds in relation to coda /r/ sequences, then, is dissimilatory. The only following context that does not follow this pattern is that for pauses, for which /r/ has a lower articulation for vowel+/r/ than vowel+/l/. Without a following context, as is the case for pauses, the
188 tongue position for sequences with coda /r/ is low, much like we saw in the case of the liquid behavior in accordance with low vowels.
In terms of duration, following vowels and approximants have significant effects on the difference between sequences with coda /r/ vs. /l/. Before vowels, sequences with
/r/ are significantly shorter than vowel+/l/, whereas this direction is reversed when there is a following approximant. The more rapid tongue movement for /r/ before vowels may indicate that the gesture is more tap-like than approximant /r/ sounds in other contexts.
The finding in this study that taps are more prevalent in prevocalic than other contexts lends support to this hypothesis. On the other hand, in the context of a following approximant, the direction of the effect is the opposite such that vowel+/r/ sequences are longer than vowel+/l/ sequences. The opposite effect for duration for these two contexts is interesting since these contexts have the same direction of effect on the F1 values of the liquids. Thus, the high tongue position for prevocalic contexts, which is usually associated with /l/-like approximants could be viewed as a more tap-like, and therefore more rhotic production, in terms of duration. This observation underlines the importance of considering multiple acoustic dimensions in determining whether two sounds are neutralized. The degree of neutralization of liquid sounds is greater for following approximants than vowels, since both closure and duration are more /l/-like in the approximant context.
In his impressionistic study, López Morales (1983b) found pre-vocalic contexts to be less lateralizing than all other following sound environments. He might have perceived this because of the effect of following vowel on duration. His observations about manner of articulation for /r/ run counter to the findings of the present study. Although following
189 vowel contexts disfavor the approximant manner of articulation of rhotics, they do not favor fricative realizations like López Morales claims. In the present study, taps are more frequent in this context than fricative /r/, which tend to precede other fricative sounds.
The findings for approximant liquids in this study also contradict Shouse de Vivas's
(1978) findings that following stops are most likely to cause lateralization, followed by nasals, and then fricatives. None of these sounds showed an effect in the acoustic analysis for the formants or duration for these following sounds.30
5.1.5 Extra-linguistic factors and the social status of liquid neutralization
As discussed in Chapter 3, the two extra-linguistic factors considered in the production study are gender and age. Gender is selected as important for the conditional inference trees for rhotic manner of articulation - men more frequently delete /r/ than do women, whereas women are more likely to pronounce /r/ as a tap or fricative than are men. There is no significant result for gender for the approximant manner of articulation, meaning that the approximant production of /r/ is not preferred by one gender more than the other. In the comparison of coda /r/ and /l/ sequences within the approximant productions, gender did not have an effect on any of the acoustic measures. Past studies that have considered gender, which were all impressionistic, have mixed reports for its effect on lateralization in SJS; López Morales (1983a, 1983b) finds that men lateralize
30 Shouse de Vivas claims that following laterals are the most likely sound to cause lateralization. The present study did not consider formant and duration measurements for liquids before lateral sounds due to the impossibility of determining the end of the liquid segment under study and the beginning of the following lateral sound. This author does not consider pauses as possible following sounds, but she does consider them as a word-position option and lateralization is more common pre-pausally than in word-final contexts sentence-internally in her study. 190 more, an earlier study by Alonso & Lida (1945) cited women as having more frequent lateralization, and Medina Rivera (1995) found no effect.
The findings that women use tap and fricative pronunciations more frequently than men and that men delete /r/ more often than women are not surprising given the purported status of these variants. Taps are the normative pronunciation for coda /r/ segments across dialects of Spanish and fricatives, according to López Morales (1983b) are evaluated positively in SJS. The lack of an effect for gender in the production of approximant liquids is surprising, given the descriptions in former studies of lateralization as a stigmatized pronunciation (López Morales 1983a, 1983b; Ramos
Pellicia 2007). Sociolinguistic theory holds that in the instance of a nonstandard speech feature that is not undergoing change, men have a higher frequency of these forms, whereas changes in progress tend to be led by women (Labov 1990). The lack of an effect for gender in this study, then, could be interpreted as evidence that liquid neutralization is neither a stable stigmatized variant nor a change in progress in San Juan.
The findings for age, however, may challenge this notion.
The present study finds that younger people lateralize less than older people, which matches López Morales's (1983b) findings that younger people disfavor lateralization. Specifically, younger speakers have lower F3 values for orthographic /r/ sequences than do older speakers. Two of the time points in the regression models for F3 reveal that younger speakers have a significant difference between the liquid sounds, whereas older speakers neutralize the liquid sounds. The SS ANOVAs in Figures 29 and
30 below show the results for F3 for older and younger speakers respectively. The overlapping confidence intervals for older speakers suggests that they have more
191 neutralized sounds than younger people, who have distinct vowel+/r/ and vowel+/l/ sounds.
old /r/
old /l/
0
0
6
2
0
5
5
2
z
0
H
0
5
2
0
5
4
2
0
0
4 2 1 2 3 4 5 6 7
Time Points
Figure 29. Smoothing spline ANOVA of F3 values for vowel+/r/ and vowel+/l/ for older speakers
young /r/
young /l/
0
0
6
2
0
5
5
2
0
0
5
2
0
5
4
2
0
0
4 2 1 2 3 4 5 6 7
Figure 30. Smoothing spline ANOVA of F3 values for vowel+/r/ and vowel+/l/ for younger speakers
192 Age has been discussed from two angles in the sociolinguistic literature: 1) differences in linguistic structures with age groups can show historical change of language, 2) changes with age can be due to the linguistic choices of groups as they move through life (Eckert 1997). Younger people in San Juan, then, in the case of (1), may represent a move in the dialect away from neutralization of /r/ and /l/ and towards the more differentiated pronunciations seen in other parts of the Spanish-speaking world.
However, it is also possible that lateralization is not undergoing any sort of historical change and that the effect for age is a result of (2), which is commonly referred to in the literature as age grading (Eckert 1997). Further research would be necessary to determine whether the difference in age groups constitutes a historical change away from coda liquid neutralization or a temporary strategy during the years that are most crucial for these speakers to establish a secure career, or some combination of the two.
A closer examination of the linguistic behavior of this variable according to social class and speech situation would help determine the status of liquid neutralization as a stigmatized or socially accepted variable. Past impressionistic studies have found social class to be an important predictor in liquid neutralization; lower class men lateralize /r/ more than any other group in López Morales's (1983b) study. The current study contains mostly upper-class speakers, so the lack of effect for gender may be due to the class sample. Although the data from the reading task that the participants performed after the interview was not acoustically analyzed, my impression was that /r/ and /l/ were kept very distinct when the participants were reading. I would expect, therefore, to find that speech situation strongly affects the degree of neutralization, much like Medina Rivera
(1999) found in his study on lateralization in situations with different levels of formality.
193 The dramatic difference that I observed between the reading task and the interview for both young and old speakers alike supports the hypothesis that lateralization carries some degree of stigma. The comments that informants made about coda liquids in their dialect further supports this view. Towards the end of the interview, I asked participants about the Puerto Rican dialect both in terms of how they think it differs from other dialects of Spanish and in terms of what they like and dislike about SJS. Coda liquid switching, which participants described as "saying /l/ instead of /r/", was frequently mentioned, alongside /s/ aspiration. Most of the participants seemed to view coda liquid neutralization and /s/ aspiration negatively, but there was a sense from many of them of a sort of affection for these "bad habits", since these speech features are symbols of Puerto
Rican identity. According to other informants, these hallmarks of SJS speech, along with their complaints about poor spelling and a lack of appreciation for the Spanish language among Puerto Ricans, are signs of a backwards society that needs to be remedied by better education. Although language attitudes were not quantified as a variable for this study and my observations are entirely impressionistic, speakers with more negative attitudes towards lateralization did not seem to behave differently in their production of these sounds from participants with more positive evaluations.
5.2 MAIN FINDINGS FROM THE PERCEPTION EXPERIMENT
The overall findings for the perception experiment in Chapter 4 seem to suggest that Puerto Rican listeners are not better able to distinguish liquid spellings than Castilian
Spanish (CS) listeners, since both groups correctly identify the orthography of the stimuli at about chance. Although the overall performance of the two groups is remarkably 194 similar, SJS listeners perform significantly better than the CS group in particular contexts. The results from the regression analysis with the orthography of the liquid and the vowel in the vowel+liquid stimuli show that SJS listeners are sensitive to differences between rhotics and laterals with mid and back vowels (/a/ and /o/), whereas there was no significant effect for CS listeners. There were no differences between SJS and CS participants in their ability to identify /r/ vs. /l/ when the preceding vowel was /e/. The results from the models which consider the four formants at each time point in the stimuli as the independent variables help understand why SJS listeners better hear differences in the liquids for /a/ and /o/ but not for /e/. The lack of effect for /e/ is probably due to the higher degree of liquid neutralization that occurs with vowels produced with more anterior articulations (see section 5.1.4.2 for details).
As mentioned previously, the results for F2 from the perception experiment, in which higher values led SJS listeners to select /l/, and the regression results for the production of F2, which suggests higher overall values for vowel+/r/ sequences than for vowel+/l/ sequences, are in conflict. However, the results for the smoothing spline analysis for F2 production are in line with the perception results in that values for vowel+/l/ are higher than those for vowel+/r/. The discussion provided in section 5.1.4.3, according to which the high frequency of preceding contexts conditions F2 values for vowel+/l/ to exceed those for vowel+/r/, is a convincing explanation for what is happening in the perception of these sounds. Frequency has been shown to play a role in how sounds are produced and also how they are perceived. The finding that SJS listeners hear words with a lower F2 as rhotics is not surprising given that the production of
195 orthographic /r/ is more frequently articulated with a more posterior tongue position, which correlates with a lowered F2.
In terms of the perception of tap stimuli, SJS listeners correctly identify the tap stimuli as /r/ a little over half of the time and CS listeners perform slightly worse. Neither the SJS group nor the CS listeners perform above chance. However, when comparing within each dialect whether listeners perform better on tap than approximant productions of /r/, the SJS group performs significantly better on taps than approximants, whereas there is no significant difference between tap and approximant performance for the CS group. The poor performance of both groups may be an indication of how difficult it is to identify coda sounds without surrounding context. This finding aligns with that of
Massone (1988) who found that taps in coda are frequently confused with /l/ in their perception.
The only other research on the perception of coda liquids in SJS prior to the present study is an unpublished presentation by Paz (2005), which is discussed in
Simonet et al.'s production study (2008). For her stimuli, Paz had a phonetically-trained
SJS speaker read words that are minimal pairs with differing coda liquids such as arma
'weapon' and alma 'soul'. The SJS listeners in her experiment correctly identified the coda liquids according to their orthography at a rate well above chance, whereas her control group of Argentinian listeners performed at chance. In the present study, the participants in both groups perform about at chance overall. The differences in results between the two studies could be due to a number of factors. First, since the study is unpublished, it is unclear what lexical items were used as stimuli; the sample stimuli given by Simonet et al. all have /a/ as the vowel before the liquid. The results of the present study shows that
196 SJS listeners perform above chance for liquids following /a/. Second, the stimuli for the present experiment are extracted from natural speech, unlike the word-list stimuli read by a speaker who was probably not naive to the purpose of the experiment.
Despite the laboratory nature of the stimuli for Paz's (2005) experiment, the result that SJS and Argentinian listeners respond so differently to liquid stimuli is impressive.
Another important possibility to consider in comparing the findings of the two studies is that the stimuli for the present study are short vowel+liquid sequences which lack the contextual information, such as preceding and following sounds, that is present in entire words. Although it is not clear from the description of Paz's study whether or not the listeners were explicitly informed that they were listening to a SJS speaker, the use of words rather than syllables is likely to give listeners clues as to the dialect that they are listening to. In the present study, listeners knew that they were listening to language sounds from Spanish, but they were not given any information about the specific dialect for the stimuli. Also, the nature of identifying syllables is very different than the task to identify words. Since words carry meaning that syllables do not, in Paz's experiment, listeners were probably trying to discern which word the speaker wanted to say. In my study, listeners were simply asked to identify which sound they heard, which allows for more prescriptivism than did Paz's word task. In other words, SJS listeners may be able to reconstruct the orthography of the stimulus as /r/, but they perceive the speaker as
"switching" the /r/ for /l/, and thus respond as hearing /l/. It is possible that SJS listeners would perform differently, and more similarly to Paz's findings, if they were informed that the stimuli were from SJS speech and were asked how the syllable was spelled. In short, the insignificant findings for both groups in the present study for the overall ability
197 to correctly identify how sounds are spelled should not lead to the conclusion that SJS speakers have the same perception of liquid sounds as speakers of non-neutralizing dialects. The significant findings for specific vowel+liquid combinations for SJS listeners, the difference in how frequently listeners hear /r/ when listening to orthographic
/r/ vs. orthographic /l/, and the potential drawbacks of the present study should not be discounted. In conclusion, the results from this study support the claim that SJS listeners perceive coda liquids from their own dialect more accurately than listeners from other dialects.
5.3 IMPLICATIONS
5.3.1 Incomplete neutralization of coda liquids
The central question of this investigation is whether or not coda liquid sounds in
SJS are neutralized, or merged, into a single sound. To answer this question, this study considers both the production and perception of orthographic vowel+/r/ and vowel+/l/.
The production analysis in Chapter 3 and the perception experiment in Chapter 4 both point to the existence of contexts in which the two sounds are significantly distinct as well as contexts where no significant differences between /r/ and /l/ exist for the acoustic parameters explored. As Dinnson & Charles-Luce (1984) point out in their study about final stop voicing neutralization in Catalan, the bar to clear in order to claim complete neutralization is higher than that for incomplete neutralization, since the assertion that two sounds are completely merged necessarily requires that all phonetic dimensions of the two sounds must be the same. I would add that in order to be sure that neutralization is complete, all possible independent linguistic and extralinguistic variables in the 198 production and perception of these sounds would have to be considered, and the existence of any significant differences in the two sounds would be evidence to refute complete neutralization. This study thoroughly examines duration and formant structure in the context of several linguistic and social predictors and finds that there are many contexts in which the two orthographic liquids are significantly distinct. It is possible that the two sounds are also distinct in terms of a number of other acoustic variables that are not examined in the present study, which would provide further evidence that these sounds are in fact incompletely neutralized. The conclusion, then, that coda liquid sounds in SJS are incompletely neutralized, is not a radical assertion. The presence of significant differences between vowel+/r/ and vowel+/l/ in particular contexts supports the claim from the two previous acoustic studies on liquid switching in SJS, Simonet et al. (2008) and Luna (2010), that coda contexts are a site of incomplete neutralization in this dialect.
Another important argument for incomplete neutralization in SJS, which is often overlooked, is the variation in the manner of articulation of coda /r/. While the coda /l/ segments in the data are almost all produced as approximants, /r/ segments are articulated not only as approximants, but as taps, fricatives, deletions, trills, and vocalizations as well. These other manners of articulation of /r/ are certainly not neutralized with /l/.
Furthermore, the analysis of approximant liquids adds a more nuanced layer to the case for incomplete neutralization since these /r/ sounds are certainly more similar to approximant /l/ than the rhotics produced with other manners of articulation. Even in the case of approximant realizations, the acoustic parameters of vowel+/r/ sequences still differ in formant structure and duration in a variety of linguistic and social environments.
An important element to the methodology of the present study is the inclusion of all
199 approximant liquid tokens, regardless of the degree of neutralization. In their study of incomplete neutralization of coda liquids, Simonet et al. (2008) had a pre-screening of the approximant tokens by a native SJS speaker in which occurrences of approximant /r/ that was perceived as totally neutralized were excluded from the acoustic analysis. By considering all of the approximant liquid sounds in the data set, the present study better represents the wide variation of liquid sounds from the most distinct to the most neutralized. I propose that realizations of orthographic coda /r/ sounds in this dialect should be conceptualized as a continuum in which tap and fricative pronunciations at one end are the most distinctly rhotic sounds, whereas at the other end of the continuum, the two approximant liquid sounds are identical. The approximant sounds range from the more rhotic to more lateral ends of the continuum, and it is this middle ground that the acoustic analysis of formants and duration helps to sort out what factors play a role in the realizations of more rhotic or more lateral liquids.
One of the most contentious debates among researchers who study neutralization phenomena is whether or not spelling is responsible for the small phonetic differences in the experiments that test sounds that were previously assumed to be neutralized. Many scholars (e.g. Fourakis & Iverson 1984, Jassem & Richter 1989, Warner et al. 2006) argue that the differences observed in experiments that lead to claims of incomplete neutralization are actually artifacts of the experimental design. In other words, these researchers believe that speakers behave differently when they are confronted with spelling differences in reading tasks, but completely neutralize in everyday speech. As discussed in Chapter 2, a number of experiments strategically attempt to find a way around the spelling bias by designing experiments that do not require reading. The fact
200 remains, though, that unless the pool of participants in a study is composed entirely of completely illiterate speakers, the participants might still be influenced by how words are spelled, especially in a laboratory experiment, which is likely to induce a degree of prescriptive carefulness on the speech of the participants.
In the case of SJS, the possibilities for the influence of orthography on coda liquids is arguably even stronger than for the languages that have been studied the most thoroughly in neutralization research. To my knowledge, stop devoicing in German and
Catalan is not a dialect-specific phenomenon. Although liquids in Spanish are highly variable, few dialects in the Spanish-speaking world feature coda liquid neutralization.
Furthermore, SJS speakers seem to be highly aware of this feature of their dialect and readily contrast it with other varieties of Spanish. As discussed previously, the evaluation of this feature varies from a positive marker of national identity to a highly undesirable trait that speakers feel should be eradicated from Puerto Rican speech. Many of the participants in the production study expressed in the interview that they are proud of their
Spanish language heritage and this linguistic difference gives them a sort of pride and separate identity from the United States, despite their status as a commonwealth and the inundation of American influence. Participants who expressed this pride for the Spanish language also often voiced concern about the SJS features, such as liquid switching, that they viewed as undermining the purity and strength of the language. The finding from the present study that young people show significant differences in their production of coda liquids in terms of F3, which was discussed in depth in section 5.1.5, may represent a move away from this stigmatized Puerto Rican feature towards a more pan-dialectal liquid production.
201 Because of the potential pressures to conform to the non-neutralizing Spanish language standard, the type of speech situation analyzed is very important. SJS has a tendency to neutralize coda liquids to a far greater degree than other varieties of Spanish, and Puerto Ricans are highly aware of this dialect feature. Therefore, the tendency towards neutralization is likely to be greater in casual contexts whereas the prescriptivism from the language standard may induce more distinction, and thus incomplete neutralization, between liquid sounds in more formal situations. In fact, this is precisely what Medina Rivera (1999) found in his impressionistic study on coda liquid switching in
SJS according to formality and familiarity. Participants who were friends and family of the interviewer lateralized /r/, and therefore presumably were more neutralizing, at a higher rate than participants who did not know the interviewer well. Although the sociolinguistic interview cannot infallibly replicate everyday speech, it is arguably a much closer approximation to natural speech than the highly artificial laboratory experiments that constitute most of the research on incomplete neutralization phenomena across languages. The results from this relatively natural speech sample that certain linguistic environments favor distinction between the productions of vowel+/r/ and vowel+/l/, while other contexts are neutralizing, lend credence to the existence of a range of liquid sounds in everyday Puerto Rican speech. The contention that the influence of orthography makes incomplete neutralization somehow less representative of 'real' speech is valid only in the case that incomplete neutralization only exists in experimental contexts, and speakers neutralize in every other context. The claim made by neutralization researchers that the influence of orthography causes speakers to behave in a way counter to their linguistic system, which is neutralizing, is a valid criticism in the
202 context of highly artificial experiments, but falls short in view of the current research. It is probably the case that the degree and frequency of liquid neutralization in SJS, and by extension similar phenomena in other languages, is highly context-dependent. However, to deny that speakers have recourse to the incomplete neutralization behavior they exhibit in experiments in their everyday interactions is by extension a denial that speakers are able to exploit their knowledge of spelling as a linguistic resource. In other words, even if speakers totally neutralize coda liquids in their everyday speech, which is possible but dubious, the ability to produce and perceive the sounds with small phonetic differences in the interview and perception contexts is all the more interesting to linguistic and social theory.
5.3.2 Methodological considerations
In order to more fully understand coda liquid behavior in SJS, the present investigation employs a range of techniques and tools for data collection and analysis.
One of the main features that sets this study apart from past research is its sociophonetic approach. I followed sociolinguistic methodology in the data collection process by eliciting naturalistic speech through interviews that were designed to encourage conversational speech. The use of sociolinguistic interviews to elicit coda liquids in SJS is not unique to this study, but this is the first time that coda liquids in natural speech data from San Juan has been analyzed acoustically. Past sociolinguistic research on this phenomenon classified liquid sounds impressionistically, rather than using spectrographic evidence for categorizing manners of articulation and for measuring differences between coda /r/ and /l/. While these studies were certainly an improvement on the cursory 203 dialectological accounts that existed previous to them, the perception experiment in this study provides evidence that the ability to perceive and categorize liquid sounds is highly dependent upon both the native dialect of the listener and the particular linguistic context of the liquid sounds. Another disadvantage of impressionistic analysis is the necessity of binning approximant coda liquids into distinct groups. This type of analysis necessarily loses sight of the rich diversity of liquid sounds in this dialect that can be appreciated through acoustic analysis.
Simonet et al.'s (2008) acoustic study on coda liquids in laboratory speech served in many ways as inspiration for the current study. Particularly, I adapted the authors' side- by-side use of regression analyses, which have long been a tool for linguistic research, along with smoothing spline (henceforth, SS) ANOVAs, which are new to the field.
Despite the important precedent set by their statistical analysis, Simonet et al.'s study considers only four informants and only vowels /a/ and /i/ and following contexts /p/, /t/, and /k/ are analyzed. The present study includes the speech of 24 informants and considers the conditioning of a wide range of linguistic and social factors on the realization of coda liquids.
There are a number of advantages that SS ANOVAs offer for the analysis of formants and other linguistic data with curve-like structures. SS ANOVAs make it possible to quickly compare the shapes of two or more curves in a holistic manner. It is all too common in phonetic research that scholars sample acoustic or articulatory data at only one or two points along the trajectory of curves and then use a regression model or some other technique to calculate whether the point from a curve has a significant distance from the same point in time for another curve. This kind of sampling is
204 problematic both because the researcher often has to make arbitrary decisions as to where along the trajectories to take measurements that will accurately represent the overall curves, and because statistical tools commonly used in linguistic analysis, such as regression, are limited to analyzing the distance between two points and do not offer a comparison of the shapes of the curves. SS ANOVAs can provide sociolinguistic research with a tool for viewing frequency trends that may be lost to the conservative estimates of regression models that result from having a number of independent variables in the model. In the present study, the SS ANOVA for F2, in its apparent contradiction to the results for the regression analyses for the seven time points, provoked a more thorough investigation into the behavior of this formant and a deeper understanding of the interplay of the frequency of preceding context environments and the directionality of vowel+/r/ and vowel+/l/ values (see section 5.1.3 for details). Linguists are beginning to understand the usefulness of smoothing spline analyses and have begun advocating for the use of this statistical tool to compare dynamic trajectories of tongue movement and formants (Davidson 2006, Fruehwald 2010, Nycz & De Decker 2006, Haddican et al.
2013, among others).
Despite the usefulness of SS ANOVAs, there are some limitations that must be kept in mind when using them. The smoothing spline analyses in the present study are representative of all of the speakers and multiple linguistic contexts taken together, unlike those in Simonet et al's (2008) study in which each vowel context and speaker was considered separately. The use of regression models and SS ANOVAs in tandem is valuable for both research situations, but it is important to explore the potential drawbacks of these techniques for the data at hand. Since spline models do not have the
205 same capability of regression models for inputting a large number of predictors, there are limitations to the use of splines for large-scale studies with uneven data distributions. In a controlled experiment with few linguistic variables and few participants, the output from
SS ANOVAs can be interpreted in a straightforward manner. For example, the non- overlapping confidence intervals for vowel+/r/ and vowel+/l/ for a particular speaker in the context of the vowel /a/ before a following voiceless stop can be interpreted as strong evidence that the liquid sequences are different for this speaker in this particular linguistic context. However, the SS ANOVAs in the present study did not permit the addition of random effects due to the complexity of the models.31 Since individuals vary in their speech production, combining multiple individuals together risks violating the assumption made by the model that the data among speakers is uniform. Another potential disadvantage of splines over regression is that since the principal use of splines is as a visualization of the data, the incorporation of more independent variables complicates interpretation by adding a new spline for each combination of variables. By way of example, the independent variable for preceding sound in the present experiment has 7 different factor levels. Including a variable with this many levels to look at vowel+/r/ and vowel+/l/ means having 14 splines. If the researcher is interested in another independent variable, the number of splines multiplies by the number of factors in the predictor. In sum, the SS ANOVA provides a useful tool for all kinds of linguistic research on curves, but should be used alongside regression models in investigations that involve uneven data, a large number of informants, or a wide variety of independent
31 Some studies claim that the addition of random effects is possible with simpler data sets. See Gu 2002 and Wang 2011. 206 variables. The regression and SSANOVA models in the current study complement each other and provide a more detailed picture of the behavior of liquids in SJS.
5.4 UNDERSTANDING THE LIQUID SWITCHING PHENOMENON
Until recently, descriptions of the speech in parts of Puerto Rico, the Dominican
Republic, Cuba, and Spain with more neutralized coda liquid pronunciations than those of other dialects assumed that the process was as simple as one liquid sound being
'switched', or in some descriptions, 'confused' with the other. The results from this dissertation show that in SJS and presumably in other dialects with similar phenomena, the picture is far more complex than the name trueque de líquidas ('liquid switching') implies. My research indicates that SJS is better understood as exhibiting a liquid continuum rather than undergoing liquid switching.
Cross-linguistically, codas tend to have more variable articulation than do onsets.
The most studied example of coda variation in Spanish is that of /s/, which can be realized as [s], aspirated ([h]), deleted, or even inserted in cases of hypercorrection, depending on the dialect and a number of linguistic and social factors within each dialect
(e.g. Lipski 1994, Hammond 2001). A number of other sounds in Spanish, including nasals and stops, also tend to display a greater deal of variability in coda position than in onset. Besides liquid switching, other phenomena involving liquids have been documented in Spanish exclusively in coda position, such as gemination with following consonants in Cuban Spanish, liquid gliding in Cibaeño Spanish, and /r/ and /l/ deletion in a number of dialects (see Guitart 1994 for a list of phenomena).
207 Several theories have been posited as to why codas tend to be more variant than onsets, and particularly why neutralization is most often found in this position. Some scholars claim that onsets have to be more consistently articulated than do vowels or codas because the auditory system in the brain is primed to hear the beginning of events more clearly than what comes next (Nicholls et al. 2001). From this standpoint, it follows that sounds in coda position are freer to vary since departure from a uniform articulation in this part of the syllable is less likely to impede communication. In her licensing by cue framework, Steriade (1997) claims that postvocalic environments offer less perceptual information to the listener than do prevocalic contexts, which results in the tendency for neutralization in this position. Physiological reasons, such as the jaw cycle hypothesis, are also posited as reasons why the gestural timing for onsets tends to be more reliably organized than that for codas. In other words, the temporal order of the articulations made by different parts of the vocal apparatus (such as the tongue tip, the tongue body, the velum, etc.) in onset tend to be more reliably ordered across speakers and dialects than in coda position (see Redford 1999). As I will discuss below, the broad gestural variability of coda segments allows for a more diverse range of sounds in this position than in onset.
Based on the assumption that /r/ was neutralized to /l/, the term liquid switching came into use in the literature on SJS. At first glance, SJS approximant liquids do in fact appear to be completely neutralized; the results for this study from the regression models when all of the linguistic and social factors are taken together are insignificant at the majority of the time points across the four formants, as well as in the model for duration.
However, a more careful examination reveals that the end result of insignificant differences results from the pull of linguistic factors on the directionality of /r/ and /l/
208 such that in some environments /r/ has a higher formant or durational value than /l/, whereas /l/ has a higher value than /r/ in another environment. The SJS pronunciations of the two liquids in coda are assuredly more similar than those of other dialects, but these small differences are often significantly distinct in production within particular environments and these distinctions can be utilized by Puerto Ricans in perception in the context of certain vowels, as the results from the perception experiment in Chapter 4 show. Previous studies (Luna 2010, Paz 2005, Simonet et al. 2008) have appropriately characterized the phenomenon as incomplete neutralization and my research builds upon their work by providing and analyzing a more exhaustive range of linguistic contexts.
Given the broad variability of gestural timing in codas cross-linguistically and the continuum of liquid sounds in SJS across linguistic and social conditions, I propose an articulatory account of liquid production in SJS to explain the liquid continuum in coda position. The data for the present study shows a considerable amount of variation in the production of both orthographic coda rhotics and laterals in SJS, but the vowel+/r/ sequences show a greater degree of variability than vowel+/l/ sequences both in manner of articulation and in the formant structure of approximant /r/. I propose that one of the main reasons for variability of liquids in SJS is the particular coarticulatory flexibility of
/r/ in this dialect. Specifically, the greater coarticulatory flexibility of rhotics in SJS is manifested in their lesser resistance to coarticulation with the preceding vowel than laterals. This idea of certain sounds being more susceptible to coarticulation than others has been observed and quantified in Recasens, Pallarès, & Fontdevila (1997 and in subsequent work by Recasens). In their framework, coda rhotics in SJS would be said to have a lower Degree of Articulatory Constraint (DAC) than laterals. In dialects with
209 liquid switching involving rhotacization instead of lateralization, the DAC of the two liquids could very well be reversed.
The coarticulatory effects of the vowel on rhotics in SJS, along with the effects of other linguistic and social factors, create such diversity in the articulation of /r/ that the implication of a binary system that comes with the idea of liquid switching falls short in its descriptive power. The formant values (F1, F2, F3, F4) for liquids studied in this dissertation reveal that the liquid sounds in SJS can be more /l/-like in terms of one formant but more /r/-like in terms of another formant. For example, after the vowel /o/, the articulation of orthographic /r/ is likely to be more /r/-like in height (F1), but more /l/- like in frontness (F2) since the articulation of this vowel has a similar height to /r/ and a similar frontness to /l/. To say then, that this /r/ sound is switched to an /l/ only accounts for the frontness dimension and misses fine-grained distinctions among the continuum of liquid sounds in this dialect. The "intermediate" or "mixed" sounds noted in impressionistic studies (Navarro Tomás 1948, López Morales 1983a, 1983b) that were later described as incompletely neutralized in acoustic studies (Paz 2005, Simonet et al.
2008) are revealed to be strongly conditioned by the linguistic environment in the present study.
The effect of coarticulation of /r/, and to a lesser extent, /l/, with the vowel is observed for all of the formants. The first two formants, F1 and F2, are particularly interesting in light of the preceding vowel, as these are the most studied formants for the pronunciation of vowels since they correspond with tongue height and frontness. There are important implications in this study for both of these formants. The second formant
(F2) presented the defining difference in how Puerto Rican and Castilian listeners
210 differed in their perception of approximant liquids. For central and back vowels, SJS listeners hear a distinct /r/, due to its back pronunciation not unlike English [ɹ], whereas with the more fronted vowel /e/, SJS perception did not differ from that of CS listeners.
In essence, the fronted vowel made the liquid fronted to the degree that it sounded /l/-like to both groups. The first formant (F1) was useful to both groups of listeners in the perception experiment. The pattern for F1 found in the production data for this study can be understood not only in terms of coarticulation, but also in terms of a sonority framework. I will first outline the concepts that are important to understand the notion of sonority and then show how SJS liquid production patterns within theories of sonority.
A number of theorists claim that syllable structure is governed by sonority requirements (Blevins 1995, Clements 1990 Harris 1983, Hooper 1976, Selkirk 1984). In a syllable, the sonority rises up from the onset to the nucleus and then falls to the coda.
Onsets prefer to rise maximally in sonority whereas codas prefer to fall minimally in sonority. This idea is captured by the Sonority Sequencing Principle (SSP), which posits that some sounds in a given inventory of a language are more sonorous than others and that syllables are organized with the most sonorous sounds in the middle, which are usually vowels, with less sonorous sounds towards the periphery. Specifically, post- vocalic environments tend to be more sonorous than pre-vocalic contexts and the tendency to maximize sonority, i.e. have a segment in coda that is as sonorous as possible without being more sonorous than the nucleus, is common across languages (Clements
1990, Prince & Smolensky 1993). One of the main phonetic correlates to sonority, as will be discussed below, is openness of the vocal apparatus, which is acoustically manifested in F1 values. The use of rhotics produced as approximants rather than taps and the
211 coarticulation of coda liquids with vowels in SJS allows for more open, sonorous productions of /r/ than in many other dialects of Spanish.
Although sonority was first conceived of by phonologists (Pike 1943, Chomsky &
Halle 1968) who did not use phonetic evidence to support their claims, phonetic correlates have been proposed for sonority both in terms of articulation (Keating 1983,
Lindblom 1983) and acoustics (Mattingly 1981, Price 1980).32 In an investigation of sonority with emphasis on Spanish and English, Parker (2002) considers F1, intensity, intraoral air pressure, total air flow, and duration as measurements of sonority for various vowels and consonants, including liquids. Although Parker does not look at SJS specifically, the cross-linguistic hierarchy that the author derives from the Colombian
Spanish and American English data in his study along with an extensive literature review of other languages yields useful insights for the present study. Figure 31 shows Parker's sonority hierarchy; from top to bottom, it lists the most sonorous to the least sonorous sounds. The most sonorous sounds are vowels, with low vowels being the most sonorous and high vowels being the least sonorous; this ranking corresponds with F1 values, in that higher F1 values represent a lower and therefore more sonorous articulation. Glides are less sonorous than vowels, but more sonorous than other consonants. After the glides,
Parker lists four kinds of liquids. The most sonorous liquid, according to his study, is the
English approximant retroflex [ɹ], which is represented as "/r/". Laterals are the second most sonorous liquid, followed by taps (labeled "flaps") and finally trills.
32 Mattingly's (1981) account considers both articulatory and acoustic data. 212
Figure 31. Parker's (2002:240) universal sonority hierarchy. The occurrence of more open /r/ sounds in the environment of more open vowels in SJS can thus be described not only in terms of coarticulation, but also in terms of sonority sequencing. The production data in the present study shows that rhotics accompanied by low vowels, which are the most sonorous vowels, tend to have low, back, bunched productions that resemble the retroflex American [ɹ], which Parker finds to be more sonorous than laterals or the Spanish tap. Less sonorous vowels, such as the high vowel /i/, tend to condition more /l/-like rhotic productions in SJS. Both of the approximant liquids in Parker's hierarchy are more sonorous than taps. Tap productions of /r/ in SJS are most likely in the context of less sonorous vowels. Given the occurrence of more sonorous rhotics with more sonorous vowels and the use of taps in less sonorous vocalic environments, I propose that sonority sequencing, alongside coarticulation, motivates liquid production in SJS. That is, the drive for maximally sonorous codas works alongside the coarticulatory flexibility of rhotics in SJS.
Coda liquids in SJS are sensitive not only to the preceding vowel, but also to the surrounding consonant sounds. Unlike with vowels, the degree of sonority of the 213 surrounding consonants inversely affects liquid production such that less sonorous surrounding sounds elicit more sonorous liquids and vice versa (see sections 5.1.4.3 and
5.1.4.4). Since these surrounding consonants are onsets to the syllable containing the liquid in the case of previous sounds and onsets of the syllable following the liquid in the case of following sounds, the result is that liquids contrast with the onsets. This pattern fits sonority sequencing theories in that liquid codas are maximally distinguishable from the surrounding onsets.
To summarize, coarticulatory effects and sonority could be described as working together as central driving forces behind the variability of coda liquids in SJS. Taken separately, coarticulation and sonority each account for part of the picture. A physiological, articulatory view has more explanatory power in terms of the coarticulation observed across formants. The sonority patterning of liquids with vowels in
SJS can be seen as emergent from the coarticulation of liquids with vowel openness.
While the articulatory hypothesis covers vowel and liquid assimilation across formants, viewing liquid tendencies from a sonority perspective only accounts for F1. However, the dissimilation of liquids from surrounding consonants follows from a sonority perspective in which sonority of coda liquids is maximally different from the surrounding sounds.
Moving beyond the motivation for the variability of liquids, the present study has interesting implications for the debate surrounding whether or not /r/ sounds should be included in the liquid class with /l/ sounds. As discussed in section 2.2, orthographic /r/ has many productions cross-linguistically that stray from the vowel-like nature of the liquid class, such as fricatives and assibilated variants. Some scholars posit that rhotics should not be considered as liquids due to their obstruent nature in certain languages (see
214 Colantoni & Steele 2005). However, the similar phonological patterning of these sounds, the historical developments involving changes from rhotics to laterals (or vice versa), and the fairly stable variation between these sounds in phenomena such as liquid switching have led most researchers to argue for the inclusion of rhotics in the liquid class
(Ladefoged & Maddieson 1996, Lindau 1985, Maddieson 1980, Walsh Dickey 1997).
The similarities found between SJS coda /r/ and /l/ in this dissertation in many environments, such as with high, front vowels, lends support to the claim that rhotics should be considered alongside laterals as a liquid class. However, it is also the case that rhotics are less stable than laterals in coda environments in SJS. Coda /l/ in SJS is nearly categorically pronounced as an approximant with a high, front articulation whereas /r/ is more variant in its place of articulation as an approximant and can also be articulated as a tap or fricative, or can be deleted entirely. Despite their variability, rhotics in this dialect are approximant and thus liquid-like most of the time in coda position. The argument to exclude rhotics from the liquid class would gain more support from a consideration of these sounds in onset, which are pronounced as fricatives in many dialects of Puerto
Rican Spanish.
215
CHAPTER 6: CONCLUSION
6.1 CONCLUSIONS AND CONTRIBUTIONS
This dissertation is the first analysis to consider acoustic data on coda liquids in natural speech for San Juan Spanish (SJS). Furthermore, this study takes into account linguistic and social independent variables conditioning liquid variation that had never been acoustically examined for this dialect and as a result, offers a more complete picture of the behavior of coda liquids in SJS. This dissertation considers all of the vowels in the
Spanish language as well as all of the preceding and following environments for vowel+liquid sequences. In addition, the effects of word position, stress, gender, and age on coda liquid production are explored. Previous to this study, only the effect of the vowels /a/ and /i/ and following voiceless stops had been examined in an acoustic study
(Simonet et al. 2008). The results from the present study lend support to previous claims of incomplete neutralization for coda liquids in Puerto Rican Spanish (Paz 2005, Simonet et al. 2008, Luna 2010). This study goes a step further and proposes the physiological and phonological grounds of coarticulation and sonority principles as forces motivating the pronunciation of coda liquids that results in what has been referred to in the literature as liquid switching. Due to the highly variable nature of coda liquids in SJS, I claim that /r/ and /l/ should be viewed not as "switched" or "neutralized", but rather as displaying a
216 liquid continuum of possible productions. Importantly, the linguistic and social context determine where the liquids fall on this continuum.
When all of the linguistic and social independent variables are considered together, formant and duration values for vowel+/r/ and vowel+/l/ sequences are rarely significantly different. That is, over seven time points for each of the four formants (F1,
F2, F3, F4), only two points for F2 and one point for F3 showed overall significant differences. However, the overall insignificant effects are often created by the pull of particular contexts in opposite directions. For example, in the context of one preceding sound, formant values for vowel+/r/ are higher than those for vowel+/l/, whereas with a different preceding environment, values for vowel+/l/ are higher. Thus, the articulation of coda liquids in this study was shown to be strongly influenced by the surrounding linguistic environment. Vowel+/r/ sequences are more responsive to the tongue position of the vowel than are vowel+/l/ sequences. When the vowel in the sequence is low or mid, /r/ is produced with a more open articulation, whereas /r/ responds to the more closed tongue position of high vowels with a more closed, more /l/-like percept.
Similarly, vowel+/r/ sequences are more adaptive to the frontness of the vowel than are vowel+/l/ sequences, meaning that the articulation varies from a more back position following back vowels and a more front position after front vowels. As a result, the degree of neutralization varies according to the vowel in the sequence, i.e. the neutralization between coda /r/ and /l/ is greater with a preceding high vowel. On the other hand, the assimilation pattern in relation to the preceding vowel that /r/ sequences display stands in stark contrast to the behavior of liquids in relation to the sounds surrounding the vowel+liquid sequences. In the context of occlusive preceding sounds
217 (stops, fricatives, and nasals), vowel+/r/ sequences prefer a more open and back pronunciation. After approximants (/β/, /ð/, /ɣ/), which are produced with more opening, vowel+/r/ is more closed and further forward. Following sounds only played a significant role in the height of the liquids and these pattern similarly to preceding sounds; /r/ articulation is more open with following stops, fricatives, and nasal articulations and more closed before approximants and vowels. The dissimilation pattern of surrounding consonants shows that more closed surrounding sounds correlate with more open liquids and vice versa, which means that the degree of neutralization is greater in the context of open consonant sounds such as approximants since /r/ has a more /l/-like production in this context.
Word position also plays an interesting role in the degree of neutralization of coda
/r/ and /l/ sequences. Vowel+/r/ sequences are significantly more low and front than vowel+/l/ in word-medial position whereas the liquids are distinct only in terms of tongue bunching in word-final position - coda /r/ sequences have more bunching than vowel+/l/.
In other words, the difference between the two liquids is maintained through different articulations, which are manifested in distinct acoustic cues depending on the word position. By considering the effect of word position acoustically, this study was able to pinpoint how vowel+/r/ and vowel+/l/ are neutralized in terms of some formant measures and distinct in others depending on whether the liquids are in word-medial or word-final position. This more accurate and nuanced description of liquid switching greatly improves upon findings from impressionistic studies.
In terms of social variables, vowel+/r/ and vowel+/l/ produced as approximants are significantly different according to the age of the speakers, whereas gender does not
218 play a role for these realizations. Younger speakers pronounce coda liquids distinctly, unlike older speakers who neutralize the two sounds. Specifically, younger speakers maintain a distinction between coda liquids by producing vowel+/r/ with a more retracted tongue root evinced by a lower F3, which is associated cross-linguistically with rhoticity.
The older speakers in this study did not produce significantly different F3 values for vowel+/r/ and vowel+/l/. This study contributes acoustic evidence to the impressionistic observation made by López Morales (1983b) about the higher rate of differenciation between liquid sounds by younger speakers and the present study specifies that younger people are distinguishing liquids by manipulating tongue bunching, but do not display different behavior from older speakers in the height or frontness of their liquid sounds.
This effect for age may be a result of a negative social evaluation of liquid switching. As discussed in section 5.1.5, many participants brought up that "saying /l/ for /r/" was a common trait of Puerto Rican Spanish. While a couple of these speakers cited this accent feature as a unique identity marker and thus felt positively about it, most of the participants who brought it up referred to it as a "bad habit" that could be attributed to a lack of education or to laziness. The difference observed between the age groups in this study could be evidence of historical change or of age grading. In either case, the negative sentiment surrounding liquid switching may very well play a role in the liquid production of young San Juan residents.
The perception study in this dissertation advances our understanding of liquid perception in Spanish, which has received very little attention in the literature. The results of the perception experiment show that Puerto Ricans are better able to hear the differences between vowel+/r/ and vowel+/l/ and they are able to use the first three
219 formants (F1, F2, and F3), whereas the Spaniards in the control group were not able to utilize F2, which correlates with the anteriority of the tongue. This finding shows that SJS speakers are able to use the cues for frontness present in the production in their perception of liquid sounds. The finding that SJS listeners use a feature unique to the articulation of these sounds in their dialect to distinguish /r/ from /l/ advances the notion put forward by research on incomplete neutralization that listeners are able to utilize small differences in production to help them classify sounds. To listeners from dialects without liquid switching, the liquid sounds in the experiment were perceived as more completely neutralized. That is, there were no significant differences between the perception of vowel+/r/ and vowel+/l/ to Spaniards for any of the vowel+liquid combinations, whereas SJS listeners responded differently for the two liquid sounds with the vowels /a/ and /o/. With the more forward and high vowel /e/, SJS listeners heard the two sounds as neutralized. This result follows from the findings in the production study in that greater degrees of neutralization are found in the context of vowels with a high, front tongue position due to the coarticulation of rhotics with vowels with /l/-like articulations.
In terms of the methodological tools, this study shows that the use of regression and SS ANOVA models provides a more thorough understanding of the behavior of liquids in SJS. SS ANOVAs, unlike regression models, allow for the analysis and comparison of entire curves, rather than fixed points. Since sound production involves movement over time, the utilization of models that capture this is enormously helpful in understanding formant trajectories. The use of regression models at various points throughout the liquid trajectories, unlike many studies that sample at one or two points,
220 also provides a more detailed understanding of the factors conditioning the articulatory variation. By considering phonetic detail in the production and perception of liquids in this dialect, this study provides an understanding of the interplay of linguistic environment with liquid neutralization and distinction. More broadly, the findings presented in this study in relation to the behavior of coda liquids in SJS in different environments contribute to the growing body of work on incomplete neutralization. SJS presents an interesting case of incomplete neutralization in that the two liquid sounds range from very similar to very distinct due to the articulatory flexibility of rhotics, which are very sensitive to the surrounding linguistic context. This sensitivity, along with the fact that this pronunciation feature is salient to speakers and carries social meaning, results in a liquid continuum in which rhotics are more variable than laterals. The linguistic and social structuring of liquid sounds in this dialect has implications for future research on incomplete neutralization phenomena.
6.2 FUTURE RESEARCH
This study is the first to consider the conditioning of both word position and the complete range of vowels and surrounding sounds on approximant liquid variation in this dialect. The strong influence of these linguistic variables on liquid production merits further investigation into the potential interactions of these variables with each other and the effects of the frequency of the particular factor levels on the variation. Studies of both type and token frequencies - i.e. common particular environments, such as preceding stops, and particular words and phrases, such as Puerto Rico - have yielded interesting
221 results in variable phenomena across languages (see Bybee 2001) and we might expect similar effects in the production and perception of liquids in SJS.
The findings for F2, the acoustic correlate for the anteriority/posteriority of articulation, both in the production and perception experiments deserve further study. The frequency of particular preceding sounds, and perhaps also vowels, may play into the perception of these sounds. One way to better understand the effects of front/back articulations is to analyze the independent variables of preceding and following sounds by place, rather than manner, of articulation.
To further explore the findings for F2, as well as the other formant values and duration, instrumental analysis using tools such as electropalatography or MRI could help understand the movement of the articulators and the amount of closure that correspond with orthographic /r/ vs. /l/ in different linguistic environments. In a number of contexts, values for approximant /r/ compared to /l/ suggest that rhotic sounds in this dialect are more anterior and more closed than laterals, which contradicts the usual profile of approximant rhotics. Further investigation into the nature of the articulation of these two sounds would serve to clarify the movements of the tongue that produce these confounding findings.
Few articulatory studies have been conducted for any dialect of Spanish, and few of these for coda segments. Articulatory data could determine if laterals in this dialect behave in the way that is assumed for dialects of Spanish. Most descriptions of lateral articulation in the Spanish language agree that /l/ is produced as in a 'clear' manner, meaning that the articulation lacks a dorsal gesture. Some degree of dorsal constriction, however, was one of the few universal findings in Gick et al.'s (2006) articulatory study
222 of coda laterals across six languages. The presence or lack of dorsal activity for /l/ in this dialect would further our understanding of the direction of the neutralization. The assumption that laterals have the more stable articulation of the two liquid sounds and therefore it is rhotics that become lateralized, rather than some degree of rhotacization on the part of laterals, may very well be challenged with articulatory evidence. A cross- dialectal approach in which coda liquids of non-neutralizing varieties of Spanish were compared to SJS productions would be ideal.
More investigation into the effects of social variables and language attitudes surrounding the liquid switching phenomenon would also contribute to the understanding of these sounds. The data for the current study lacks representation from speakers in lower social classes, who, according to impressionistic studies, lateralize /r/ more frequently than speakers of higher classes. The lack of effects for gender in the present study could very well turn out differently in terms of the intersection of class with this variable. The findings for F3 for age suggest a potential reorganization of liquid behavior in this dialect. Considering how younger and older speakers from lower social classes treat this variable, along with more in-depth investigations of language attitudes using experimental perceptual methods such as matched-guise tests, would help understand the social evaluation of this variable and potential for language change towards a more pan- dialectal realization of coda liquids. The analysis of the liquid sounds in the reading task that the participants for this study performed after the sociolinguistic interview would also provide useful insights into how more normative production of liquid sounds differs from more casual speech situations.
223 The findings in this dissertation sets a precedent for future research on phonetically gradient phenomena beyond SJS. This study supports the assertion made by many researchers that sounds that were before considered to be neutralized are in fact different (see section 2.5 for a review of these studies) and shows just how important the linguistic and social context of a sound is for its production. In natural speech, and probably also in the case of read speech, factors such as the vowel in the syllable and the age of the person speaking determine whether or not /r/ and /l/ are neutralized or distinct in SJS. In light of these findings, it would be worth revisiting previous studies that found complete neutralization to test a wider range of linguistic environments and speakers.
The methodological principles set out in this project could be applied to better understand phonetic gradience across languages. This dissertation considers not only the production of small phonetic differences, but also how listeners behave in their perception, which is often overlooked. The inter-dialectal comparison of perceptual abilities employed in this project could also be applied to geographical and social lects of many languages. The sociophonetic approach to data collection and the analysis of several points during the trajectory of a sound would be useful for developing a nuanced account of phonetic phenomena in other languages. Through the analysis of coda liquid variation in SJS, this project paves the way for a variety of future phonetic studies.
224
REFERENCES
Alba, Orlando. 1988. Estudio sociolingüístico de la variación de las líquidas finales de palabra en el español cibaeño. Studies in Caribbean Spanish Dialectology, ed. by Robert Hammond and Melvyn Resnick, 1-12. Washington, D.C.: Georgetown University Press.
Alfaraz, Gabriela G. 2007. Effects of Age and Gender on Liquid Assimilation in Cuban Spanish. Selected Proceedings of the Third Workshop on Spanish Sociolinguistics, ed. by Jonathan Holmquist, Augusto Lorenzino, and Lotfi Sayahi. Somerville, MA: Cascadilla Proceedings Project.
Alonso, Amado and Raimundo Lida. Geografía fonética: -l y -r implosivas en español. RFH VII, 313-345.
Álvarez Nazario, Manuel. 1990. El habla campesina del país. Río Piedras: Editorial de la Universidad de Puerto Rico.
Almeida Suárez, Manuel V. & Dorta Luis, Josefa. 1993. Datos acústicos de las líquidas españolas,Homenaje a José Pérez Pidal, ed. by Carmen Díaz Alayó, 97-110.
Argüello, Fanny. 1978. The Žeísta Dialect of Spanish Spoken in Ecuador: A phonetic and phonological study. State College, PA: The Pennsylvania State University, doctoral dissertation.
Ash, Sharon. 2002. Social class. The handbook of language variation and change, ed. by J. K. Chambers, Peter Trudgill, and Natalie Schilling-Estes, 402-22. Oxford: Blackwell.
Azevedo, Milton. 2005. A Linguistic Introduction to Portuguese. Cambridge: Cambridge University Press.
Baranowski, Maciej. 2013. Sociophonetics. The Oxford Handbook of Sociolinguistics, ed. by Robert Bayley, Richard Cameron and Ceil Lucas. Oxford: Oxford UP.
Bates, Douglas, Martin Maechler, and Ben Bolker. 2011. lme4 package for R. Madison: University of Wisconsin-Madison, R Foundation for Statistical Computing and McMaster University.
Baayen, R. Harald. 2011. LanguageR package for R. Tuebingen: University of Tuebingen. 225 Best, Catherine T. and Winnifred Strange. 1992. Effects of phonological and phonetic factors on cross-language perception of approximants. Journal of Phonetics 20, 305-330.
Bishop, Jason B. 2007. Incomplete neutralization in Eastern Andalusian Spanish: perceptual consequences of durational differences involved in s-aspiration. Proceedings of the International Congress of Phonetic Sciences XVI, 1765-1768.
Bortoni-Ricardo, Stella Maris. 1985. The urbanization of rural dialect speakers. Cambridge: Cambridge University Press.
Bradley, Travis G. 2001a. A Typology of Rhotic Duration Contrast and Neutralization, ed. by Minjoo Kim and Uri Strauss, Proceedings of the North East Linguistics Society (NELS 31), GSLA Publications, 79-97.
Bradley, Travis G. 2001b. The Phonetics and Phonology of Rhotic Duration Contrast and Neutralization. State College, PA: The Pennsylvania State University, doctoral dissertation.
Bradley, Travis G. 2004. Gestural Timing and Rhotic Variation in Spanish Codas, ed. by Timothy L. Face Laboratory Approaches to Spanish Phonology, 197-224. Berlin: Mouton de Gruyter.
Bradley, Travis G. 2005. Systemic Markedness and Phonetic Detail in Phonology, ed. by Randall Gess and Edward Rubin, Experimental and Theoretical Approaches to Romance Linguistics, 41-62. Amsterdam: John Benjamins.
Bradley, Travis G. 2006. Spanish complex onsets and the phonetics-phonology interface, ed. by Fernando Martínez-Gil and Sonia Colina, Optimality-Theoretic Studies in Spanish Phonology, 15-38. Amsterdam: John Benjamins.
Bybee, Joan. 2001. Phonology and Language Use. Cambridge: Cambridge University Press.
Charles-Luce, Jan. & Walker, Laurie Ann. 1981. Effects of linguistic context on the durations of lexical categories. Research on Speech Perception: Progress Report 7. Speech Research Laboratory, Indiana University, 263-272.
Charles-Luce, Jan. 1993. Effects of semantic context on voicing neutralization. Phonetica 50, 28-43.
Chela Flores, Godsuno. 1986. Las teorías fonológicas y los dialectos del Caribe Hispánico, ed. by Núñez Cedeño, Páez Urdaneta and Jorge Guitart. Estudios sobre la fonología del español del caribe, 21-30.
226 Chomsky, Noam and Morris Halle. 1968. The sound pattern of English. New York: Harper & Row Publishing.
Clements, George Nick. 1990. The role of the sonority cycle in core syllabification, ed. by John Kingston and Mary E. Beckman, Papers in Laboratory Phonology I: Between the grammar and the physics of speech. Cambridge: Cambridge University Press. 283- 333.
Colantoni, Laura and Jeffrey Steele. 2005. Liquid Asymmetries in French and Spanish. Toronto Working Papers in Linguistics 24, 1-14.
Colantoni, Laura. 2011. Laboratory Approaches to Sound Variation and Change, ed. by Manuel Díaz-Campos, Handbook of Hispanic Sociolinguistics. West Sussex: Wiley- Blackwell.
Davidson, Lisa. 2006. Comparing tongue shapes from ultrasound imaging using smoothing spline analysis of variance. Journal of the Acoustical Society of America 120(1), 407-415.
Delattre, Pierre. 1958. Les indices acoustiques de la parole: Premier rapport. Phonetica 2. 108-118, 226-251.
Di Paolo, Marianna and Malcah Yaeger-Dror (eds). 2011. Sociophonetics: A student's guide. New York, NY: Routledge.
Dinnsen, Daniel A. and Jan Charles-Luce. 1984. Phonological neutralization, phonetic implementation and individual differences. Journal of Phonetics 12, 49-60.
Docherty, Gerard, Simón Gonzalez, and Nathaniel Mitchell. 2015. Static vs. dynamic perspectives on the realization of vowel nuclei in West Australian English. Paper presented at 18th International Congress of Phonetic Sciences. Glasgow, Scotland.
Eckert, Penelope. 1997. Age as a sociolinguistic variable, ed. by Florian Coulmas, The Handbook of Sociolinguistics. Oxford: Blackwell. 151-167.
Fontanella de Weinberg, María Beatriz. 1984. Confusión de líquidas en el español rioplatense (siglos XVI a XVIII). Romance Philology 37, 432-45.
Fourakis, Marios and Gregory K. Iverson. 1984. On the 'incomplete neutralization' of German final obstruents. Phonetica 41, 140-149.
Fruehwald, Josef. 2010. SS ANOVA. Accessible online: http://www.ling.upenn.edu/~joseff/papers/fruehwald_ssanova.pdf
Fujimura, Osamu and Donna Erickson. 1997. Acoustic phonetics, ed. by William J. Hardcastle & John Laver, The handbook of phonetic sciences. Oxford: Blackwell. 227 Gerfen, Chip and Kathleen Currie Hall. 2001. Coda aspiration and incomplete neutralization in Eastern Andalusian Spanish. University of North Carolina at Chapel Hill, unpublished manuscript.
Gick, Bryan. 2002. The American intrusive l. American Speech 77(2), 167-183.
Gick, Bryan, A. Min Kang, and Douglas H. Whalen. 2002. MRI evidence for commonality in post-oral articulations of English vowels and liquids. Journal of Phonetics 30. 357-371.
Gick, Bryan, Fiona Campbell, Sunyoung Oh, and Linda Tamburri-Watt. 2006. Toward universals in the gestural organization of syllables: A cross-linguistic study of liquids. Journal of Phonetics 34, 49-72.
Giles, Stephen B. and Kenneth L. Moll. 1975. Cinefluorographic Study of Selected Allophones of English /l/. Phonetica: International Journal of Speech Science 31, 206- 227.
Gu, Chong. 2002. Smoothing spline ANOVA models. Springer Series in Statistics. New York: Springer-Verlag.
Gu, Chong. 2004. Model Diagnostics for Smoothing Spline ANOVA Models. The Canadian Journal of Statistics 32(4), 347-358.
Gu, Chong. 2009. gss: General Smoothing Splines. R package version 1.0-5.
Guirao, Miguelina and María Amalia García Jurado. 1991. Los perfiles acústicos y la identificación de /l/ y /r/. Revista argentina de lingüística 7, 21-42.
Guitart, Jorge M. 1994. Liquids in Caribbean Spanish and variation as code-switching. Thesaurus 49(2), 229-244.
Haddican, Bill, Paul Foulkes, Vincent Hughes, and Hazel Richards. 2013. Interaction of social and linguistic constraints on two vowel changes in northern England. Language Variation and Change 25, 371–403.
Hammond, Robert M. 2001. The sounds of Spanish: analysis and application. Somerville, MA: Cascadilla.
Heselwood, Barry. 2009. Rhoticity without F3: Lowpass Filtering, F1-F2 Relations and the Perception of Rhoticity in 'North-Force,' 'Start,' and 'Nurse' Words. Leeds Working Papers in Linguistics & Phonetics 14, 49-64.
Holmquist, Jonathan Carl. 2003. Coffee Farmers, Social Integration and Five Phonologial Features: Regional Socio-Dialectology in West-Central Puerto Rico, ed. by Lofti Sayahi and Maurice Westmoreland, Selected Proceedings of the First Workshop on Spanish 228 Sociolingustics, 70-76. Sommerville, MA: Cascadilla Proceedings Project.
Holmquist, Jonathan Carl. 2005. Social Stratification in Women’s Speech in Rural Puerto Rico: A Study of Five Phonological Features, ed. by Lofti Sayahi and Maurice Westmoreland, Selected Proceedings of the Second Workshop on Spanish Linguistics, 109-119. Somerville, MA: Cascadilla Proceedings Project.
Hothorn, Torsten & Achim Zeileis. 2015. partykit: A Modular Toolkit for Recursive Partytioning in R. R package version 1.0-20.
Horvath, Barbara M. and Horvath, Ronald J. 2001. A multilocality study of a sound change in progress: The case of /l/ vocalization in New Zealand and Australian English. Language Variation and Change, 13(01), 37-57.
Hualde, José Ignacio. 2005. The Sounds of Spanish. New York: Cambridge University Press.
Jassem, Wiktor and Lutoslawa Richter. 1989. Neutralization of voicing in Polish obstruents. Journal of Phonetics 17, 317–325.
Johnson, Keith. 2003. Acoustic and Auditory Phonetics. 2nd edition. Malden, MA: Blackwell.
Kleber, Felicitas, Tina John, and Jonathan Harrington. 2010. The implications for speech perception of incomplete neutralization of final devoicing in German. Journal of Phonetics 38. 185–196.
Keating, Patricia. 1983. Comments on the jaw and syllable structure. Journal of Phonetics 11, 401-406.
Kleber, Felicitas, Tina John, Jonathan Harrington. 2010. The implications for speech perception of incomplete neutralization of final devoicing in German. Journal of Phonetics 38, 185–196.
Labov, William. 1966. The Social Stratification of English in New York City. Washington, D.C.: Center for Applied Linguistics. Cambridge: Cambridge University Press.
Labov, William. 1984. Field methods of the Project on Linguistic Change and Variation, ed. by John Baugh & Joel Sherzer, Language in Use. Englewood Cliffs: Prentice Hall.
Labov, William. 1990. The interaction of sex and social class in the course of linguistic change. Language Variation and Change 2, 205-254.
Ladefoged, Peter and Ian Maddieson. 1996. The sounds of the world's languages. Oxford: Blackwell. 229 Lindblom, Bjorn. 1963. Spectrographic study of vowel reduction. Journal of the Acoustical Society of America 35, 1773-1781.
Lindblom, Bjorn. 1983. Economy of speech gestures, ed. by Peter F. MacNeilage, The Production of Speech, 217-245. New York: Springer-Verlag.
Lindau, Mona. 1985. The story of /r/, ed. by Victoria Fromkin, Phonetic linguistics: Essays in honor of Peter Ladefoged, 157–168. Orlando: Academic Press.
Lipski, John. 1994. Latin American Spanish. New York: Longman.
Lipski, John M. 2011. Socio-phonological Variation in Latin American Spanish, ed. by Manuel Díaz-Campos, Handbook of Hispanic Sociolinguistics. West Sussex: Wiley- Blackwell.
Lisker, Leigh. 1983. “Voicing” in English: A Catalogue of Acoustic Features Signaling /b/ Versus /p/ in Trochees. Language and Speech 29, 3-11.
López Morales, Humberto. 1983a. Lateralización de /-r/ en el español de Puerto Rico. Philologia Hispanensia in honorem Manuel Alvar, 387-398. Madrid: Gredos.
López Morales, Humberto. 1983b. Estratificación Social del Español de San Juan de Puerto Rico. México: Universidad Nacional Autónoma de México.
Luna, Kenneth Vladimir. 2010. The Spanish of Ponce, Puerto Rico: A Phonetic, Phonological, and Intonational Analysis. University of California, Los Angeles, doctoral dissertation.
Maddieson, Ian. 1980. A survey of liquids. University of California Working Papers in Phonetics, 50, 93-112.
Martínez Celdrán, Eugenio. 1984. Fonética. Barcelona: Teide.
Martínez Celdrán, Eugenio. 1993. La percepción categorial de /b-p/ en español basada en las diferencias de duración. Estudios de Fonética Experimental 5, 223-239.
Martínez Celdrán, Eugenio and Ana María Fernández Planas. 2007. Manual de fonética española. Ariel Lingüística.
Massone, María Ignacia. 1988. Estudio acústico y perceptivo de las consonantes nasales y líquidas del español. Estudios de Fonética Experimental 3, 13-34.
Mattingly, Ignatius. 1981. Phonetic representation and speech synthesis by rule, ed. by Terry Myers, John Laver and John Anderson, The Cognitive Representation of Speech, 415-420. Phildelphia: Elsevier.
230 Medina-Rivera, Antonio. 1999. Variación fonológica y estilística en el español de Puerto Rico. Hispania 82, 529-541.
Megenney, William W. 1978. El problema de R velar en Puerto Rico. Thesaurus-Boletín del Instituto Caro y Cuervo 33, 72-86.
Minagawa-Kawai, Yasuyo, Koichi Mori, and Yutaka Sato. 2005. Different brain strategies underlie the categorical perception of foreign and native phonemes. Journal of Cognitive Neuroscience 17(9), 1376-1385.
Navarro Tomás, Tomás. 1948. El español en Puerto Rico. San Juan: Editorial de la Universidad de Puerto Rico.
Navarro Tomás, Tomás. 1970. Manual de pronunciación española. Madrid: CSIC.
Newton, David E. 1996. The nature of resonance in English: an investigation into lateral articulations. York Papers in Linguistics 17, 167-190.
Nicholls, John G., Bruce G. Wallace, Paul A. Fuchs, & A. Robert Martin. 2001. From Neuron to Brain: A Cellular and Molecular Approach to the Function of the Nervous System. Fourth edition. Sunderland, MA: Sinauer.
Nycz, Jennifer, and Paul De Decker. 2006. A New Way of Analyzing Vowels: Comparing Formant Contours Using Smoothing Spline ANOVA. Poster presented at NWAV 35, Columbus, OH.
O'Brien, Valerie. 2013. Rhotic Production and Contrast in Equatoguinean Spanish. The Ohio State University, MA thesis.
Pike, Kenneth. 1943. Phonetics: a critical analysis of phonetic theory and technic for the practical description of sounds. Ann Arbor, MI: University of Michigan Press.
O’Connor Joseph Desmond, Louis J. Gerstman, Alvin Meyer Liberman, Pierre Charles Delattre & Franklin Seaney Cooper. 1957. Acoustic Cues for the Perception of Initial /w,j,r,l/ in English. Word 13. 22-43.
Parker, Stephen G. 2002. Quantifying the Sonority Hierarchy. Amherst, MA: University of Massachusetts, doctoral dissertation.
Paz, Mercedes. 2005. Retroflection of post-nuclear /r/ in Puerto Rican Spanish. Paper presented at the 9th Hispanic Linguistics Symposium, November. University Park, PA: Pennsylvania State University.
Pollack, Irwin & Pickett, James M. 1964. Intelligibility of Excerpts from fluent speech: auditory vs. structural context. Journal of Verbal Learning and Verbal Behavior 3. 79-84.
231 Port, Robert & Penny Crawford. 1989. Incomplete neutralization and pragmatics in German. Journal of Phonetics 17, 257-282.
Price, Patti J. 1980. Sonority and syllabicity: acoustics correlates of perception. Phonetica 37, 327-343.
Prince, Alan and Paul Smolensky. 1993. Optimality Theory: Constraint Interaction in Generative Grammar. RuCCS Technical Report #2. Rutgers University (ROA 537).
Proctor, Michael. 2009. Gestural Characterization of a Phonological Class: the Liquids. Hartford, CT: Yale University, doctoral dissertation.
Prosper-Sánchez, Gloria D. 1995. Neutralización Homofonética de Líquidas a Final de Sílaba: Aspectos Sociolingüísticos en el Español de Puerto Rico. Amherst, MA: University of Massachusetts, doctoral dissertation.
Quilis, Antonio. 1981. Fonética acústica de la lengua española. Madrid: Gredos.
Quilis, Antonio. 1993. Tratado de fonología y fonética españolas. Madrid: Gredos.
R Core Team. 2015. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL http://www.R-project.org/.
Ramos-Pellicia, Michelle F. 2007. Lorain Puerto Rican Spanish and ‘r’ in Three Generations, ed. by Jonathan Holmquist, Augusto Lorenzino and Lotfi Sayahi, Selected Proceedings of the Third Workshop on Spanish Sociolinguistics. Somerville, MA: Cascadilla Proceedings Project.
Recasens, Daniel. 2004. Darkness in 1 as a scalar phonetic property: implications for phonology and articulatory control. Clinical Linguistics & Phonetics 18, 593-603.
Redford, Melissa Annette. 1999. An articulatory basis for the syllable. Austin, Texas: University of Texas, doctoral dissertation.
Röttger, Timo. 2012. Incomplete Neutralization: A cognitive artifact of lexical co- activation? University of Cologne, MA thesis.
Roettger, Timo, Bodo Winter, Sven Grawunder, James Kirby, and Martine Grice. 2014. Assessing incomplete neutralization of final devoicing in German. Journal of Phonetics 43. 11-25.
Ruiz-Peña Sánchez de Ibargüen, Esperanza. 2013. “Alma” o “arma”, evidencia de la neutralización /l/ /r/ en la variedad dialectal andaluza de Sevilla. University of Western Ontario, MA thesis.
232 Sessarego, Sandro. 2011. Phonetic analysis of /sr/ clusters in Cochabambino Spanish, ed. by Luis A. Ortiz-López, Selected Proceedings of the 13th Hispanic Linguistics Symposium, 251-263. Somerville, MA: Cascadilla Proceedings Project.
Shouse de Vivas, Dolores. 1978. El uso de [l] variante de ‘r’ en el habla de Puerto Rico. Actas del V Congreso Internacional de la Asociación de Lingüística y Filología de la América Latina (ALFAL): Caracas, Venezuela.
Simonet, Miquel, Marcos Rohena-Madrazo and Mercedes Paz. 2008. Preliminary evidence of incomplete neutralization of coda liquids in Puerto Rican Spanish,ed. by Laura Colantoni and Jeffrey Steele, Laboratory approaches to Spanish phonology III, 72- 86. Somerville, MA: Cascadilla.
Solé, Maria-Josep. 2002. Aerodynamic characteristics of trills and phonological patterning. Journal of Phonetics 30, 655-688.
Sproat, Richard and Osamu Fujimura. 1993. Allophonic variation in English /l/ and its implications for phonetic implementation. Journal of Phonetics 21, 291–311.
Steriade, Donca. 1997. Phonetics in Phonology: the case of laryngeal neutralization. UCLA, unpublished manuscript.
Straka, Georges. 1979. Contribution à la description et à l'histoire des consonnes L, ed. by Georges Straka, Les sons et les mots, 363-422. Paris: Éditions Klincksieck.
Sundberg, Johan. 1995. The singer's formant revisited. Voice 4, 106-19.
Thomas, Erik R. 2011. Sociophonetics: An Introduction. New York: Palgrave.
Vaquero, María and Antonio Quilis. 1984. Datos acústicos de /r/ en el español de Puerto Rico. Actas del VII Congreso de la Asociación de Lingüística y Filología de América Latina 2, 115–142.
Valentín-Márquez, Wilfredo. 2007. Doing being Borícua: Perceptions of National Identity and the Sociolinguistic Distribution of Liquid Variables in Puerto Rican Spanish. Ann Arbor, MI: University of Michigan, doctoral dissertation.
Vanek, Christian and Scott McDaniel. 2006. Surveygizmo. Widgix, LLC.
Wahba, Grace. 1983. Bayesian "Confidence Intervals" for the Cross-Validated Smoothing Spline. Journal of the Royal Statistical Society 45(1), 133-150.
Walsh Dickey, Laura. 1997. The phonology of liquids. Amherst, MA: University of Massachusetts, doctoral dissertation.
233 Warner, Natasha, Allard Jongman, Joan Sereno and Rachèl Kemps. 2004. Incomplete neutralization and other sub-phonemic durational differences in production and perception: evidence from Dutch. Journal of Phonetics 32, 251-276.
Warner, Natasha, Erin Good, Allard Jongman and Joan Sereno. 2006. Orthographic vs. morphological incomplete neutralization effects. Journal of Phonetics 34: 285-293.
Willis, Erik W. 2007. An Acoustic Study of the “Pre-aspirated Trill” in Narrative Cibaeño Dominican Spanish. Journal of the International Phonetic Association 37, 33- 49.
Willis, Erik W. and Travis Bradley. 2008. Contrast Maintenance of Taps and Trills in Dominican Spanish: Data and Analysis, ed. by Laura Colantini and Jeffrey Steele, Selected Proceedings of the 3rd Conference on Laboratory Approaches to Spanish Phonology, 87-100. Somerville: Cascadilla Proceedings Project.
Wong, Yuedong. 2011. Smoothing Splines: Methods and Applications. New York: CRC Press.
234
APPENDIX A: INTERVIEW PROTOCOL General information ¿Dónde naciste?¿Creciste en el mismo lugar? – Where were you born, and did you grow up in the same place? ¿Cuándo naciste? – When were you born? ¿Cuáles idiomas hablabas durante la niñez? – What languages did you speak growing up? ¿Has vivido en otros lugares? – Have you lived in other places? ¿Asististe a la escuela? ¿Dónde? – Did you attend school? If so, where?
Language use: ¿Cuáles lenguas hablas? – What languages do you speak? ¿Con quienes hablas español/inglés/etc.? – Who do you speak Spanish/English/etc. with?
School/Work: ¿En qué estás trabajando/quieres trabajar? – What do you do/want to do? ¿Qué pretendes hacer después de graduarte/en su trabajo en el futuro? – What do you plan to do after you graduate/in your future work? ¿Qué sueñas hacer profesionalmente? – What do you dream about doing professsionally?
Family ¿Sus padres viven en Puerto Rico? ¿Ves a tus padres a menudo? – Do your parents live in PR? Do you see them often? ¿Tienes hermanos? ¿Viven en Puerto Rico? ¿Te llevas bien con ellos? – Do you have siblings? Do they live in PR? Do you get along well with them? ¿Tienes pareja? ¿Estás casado? ¿Cómo conociste a tu pareja/esposo/a? – Do you have a significant other? Are you married? How did you meet your partner/spouse? ¿Tienes hijos? – Do you have children? ¿Cuáles son tus responsabilidades familiares? En otras palabras, qué tienes que hacer diariamente? – What are your familial responsibilities? In other words, what do you have to do daily? ¿Cuáles valores quieres pasar a tus hijos? – What values do you want to pass on to your children?
235 Leisure Time ¿Qué te gusta hacer en tu tiempo libre? ¿Con quiénes? – What do you like to do in your free time? With whom? ¿Hay actividades que te gustaría aprender a hacer? – Are there activities that you would like to learn to do?
Travel ¿Has viajado dentro de Puerto Rico? – Have you traveled within Puerto Rico? ¿Has viajado fuera de Puerto Rico? – Have you traveled outside of Puerto Rico? ¿Cómo fue la experiencia? ¿Cómo te trataron las personas? – How was the experience? How did people treat you?
US-Puerto Rico Relations ¿Cómo ves la relación entre los EEUU y Puerto Rico? – How do you see the relationship between the US and Puerto Rico? ¿Te gustaría que fuera PR un estado? ¿Por qué sí/no? – Would you like for Puerto Rico to be a state? Why or why not? ¿Cómo es diferente la cultura y la gente de PR de los EEUU? – How are Puerto Rican culture/people different than US culture/people? ¿PR es más como los EEUU o como América Latina? – Is Puerto Rico more like the US or Latin America? ¿Cómo son las relaciones entre PR y las otras islas caribeñas? – How are relations between Puerto Rico and the other Caribbean islands? ¿Tienes parientes o amigos que viven en los EEUU? ¿Le gusta/n vivir allá? ¿Por qué si o no? – Do you have relatives that live in the US? Do they like to live there? Why or why not?
Future Plans ¿Quieres quedarte en (ciudad)? – Do you want to stay in (city)? ¿Cómo sería tu vida ideal? – What would your ideal life be like?
Language Ideologies ¿Crees que es importante ser bilingüe para los puertorriqueños? ¿Por qué? – Do you think that it is important for Puerto Ricans to be bilingual? Why? ¿Cómo es diferente el español puertorriqueño de otros dialectos del español? – How is Puerto Rican Spanish different from other dialects of Spanish? ¿Cómo es diferente el habla de (ciudad) del habla de otras zonas de la isla? – How is speech here in (city) different from speech on other parts of the island? ¿Dónde se habla el mejor español? – Where is the best Spanish spoken? ¿Cuáles cosas te molestan del habla de los puertorriqueños? – What bothers you about Puerto Rican speech?
236
APPENDIX B: LINEAR REGRESSION MODELS FOR FOUR FORMANTS AT SEVEN TIME POINTS AND FOR DURATION
The following tables are the raw number results from the linear regression models. In order to derive interaction values, the intercept, main effect, liquid type, and interaction must be added together. For example, the values for the significant effect for nasals for F1 at the first time point (F1.1) below is: vowel+/r/ with preceding nasal: 521.08+24.04+-2.41+14.27 = 556.98 Hz vowel+/l/ with preceding nasal: 521.08+24.04+ 2.41 +-14.27 = 533.26 Hz
Note that the interaction term for /l/ is inverse that given in the table, as the interaction terms represent the distance of /r/.
F1.1 Estimate Standard Error t-Value p-value (Intercept) 521.0819 7.368699 70.71560 0.0000 Stress stressed 5.7218 1.995402 2.86750 0.0042 unstressed -5.7218 1.995402 -2.86750 0.0042 Word Position word-medial 4.7400 2.360557 2.00802 0.0447 word-final -4.7400 2.360557 -2.00802 0.0447 Vowel /a/ 149.0745 3.521725 42.32996 0.0000 glide+/a/ 43.4791 6.208260 7.00343 0.0000 glide +/e/ -7.3592 7.740359 -0.95075 0.3418 /e/ 3.9204 4.529483 0.86554 0.3868 /i/ -97.6015 6.590412 -14.80962 0.0000 /o/ 4.2953 4.661061 0.92153 0.3569 /u/ -95.8087 7.992076 -11.98796 0.0000 Preceding Sound stop -0.1328 4.784164 -0.02776 0.9779 pause 4.6713 11.947666 0.39098 0.6958 approximant -8.2318 5.661996 -1.45386 0.1461 fricative -2.2481 5.101143 -0.44070 0.6595 lateral -17.5916 18.137455 -0.96990 0.3322 nasal 24.0402 5.412836 4.44134 0.0000 237 rhotic -0.5073 7.697089 -0.06591 0.9475 Following Sound stop -4.0246 3.207387 -1.25479 0.2097 pause -7.2326 4.237123 -1.70696 0.0879 approximant 1.0569 4.130877 0.25584 0.7981 fricative -8.5138 6.466715 -1.31656 0.1881 nasal -0.2775 4.762731 -0.05826 0.9535 vowel 18.9916 4.387969 4.32812 0.0000 Gender female 33.8414 5.978271 5.66074 0.0000 male -33.8414 5.978271 -5.66074 0.0000 Age younger 8.2322 6.175408 1.33306 0.1968 older -8.2322 6.175408 -1.33306 0.1968 Liquid Type /r/ -2.4064 4.496149 -0.53521 0.5925 /l/ 2.4064 4.496149 0.53521 0.5925 Interaction: Word Position and Liquid Type word-medial -1.2924 2.308014 -0.55996 0.5756 word-final 1.2924 2.308014 0.55996 0.5756 Interaction: Preceding Sound and Liquid Type stop -6.0009 4.755409 -1.26191 0.2071 pause -15.8427 11.936734 -1.32722 0.1846 approximant -8.7726 5.658531 -1.55034 0.1212 fricative -1.1239 5.069450 -0.22170 0.8246 lateral 27.7831 18.178656 1.52834 0.1266 nasal 14.2684 5.382641 2.65081 0.0081 rhotic -10.3114 7.671232 -1.34417 0.1790 Interaction: Following Sound and Liquid Type stop -1.6452 3.195765 -0.51482 0.6067 pause 5.8417 4.229626 1.38114 0.1674 approximant -4.6540 4.125761 -1.12804 0.2594 fricative 8.1594 6.468927 1.26132 0.2073 nasal 1.6080 4.766517 0.33735 0.7359 vowel -9.3098 4.397293 -2.11717 0.0343 238 Interaction: Vowel and Liquid Type /a/ -0.1748 3.497931 -0.04999 0.9601 glide+/a/ -17.1022 6.184942 -2.76513 0.0057 glide+/e/ -6.7353 7.506359 -0.89728 0.3697 /e/ 14.9476 4.367411 3.42254 0.0006 /i/ -1.8552 6.607041 -0.28079 0.7789 /o/ 7.3420 4.677475 1.56965 0.1166 /u/ 3.5779 7.979120 0.44841 0.6539
F1.2 Estimate Standard Error t-value p-value (Intercept) 516.0505 7.985855 64.62057 0.0000 Stress stressed 12.4181 2.150656 5.77408 0.0000 unstressed -12.4181 2.150656 -5.77408 0.0000
Word Position word-medial 1.6945 2.544224 0.66604 0.5054 word-final -1.6945 2.544224 -0.66604 0.5054 Vowel /a/ 122.7001 3.795750 32.32565 0.0000 glide+/a/ 72.3501 6.691316 10.81254 0.0000 glide +/e/ -4.0641 8.342681 -0.48714 0.6262 /e/ -6.2775 4.881919 -1.28587 0.1986 /i/ -101.6154 7.103210 -14.30556 0.0000 /o/ 5.4163 5.023733 1.07813 0.2811 /u/ -88.5095 8.613930 -10.27516 0.0000 Preceding Sound stop -5.2393 5.156409 -1.01607 0.3097 pause -1.3434 12.877264 -0.10432 0.9169 approximant -10.1776 6.102531 -1.66776 0.0955 fricative -3.3507 5.498056 -0.60944 0.5423 lateral -6.3686 19.548649 -0.32578 0.7446 nasal 26.6165 5.833998 4.56230 0.0000 rhotic -0.1369 8.295962 -0.01650 0.9868 Following Sound stop -13.6219 3.456956 -3.94044 0.0001 pause -18.1595 4.566793 -3.97642 0.0001 approximant 0.6096 4.452286 0.13692 0.8911 fricative 1.2655 6.969871 0.18156 0.8559 239 nasal -1.0810 5.133285 -0.21058 0.8332 vowel 30.9873 4.729373 6.55210 0.0000 Gender female 31.8685 6.497348 4.90484 0.0001 male -31.8685 6.497348 -4.90484 0.0001 Age younger 6.7685 6.711581 1.00848 0.3247 older -6.7685 6.711581 -1.00848 0.3247 Liquid Type /r/ -2.1290 4.845972 -0.43933 0.6605 /l/ 2.1290 4.845972 0.43933 0.6605 Interaction: Word Position and Liquid Type word-medial 3.3376 2.487599 1.34171 0.1798 word-final -3.3376 2.487599 -1.34171 0.1798 Interaction: Preceding Sound and Liquid Type stop -3.4934 5.125407 -0.68158 0.4956 pause -7.7204 12.865500 -0.60009 0.5485 approximant -0.9222 6.098795 -0.15121 0.8798 fricative -3.9790 5.463888 -0.72824 0.4665 lateral 9.0302 19.593075 0.46089 0.6449 nasal 12.5203 5.801444 2.15813 0.0310 rhotic -5.4354 8.268081 -0.65740 0.5110 Interaction: Following Sound and Liquid Type stop -3.8245 3.444409 -1.11036 0.2669 pause 1.8549 4.558710 0.40690 0.6841 approximant -8.9668 4.446781 -2.01648 0.0439 fricative 9.4283 6.972269 1.35225 0.1764 nasal 6.2940 5.137379 1.22514 0.2206 vowel -4.7859 4.739435 -1.00980 0.3127 Interaction: Vowel and Liquid Type /a/ 3.9580 3.770085 1.04985 0.2939 glide+/a/ -11.8146 6.666167 -1.77232 0.0765 glide+/e/ -7.3299 8.090444 -0.90600 0.3650 /e/ 12.9758 4.707239 2.75656 0.0059 240 /i/ 2.3489 7.121159 0.32985 0.7415 /o/ 9.6473 5.041430 1.91360 0.0558 /u/ -9.7855 8.599953 -1.13785 0.2553
F1.3 Estimate Standard Error t-value p-value (Intercept) 498.9426 8.183623 60.96843 0.0000 Stress stressed 8.9721 2.164132 4.14582 0.0000 unstressed -8.9721 2.164132 4.14582 0.0000 Word Position word-medial -1.8909 2.560172 -0.73859 0.4602 word-final 1.8909 2.560172 0.73859 0.4602 Vowel /a/ 95.3274 3.819578 24.95758 0.0000 glide+/a/ 91.0751 6.733295 13.52608 0.0000 glide +/e/ 3.4130 8.395195 0.40655 0.6844 /e/ -14.5970 4.912554 -2.97137 0.0030 /i/ -96.5124 7.147795 -13.50240 0.0000 /o/ -0.9782 5.055254 -0.19351 0.8466 /u/ -77.7278 8.667974 -8.96724 0.0000 Preceding Sound stop -10.3485 5.188745 -1.99441 0.0462 pause -7.4195 12.957945 -0.57258 0.5670 approximant -7.5351 6.140761 -1.22706 0.2199 fricative -4.4283 5.532550 -0.80040 0.4235 lateral 9.4770 19.671110 0.48177 0.6300 nasal 22.2993 5.870589 3.79848 0.0001 rhotic -2.0451 8.347920 -0.24498 0.8065 Following Sound stop -20.7340 3.478664 -5.96033 0.0000 pause -23.7638 4.595394 -5.17121 0.0000 approximant -7.8146 4.480189 -1.74425 0.0812 fricative 4.6817 7.013567 0.66753 0.5045 nasal -0.2289 5.165402 -0.04432 0.9647 vowel 47.8595 4.758988 10.05666 0.0000 Gender female 26.6779 6.718763 3.97065 0.0007 male -26.6779 6.718763 -3.97065 0.0007 Age younger 7.8382 6.940234 1.12938 0.2715 older -7.8382 6.940234 -1.12938 0.2715 Liquid Type 241 /r/ 4.9470 4.876320 1.01450 0.3104 /l/ -4.9470 4.876320 -1.01450 0.3104 Interaction: Word Position and Liquid Type word-medial 6.2840 2.503210 -2.51036 0.0121 word-final -6.2840 2.503210 -2.51036 0.0121 Interaction: Preceding Sound and Liquid Type stop -4.5971 5.157520 -0.89135 0.3728 pause 6.4928 12.946169 0.50152 0.6160 approximant -4.3298 6.136996 -0.70552 0.4805 fricative -5.8367 5.498135 -1.06159 0.2885 lateral 6.6993 19.715877 0.33979 0.7340 nasal 8.0879 5.837796 1.38543 0.1660 rhotic -6.5163 8.319831 -0.78323 0.4336 Interaction: Following Sound and Liquid Type stop -3.4106 3.465975 -0.98402 0.3252 pause 5.2609 4.587254 1.14684 0.2516 approximant -19.5475 4.474679 -4.36848 0.0000 fricative 16.0099 7.016024 2.28191 0.0226 nasal 7.1945 5.169564 1.39171 0.1641 vowel -5.5072 4.769150 -1.15475 0.2483 Interaction: Vowel and Liquid Type /a/ 2.8846 3.793690 0.76036 0.4471 glide+/a/ 2.9133 6.707933 0.43431 0.6641 glide+/e/ -14.9346 8.141281 -1.83443 0.0667 /e/ 14.0090 4.736787 2.95750 0.0031 /i/ 8.1271 7.165943 1.13413 0.2568 /o/ 10.2886 5.073080 2.02807 0.0427 /u/ -23.2880 8.653871 -2.69105 0.0072
F1.4 Estimate Standard Error t-value p-value (Intercept) 476.2681 8.170378 58.29206 0.0000 Stress stressed 6.9708 2.386333 2.92112 0.0035 242 unstressed -6.9708 2.386333 -2.92112 0.0035 Word Position word-medial -0.5094 2.822996 -0.18046 0.8568 word-final 0.5094 2.822996 0.18046 0.8568 Vowel /a/ 70.7927 4.211438 16.80962 0.0000 glide+/a/ 78.1372 7.424265 10.52457 0.0000 glide +/e/ 3.6464 9.255446 0.39397 0.6936 /e/ -15.3107 5.416624 -2.82661 0.0047 /i/ -74.1844 7.881141 -9.41290 0.0000 /o/ -4.1172 5.574000 -0.73865 0.4602 /u/ -58.9639 9.557456 -6.16942 0.0000 Preceding Sound stop -10.5293 5.721310 -1.84036 0.0658 pause -17.3938 14.288444 -1.21733 0.2236 approximant -5.6469 6.771318 -0.83394 0.4044 fricative -0.9003 6.100291 -0.14758 0.8827 lateral 17.2156 21.691044 0.79367 0.4275 nasal 12.7629 6.473096 1.97168 0.0488 rhotic 4.4918 9.205208 0.48796 0.6256 Following Sound stop -21.9985 3.835501 -5.73549 0.0000 pause -20.3545 5.067323 -4.01681 0.0001 approximant -18.0250 4.940153 -3.64866 0.0003 fricative 15.2208 7.733524 1.96816 0.0492 nasal -1.8978 5.696103 -0.33318 0.7390 vowel 47.0549 5.247753 8.96667 0.0000 Gender female 21.0195 6.341656 3.31451 0.0033 male -21.0195 6.341656 -3.31451 0.0033 Age younger 9.8342 6.551086 1.50116 0.1482 older -9.8342 6.551086 -1.50116 0.1482 Liquid Type /r/ 5.1220 5.377113 0.95256 0.3409 /l/ -5.1220 5.377113 -0.95256 0.3409 Interaction: Word Position and Liquid Type word-medial 7.2280 2.760058 -2.61877 0.0089 word-final -7.2280 2.760058 -2.61877 0.0089 Interaction: 243 Preceding Sound and Liquid Type stop 1.0092 5.687085 0.17746 0.8592 pause 6.7244 14.275021 0.47106 0.6376 approximant 3.2577 6.767206 0.48140 0.6303 fricative -6.2250 6.062574 -1.02678 0.3046 lateral -6.8698 21.739961 -0.31600 0.7520 nasal 7.9517 6.437183 1.23527 0.2168 rhotic -5.8483 9.174482 -0.63746 0.5239 Interaction: Following Sound and Liquid Type stop -5.9464 3.821961 -1.55584 0.1199 pause 13.8659 5.058394 2.74116 0.0062 approximant -16.7690 4.933866 -3.39876 0.0007 fricative 3.5276 7.735919 0.45600 0.6484 nasal 4.8950 5.700389 0.85872 0.3906 vowel 0.4269 5.258694 0.08119 0.9353 Interaction: Vowel and Liquid Type /a/ 6.0733 4.183335 1.45178 0.1467 glide+/a/ 11.5133 7.396695 1.55655 0.1197 glide+/e/ -18.3082 8.976155 -2.03965 0.0415 /e/ 8.8587 5.222753 1.69618 0.0900 /i/ -0.2381 7.900536 -0.03014 0.9760 /o/ 8.1472 5.593523 1.45655 0.1454 /u/ -16.0462 9.542182 -1.68161 0.0928
F1.5 Estimate Standard Error t-value p-value (Intercept) 446.6112 7.978770 55.97495 0.0000 Stress stressed 5.7344 2.275465 2.52009 0.0118 unstressed -5.7344 2.275465 -2.52009 0.0118 Word Position word-medial 0.6390 2.691852 0.23738 0.8124 word-final -0.6390 2.691852 -0.23738 0.8124 Vowel /a/ 58.6881 4.015861 14.61407 0.0000 glide+/a/ 57.1172 7.079436 8.06804 0.0000 glide +/e/ -1.6852 8.825905 -0.19094 0.8486 /e/ -2.1781 5.165058 -0.42170 0.6733 244 /i/ -55.9804 7.515136 -7.44902 0.0000 /o/ 0.3589 5.315115 0.06753 0.9462 /u/ -56.3205 9.113556 -6.17986 0.0000 Preceding Sound stop -3.0059 5.455552 -0.55097 0.5817 pause -10.1788 13.624600 -0.74709 0.4551 approximant 1.2166 6.456712 0.18842 0.8506 fricative -0.1795 5.816958 -0.03085 0.9754 lateral 10.0270 20.683236 0.48479 0.6279 nasal 2.5235 6.172428 0.40883 0.6827 rhotic -0.4029 8.777495 -0.04590 0.9634 Following Sound stop -10.2088 3.657395 -2.79128 0.0053 pause -19.8909 4.831872 -4.11660 0.0000 approximant -21.3943 4.710646 -4.54170 0.0000 fricative 16.1058 7.374271 2.18405 0.0290 nasal -4.5187 5.431373 -0.83196 0.4055 vowel 39.9070 5.003909 7.97516 0.0000 Gender female 19.0896 6.287275 3.03623 0.0063 male -19.0896 6.287275 -3.03623 0.0063 Age younger 11.4715 6.494800 1.76626 0.0919 older -11.4715 6.494800 -1.76626 0.0919 Liquid Type /r/ 5.9074 5.127264 1.15215 0.2494 /l/ -5.9074 5.127264 -1.15215 0.2494 Interaction: Word Position and Liquid Type word-medial 5.2960 2.631872 2.01227 0.0443 word-final -5.2960 2.631872 -2.01227 0.0443 Interaction: Preceding Sound and Liquid Type stop 2.5142 5.422862 0.46363 0.6430 pause 1.0094 13.611918 0.07415 0.9409 approximant 6.1659 6.452780 0.95554 0.3394 fricative 0.7914 5.780932 0.13690 0.8911 lateral -10.2842 20.730000 -0.49610 0.6199 nasal -0.6733 6.138117 -0.10969 0.9127 245 rhotic 0.4766 8.748130 0.05448 0.9566 Interaction: Following Sound and Liquid Type stop -4.4818 3.644364 -1.22978 0.2189 pause 14.1891 4.823345 2.94176 0.0033 approximant -15.0798 4.704708 -3.20525 0.0014 fricative -2.6800 7.376638 -0.36331 0.7164 nasal 3.4643 5.435541 0.63734 0.5240 vowel 4.5882 5.014412 0.91499 0.3603 Interaction: Vowel and Liquid Type /a/ 9.9102 3.988945 2.48442 0.0130 glide+/a/ 2.3263 7.053042 0.32983 0.7416 /e/ -11.6764 8.559404 -1.36417 0.1726 glide+/e/ 3.3265 4.980209 0.66795 0.5042 /i/ -0.1786 7.533794 -0.02370 0.9811 /o/ 6.4651 5.333767 1.21211 0.2256 /u/ -10.1732 9.098918 -1.11806 0.2636
F1.6 Estimate Standard Error t-value p-value (Intercept) 434.7215 8.072167 53.85438 0.0000 Stress stressed 3.0241 2.173081 1.39162 0.1642 unstressed -3.0241 2.173081 -1.39162 0.1642 Word Position word-medial 1.6077 2.570754 0.62539 0.5318 word-final -1.6077 2.570754 -0.62539 0.5318 Vowel /a/ 52.6925 3.835331 13.73872 0.0000 glide+/a/ 44.6015 6.761090 6.59679 0.0000 glide +/e/ 2.7949 8.429678 0.33155 0.7403 /e/ -1.2692 4.932825 -0.25730 0.7970 /i/ -53.9262 7.177280 -7.51346 0.0000 /o/ 4.6317 5.076118 0.91245 0.3616 /u/ -49.5253 8.703752 -5.69010 0.0000 Preceding Sound stop 2.3265 5.210177 0.44654 0.6552 pause -10.7566 13.011540 -0.82670 0.4085 approximant 2.7852 6.166164 0.45169 0.6515 fricative 3.5502 5.555387 0.63905 0.5228 246 lateral 22.1601 19.752489 1.12189 0.2620 nasal -5.0422 5.894832 -0.85536 0.3924 rhotic -15.0232 8.382466 -1.79222 0.0732 Following Sound stop 2.7575 3.493004 0.78945 0.4299 pause -27.1096 4.614412 -5.87499 0.0000 approximant -24.0486 4.498712 -5.34566 0.0000 fricative 19.4878 7.042549 2.76715 0.0057 nasal -14.8731 5.186810 -2.86749 0.0042 vowel 43.7860 4.778687 9.16276 0.0000 Gender female 18.5394 6.568833 2.82233 0.0102 male -18.5394 6.568833 -2.82233 0.0102 Age younger 10.9551 6.785423 1.61450 0.1213 older -10.9551 6.785423 -1.61450 0.1213 Liquid Type /r/ 7.5566 4.896502 1.54326 0.1229 /l/ -7.5566 4.896502 -1.54326 0.1229 Interaction: Word Position and Liquid Type word-medial 5.8682 2.513539 2.33465 0.0196 word-final -5.8682 2.513539 -2.33465 0.0196 Interaction: Preceding Sound and Liquid Type stop -5.5792 5.178852 -1.07731 0.2814 pause -1.6777 12.999654 -0.12906 0.8973 approximant 6.6780 6.162389 1.08366 0.2786 fricative -2.3596 5.520862 -0.42739 0.6691 lateral -10.2160 19.797380 -0.51603 0.6059 nasal 0.9236 5.861937 0.15756 0.8748 rhotic 12.2310 8.354294 1.46404 0.1433 Interaction: Following Sound and Liquid Type stop 0.3561 3.480325 0.10230 0.9185 pause 3.0756 4.606245 0.66770 0.5044 approximant -10.5457 4.493149 -2.34706 0.0190 fricative -6.0443 7.044973 -0.85796 0.3910 247 nasal 6.0258 5.190948 1.16084 0.2458 vowel 7.1326 4.788855 1.48941 0.1365 Interaction: Vowel and Liquid Type /a/ 10.2812 3.809396 2.69891 0.0070 glide+/a/ -0.1350 6.735677 -0.02004 0.9840 glide+/e/ -14.8888 8.174809 -1.82130 0.0687 /e/ 1.5507 4.756324 0.32603 0.7444 /i/ 1.9671 7.195417 0.27338 0.7846 /o/ -5.7518 5.094000 -1.12914 0.2589 /u/ 6.9766 8.689628 0.80286 0.4221
F1.7 Estimate Standard Error t-value p-value (Intercept) 435.6355 8.151094 53.44504 0.0000 Stress stressed 1.9154 2.414456 0.79329 0.4277 unstressed -1.9154 2.414456 -0.79329 0.4277 Word Position word-medial 1.4688 2.856258 0.51423 0.6071 word-final -1.4688 2.856258 -0.51423 0.6071 Vowel /a/ 55.2236 4.261014 12.96021 0.0000 glide+/a/ 41.8058 7.511694 5.56543 0.0000 glide +/e/ -7.0421 9.364208 -0.75203 0.4521 /e/ 3.5639 5.480401 0.65030 0.5156 /i/ -42.8117 7.973921 -5.36897 0.0000 /o/ -2.9533 5.639637 -0.52366 0.6006 /u/ -47.7862 9.670002 -4.94170 0.0000 Preceding Sound stop 4.6290 5.788703 0.79966 0.4240 pause -6.8892 14.456845 -0.47653 0.6337 approximant 7.5058 6.851131 1.09555 0.2734 fricative 6.3760 6.172127 1.03304 0.3017 lateral -3.7565 21.946716 -0.17117 0.8641 nasal -5.5867 6.549337 -0.85303 0.3937 rhotic -2.2784 9.313724 -0.24462 0.8068 Following Sound stop 10.4957 3.880643 2.70464 0.0069 pause -16.0660 5.127060 -3.13357 0.0017 approximant -26.5279 4.998367 -5.30731 0.0000 fricative 7.8756 7.824638 1.00652 0.3143 248 nasal -39.1796 5.763296 -6.79813 0.0000 vowel 63.4022 5.309625 11.94099 0.0000 Gender female 19.3842 6.266853 3.09314 0.0055 male -19.3842 6.266853 -3.09314 0.0055 Age younger 10.1366 6.473882 1.56577 0.1323 older -10.1366 6.473882 -1.56577 0.1323 Liquid Type /r/ 9.4117 5.440506 1.72993 0.0838 /l/ -9.4117 5.440506 -1.72993 0.0838 Interaction: Word Position and Liquid Type word-medial -6.3465 2.792555 -2.27263 0.0231 word-final 6.3465 2.792555 2.27263 0.0231 Interaction: Preceding Sound and Liquid Type stop -4.9762 5.754112 -0.86481 0.3872 pause 12.2204 14.443184 0.84610 0.3976 approximant 0.1111 6.846977 0.01622 0.9871 fricative -5.4276 6.134008 -0.88483 0.3763 lateral -11.2695 21.996129 -0.51234 0.6085 nasal 10.7565 6.513045 1.65153 0.0988 rhotic -1.4146 9.282682 -0.15239 0.8789 Interaction: Following Sound and Liquid Type stop -1.9483 3.867026 -0.50381 0.6144 pause 5.0946 5.118035 0.99543 0.3196 approximant -7.7032 4.991968 -1.54312 0.1229 fricative -4.5341 7.827004 -0.57929 0.5624 nasal 8.0164 5.767577 1.38991 0.1647 vowel 1.0746 5.320646 0.20196 0.8400 Interaction: Vowel and Liquid Type /a/ 8.2499 4.232660 1.94910 0.0514 glide+/a/ -8.5785 7.483872 -1.14626 0.2518 glide+/e/ -3.1089 9.081753 -0.34233 0.7321 /e/ -3.2360 5.284235 -0.61240 0.5403 249 /i/ -4.5285 7.993432 -0.56653 0.5711 /o/ -0.4408 5.659366 -0.07788 0.9379 /u/ 11.6428 9.654599 1.20594 0.2280
F2.1 Estimate Standard Error t-Value p-value (Intercept) 1534.6331 29.77962 51.53300 0.0000 Stress stressed -11.0276 8.99606 -1.22583 0.2204 unstressed 11.0276 8.99606 1.22583 0.2204 Word Position word-medial -31.6249 10.35073 -3.05533 0.0023 word-final 31.6249 10.35073 3.05533 0.0023 Vowel /a/ -50.2853 15.56852 -3.22994 0.0013 glide+/a/ 72.3229 27.75254 2.60599 0.0092 glide +/e/ -47.0438 34.85097 -1.34986 0.1772 /e/ 172.4813 20.26838 8.50987 0.0000 /i/ 477.5433 29.66436 16.09822 0.0000 /o/ -262.8355 20.64748 -12.72966 0.0000 /u/ -362.1830 35.22868 -10.28091 0.0000 Preceding Sound stop -70.6263 21.49621 -3.28552 0.0010 pause 122.1805 53.91863 2.26602 0.0235 approximant -15.4611 25.21758 -0.61311 0.5399 fricative 31.0163 22.98296 1.34953 0.1773 lateral -164.1309 81.65158 -2.01014 0.0445 nasal 40.0082 24.39975 1.63970 0.1012 rhotic 57.0134 34.71266 1.64244 0.1006 Following Sound stop 29.5410 12.20119 2.42116 0.0155 pause 10.4866 18.18189 0.57676 0.5641 approximant 10.8154 17.08277 0.63312 0.5267 fricative -69.5862 20.54631 -3.38680 0.0007 nasal 50.4654 16.94804 2.97765 0.0029 vowel -31.7223 18.32597 -1.73100 0.0836 Gender female 79.6223 23.20913 3.43064 0.0025 male -79.6223 23.20913 -3.43064 0.0025 Age younger 21.3564 23.97582 0.89075 0.3832 older -21.3564 23.97582 -0.89075 0.3832 Liquid Type 250 /r/ 73.5592 19.48754 3.77468 0.0002 /l/ -73.5592 19.48754 -3.77468 0.0002 Interaction: Word Position and Liquid Type word-medial 26.8438 9.27873 2.89305 0.0038 word-final -26.8438 9.27873 -2.89305 0.0038 Interaction: Preceding Sound and Liquid Type stop -43.3452 21.34738 -2.03047 0.0424 pause -49.0152 53.76631 -0.91163 0.3620 approximant 48.9818 25.28020 1.93755 0.0528 fricative -67.3185 22.84397 -2.94688 0.0032 lateral 113.0080 81.87885 1.38019 0.1676 nasal -47.1044 24.25019 -1.94244 0.0522 rhotic 44.7936 34.52355 1.29748 0.1946 Interaction: Vowel and Liquid Type /a/ -19.9261 15.48326 -1.28695 0.1982 glide+/a/ 124.8506 27.71657 4.50455 0.0000 glide+/e/ 7.6019 33.70828 0.22552 0.8216 /e/ -43.5996 19.38775 -2.24882 0.0246 /i/ 26.6964 29.69169 0.89912 0.3687 /o/ -119.7937 20.94813 -5.71859 0.0000 /u/ 24.1705 35.49187 0.68102 0.4959
F2.2 Estimate Standard Error t-Value p-value (Intercept) 1567.8434 29.40705 53.31523 0.0000 Stress stressed -7.6742 8.87251 -0.86494 0.3872 unstressed 7.6742 8.87251 0.86494 0.3872 Word Position word-medial -10.7114 10.20858 -1.04926 0.2942 word-final 10.7114 10.20858 1.04926 0.2942 Vowel /a/ -33.5013 15.35472 -2.18183 0.0292 glide+/a/ -39.0087 27.37141 -1.42516 0.1542 glide +/e/ 25.1658 34.37243 0.73215 0.4641 /e/ 159.5238 19.99003 7.98017 0.0000 /i/ 490.0979 29.25698 16.75149 0.0000 251 /o/ -241.1332 20.36392 -11.84120 0.0000 /u/ -361.1443 34.74487 -10.39418 0.0000 Preceding Sound stop -40.8855 21.20100 -1.92847 0.0539 pause 117.0894 53.17811 2.20183 0.0278 approximant -36.1770 24.87124 -1.45457 0.1459 fricative 9.5944 22.66733 0.42327 0.6721 lateral -151.3312 80.53016 -1.87919 0.0603 nasal 41.5823 24.06466 1.72794 0.0841 rhotic 60.1276 34.23591 1.75627 0.0792 Following Sound stop 16.2414 12.03364 1.34967 0.1772 pause 36.7350 17.93217 2.04855 0.0406 approximant 3.4447 16.84816 0.20446 0.8380 fricative -53.7362 20.26412 -2.65179 0.0081 nasal 21.8808 16.71526 1.30903 0.1906 vowel -24.5658 18.07429 -1.35916 0.1742 Gender female 86.0145 22.93706 3.75002 0.0012 male -86.0145 22.93706 -3.75002 0.0012 Age younger 21.4554 23.69474 0.90549 0.3755 older -21.4554 23.69474 -0.90549 0.3755 Liquid Type /r/ 45.3534 19.21989 2.35971 0.0184 /l/ -45.3534 19.21989 -2.35971 0.0184 Interaction: Word Position and Liquid Type word-medial 19.0798 9.15131 2.08492 0.0372 word-final -19.0798 9.15131 -2.08492 0.0372 Interaction: Preceding Sound and Liquid Type stop -48.7374 21.05419 -2.31485 0.0207 pause -33.9317 53.02790 -0.63988 0.5223 approximant 28.1576 24.93299 1.12933 0.2589 fricative -59.8702 22.53023 -2.65733 0.0079 lateral 144.8543 80.75434 1.79377 0.0730 nasal -48.5983 23.91713 -2.03195 0.0423 rhotic 18.1257 34.04937 0.53234 0.5945 252 Interaction: Vowel and Liquid Type /a/ 10.0130 15.27060 0.65570 0.5121 glide+/a/ 16.9777 27.33591 0.62108 0.5346 glide+/e/ -16.6141 33.24539 -0.49974 0.6173 /e/ -15.2645 19.12150 -0.79829 0.4248 /i/ 56.7303 29.28398 1.93725 0.0528 /o/ -59.5837 20.66044 -2.88395 0.0040 /u/ -361.1443 34.74487 -10.39418 0.0000
F2.3 Estimate Standard Error t-Value p-value (Intercept) 1589.0682 31.69543 50.13556 0.0000 Stress stressed -4.1160 8.94298 -0.46024 0.6454 unstressed 4.1160 8.94298 0.46024 0.6454 Word Position word-medial -8.8331 10.28984 -0.85843 0.3907 word-final 8.8331 10.28984 0.85843 0.3907 Vowel /a/ -54.3805 15.47750 -3.51352 0.0004 glide+/a/ -30.0434 27.58990 -1.08893 0.2763 glide +/e/ 90.8139 34.65041 2.62086 0.0088 /e/ 126.5855 20.14959 6.28229 0.0000 /i/ 347.1687 29.49099 11.77203 0.0000 /o/ -184.2897 20.52633 -8.97821 0.0000 /u/ -295.8546 35.02194 -8.44769 0.0000 Preceding Sound stop -11.2877 21.36995 -0.52821 0.5974 pause 171.9063 53.60050 3.20718 0.0014 approximant -22.5878 25.06846 -0.90104 0.3676 fricative -3.7960 22.84820 -0.16614 0.8681 lateral -166.1794 81.16943 -2.04731 0.0407 nasal 24.0043 24.25651 0.98960 0.3225 rhotic 7.9404 34.50743 0.23011 0.8180 Following Sound stop 21.1896 12.13042 1.74681 0.0808 pause 70.4044 18.07440 3.89526 0.0001 approximant -24.4411 16.98250 -1.43919 0.1502 fricative -31.2921 20.42468 -1.53208 0.1256 nasal -27.1467 16.84726 -1.61134 0.1072 vowel -8.7141 18.21820 -0.47832 0.6325 253 Gender female 91.5418 25.69992 3.56195 0.0018 male -91.5418 25.69992 -3.56195 0.0018 Age younger 13.0223 26.54770 0.49053 0.6288 older -13.0223 26.54770 -0.49053 0.6288 Liquid Type /r/ 31.7698 19.37219 1.63997 0.1011 /l/ -31.7698 19.37219 -1.63997 0.1011 Interaction: Word Position and Liquid Type word-medial 23.1866 9.22461 2.51356 0.0120 word-final -23.1866 9.22461 -2.51356 0.0120 Interaction: Preceding Sound and Liquid Type stop -62.3449 21.22141 -2.93783 0.0033 pause -8.2789 53.45000 -0.15489 0.8769 approximant 19.6215 25.13093 0.78077 0.4350 fricative -50.1965 22.70941 -2.21038 0.0272 lateral 177.4716 81.39643 2.18034 0.0293 nasal -35.3215 24.10713 -1.46519 0.1430 rhotic -40.9511 34.31872 -1.19326 0.2329 Interaction: Vowel and Liquid Type /a/ 17.0339 15.39150 1.10671 0.2685 glide+/a/ 15.1158 27.55303 0.54861 0.5833 glide+/e/ -38.9565 33.51237 -1.16245 0.2452 /e/ 16.4529 19.27433 0.85362 0.3934 /i/ 95.3493 29.52020 3.22997 0.0013 /o/ -47.9474 20.82559 -2.30233 0.0214 /u/ -57.0479 35.28395 -1.61682 0.1060 F2.4 Estimate Standard Error t-Value p-value (Intercept) 1616.4573 29.38053 55.01798 0.0000 Stress stressed -7.0336 9.17041 -0.76699 0.4432 unstressed 7.0336 9.17041 0.76699 0.4432 Word Position word-medial -8.8944 10.55121 -0.84298 0.3993 word-final 8.8944 10.55121 0.84298 0.3993 254 Vowel /a/ -48.7584 15.86971 -3.07242 0.0021 glide+/a/ -14.9269 28.28968 -0.52765 0.5978 glide +/e/ 36.9770 35.52324 1.04092 0.2980 /e/ 93.0471 20.66068 4.50358 0.0000 /i/ 303.4031 30.23822 10.03376 0.0000 /o/ -142.5483 21.04721 -6.77279 0.0000 /u/ -227.1936 35.91071 -6.32663 0.0000 Preceding Sound stop -9.6777 21.91245 -0.44165 0.6588 pause 134.6518 54.96353 2.44984 0.0144 approximant -24.9560 25.70649 -0.97081 0.3317 fricative 10.6702 23.42784 0.45545 0.6488 lateral -190.3265 83.23417 -2.28664 0.0223 nasal 8.3225 24.87215 0.33461 0.7379 rhotic 71.3158 35.38562 2.01539 0.0440 Following Sound stop 36.9527 12.43689 2.97122 0.0030 pause 62.8218 18.53437 3.38947 0.0007 approximant -2.4636 17.41350 -0.14148 0.8875 fricative -15.4510 20.94473 -0.73770 0.4608 nasal -25.4552 17.27696 -1.47336 0.1408 vowel -56.4047 18.68091 -3.01937 0.0026 Gender female 101.1035 22.39327 4.51491 0.0002 male -101.1035 22.39327 -4.51491 0.0002 Age younger 2.9980 23.13363 0.12960 0.8981 older -2.9980 23.13363 -0.12960 0.8981 Liquid Type /r/ 25.8306 19.86542 1.30028 0.1936 /l/ -25.8306 19.86542 -1.30028 0.1936 Interaction: Word Position and Liquid Type word-medial 17.7718 9.45816 1.87900 0.0604 word-final -17.7718 9.45816 -1.87900 0.0604 Interaction: Preceding Sound and Liquid Type stop -54.8502 21.76108 -2.52056 0.0118 255 pause 38.6617 54.80769 0.70541 0.4806 approximant 22.0633 25.77017 0.85616 0.3920 fricative -46.8235 23.28653 -2.01075 0.0445 lateral 138.0970 83.46519 1.65455 0.0981 nasal -19.6314 24.72013 -0.79414 0.4272 rhotic -77.5170 35.19328 -2.20261 0.0277 Interaction: Vowel and Liquid Type /a/ 19.0750 15.78356 1.20854 0.2269 glide+/a/ 24.5429 28.25370 0.86866 0.3851 glide+/e/ -30.5021 34.35970 -0.88773 0.3748 /e/ 9.2139 19.76287 0.46622 0.6411 /i/ 93.9531 30.26483 3.10437 0.0019 /o/ -68.1955 21.35344 -3.19365 0.0014 /u/ -48.0873 36.17877 -1.32916 0.1839
F2.5 Estimate Standard Error t-Value p-value (Intercept) 1595.5752 28.81203 55.37879 0.0000 Stress stressed 4.0720 9.26843 0.43934 0.6604 unstressed -4.0720 9.26843 -0.43934 0.6604 Word Position word-medial 12.2256 10.66387 1.14645 0.2517 word-final -12.2256 10.66387 -1.14645 0.2517 Vowel /a/ -51.6755 16.03877 -3.22191 0.0013 glide+/a/ -19.7653 28.59132 -0.69130 0.4894 glide +/e/ -2.0431 35.89953 -0.05691 0.9546 /e/ 123.0436 20.88098 5.89262 0.0000 /i/ 249.5991 30.56030 8.16743 0.0000 /o/ -126.5560 21.27173 -5.94949 0.0000 /u/ -77.7278 8.667974 -8.96724 0.0000 Preceding Sound stop -25.2090 22.14627 -1.13829 0.2551 pause 122.9631 55.55100 2.21352 0.0269 approximant -4.2281 25.98148 -0.16274 0.8707 fricative 21.4747 23.67769 0.90696 0.3645 lateral -142.9206 84.12406 -1.69893 0.0895 nasal 24.3506 25.13751 0.96870 0.3328 rhotic 3.5693 35.76412 0.09980 0.9205 Following Sound 256 stop 22.5548 12.56900 1.79448 0.0729 pause 68.2968 18.73262 3.64588 0.0003 approximant -34.2889 17.59928 -1.94831 0.0515 fricative 19.3469 21.16887 0.91393 0.3608 nasal -71.1754 17.46214 -4.07598 0.0000 vowel -4.7342 18.88035 -0.25075 0.8020 Gender female 98.6926 21.46241 4.59839 0.0002 male -98.6926 21.46241 -4.59839 0.0002 Age younger -10.8293 22.17263 -0.48841 0.6303 older 10.8293 22.17263 0.48841 0.6303 Liquid Type /r/ 27.8162 20.07799 1.38541 0.1660 /l/ -27.8162 20.07799 -1.38541 0.1660 Interaction: Word Position and Liquid Type word-medial 9.4780 9.55883 0.99154 0.3215 word-final -9.4780 9.55883 -0.99154 0.3215 Interaction: Preceding Sound and Liquid Type stop -29.3433 21.99368 -1.33417 0.1823 pause 1.0813 55.39287 0.01952 0.9844 approximant 44.5308 26.04569 1.70972 0.0874 fricative -66.7458 23.53528 -2.83599 0.0046 lateral 175.4509 84.35684 2.07987 0.0376 nasal -84.8802 24.98433 -3.39734 0.0007 rhotic -40.0937 35.57020 -1.12717 0.2598 Interaction: Vowel and Liquid Type /a/ 23.1663 15.95253 1.45220 0.1466 glide+/a/ -17.5806 28.55569 -0.61566 0.5382 glide+/e/ 16.1075 34.72497 0.46386 0.6428 /e/ -13.7051 19.97346 -0.68617 0.4927 /i/ 109.7872 30.58585 3.58948 0.0003 /o/ -33.7917 21.58099 -1.56581 0.1175 /u/ -83.9836 36.56448 -2.29686 0.0217
F2.6 Estimate Standard Error t-Value p-value 257 (Intercept) 1587.6002 29.17723 54.41230 0.0000 Stress stressed 20.3618 9.25964 2.19898 0.0280 unstressed -20.3618 9.25964 -2.19898 0.0280 Word Position word-medial 10.0227 10.65382 0.94076 0.3469 word-final -10.0227 10.65382 -0.94076 0.3469 Vowel /a/ -151.7853 36.25974 -4.18606 0.0000 glide+/a/ -10.1161 28.56457 -0.35415 0.7233 glide +/e/ 19.8664 35.86709 0.55389 0.5797 /e/ 116.1150 20.86144 5.56601 0.0000 /i/ 157.7999 30.53186 5.16837 0.0000 /o/ -86.3674 21.25177 -4.06401 0.0000 /u/ -151.7853 36.25974 -4.18606 0.0000 Preceding Sound stop 35.9502 35.73009 1.00616 0.3144 pause 144.5230 55.49835 2.60410 0.0093 approximant -28.4775 25.95675 -1.09711 0.2727 fricative 31.9078 23.65551 1.34885 0.1775 lateral -181.5121 84.04422 -2.15972 0.0309 nasal 0.9863 25.11391 0.03927 0.9687 rhotic 35.9502 35.73009 1.00616 0.3144 Following Sound stop 24.3843 12.55747 1.94181 0.0523 pause 6.4711 18.71480 0.34578 0.7295 approximant -18.2307 17.58276 -1.03685 0.2999 fricative 91.0551 21.14868 4.30547 0.0000 nasal -94.4131 17.44535 -5.41194 0.0000 vowel -9.2667 18.86256 -0.49127 0.6233 Gender female 107.7585 21.96583 4.90574 0.0001 male -107.7585 21.96583 -4.90574 0.0001 Age younger -17.4857 22.69240 -0.77055 0.4496 older 17.4857 22.69240 0.77055 0.4496 Liquid Type /r/ 30.3047 20.05885 1.51079 0.1310 /l/ -30.3047 20.05885 -1.51079 0.1310 Interaction: Word Position and Liquid Type 258 word-medial 8.5932 9.54997 0.89982 0.3683 word-final -8.5932 9.54997 -0.89982 0.3683 Interaction: Preceding Sound and Liquid Type stop -73.2644 21.97283 -3.33432 0.0009 pause 19.6806 55.34066 0.35563 0.7221 approximant 30.3083 26.02097 1.16477 0.2442 fricative -66.0048 23.51304 -2.80716 0.0050 lateral 246.1030 84.27711 2.92016 0.0035 nasal -101.5109 24.96066 -4.06684 0.0000 rhotic -55.3118 35.53614 -1.55650 0.1197 Interaction: Vowel and Liquid Type /a/ 23.1797 15.93728 1.45443 0.1459 glide+/a/ -0.9819 28.52863 -0.03442 0.9725 glide+/e/ 17.8334 34.69299 0.51404 0.6073 /e/ 0.2849 19.95483 0.01428 0.9886 /i/ 84.4645 30.55801 2.76407 0.0057 /o/ -33.8573 21.56086 -1.57031 0.1165 /u/ -90.9233 36.53029 -2.48898 0.0129
F2.7 Estimate Standard Error t-Value p-value (Intercept) 1627.4840 28.61633 56.87256 0.0000 Stress stressed 4.2571 9.10960 0.46731 0.6403 unstressed -4.2571 9.10960 -0.46731 0.6403 Word Position word-medial -6.0590 10.48118 -0.57808 0.5633 word-final 6.0590 10.48118 0.57808 0.5633 Vowel /a/ -20.2198 15.76414 -1.28265 0.1997 glide+/a/ 21.2185 28.10165 0.75506 0.4503 glide +/e/ 14.2095 35.28558 0.40270 0.6872 /e/ 58.7265 20.52336 2.86145 0.0043 /i/ 121.0871 30.03703 4.03126 0.0001 /o/ -78.8153 20.90738 -3.76974 0.0002 /u/ -116.2064 35.67214 -3.25762 0.0011 Preceding Sound stop -31.6586 21.76692 -1.45444 0.1459 pause 109.5921 54.59909 2.00721 0.0448 259 approximant -58.8930 25.53618 -2.30626 0.0212 fricative -32.6146 23.27216 -1.40144 0.1612 lateral 37.3845 82.68245 0.45215 0.6512 nasal -10.7249 24.70694 -0.43409 0.6643 rhotic -13.0856 35.15117 -0.37227 0.7097 Following Sound stop 0.4033 18.55690 0.02174 0.9827 pause 32.1071 18.41157 1.74385 0.0813 approximant -17.9842 17.29783 -1.03968 0.2986 fricative 129.7859 20.80603 6.23790 0.0000 nasal -153.3764 17.16273 -8.93660 0.0000 vowel 0.4033 18.55690 0.02174 0.9827 Gender female 107.2050 21.49279 4.98795 0.0001 male -107.2050 21.49279 -4.98795 0.0001 Age younger -12.8056 22.20377 -0.57673 0.5703 older 12.8056 22.20377 0.57673 0.5703 Liquid Type /r/ -0.0336 19.73385 -0.00171 0.9986 /l/ 0.0336 19.73385 0.00171 0.9986 Interaction: Word Position and Liquid Type word-medial 20.1383 9.39518 2.14348 0.0322 word-final -20.1383 9.39518 -2.14348 0.0322 Interaction: Preceding Sound and Liquid Type stop -18.5185 21.61680 -0.85667 0.3917 pause 108.2029 54.44390 1.98742 0.0470 approximant 53.1692 25.59935 2.07697 0.0379 fricative -7.1608 23.13203 -0.30956 0.7569 lateral -76.1822 82.91149 -0.91884 0.3583 nasal -54.0201 24.55621 -2.19985 0.0279 rhotic -5.4904 34.96041 -0.15705 0.8752 Interaction: Vowel and Liquid Type /a/ 23.1947 15.67907 1.47934 0.1392 glide+/a/ 1.8492 28.06637 0.06589 0.9475 glide+/e/ 10.2142 34.13064 0.29927 0.7648 260 /e/ 36.9229 19.63143 1.88080 0.0601 /i/ 53.6716 30.06262 1.78533 0.0743 /o/ -54.2147 21.21143 -2.55592 0.0106 /u/ -71.6379 35.93828 -1.99336 0.0463
F3.1 Estimate Standard Error t-Value p-value (Intercept) 2524.8786 41.39290 60.99787 0.0000 Stress stressed 5.8331 7.69741 0.75781 0.4486 unstressed -5.8331 7.69741 -0.75781 0.4486 Word Position word-medial -27.4468 8.86014 -3.09778 0.0020 word-final 27.4468 8.86014 3.09778 0.0020 Vowel /a/ -128.9651 13.37044 -9.64554 0.0000 glide+/a/ -8.1558 23.75288 -0.34336 0.7314 glide +/e/ -34.5176 29.93638 -1.15303 0.2490 /e/ 4.9083 17.37436 0.28250 0.7776 /i/ 94.7682 25.38553 3.73316 0.0002 /o/ -13.2181 17.68304 -0.74750 0.4548 /u/ 85.1802 30.15094 2.82512 0.0048 Preceding Sound stop -39.0818 18.40032 -2.12397 0.0338 pause 53.5675 46.16842 1.16026 0.2460 approximant 14.1539 21.58230 0.65581 0.5120 fricative 43.3606 19.69181 2.20196 0.0278 lateral 35.6191 70.06246 0.50839 0.6112 nasal -26.0701 20.89912 -1.24742 0.2124 rhotic -81.5493 29.70196 -2.74559 0.0061 Following Sound stop 11.1049 10.44755 1.06292 0.2879 pause -27.2676 15.55538 -1.75294 0.0797 approximant 15.9246 14.61919 1.08930 0.2761 fricative -54.9399 17.58126 -3.12491 0.0018 nasal 46.0302 14.49816 3.17490 0.0015 vowel 9.1478 15.68296 0.58330 0.5597 Gender female 113.2261 38.18209 2.96542 0.0074 male -113.2261 38.18209 -2.96542 0.0074 Age younger -6.3219 39.56373 -0.15979 0.8746 older 6.3219 39.56373 0.15979 0.8746 261 Liquid Type /r/ 9.9383 16.72450 0.59424 0.5524 /l/ -9.9383 16.72450 -0.59424 0.5524 Interaction: Word Position and Liquid Type word-medial 10.1525 7.94083 1.27851 0.2012 word-final -10.1525 7.94083 -1.27851 0.2012 Interaction: Preceding Sound and Liquid Type stop 2.4574 18.26691 0.13453 0.8930 pause 42.0271 46.08483 0.91195 0.3619 approximant 5.2638 21.63374 0.24332 0.8078 fricative -11.4335 19.55221 -0.58477 0.5588 lateral -63.3328 70.21279 -0.90201 0.3671 nasal 16.3820 20.77348 0.78860 0.4304 rhotic 8.6359 29.53589 0.29239 0.7700 Interaction: Age and Liquid Type younger -10.1784 6.33600 -1.60643 0.1083 older 10.1784 6.33600 1.60643 0.1083 Interaction: Vowel and Liquid Type /a/ -3.7444 13.28567 -0.28183 0.7781 glide+/a/ 38.3335 23.75126 1.61396 0.1067 glide+/e/ -11.1964 29.03785 -0.38558 0.6998 /e/ -45.6888 16.61094 -2.75052 0.0060 /i/ 26.5655 25.41550 1.04525 0.2960 /o/ -43.1660 17.95111 -2.40464 0.0163 /u/ 38.8964 30.39081 1.27988 0.2007
F3.2 Estimate Standard Error t-Value p-value (Intercept) 2493.4998 40.45592 61.63498 0.0000 Stress stressed 18.7977 7.72533 2.43325 0.0150 unstressed -18.7977 7.72533 -2.43325 0.0150 Word Position word-medial -24.0419 8.89227 -2.70368 0.0069 word-final 24.0419 8.89227 2.70368 0.0069 262 Vowel /a/ -79.8161 13.41887 -5.94805 0.0000 glide+/a/ -37.7040 23.83896 -1.58161 0.1139 glide +/e/ -34.4710 30.04457 -1.14733 0.2514 /e/ 41.4471 17.43733 2.37692 0.0175 /i/ 94.4438 25.47749 3.70695 0.0002 /o/ -15.5635 17.74714 -0.87696 0.3806 /u/ 31.6637 30.26023 1.04638 0.2955 Preceding Sound stop -25.1532 18.46702 -1.36206 0.1733 pause 52.8339 46.33591 1.14024 0.2543 approximant 0.2036 21.66061 0.00940 0.9925 fricative 38.5518 19.76317 1.95069 0.0512 lateral 32.2675 70.31664 0.45889 0.6464 nasal -20.9362 20.97488 -0.99815 0.3183 rhotic -77.7675 29.80974 -2.60879 0.0091 Following Sound stop 1.4167 10.48535 0.13511 0.8925 pause 7.7258 15.61182 0.49487 0.6207 approximant 13.1148 14.67218 0.89386 0.3715 fricative -43.9356 17.64506 -2.48997 0.0128 nasal 22.2132 14.55081 1.52660 0.1270 vowel -0.5349 15.73983 -0.03398 0.9729 Gender female 103.9074 37.13930 2.79778 0.0108 male -103.9074 37.13930 -2.79778 0.0108 Age younger -2.1281 38.49133 -0.05529 0.9564 older 2.1281 38.49133 0.05529 0.9564 Liquid Type /r/ 2.5431 16.78521 0.15151 0.8796 /l/ -2.5431 16.78521 -0.15151 0.8796 Interaction: Word Position and Liquid Type word-medial 18.6623 7.96959 2.34169 0.0193 word-final -18.6623 7.96959 -2.34169 0.0193 Interaction: Preceding Sound and Liquid Type stop 6.5024 18.33318 0.35468 0.7229 263 pause 34.9155 46.25194 0.75490 0.4504 approximant 29.5610 21.71223 1.36149 0.1735 fricative -21.1101 19.62312 -1.07578 0.2821 lateral -67.2223 70.46744 -0.95395 0.3402 nasal 13.6547 20.84883 0.65494 0.5126 rhotic 3.6988 29.64313 0.12478 0.9007 Interaction: Age and Liquid Type younger -19.4586 6.35900 -3.06001 0.0022 older 19.4586 6.35900 3.06001 0.0022 Interaction: Vowel and Liquid Type /a/ 3.8491 13.33389 0.28867 0.7729 glide+/a/ 19.6116 23.83742 0.82272 0.4107 glide+/e/ 10.3105 29.14293 0.35379 0.7235 /e/ -47.8303 16.67113 -2.86905 0.0042 /i/ 42.9296 25.50741 1.68303 0.0925 /o/ -40.0614 18.01615 -2.22364 0.0263 /u/ 11.1910 30.50094 0.36691 0.7137
F3.3 Estimate Standard Error t-Value p-value (Intercept) 2488.6179 39.38174 63.19218 0.0000 Stress stressed 12.9459 8.07475 1.60325 0.1090 unstressed -12.9459 8.07475 -1.60325 0.1090 Word Position word-medial -14.2234 9.29443 -1.53032 0.1261 word-final 14.2234 9.29443 1.53032 0.1261 Vowel /a/ -52.5596 14.02559 -3.74741 0.0002 glide+/a/ -24.2031 24.91694 -0.97135 0.3315 glide +/e/ -33.1688 31.40221 -1.05626 0.2909 /e/ 44.8489 18.22582 2.46074 0.0139 /i/ 95.2987 26.62944 3.57870 0.0004 /o/ -23.7100 18.54969 1.27819 0.2013 /u/ -6.5062 31.62864 -0.20571 0.8370 Preceding Sound stop -22.6456 19.30215 -1.17322 0.2408 pause 49.3180 48.43173 1.01830 0.3086 approximant -11.0328 22.64042 -0.48731 0.6261 fricative 36.9485 20.65683 1.78868 0.0738 264 lateral 55.2537 73.49714 0.75178 0.4523 nasal -25.6627 21.92341 -1.17056 0.2419 rhotic -82.1792 31.15815 -2.63749 0.0084 Following Sound stop 0.9569 10.95929 0.08731 0.9304 pause 44.4941 16.31801 2.72669 0.0064 approximant -11.9883 15.33567 -0.78173 0.4344 fricative -48.6456 18.44325 -2.63758 0.0084 nasal -3.2072 15.20915 -0.21087 0.8330 vowel 18.3901 16.45166 1.11783 0.2637 Gender female 107.2722 35.63385 3.01040 0.0067 male -107.2722 35.63385 -3.01040 0.0067 Age younger -10.5647 36.95505 -0.28588 0.7778 older 10.5647 36.95505 0.28588 0.7778 Liquid Type /r/ -11.7280 17.54452 -0.66847 0.5039 /l/ 11.7280 17.54452 0.66847 0.5039 Interaction: Word Position and Liquid Type word-medial 23.6357 8.32990 2.83745 0.0046 word-final -23.6357 8.32990 -2.83745 0.0046 Interaction: Preceding Sound and Liquid Type stop 5.5933 19.16240 0.29189 0.7704 pause 1.2662 48.34372 0.02619 0.9791 approximant 59.2598 22.69432 2.61122 0.0091 fricative -12.3538 20.51062 -0.60231 0.5470 lateral -63.0782 73.65452 -0.85641 0.3919 nasal 28.5141 21.79183 1.30848 0.1908 rhotic -19.2015 30.98419 -0.61972 0.5355 Interaction: Age and Liquid Type younger -13.6620 6.64667 -2.05546 0.0399 older 13.6620 6.64667 2.05546 0.0399 Interaction: Vowel and Liquid Type 265 /a/ -3.8323 13.93708 -0.27497 0.7834 glide+/a/ 12.7420 24.91560 0.51141 0.6091 glide+/e/ 31.8770 30.46026 1.04651 0.2954 /e/ -16.0621 17.42492 -0.92179 0.3567 /i/ 55.7927 26.66020 2.09273 0.0365 /o/ -60.6290 18.83076 -3.21968 0.0013 /u/ -19.8881 31.88014 -0.62384 0.5328
F3.4 Estimate Standard Error t-Value p-value (Intercept) 2524.1920 38.23903 66.01088 0.0000 Stress stressed 6.3008 8.46582 0.74426 0.4568 unstressed -6.3008 8.46582 -0.74426 0.4568 Word Position word-medial -19.0862 9.74450 -1.95867 0.0503 word-final 19.0862 9.74450 1.95867 0.0503 Vowel /a/ -51.2918 14.70452 -3.48817 0.0005 glide+/a/ -23.8894 26.12328 -0.91449 0.3605 glide+/e/ -1.9643 32.92120 -0.05967 0.9524 /e/ 54.6549 19.10821 2.86028 0.0043 /i/ 97.9444 27.91853 3.50822 0.0005 /o/ -62.2106 19.44782 -3.19885 0.0014 /u/ -13.2431 33.16002 -0.39937 0.6897 Preceding Sound stop -24.8949 20.23674 -1.23018 0.2187 pause 23.6236 50.77730 0.46524 0.6418 approximant -31.6787 23.73701 -1.33457 0.1821 fricative 31.7615 21.65691 1.46658 0.1426 lateral 78.7555 77.05666 1.02205 0.3069 nasal -19.2154 22.98491 -0.83600 0.4032 rhotic -58.3518 32.66727 -1.78625 0.0742 Following Sound stop 8.2809 11.48961 0.72073 0.4711 pause 60.6344 17.10836 3.54414 0.0004 approximant 9.3771 16.07819 0.58322 0.5598 fricative -61.8834 19.33659 -3.20032 0.0014 nasal -4.6398 15.94600 -0.29097 0.7711 vowel -11.7693 17.24828 -0.68234 0.4951 Gender female 117.9018 33.95876 3.47191 0.0023 male -117.9018 33.95876 -3.47191 0.0023 266 Age younger -11.1413 35.24837 -0.31608 0.7551 older 11.1413 35.24837 0.31608 0.7551 Liquid Type /r/ -23.8816 18.39435 -1.29831 0.1943 /l/ 23.8816 18.39435 1.29831 0.1943 Interaction: Word Position and Liquid Type word-medial 21.0665 8.73309 2.41226 0.0159 word-final -21.0665 8.73309 -2.41226 0.0159 Interaction: Preceding Sound and Liquid Type stop 19.3704 20.09043 0.96416 0.3351 pause 33.8746 50.68469 0.66834 0.5040 approximant 63.5938 23.79345 2.67274 0.0076 fricative -14.4623 21.50386 -0.67254 0.5013 lateral -82.0660 77.22133 -1.06274 0.2880 nasal 5.3578 22.84719 0.23451 0.8146 rhotic -25.6683 32.48516 -0.79015 0.4295 Interaction: Age and Liquid Type younger -11.3468 6.96863 -1.62826 0.1036 older 11.3468 6.96863 1.62826 0.1036 Interaction: Vowel and Liquid Type /a/ 8.5883 14.61218 0.58775 0.5568 glide+/a/ 24.8779 26.12225 0.95236 0.3410 glide+/e/ -14.6605 31.93429 -0.45908 0.6462 /e/ -19.2537 18.26846 -1.05393 0.2920 /i/ 41.6858 27.95006 1.49144 0.1360 /o/ -51.8892 19.74235 -2.62832 0.0086 /u/ 10.6516 33.42358 0.31868 0.7500
F3.5 Estimate Standard Error t-Value p-value (Intercept) 2541.0154 37.34717 68.03770 0.0000 Stress stressed 8.2526 8.58699 0.96106 0.3366 unstressed -8.2526 8.58699 -0.96106 0.3366 267 Word Position word-medial -4.6874 9.88394 -0.47424 0.6354 word-final 4.6874 9.88394 0.47424 0.6354 Vowel /a/ -54.2824 14.91481 -3.63950 0.0003 glide+/a/ -6.6506 26.49697 -0.25099 0.8018 glide +/e/ -21.4692 33.39134 -0.64296 0.5203 /e/ 51.9019 19.38154 2.67790 0.0075 /i/ 96.9010 28.31780 3.42191 0.0006 /o/ -58.2793 19.72605 -2.95443 0.0032 /u/ -8.1215 33.63443 -0.24146 0.8092 Preceding Sound stop -38.0458 20.52628 -1.85352 0.0639 pause 59.8186 51.50412 1.16143 0.2456 approximant -25.6128 24.07684 -1.06379 0.2875 fricative 21.7702 21.96670 0.99105 0.3218 lateral 60.0835 78.15964 0.76873 0.4421 nasal -23.4712 23.31376 -1.00675 0.3141 rhotic -54.5424 33.13494 -1.64607 0.0999 Following Sound stop 8.2217 11.65381 0.70550 0.4806 pause 50.8110 17.35328 2.92803 0.0034 approximant -28.0378 16.30821 -1.71924 0.0857 fricative -4.2655 19.61344 -0.21748 0.8279 nasal -32.3148 16.17441 -1.99790 0.0458 vowel 5.5854 17.49509 0.31925 0.7496 Gender female 115.9230 32.81592 3.53252 0.0020 male -115.9230 32.81592 -3.53252 0.0020 Age younger -21.6449 34.07938 -0.63513 0.5322 older 21.6449 34.07938 0.63513 0.5322 Liquid Type /r/ -39.4856 18.65772 -2.11631 0.0344 /l/ 39.4856 18.65772 2.11631 0.0344 Interaction: Word Position and Liquid Type word-medial 11.3396 8.85796 1.28016 0.2006 word-final -11.3396 8.85796 -1.28016 0.2006 Interaction: Preceding 268 Sound and Liquid Type stop 36.1891 20.37800 1.77589 0.0759 pause 23.3504 51.40998 0.45420 0.6497 approximant 41.4947 24.13404 1.71934 0.0857 fricative -21.4031 21.81160 -0.98127 0.3266 lateral -55.8207 78.32648 -0.71267 0.4761 nasal -37.0294 23.17420 -1.59787 0.1102 rhotic 13.2190 32.95038 0.40118 0.6883 Interaction: Age and Liquid Type younger -7.5957 7.06841 -1.07460 0.2827 older 7.5957 7.06841 1.07460 0.2827 Interaction: Vowel and Liquid Type /a/ 20.2521 14.82140 1.36641 0.1719 glide+/a/ 0.2612 26.49615 0.00986 0.9921 glide+/e/ 28.8115 32.39069 0.88950 0.3738 /e/ -7.5865 18.52974 -0.40942 0.6823 /i/ 41.5044 28.34935 1.46403 0.1433 /o/ -24.7208 20.02471 -1.23451 0.2171 /u/ -58.5219 33.90168 -1.72622 0.0844
F3.6 Estimate Standard Error t-Value p-value (Intercept) 2582.7979 38.02956 67.91553 0.0000 Stress stressed 11.7246 8.54980 1.37133 0.1704 unstressed -11.7246 8.54980 -1.37133 0.1704 Word Position word-medial -9.9873 9.84115 -1.01486 0.3103 word-final 9.9873 9.84115 1.01486 0.3103 Vowel /a/ -66.3388 14.85032 -4.46716 0.0000 glide+/a/ -8.9564 26.38234 -0.33948 0.7343 glide +/e/ 72.8485 33.24736 2.19111 0.0285 /e/ 21.6282 19.29769 1.12077 0.2625 /i/ 36.6431 28.19535 1.29961 0.1938 /o/ -68.3115 19.64069 -3.47806 0.0005 /u/ 12.4869 33.48888 0.37287 0.7093 Preceding Sound stop -36.7628 20.43744 -1.79880 0.0722 269 pause 83.6644 51.28102 1.63149 0.1029 approximant -41.6269 23.97251 -1.73644 0.0826 fricative 26.5094 21.87167 1.21205 0.2256 lateral 35.9919 77.82108 0.46250 0.6438 nasal -30.5168 23.21287 -1.31465 0.1887 rhotic -37.2593 32.99136 -1.12936 0.2588 Following Sound stop -13.8791 11.60349 -1.19612 0.2318 pause 22.1593 17.27809 1.28251 0.1998 approximant -23.4545 16.23764 -1.44445 0.1487 fricative 71.8648 19.52844 3.68001 0.0002 nasal -42.2016 16.10425 -2.62053 0.0088 vowel -14.4888 17.41935 -0.83177 0.4056 Gender female 111.8368 33.63129 3.32538 0.0032 male -111.8368 33.63129 -3.32538 0.0032 Age younger -26.9975 34.91540 -0.77323 0.4480 older 26.9975 34.91540 0.77323 0.4480 Liquid Type /r/ -31.4637 18.57685 -1.69370 0.0904 /l/ 31.4637 18.57685 1.69370 0.0904 Interaction: Word Position and Liquid Type word-medial 20.5064 8.81967 2.32507 0.0201 word-final -20.5064 8.81967 -2.32507 0.0201 Interaction: Preceding Sound and Liquid Type stop 6.2317 20.28973 0.30713 0.7588 pause 69.6334 51.18741 1.36036 0.1738 approximant 51.9766 24.02949 2.16303 0.0306 fricative -32.2062 21.71716 -1.48298 0.1382 lateral -18.5448 77.98731 -0.23779 0.8121 nasal -53.0725 23.07383 -2.30012 0.0215 rhotic -24.0181 32.80751 -0.73209 0.4642 Interaction: Age and Liquid Type younger 2.7739 7.03777 0.39415 0.6935 older -2.7739 7.03777 -0.39415 0.6935 270 Interaction: Vowel and Liquid Type /a/ 21.3278 14.75716 1.44525 0.1485 glide+/a/ 27.5095 26.38138 1.04276 0.2972 glide+/e/ -42.9450 32.25081 -1.33159 0.1831 /e/ 14.2408 18.44961 0.77187 0.4403 /i/ 68.1043 28.22702 2.41274 0.0159 /o/ -9.6525 19.93811 -0.48412 0.6283 /u/ -78.5850 33.75502 -2.32810 0.0200
F3.7 Estimate Standard Error t-Value p-value (Intercept) 2593.7180 34.58039 75.00545 0.0000 Stress stressed -0.6559 8.44652 -0.07765 0.9381 unstressed 0.6559 8.44652 0.07765 0.9381 Word Position word-medial -29.4158 9.72218 -3.02564 0.0025 word-final 29.4158 9.72218 3.02564 0.0025 Vowel /a/ -42.9765 14.67046 -2.92946 0.0034 glide+/a/ -15.6844 26.06308 -0.60179 0.5474 glide +/e/ 75.2672 32.84308 2.29172 0.0220 /e/ -22.6877 19.06415 -1.19007 0.2341 /i/ 14.6150 27.85390 0.52470 0.5998 /o/ -63.0802 19.40309 -3.25104 0.0012 /u/ 54.5467 33.08375 1.64875 0.0993 Preceding Sound stop -27.4034 20.19025 -1.35726 0.1748 pause 109.4229 50.66156 2.15988 0.0309 approximant 0.8286 23.68308 0.03499 0.9721 fricative 16.7611 21.60697 0.77572 0.4380 lateral -27.9690 76.88105 -0.36380 0.7160 nasal -21.1984 22.93209 -0.92440 0.3554 rhotic -50.4417 32.59302 -1.54762 0.1218 Following Sound stop -26.6163 11.46268 -2.32200 0.0203 pause 35.1414 17.06947 2.05873 0.0396 approximant -30.1540 16.04121 -1.87979 0.0602 fricative 147.2262 19.29272 7.63118 0.0000 nasal -87.6899 15.91010 -5.51159 0.0000 vowel -37.9073 17.20874 -2.20279 0.0277 271 Gender female 104.7403 29.80288 3.51444 0.0021 male -104.7403 29.80288 -3.51444 0.0021 Age younger -29.6890 30.97968 -0.95834 0.3488 older 29.6890 30.97968 0.95834 0.3488 Liquid Type /r/ -2.5792 18.35266 -0.14054 0.8882 /l/ 2.5792 18.35266 0.14054 0.8882 Interaction: Word Position and Liquid Type word-medial 30.3634 8.71280 3.48492 0.0005 word-final -30.3634 8.71280 -3.48492 0.0005 Interaction: Preceding Sound and Liquid Type stop -12.1993 20.04463 -0.60861 0.5428 pause 65.7822 50.56860 1.30085 0.1934 approximant 0.1014 23.73927 0.00427 0.9966 fricative -31.9772 21.45468 -1.49045 0.1362 lateral -15.0273 77.04479 -0.19505 0.8454 nasal -16.2133 22.79506 -0.71126 0.4770 rhotic 9.5334 32.41177 0.29413 0.7687 Interaction: Age and Liquid Type younger 4.7410 6.95284 0.68188 0.4954 older -4.7410 6.95284 -0.68188 0.4954 Interaction: Vowel and Liquid Type /a/ 16.8074 14.57907 1.15284 0.2491 glide+/a/ -3.4966 26.06268 -0.13416 0.8933 glide+/e/ -31.8991 31.85953 -1.00124 0.3168 /e/ 26.6912 18.22623 1.46444 0.1432 /i/ 54.7802 27.88414 1.96456 0.0496 /o/ -9.1157 19.69670 -0.46280 0.6435 /u/ -53.7673 33.34650 -1.61238 0.1070
F4.1 Estimate Standard Error t-Value p-value (Intercept) 3496.434 42.51462 82.24074 0.0000 272 Stress stressed 17.830 10.79854 1.65119 0.0988 unstressed -17.830 10.79854 -1.65119 0.0988 Word Position word-medial -38.482 12.43562 -3.09447 0.0020 word-final 38.482 12.43562 3.09447 0.0020 Vowel /a/ -54.892 18.78311 -2.92239 0.0035 glide+/a/ 41.578 33.31143 1.24817 0.2121 /e/ -11.738 41.97218 -0.27966 0.7798 glide +/e/ 4.244 24.37498 0.17409 0.8618 /i/ 137.818 35.77460 3.85241 0.0001 /o/ -82.020 24.79157 -3.30839 0.0010 /u/ -34.991 42.27567 -0.82768 0.4079 Preceding Sound stop 2.419 25.80697 0.09375 0.9253 pause 89.239 64.71619 1.37892 0.1680 approximant 83.634 30.27027 2.76290 0.0058 fricative 119.227 27.65001 4.31202 0.0000 lateral -170.924 98.20827 -1.74042 0.0819 nasal -17.692 29.30388 -0.60374 0.5461 rhotic -105.903 41.86365 -2.52972 0.0115 Following Sound stop 19.186 14.67919 1.30702 0.1913 pause -40.235 21.86093 -1.84049 0.0658 approximant 33.708 20.52266 1.64249 0.1006 fricative -78.184 24.68407 -3.16739 0.0016 nasal 23.091 20.38793 1.13257 0.2575 vowel 42.434 21.99590 1.92917 0.0538 Gender female 120.988 36.12422 3.34922 0.0030 male -120.988 36.12422 -3.34922 0.0030 Age younger -50.445 37.57909 -1.34237 0.1938 older 50.445 37.57909 1.34237 0.1938 Liquid Type /r/ 35.098 23.46198 1.49594 0.1348 /l/ -35.098 23.46198 -1.49594 0.1348 Interaction: Word Position and Liquid Type word-medial 17.722 11.14150 1.59065 0.1118 273 word-final -17.722 11.14150 -1.59065 0.1118 Interaction: Preceding Sound and Liquid Type stop -22.501 25.61998 -0.87825 0.3799 pause 54.760 64.59932 0.84769 0.3967 approximant -40.805 30.34334 -1.34479 0.1788 fricative -37.311 27.45675 -1.35889 0.1743 lateral 104.230 98.41665 1.05907 0.2897 nasal 4.344 29.12829 0.14913 0.8815 rhotic -62.717 41.62585 -1.50668 0.1320 Interaction: Age and Liquid Type younger -50.445 37.57909 -1.34237 0.1938 older 50.445 37.57909 1.34237 0.1938 Interaction: Vowel and Liquid Type /a/ -22.501 25.61998 -0.87825 0.3799 glide+/a/ 54.760 64.59932 0.84769 0.3967 glide+/e/ -40.805 30.34334 -1.34479 0.1788 /e/ -37.311 27.45675 -1.35889 0.1743 /i/ 104.230 98.41665 1.05907 0.2897 /o/ 15.451 42.60828 0.36263 0.7169 /u/ -13.153 44.16085 -0.29784 0.7659
F4.2 Estimate Standard Error t-Value p-value (Intercept) 3487.073 40.48094 86.14112 0.0000 Stress stressed 28.925 11.18591 2.58586 0.0098 unstressed -28.925 11.18591 -2.58586 0.0098 Word Position word-medial -35.867 12.88263 -2.78414 0.0054 word-final 35.867 12.88263 2.78414 0.0054 Vowel /a/ 0.999 19.44841 0.05135 0.9590 glide+/a/ -25.399 34.70040 -0.73195 0.4643 glide +/e/ -8.571 43.48382 -0.19711 0.8438 /e/ 28.204 25.26886 1.11615 0.2645 /i/ 195.887 36.93449 5.30365 0.0000 /o/ -118.238 25.76142 -4.58973 0.0000 /u/ -72.882 43.80989 -1.66360 0.0963 274 Preceding Sound stop -10.495 26.73398 -0.39258 0.6947 pause 6.100 67.08544 0.09094 0.9276 approximant 53.483 31.38295 1.70420 0.0885 fricative 82.265 28.61665 2.87472 0.0041 lateral -26.677 101.80279 -0.26204 0.7933 nasal -43.098 30.36871 -1.41917 0.1560 rhotic -61.578 43.16754 -1.42648 0.1539 Following Sound stop 6.787 15.19434 0.44666 0.6552 pause 0.460 22.68912 0.02027 0.9838 approximant 23.478 21.26888 1.10386 0.2698 fricative -65.345 25.55092 -2.55744 0.0106 nasal -3.364 21.14926 -0.15908 0.8736 vowel 37.985 22.83204 1.66367 0.0963 Gender female 105.017 33.15426 3.16754 0.0046 male -105.017 33.15426 -3.16754 0.0046 Age younger -41.899 34.55370 -1.21257 0.2388 older 41.899 34.55370 1.21257 0.2388 Liquid Type /r/ -9.072 24.31055 -0.37317 0.7090 /l/ 9.072 24.31055 0.37317 0.7090 Interaction: Word Position and Liquid Type word-medial 29.917 11.53955 2.59257 0.0096 word-final -29.917 11.53955 -2.59257 0.0096 Interaction: Preceding Sound and Liquid Type stop 5.269 26.54285 0.19852 0.8427 pause 42.036 66.95967 0.62778 0.5302 approximant 23.719 31.45276 0.75411 0.4509 fricative -13.895 28.41732 -0.48897 0.6249 lateral -76.813 102.02038 -0.75292 0.4516 nasal 33.769 30.18889 1.11859 0.2634 rhotic -14.085 42.92501 -0.32813 0.7428 Interaction: Age and Liquid 275 Type younger -41.899 34.55370 -1.21257 0.2388 older 41.899 34.55370 1.21257 0.2388 Interaction: Vowel and Liquid Type /a/ -21.160 19.33246 -1.09456 0.2738 glide+/a/ -0.914 34.70752 -0.02633 0.9790 glide+/e/ -17.218 42.18863 -0.40812 0.6832 /e/ 4.330 24.15265 0.17928 0.8577 /i/ 62.878 36.95872 1.70132 0.0890 /o/ -14.764 26.14224 -0.56474 0.5723 /u/ -13.153 44.16085 -0.29784 0.7659
F4.3 Estimate Standard Error t-Value p-value (Intercept) 3462.914 40.77456 84.92828 0.0000 Stress stressed 26.430 11.74339 2.25062 0.0245 unstressed -26.430 11.74339 -2.25062 0.0245 Word Position word-medial -21.104 13.54120 -1.55849 0.1192 word-final 21.104 13.54120 1.55849 0.1192 Vowel /a/ 16.065 20.41947 0.78675 0.4315 glide+/a/ 6.053 36.21924 0.16713 0.8673 glide +/e/ -27.836 45.62827 -0.61006 0.5419 /e/ 41.201 26.50553 1.55444 0.1202 /i/ 127.355 38.78318 3.28378 0.0010 /o/ -113.674 26.95888 -4.21658 0.0000 /u/ -49.164 46.00215 -1.06874 0.2853 Preceding Sound stop -5.775 28.07873 -0.20568 0.8371 pause -86.066 70.37695 -1.22292 0.2215 approximant 27.422 32.91319 0.83316 0.4048 fricative 100.284 30.05652 3.33652 0.0009 lateral 29.111 106.79869 0.27257 0.7852 nasal -20.431 31.86814 -0.64110 0.5215 rhotic -44.545 45.65625 -0.97566 0.3293 Following Sound stop 21.305 15.93577 1.33693 0.1814 pause 80.883 23.75982 3.40417 0.0007 approximant -21.277 22.31339 -0.95357 0.3404 276 fricative -78.878 26.80361 -2.94281 0.0033 nasal -36.598 22.14074 -1.65297 0.0985 vowel 34.565 23.91403 1.44540 0.1485 Gender female 102.523 32.68546 3.13667 0.0050 male -102.523 32.68546 -3.13667 0.0050 Age younger -30.847 34.10987 -0.90436 0.3761 older 30.847 34.10987 0.90436 0.3761 Liquid Type /r/ -23.915 25.52070 -0.93708 0.3488 /l/ 23.915 25.52070 0.93708 0.3488 Interaction: Word Position and Liquid Type word-medial 23.416 12.12628 1.93099 0.0536 word-final -23.416 12.12628 -1.93099 0.0536 Interaction: Preceding Sound and Liquid Type stop 23.655 27.88200 0.84841 0.3963 pause -66.747 70.25200 -0.95011 0.3421 approximant 50.007 32.99595 1.51556 0.1298 fricative 31.071 29.85732 1.04066 0.2981 lateral -26.723 107.02042 -0.24970 0.8028 nasal 34.627 31.68164 1.09298 0.2745 rhotic -45.891 45.42689 -1.01022 0.3125 Interaction: Age and Liquid Type younger -16.301 9.67888 -1.68417 0.0923 older 16.301 9.67888 1.68417 0.0923 Interaction: Vowel and Liquid Type /a/ -57.132 20.30680 -2.81346 0.0049 glide+/a/ 26.978 36.21498 0.74494 0.4564 glide+/e/ 25.786 44.25663 0.58266 0.5602 /e/ 20.768 25.33298 0.81981 0.4124 /i/ 54.911 38.84476 1.41359 0.1576 /o/ -34.420 27.36570 -1.25777 0.2086 /u/ -36.891 46.36846 -0.79561 0.4263
27 7 F4.4 Estimate Standard Error t-Value p-value (Intercept) 3490.760 39.42024 88.55247 0.0000 Stress stressed 22.905 12.09602 1.89359 0.0584 unstressed -22.905 12.09602 -1.89359 0.0584 Word Position word-medial -26.410 -13.90221 -1.89968 0.0576 word-final 26.410 13.90221 1.89968 0.0576 Vowel /a/ 3.341 21.01453 0.15900 0.8737 glide+/a/ 45.895 37.25354 1.23196 0.2181 glide +/e/ -57.137 46.93184 -1.21746 0.2235 /e/ 44.117 27.43191 1.60823 0.1079 /i/ 115.544 40.01740 2.88735 0.0039 /o/ -137.970 27.76710 -4.96882 0.0000 /u/ -13.790 47.38769 -0.29101 0.7711 Preceding Sound stop 29.602 28.92446 1.02341 0.3062 pause -92.607 72.40444 -1.27902 0.2010 approximant 50.478 33.85719 1.49090 0.1361 fricative 93.308 30.88098 3.02155 0.0025 lateral -9.186 109.85924 -0.08361 0.9334 nasal -39.789 32.84837 -1.21128 0.2259 rhotic -31.806 46.58820 -0.68272 0.4948 Following Sound stop 30.129 16.40459 1.83665 0.0664 pause 82.150 24.50562 3.35229 0.0008 approximant -9.733 22.96523 -0.42380 0.6717 fricative -95.538 27.57676 -3.46445 0.0005 nasal -1.324 22.74399 -0.05822 0.9536 vowel -5.684 24.62421 -0.23084 0.8175 Gender female 108.558 30.42007 3.56863 0.0018 male -108.558 30.42007 -3.56863 0.0018 Age younger -22.653 31.81865 -0.71193 0.4843 older 22.653 31.81865 0.71193 0.4843 Liquid Type /r/ -19.302 26.23583 -0.73571 0.4620 /l/ 19.302 26.23583 0.73571 0.4620 Interaction: Word Position 278 and Liquid Type word-medial 16.715 12.46338 1.34114 0.1800 word-final -16.715 12.46338 -1.34114 0.1800 Interaction: Preceding Sound and Liquid Type stop 33.625 28.70877 1.17124 0.2416 pause -45.969 72.27653 -0.63602 0.5248 approximant 36.796 33.94477 1.08399 0.2785 fricative 18.685 30.66564 0.60932 0.5424 lateral 66.365 110.08736 0.60284 0.5467 nasal -13.445 32.64673 -0.41184 0.6805 rhotic -96.056 46.32843 -2.07337 0.0382 Interaction: Age and Liquid Type younger -18.601 9.96427 -1.86672 0.0621 older 18.601 9.96427 1.86672 0.0621 Interaction: Vowel and Liquid Type /a/ -23.351 20.88525 -1.11805 0.2636 glide+/a/ 45.477 37.26110 1.22049 0.2224 glide+/e/ 32.046 45.53110 0.70383 0.4816 /e/ 9.860 26.21772 0.37608 0.7069 /i/ 47.177 40.05193 1.17789 0.2390 /o/ -56.201 28.16677 -1.99531 0.0461 /u/ -55.008 47.75907 -1.15177 0.2495
F4.5 Estimate Standard Error t-Value p-value (Intercept) 3517.992 36.56302 96.21724 0.0000 Stress stressed 29.300 12.22179 2.39732 0.0166 unstressed -29.300 12.22179 -2.39732 0.0166 Word Position word-medial 6.324 14.05565 0.44993 0.6528 word-final -6.324 14.05565 -0.44993 0.6528 Vowel /a/ -21.741 21.19218 -1.02591 0.3050 glide+/a/ 27.814 37.70750 0.73762 0.4608 glide +/e/ -80.935 47.39385 -1.70772 0.0878 /e/ 78.497 27.59956 2.84415 0.0045 279 /i/ 106.270 40.20136 2.64343 0.0083 /o/ -117.725 28.10722 -4.18843 0.0000 /u/ 7.821 47.77152 0.16372 0.8700 Preceding Sound stop 9.751 29.15012 0.33452 0.7380 pause 41.034 73.15873 0.56089 0.5749 approximant 38.079 34.20908 1.11313 0.2658 fricative 97.753 31.18571 3.13454 0.0017 lateral -168.959 110.93851 -1.52300 0.1279 nasal 26.386 33.14816 0.79599 0.4261 rhotic -44.044 47.04677 -0.93617 0.3493 Following Sound stop 63.371 16.59001 3.81980 0.0001 pause 97.798 24.68810 3.96133 0.0001 approximant -23.514 23.15248 -1.01563 0.3099 fricative -86.442 27.85628 -3.10313 0.0019 nasal -31.779 23.03203 -1.37976 0.1678 vowel -19.434 24.87496 -0.78125 0.4347 Gender female 93.780 26.39308 3.55320 0.0019 male -93.780 26.39308 -3.55320 0.0019 Age younger -39.412 27.72849 -1.42137 0.1699 older 39.412 27.72849 1.42137 0.1699 Liquid Type /r/ 15.234 26.49442 0.57500 0.5653 /l/ -15.234 26.49442 -0.57500 0.5653 Interaction: Word Position and Liquid Type word-medial 8.927 12.60282 0.70833 0.4788 word-final -8.927 12.60282 -0.70833 0.4788 Interaction: Preceding Sound and Liquid Type stop 6.758 28.94665 0.23347 0.8154 pause 5.466 73.00417 0.07487 0.9403 approximant -5.670 34.28463 -0.16537 0.8687 fricative -53.389 30.97413 -1.72365 0.0849 lateral 220.808 111.17014 1.98622 0.0471 nasal -59.387 32.94869 -1.80240 0.0716 280 rhotic -114.588 46.78197 -2.44941 0.0144 Interaction: Age and Liquid Type younger -21.447 10.07140 -2.12954 0.0333 older 21.447 10.07140 2.12954 0.0333 Interaction: Vowel and Liquid Type /a/ 7.321 21.07164 0.34746 0.7283 glide+/a/ -9.773 37.70349 -0.25921 0.7955 glide+/e/ 71.354 45.97892 1.55189 0.1208 /e/ -16.676 26.39691 -0.63173 0.5276 /i/ 74.121 40.22709 1.84257 0.0655 /o/ -51.454 28.50039 -1.80538 0.0711 /u/ -74.894 48.15884 -1.55515 0.1200
F4.6 Estimate Standard Error t-Value p-value (Intercept) 3540.022 37.04619 95.55699 0.0000 Stress stressed 28.853 12.14601 2.37554 0.0176 unstressed -28.853 12.14601 -2.37554 0.0176 Word Position word-medial -1.183 13.95511 -0.08476 0.9325 word-final 1.183 13.95511 0.08476 0.9325 Vowel /a/ -16.284 21.07824 -0.77256 0.4399 glide+/a/ 55.846 37.32957 1.49603 0.1348 glide +/e/ 2.262 47.90335 0.04722 0.9623 /e/ 53.204 27.37961 1.94320 0.0521 /i/ 33.950 39.90077 0.85086 0.3949 /o/ -88.952 27.94846 -3.18270 0.0015 /u/ -40.027 47.51823 -0.84235 0.3997 Preceding Sound stop 23.276 28.96160 0.80368 0.4217 pause 15.025 72.48715 0.20727 0.8358 approximant 34.805 34.08885 1.02099 0.3074 fricative 114.444 30.99631 3.69217 0.0002 lateral -194.328 110.01654 -1.76636 0.0775 nasal 27.515 32.87232 0.83701 0.4027 rhotic -20.735 46.64094 -0.44457 0.6567 Following Sound 281 stop 84.673 16.43685 5.15144 0.0000 pause 21.248 24.51285 0.86680 0.3861 approximant -18.098 22.98087 -0.78753 0.4310 fricative 17.362 27.61614 0.62869 0.5296 nasal -62.021 22.83200 -2.71643 0.0066 vowel -43.164 24.62759 -1.75265 0.0798 Gender female 91.029 27.22626 3.34342 0.0031 male -91.029 27.22626 -3.34342 0.0031 Age younger -42.556 28.57678 -1.48918 0.1513 older 42.556 28.57678 1.48918 0.1513 Liquid Type /r/ 23.506 26.30602 0.89357 0.3716 /l/ -23.506 26.30602 -0.89357 0.3716 Interaction: Word Position and Liquid Type word-medial 9.701 12.50292 0.77589 0.4379 word-final -9.701 12.50292 -0.77589 0.4379 Interaction: Preceding Sound and Liquid Type stop -32.343 28.74870 -1.12502 0.2607 pause -40.138 72.35554 -0.55473 0.5791 approximant 19.562 34.16114 0.57264 0.5669 fricative -40.676 30.78626 -1.32124 0.1865 lateral 253.203 110.24738 2.29668 0.0217 nasal -70.104 32.67117 -2.14574 0.0320 rhotic -89.505 46.37809 -1.92989 0.0537 Interaction: Age and Liquid Type younger -8.330 10.00519 -0.83258 0.4052 older 8.330 10.00519 0.83258 0.4052 Interaction: Vowel and Liquid Type /a/ 3.653 20.95352 0.17434 0.8616 glide+/a/ 2.737 37.32915 0.07332 0.9416 glide+/e/ 28.200 46.58132 0.60539 0.5450 /e/ 0.411 26.16168 0.01570 0.9875 /i/ 69.153 39.93488 1.73164 0.0835 282 /o/ -30.556 28.38747 -1.07640 0.2818 /u/ -73.597 47.89344 -1.53668 0.1245
F4.7 Estimate Standard Error t-Value p-value (Intercept) 3591.888 34.02414 105.56881 0.0000 Stress stressed 14.068 11.44561 1.22913 0.2191 unstressed -14.068 11.44561 -1.22913 0.2191 Word Position word-medial -11.239 13.17338 -0.85320 0.3936 word-final 11.239 13.17338 0.85320 0.3936 Vowel /a/ 5.711 19.90940 0.28682 0.7743 glide+/a/ 8.743 35.50283 0.24628 0.8055 glide +/e/ -2.680 45.20087 -0.05929 0.9527 /e/ 9.501 25.86304 0.36736 0.7134 /i/ 13.589 37.70958 0.36037 0.7186 /o/ -25.169 26.29486 -0.95718 0.3386 /u/ -9.696 44.79901 -0.21642 0.8287 Preceding Sound stop 4.110 27.33437 0.15036 0.8805 pause -14.771 68.50077 -0.21564 0.8293 approximant 28.074 32.05716 0.87575 0.3812 fricative 27.716 29.24050 0.94788 0.3433 lateral -22.942 103.94001 -0.22073 0.8253 nasal 8.986 31.07658 0.28916 0.7725 rhotic -31.173 44.07960 -0.70720 0.4795 Following Sound stop 48.706 15.52123 3.13802 0.0017 pause 45.274 23.16609 1.95430 0.0508 approximant 27.172 21.72395 1.25079 0.2111 fricative 83.177 26.09053 3.18800 0.0014 nasal -121.460 21.59188 -5.62528 0.0000 vowel -82.868 23.29739 -3.55696 0.0004 Gender female 79.063 24.37669 3.24337 0.0039 male -79.063 24.37669 -3.24337 0.0039 Age younger -46.497 25.62106 -1.81480 0.0839 older 46.497 25.62106 1.81480 0.0839 Liquid Type /r/ -9.644 24.85580 -0.38800 0.6980 283 /l/ 9.644 24.85580 0.38800 0.6980 Interaction: Word Position and Liquid Type word-medial 2.214 11.79746 0.18769 0.8511 word-final -2.214 11.79746 -0.18769 0.8511 Interaction: Preceding Sound and Liquid Type stop 5.668 27.13887 0.20886 0.8346 pause -8.946 68.36687 -0.13086 0.8959 approximant 17.783 32.13200 0.55343 0.5800 fricative 2.967 29.03868 0.10219 0.9186 lateral 40.354 104.15626 0.38744 0.6985 nasal -5.949 30.88290 -0.19264 0.8473 rhotic -51.877 43.83584 -1.18345 0.2367 Interaction: Age and Liquid Type younger 11.642 9.41561 1.23646 0.2164 older -11.642 9.41561 -1.23646 0.2164 Interaction: Vowel and Liquid Type /a/ 28.429 19.79734 1.43600 0.1511 glide+/a/ -81.031 35.51519 -2.28159 0.0226 glide+/e/ 51.174 43.96851 1.16388 0.2446 /e/ 25.577 24.73136 1.03420 0.3011 /i/ 57.156 37.73463 1.51469 0.1300 /o/ -51.426 26.68343 -1.92726 0.0541 /u/ -29.879 45.14943 -0.66179 0.5082
Duration Estimate Standard Error t-Value p-value (Intercept) 4.896790 0.02782279 175.99926 0.0000 Stress stressed 0.069326 0.01000189 6.93133 0.0000 unstressed -0.069326 0.01000189 -6.93133 0.0000 Word Position word-medial -0.011407 0.00953179 -1.19674 0.2315 word-final 0.011407 0.00953179 1.19674 0.2315 Vowel /a/ 0.066150 0.01686841 3.92154 0.0001 284 glide+/a/ 0.135222 0.02975752 4.54412 0.0000 glide +/e/ 0.020716 0.03672668 0.56407 0.5728 /e/ 0.003876 0.02097733 0.18478 0.8534 /i/ -0.137243 0.03116855 -4.40326 0.0000 /o/ 0.003508 0.02246951 0.15610 0.8760 /u/ -0.092229 0.03760731 -2.45242 0.0143 Preceding Sound stop -0.015022 0.02279837 -0.65890 0.5100 pause -0.035249 0.05701558 -0.61824 0.5365 approximant 0.012139 0.02704879 0.44879 0.6536 fricative -0.029376 0.02425347 -1.21121 0.2259 lateral -0.038565 0.08658880 -0.44538 0.6561 nasal 0.005236 0.02578119 0.20311 0.8391 rhotic 0.100836 0.03661441 2.75401 0.0059 Following Sound stop -0.047434 0.01531756 -3.09668 0.0020 pause 0.447019 0.02006957 22.27346 0.0000 approximant -0.002463 0.01944022 -0.12668 0.8992 fricative -0.255887 0.03076295 -8.31803 0.0000 nasal -0.034354 0.02253299 -1.52462 0.1275 vowel -0.106881 0.02088934 -5.11653 0.0000 Gender female 0.026965 0.01868694 1.44298 0.1638 male -0.026965 0.01868694 -1.44298 0.1638 Age younger -0.015284 0.01930930 -0.79151 0.4375 older 0.015284 0.01930930 0.79151 0.4375 Liquid Type /r/ -0.012884 0.02160596 -0.59633 0.5510 /l/ 0.012884 0.02160596 0.59633 0.5510 Interaction: Preceding Sound and Liquid Type stop -0.042273 0.02280261 -1.85388 0.0639 pause -0.066664 0.05702281 -1.16908 0.2425 approximant 0.034648 0.02705567 1.28063 0.2004 fricative -0.029453 0.02426688 -1.21373 0.2250 lateral 0.093157 0.08675099 1.07384 0.2830 nasal 0.056043 0.02573626 2.17761 0.0295 rhotic -0.045457 0.03655794 -1.24343 0.2138 Interaction: Following 285 Sound and Liquid Type stop -0.014992 0.01512795 -0.99098 0.3218 pause -0.034022 0.01934023 -1.75913 0.0787 approximant 0.052255 0.01918992 2.72303 0.0065 fricative 0.051896 0.03089133 1.67996 0.0931 nasal 0.002195 0.02226484 0.09858 0.9215 vowel -0.057332 0.02030524 -2.82352 0.0048 Interaction: Vowel and Liquid Type /a/ -0.000517 0.01682170 -0.03071 0.9755 glide+/a/ 0.096055 0.02965286 3.23930 0.0012 glide+/e/ -0.129857 0.03572783 -3.63461 0.0003 /e/ -0.051105 0.02069460 -2.46951 0.0136 /i/ 0.124847 0.03085501 4.04624 0.0001 /o/ 0.004430 0.02244356 0.19739 0.8435 /u/ -0.043853 0.03710072 -1.18199 0.2373
286