University of Iowa
Iowa Research Online
Theses and Dissertations
Spring 2014
An optimality theoretic typology of three fricativevowel assimilations in Latin American Spanish
Jeffrey Bernard Renaud
University of Iowa
Copyright 2014 Jeffrey Bernard Renaud
This dissertation is available at Iowa Research Online: http://ir.uiowa.edu/etd/4733
Recommended Citation
Renaud, Jeffrey Bernard. "An optimality theoretic typology of three fricative-vowel assimilations in Latin American Spanish." PhD
(Doctor of Philosophy) thesis, University of Iowa, 2014.
http://ir.uiowa.edu/etd/4733.
Follow this and additional works at: http://ir.uiowa.edu/etd
Part of the Spanish and Portuguese Language and Literature Commons
AN OPTIMALITY THEORETIC TYPOLOGY
OF THREE FRICATIVE-VOWEL ASSIMILATIONS
IN LATIN AMERICAN SPANISH
by
Jeffrey Bernard Renaud
A thesis submitted in partial fulfillment of the requirements
for the Doctor of Philosophy degree in Spanish
in the Graduate College
of The University of Iowa
May 2014
Thesis Supervisors: Assistant Professor Christine Shea
Professor Jerzy Rubach
Graduate College
The University of Iowa
Iowa City, Iowa
CERTIFICATE OF APPROVAL
_______________________
PH.D. THESIS
_______________
This is to certify that the Ph.D. thesis of
Jeffrey Bernard Renaud
has been approved by the Examining Committee
for the thesis requirement for the Doctor of Philosophy degree in Spanish
at the May 2014 graduation.
Thesis Committee:
___________________________________
Christine Shea, Thesis Co-supervisor
___________________________________
Jerzy Rubach, Thesis Co-supervisor
___________________________________
Paula Kempchinsky
___________________________________
Mercedes Niño-Murcia
___________________________________
Judith Liskin-Gasparro
___________________________________
Jill Beckman
To my family: Doyle, Kathy and Jenny; my son, Parker; and my wife, Jennifer
ii
It is quite impossible to stop the progress of language. It is like the course of the
Mississippi, the motion of which, at times, is scarcely perceptible. Yet even then, it
possesses a momentum quite irresistible. Words and expressions will be forced into use
in spite of all the exertions of all the writers of the world.
Noah Webster
iii
ACKNOWLEDGEMENTS
This dissertation is, not unlike all those that have preceded it, the result of
invaluable intellectual, emotional and financial support from countless sources, many of
whom I will unintentionally neglect to mention in what follows. First and foremost, this
dissertation–and my having survived college in general–would not have been possible
without my advisor and mentor Paula Kempchinsky. My research benefited greatly from
her comments, suggestions and probing questions, and when I struggled to find answers
(or justify to myself why I chose to stay in graduate school), her constant reassurance
sustained me throughout this difficult process. (I would also be remiss if I did not, at this
point, thank her for her keen editorial eye.)
When the University of Iowa was without a Spanish phonologist, Paula assumed
the unenviable role of temporary thesis co-director until I had the privilege of meeting
Christine Shea, my eventual permanent co-director, whose numerous consultations,
brainstorming sessions and e-mails on everything from phonetics to alternative
theoretical approaches turned a limited OT analysis into (what I hope is) the multifaceted dissertation before you. I thank her for adopting this phonological orphan.
I met Jerzy Rubach, also my co-director and mentor, in [hɛɫ] (that is, H.E.L., or
History of the English Language). I once told him during a subsequent semester of
independent study in which we met for one hour once per week that I learned more from
him in those sixteen total hours than I had in a few previous (general education
undergraduate) classes that met three times per week. I feel privileged just to have
retained probably less than one-quarter of all the incredible phonological knowledge he
has imparted to me over the years.
iv
Much debt is owed to Jill Beckman for being an inspirational instructor, for the
countless hours of individual meetings and for all her grammatical corrections, as well. I
would like to thank Mercedes Niño-Murcia for being a constant source of positive energy
and kindness, and for reminding me not to neglect the importance of my sociolinguistic
background, especially when it came to preparing the experimental methodology for this
research. I thank Judy Liskin-Gasparro for her insightful methodological comments, as
well. From a non-academic perspective, the Spanish and Portuguese picnics that she
hosted are fond memories of graduate school that I will always cherish.
I cannot envision a graduate school experience that is not shared with fellow
graduate students. I thus thank my friends and colleagues who equally suffered with me:
Mike, Tiff, Jen, Tania, Libby, Gonzalo, Cortney, Tim, Cindee, Travis, Anne, Elena,
Erica, Andrea, Matt, Andrew, Jordan, Sarah, Mark, Jialing, Emily, Fernando, and all my
literature and linguistics colleagues, as well. I send special thanks to Mike for his
assistance with statistics, and for his advice, emotional support and comic relief; to Tiff
and Gonzalo not only for their assistance with participant recruitment but, beyond this,
for their encouragement and kindness; to Tania for the innumerable rides home, being, in
general, a wonderful neighbor, recording preliminary stimuli and, most of all, welcoming
me into her parents’ home while in Mexico; and to Erica for being a wonderful
sympathizer and purveyor of humor in times of need. For all your kindness, generosity,
humor and guidance, you all are true friends.
My appreciation must also be sent to Scott Sadowsky, my Chilean contact at La
Universidad de La Frontera in Temuco, Chile, for his comments, methodological
assistance and stimuli recordings. I doubt that he knew what he was getting himself into
v
when he answered an e-mail from a complete stranger; I am truly grateful for his selfless
generosity and phonetic expertise. I must also thank Vladimir Kulikov for assistance with
Praat and Liz Goodin-Mayeda for additional participant recruitment. André Zampaulo
has been a tremendous source of positive energy and intellectual discussion.
Much of my linguistic foundation is due to the incomparable influence of two
former professors: Carlos-Eduardo Piñeros, whose 2009 book contained the data that
gave me the idea for this dissertation and to whose passion for Spanish phonology I can
only aspire, and Jason Rothman, who first introduced me to second language acquisition
theory. I would also like to thank all the professors of the Spanish and Portuguese and
Linguistics departments at the University of Iowa for their support. I additionally thank
the always helpful, kind and patient Phillips Hall office staff: Merry Powell, Jeanne
Mullen, Jessie Trepanier, Beth Mellinger and Matt Lively.
Although I specifically discovered linguistics in college, my respect for
knowledge, inquiry, scholarship, academia and teaching was fostered by a wonderful
series of middle school and high school teachers, among them, Lee Raines, Colleen
Walter, Gloria Leventhal, Thelma Wemer, Sherry Dogruyusever, Dawn Kilstrom, Deena
Stanley-Dostart, Tom Wilkie and Bass Cameron.
This project would not exist but for the generous people of Temuco, Chile, and
Puebla and Cholula, Mexico, who welcomed me into their countries and their lives; I will
always carry with me the memories of those I met and the stories they shared while
conducting fieldwork in July and October, 2012. My research in Chile and Mexico was
funded in part by a 2012 Stanley Award for International Research and a 2013 Executive
Council of Graduate and Professional Students Research Grant provided by the
vi
Partnership of Student Governments at Iowa. Completion of this thesis would not have
been possible without a 2013-2014 Ballard and Seashore Fellowship from the Graduate
College at the University of Iowa, for which I must specifically thank Kathy Klein
Gerling for all her assistance.
While all the above have contributed to my life and this project in some way, I
claim full responsibility for the opinions, conclusions, arguments and errors herein. The
only question that remains now is how to put into words my gratitude to the following
irreplaceable people?
To my lifelong friends and fellow Iowa alumni, Tim Ohrt and Kirsten Kloock, for
all the laughs and fond memories. To my extended family–every aunt, uncle and cousin–
for their support, for feigning interest in my work at family functions when I tried to
explain what a fricative is and for asking me only occasionally when I was going to
finish.
To my grandmothers, Viola Renaud (requiescat in pace) and Nona Kain, for your
unconditional love; if two kinder, sincerer, more generous people exist, I have never met
them. To my sister, Jenny, for your words of encouragement. To my son, Parker, for
bringing joy into my life that I could not have imagined (and for sleeping in your rocker
by my computer while I worked through the night). To my parents, Doyle and Kathy,
words could never express my gratitude for your love, support and understanding; an
exhaustive list of your contributions would be endless. To my best friend and wife,
Jennifer, for your unconditional love and support, for forgiving the long nights that I had
to sit at the desk to write and for making this journey bearable. The completion of this
thesis is as much a reflection of your time, energy and devotion as it is of mine.
vii
ABSTRACT
The roles of phonetics (e.g., Jun 1995, Holt 1997, Steriade 2001) and Articulatory
Phonology (AP, Browman and Goldstein 1986, et seq.) in both the diachronic evolution
of and synchronic analyses for phonological processes are relatively recent incorporations
into Optimality Theory (OT) (Prince and Smolensky 1993/2004, McCarthy and Prince
1993/2001). I continue this line of inquiry by offering an AP-based OT proposal of three
fricative-vowel assimilations in Latin American Spanish: /f/→[x] velarization (fui [xu̯ i] ‘I
went’), /f/→[ɸ] bilabialization (fumo [ɸumo] ‘I smoke’) and /x/→[ç] palatalization (gente
[çen̪ t̪ e] ‘people’).
In this dissertation, I pursue three main objectives: to update and clarify via
empirical study and spectral analysis the available data; to account for the
crosslinguistically recurrent phonological patterns that affect fricative-vowel sequences;
and to explain the above processes’ genesis and diffusion in Latin American Spanish by
integrating the first two goals into an Optimality Theoretic framework.
Concerning the first task, data for the three processes are culled primarily from
sociolinguistic corpora (Perissinotto 1975, Resnick 1975, Sanicky 1988, inter alia).
Lacking from these accounts are detailed phonetic analyses. To fill this gap, I report on a
four-part perception and production study designed to update the descriptive facts and
provide spectral analyses for the allophonic variants.
Regarding the second goal, I show that fricatives are susceptible to regressive
consonant-vowel assimilation given the recurrence of assimilatory patterns nearly
identical to the Spanish processes under investigation in disparate languages throughout
the world. I argue that articulatory and acoustic facts conspire to render place features in
viii
(non-sibilant) fricatives difficult to recover given the vast interspeaker, intraspeaker and
crosslinguistic variability in production (e.g., Ladefoged and Maddieson 1996) and the
greater reliance on fricative-vowel transitional cues as opposed to cues internal to the
frication on the part of the hearer (e.g., Manrique and Massone 1981, Feijóo and
Fernández 2003). To that end, I argue that the sound changes originate(d) with the
hearer’s misperception of a speaker’s extremely coarticulated target (Baker, Archangeli
and Mielke 2011, inter alia).
The dissertation concludes with a proposal adapting Jun (1995) that encodes the
above articulatory and acoustic facts into an AP-based, typologically-minded OT
approach that accounts both diachronically and synchronically for /f/ velarization, /f/
bilabialization and /x/ palatalization in Spanish (updating previous analyses by Lipski
1995 and Mazzaro 2005, 2011).
ix
TABLE OF CONTENTS
LIST OF TABLES .......................................................................................................... xiv
LIST OF FIGURES ..........................................................................................................xx
LIST OF ABBREVIATIONS ....................................................................................... xxiv
LIST OF SYMBOLS ......................................................................................................xxv
CHAPTER 1
CONSONANT-VOWEL AFFINITIES
AND THE CASE OF SPANISH ....................................................1
1.1. Spanish data and generalizations .....................................................................3
1.2. Crosslinguistic consonant-vowel interactions ................................................11
1.3. Research questions .........................................................................................12
1.3.1. Phonetic/descriptive ........................................................................13
1.3.2. Phonological/theoretical .................................................................13
1.3.3. Sociolinguistic and diachronic ........................................................13
1.4. Motivation, theoretical relevance and contributions of the thesis .................14
1.5. Organization of the thesis ..............................................................................22
CHAPTER 2
SPANISH FRICATIVE ALLOPHONY:
PREVIOUS ACCOUNTS ........................................................... 23
2.1. Mazzaro (2005) ..............................................................................................24
2.1.1. Analysis ...........................................................................................25
2.1.2. Criticisms ........................................................................................26
2.2. Lipski (1995) ..................................................................................................30
2.2.1. Analysis ...........................................................................................31
2.2.2. Criticisms ........................................................................................37
2.3. External explanations .....................................................................................41
2.3.1. The African substratum ...................................................................42
2.3.2. The American indigenous substratum ............................................46
2.4. Conclusion .....................................................................................................50
CHAPTER 3
EXPERIMENTAL METHODOLOGY ....................................... 53
3.1. Participants .....................................................................................................55
3.1.1. Chilean production participants ......................................................57
3.1.2. Chilean perception participants .......................................................58
3.1.3. Mexican production participants .....................................................61
3.1.4. Mexican perception participants .....................................................62
3.1.5. Control groups ................................................................................63
3.1.5.1. Nativeness rating participants ..........................................64
3.1.5.2. Task 2: AX discrimination control participants ...............66
x
3.2. Task 1: sociolinguistic interview ...................................................................70
3.3. Task 2: perception ..........................................................................................72
3.3.1. Justification for tokens and conditions ...........................................75
3.3.2. Token analysis ................................................................................77
3.3.2.1. Native rating .....................................................................78
3.3.2.2. Fricative spectral analyses ...............................................81
3.3.2.3. Duration ...........................................................................85
3.3.2.4. Vowel quality ...................................................................88
3.3.3. Interim summary .............................................................................91
3.4. Task 3: sentence reading ................................................................................91
3.4.1. Justification for tokens and conditions ...........................................98
3.4.2. /f/ conditions ...................................................................................98
3.4.2.1. Assimilating .....................................................................98
3.4.2.2. Non-assimilating ............................................................100
3.4.2.3. Progressive assimilation .................................................101
3.4.2.4. Lexicalized velarization with ir and ser .........................102
3.4.3. /x/ conditions .................................................................................103
3.4.3.1. Assimilating ...................................................................103
3.4.3.2. Non-assimilating ............................................................104
3.4.3.3. Progressive assimilation .................................................105
3.5. Task 4: picture description ...........................................................................106
3.5.1. Justification for tokens and conditions .........................................108
3.5.2. /f/ and /x/ conditions .....................................................................108
3.6. Conclusion ...................................................................................................111
CHAPTER 4
STATISTICAL ANALYSIS: RESULTS,
DISCUSSION AND IMPLICATIONS ..................................... 112
4.1. Analytic tools ...............................................................................................114
4.1.1. Fast Fourier transform and Linear predictive coding ...................114
4.1.2. Center of gravity ...........................................................................115
4.1.3. Locus equations and second formants ..........................................117
4.2. Allophone identity .......................................................................................120
4.3. Productivity ..................................................................................................134
4.2.1. Frequency effects in the reading task ............................................135
4.2.2. Assimilation rates across all production tasks ..............................137
4.4. Ir and ser lexicalization ...............................................................................140
4.5. Regressive versus progressive assimilation .................................................142
4.6. Prosodic effect .............................................................................................151
4.7. Position within the word ..............................................................................154
4.8. Toward the grammatical analysis ................................................................158
4.8.1. Coarticulatory effects ....................................................................159
4.8.2. Variance in center of gravity values .............................................167
4.8.3. Perception .....................................................................................171
4.9. Conclusion ...................................................................................................180
xi
CHAPTER 5
OPTIMALITY THEORETIC ANALYSIS ............................... 183
5.1. Arguments for a synchronic analysis ...........................................................184
5.1.1. Perception .....................................................................................185
5.1.2. Coarticulation ................................................................................189
5.1.3. Systematic crosslinguistic variation ..............................................193
5.2. Theoretical components ...............................................................................194
5.2.1. A general overview of Optimality Theory ....................................194
5.2.2. Alternative analyses and featural assumptions .............................200
5.2.2.1. Positional Faithfulness ...................................................201
5.2.2.2. Preservation constraints .................................................202
5.2.2.3. Perceptual Faithfulness ..................................................204
5.2.2.4. The Sound Pattern of English ........................................208
5.2.2.5. Unified Feature Theory ..................................................210
5.2.2.6. Element Theory ..............................................................213
5.2.2.7. Parallel Structures Model ...............................................214
5.2.3. A general overview of Articulatory Phonology ............................217
5.2.4. Articulatory Phonology-based constraints
in Optimality Theory .....................................................................223
5.3. Optimality Theoretic analysis of Spanish fricative-vowel assimilations .....227
5.3.1. /f/ bilabialization ...........................................................................227
5.3.2. /f/ velarization ...............................................................................231
5.3.3. /x/ palatalization ............................................................................232
5.3.4. Restraining the grammar ...............................................................240
5.3.4.1. Manner-specific place faithfulness constraints ..............240
5.3.4.2. Constraints on coarticulatory distance ...........................245
5.3.5. An assimilatory typology ..............................................................253
5.4. Initiation versus diffusion in sound patterns ................................................258
5.4.1. Correction, hypocorrection and the role of the listener ................259
5.4.2. Literacy and the role of social factors ...........................................263
5.5. Conclusion ...................................................................................................269
CHAPTER 6
CONCLUDING DIACHRONIC
AND SYNCHRONIC ISSUES .................................................. 270
6.1. Variability/optionality ..................................................................................273
6.2. Context-free assimilation and phonological analogy ...................................278
6.3. The fate of Latin F .......................................................................................282
6.4. Theoretical and empirical contributions and future research .......................288
6.5. Conclusions ..................................................................................................290
APPENDIX A
TASK 1: SOCIOLINGUISTIC INTERVIEW ...........................291
A.1. Task 1 instructions and written questionnaire .............................................291
A.1.1. Spanish version ............................................................................291
A.1.2. English translation ........................................................................292
xii
A.2. Task 1 oral interview questions ..................................................................293
A.2.1. Spanish version ............................................................................293
A.2.2. English translation ........................................................................294
APPENDIX B
TASK 2: PERCEPTION .............................................................296
B.1. Nativeness rating task instructions ..............................................................296
B.1.1. Spanish version ............................................................................296
B.1.2. English translation ........................................................................296
B.2. Task 2 AX discrimination instructions ........................................................297
B.2.1. Spanish version ............................................................................297
B.2.2. English translation ........................................................................297
APPENDIX C
TASK 3: SENTENCE READING ..............................................299
C.1. Task 3 instructions ......................................................................................299
C.1.1. Spanish version ............................................................................299
C.1.2. English translation ........................................................................299
C.2. Frequency analyses .....................................................................................299
C.3. Task 3 Spanish sentences (non-randomized)
with corresponding translation ....................................................................303
APPENDIX D
TASK 4: PICTURE DESCRIPTION .........................................312
D.1. Task 4 instructions ......................................................................................312
D.1.1. Spanish version ............................................................................312
D.1.2. English translation ........................................................................312
D.2. Task 4 images (non-randomized) with target word
and corresponding translation .....................................................................313
REFERENCES ...............................................................................................................321
xiii
LIST OF TABLES
Table 1.1.
Grammatical /f/ velarization data ................................................................4
Table 1.2.
Ungrammatical /f/ velarization data ............................................................5
Table 1.3.
Grammatical /x/ palatalization data ............................................................8
Table 1.4.
Crosslinguistic CV assimilations ..............................................................12
Table 1.5.
Transcription conventions in /f/ allophony literature ................................15
Table 1.6.
Commonalities among the processes ........................................................20
Table 2.1.
Distribution of /f/ allophones in Corrientes, Argentina ............................24
Table 2.2.
Percentage of occurrence of /f/ allophones in Corrientes, Argentina .......27
Table 2.3.
Labial interaction vis-à-vis [round] OCP in the Spanish
of Iquitos, Peru ..........................................................................................33
Table 2.4.
Origins of colonial Latin American slaves by region ...............................42
Table 2.5.
African languages and linguistic groups by region ...................................43
Table 2.6.
Fricatives of African languages ................................................................44
Table 2.7.
Phonological processes by country ...........................................................45
Table 2.8.
Indigenous languages by population .........................................................46
Table 2.9.
Fricatives of American indigenous languages ..........................................47
Table 2.10.
Five stages of Latin F ................................................................................49
Table 2.11.
Distribution of /f/ allophones in Misiones, Argentina ..............................51
Table 3.1.
Individual biographical information for Chilean production group .........58
Table 3.2.
Group to task correspondences, Chilean and Mexican participants .........59
Table 3.3.
Individual biographical information for Chilean perception group ...........60
Table 3.4.
Chilean listeners’ exposure to English (in years) .....................................60
Table 3.5.
Individual biographical information for the Mexican production group ..61
xiv
Table 3.6.
Mexican participants’ exposure to Nahuatl (in years) ..............................62
Table 3.7.
Individual biographical information for the Mexican perception group ..63
Table 3.8.
Individual biographical information for the nativeness rating group .......64
Table 3.9.
Nativeness controls’ exposure to English (in years) .................................65
Table 3.10.
Nativeness controls’ exposure to French (in years) ..................................65
Table 3.11.
Individual biographical information for the AX
discrimination controls ..............................................................................66
Table 3.12.
Discrimination controls’ exposure to French (in years) ...........................67
Table 3.13.
Discrimination controls’ exposure to English (in years) ..........................68
Table 3.14.
Discrimination controls’ exposure to German (in years) ..........................68
Table 3.15.
Discrimination controls’ exposure to Italian (in years) ............................68
Table 3.16.
Discrimination controls’ exposure to Portuguese (in years) .....................69
Table 3.17.
Group to task correspondences, all participants .......................................69
Table 3.18.
AX discrimination task consonant pairs ...................................................75
Table 3.19.
Individual syllable mean nativeness ratings greater than 3 (‘unsure’) .....80
Table 3.20.
Average spectral mean (Hz) of token fricatives (n=5 per fricative)
in nativeness rating/AX discrimination tasks ...........................................84
Table 3.21.
Fricative duration of target CV tokens (in seconds) .................................86
Table 3.22.
Vocalic duration (in seconds) of fricative CV tokens ...............................87
Table 3.23.
Syllable duration (in seconds) of target CV tokens ..................................87
Table 3.24.
F1 and F2 values (Hz) of vowels in target CV tokens ..............................88
Table 3.25.
Ranges (two SDs) of F1 and F2 values (Hz) of vowels
in target CV tokens ...................................................................................89
Table 3.26.
Native ranges of F1 and F2 values (Hz) of Spanish vowels .....................90
xv
Table 3.27.
Length of sentences (n=110) by character (including spaces)
in task 3 .....................................................................................................93
Table 3.28.
Sample token list for /f/ assimilating condition, task 3 /fu/ tokens ...........96
Table 3.29.
Task 3 /f/ assimilating tokens (n=20) .......................................................99
Table 3.30.
Task 3 /f/ non-assimilating tokens (n=24) ..............................................100
Table 3.31.
Task 3 /f/ progressive assimilation tokens (n=8) ....................................101
Table 3.32.
Task 3 verbal paradigms for ir ‘to go’ and ser ‘to be’ (n=8) ..................102
Table 3.33.
Task 3 /x/ assimilating tokens (n=18) .....................................................104
Table 3.34.
Task 3 /x/ non-assimilating tokens (n=24) .............................................105
Table 3.35.
Task 3 /x/ progressive assimilation tokens (n=8) ...................................106
Table 3.36.
Task 4 real word /f/ conditions (n=26) ...................................................109
Table 3.37.
Task 4 real word /x/ conditions (n=25) ...................................................110
Table 4.1.
Number of fricatives analyzed (n=1667) ................................................113
Table 4.2.
Number of tokens by condition ...............................................................122
Table 4.3.
Group COG averages (Hz) by allophonic conditions .............................123
Table 4.4.
Crosslinguistic COG (Hz) averages ........................................................123
Table 4.5.
Average [x] (/x/) LPC spectral peaks by group ......................................126
Table 4.6.
Ratio of /f/ allophones by experimental participants ..............................126
Table 4.7.
COG (Hz) of /f/ allophones by experimental participants ......................127
Table 4.8.
Individual [x] (/f/) LPC spectral peaks by token
of experimental participants ....................................................................128
Table 4.9.
Average [ɸ] (/f/) LPC spectral peaks by experimental speaker ..............131
Table 4.10.
Comparison of /f/ allophone ratios by frequency:
descriptive statistics ................................................................................136
Table 4.11.
Comparison of /f/ allophone ratios by task: descriptive statistics ...........139
xvi
Table 4.12.
Verbal paradigms for ser ‘to be’ and ir ‘to go’ .......................................140
Table 4.13.
Comparison of /f/ allophone ratios ir/ser by task:
descriptive statistics ................................................................................141
Table 4.14.
Comparison of /f/ allophone ratios by following vowel:
descriptive statistics ................................................................................144
Table 4.15.
Comparison of /f/ allophone ratios by preceding vowel:
descriptive statistics ................................................................................146
Table 4.16.
Comparison of /x/ allophone ratios by following vowel:
descriptive statistics ................................................................................147
Table 4.17.
Comparison of /x/ allophone ratios by preceding vowel:
descriptive statistics ................................................................................148
Table 4.18.
Comparison of /f/ allophone ratios by stress: descriptive statistics ........153
Table 4.19.
Comparison of /x/ allophone ratios by stress: descriptive statistics .......154
Table 4.20.
Comparison of /f/ allophone ratios by position: descriptive statistics ....155
Table 4.21.
Comparison of /x/ allophone ratios by position: descriptive statistics ...156
Table 4.22.
/f/ locus equations for all participants .....................................................162
Table 4.23.
/f/ locus equations by geolect ..................................................................163
Table 4.24.
/f/ locus equations by group ....................................................................163
Table 4.25.
/x/ locus equations for all participants ....................................................165
Table 4.26.
/x/ locus equations by geolect .................................................................165
Table 4.27.
/x/ locus equations by group ...................................................................166
Table 4.28.
Maximum, minimum and average COG values
for /f/ allophones by group ......................................................................170
Table 4.29.
Maximum, minimum and average COG values
for /x/ allophones by group .....................................................................171
Table 4.30.
Fricative misperception in Galician ........................................................175
xvii
Table 5.1.
Average /f/ locus equations and
center of gravity (Hz) comparisons: data ................................................188
Table 5.2.
Palatalized labials in Moldavian .............................................................189
Table 5.3.
Differences in place/manner concerning
likelihood of target/trigger status ............................................................202
Table 5.4.
Variable cues to place depending on condition by manner ....................205
Table 5.5.
Place features in SPE /f/ bilabialization ..................................................208
Table 5.6.
Place features in UFT /f/ bilabialization .................................................211
Table 5.7.
Place features in ET /f/ bilabialization ....................................................214
Table 5.8.
Place features in PSM /f/ bilabialization .................................................216
Table 5.9.
Articulatory Phonology tract variables ...................................................218
Table 5.10.
Intrusive vowels within onset and coda clusters .....................................222
Table 5.11.
AP features for /f/ bilabialization ............................................................227
Table 5.12.
AP features for /f/ velarization ................................................................231
Table 5.13.
AP features for /x/ palatalization ............................................................233
Table 5.14.
Latin American Spanish phonemic inventory
with relevant manner features .................................................................241
Table 5.15.
Latin American Spanish dialects of /f/ and /x/ assimilation ...................245
Table 5.16.
Implicational hierarchy of articulatory efficiency ..................................248
Table 5.17.
Summary of constraint rankings and their effects ..................................252
Table 5.18.
Constraint rankings for /f, x/ assimilations
in Latin American Spanish dialects ........................................................254
Table 5.19.
Crosslinguistic CV assimilations redux ..................................................255
Table 5.20.
Rate of velar allophony in /xi, xe/ contexts by region ............................261
Table 5.21.
/xi, xe/ locus equations by geolect redux ................................................262
xviii
Table 5.22.
Rate of allophony in /fu, fo/ contexts by region (assimilators only) ......263
Table 5.23.
/fu, fo/ locus equations by geolect (assimilators only) ...........................264
Table 5.24.
Rate of allophony in /fu, fo/ contexts by group ......................................265
Table 5.25.
/fu, fo/ locus equations by group .............................................................265
Table 6.1.
Development of Latin initial /Cl/ clusters ................................................280
Table 6.2.
Upper Aragonese /Cl/ clusters .................................................................281
Table 6.3.
Latin F in Romance ..................................................................................282
Table 6.4.
Five stages of Latin F redux.....................................................................283
Table 6.5.
Orthographic evidence for F erosion .......................................................284
Table 6.6.
F erosion, document 257, 1216, Sigüenza ...............................................286
Table 6.7.
F erosion, early northern origin ...............................................................287
Table C.1.
Task 3 /fu, fuV, fo/ assimilating frequency analysis ..............................300
Table C.2.
Task 3 /fi, fe, fa/ non-assimilating frequency analysis ...........................301
Table C.3.
Task 3 /uf, of/ progressive assimilation frequency analysis ...................301
Table C.4.
Task 3 /xi, xiV, xe/ assimilating frequency analysis ..............................302
Table C.5.
Task 3 /xu, xo, xa/ non-assimilating frequency analysis ........................302
Table C.6.
Task 3 /ix, ex/ progressive assimilation frequency analysis ...................303
xix
LIST OF FIGURES
Figure 1.1.
Three possibilities for the speaker-listener exchange ...............................18
Figure 2.1.
Direction of free [round] association ........................................................36
Figure 3.1.
Mean nativeness rating by condition ........................................................79
Figure 3.2.
CV syllable [fi] with four fricative spectral divisions
(1 initial, 2 medial, 3 final, 4 transitional) and fricative
and vocalic onsets and offsets ...................................................................83
Figure 3.3.
Vowel distribution by F1 and F2 values (Hz) of target CV tokens ..........90
Figure 3.4.
Task 4 sample images (initial, stressed) /f/ condition,
(a) assimilating fútbol ‘football’ and (b) non-assimilating fila ‘line’ .....109
Figure 3.5.
Task 4 sample images (initial, unstressed) /x/ condition,
(a) assimilating jirafa ‘giraffe’ and
(b) non-assimilating juguete ‘toy’ ...........................................................110
Figure 4.1.
Fast Fourier transform, [x] trabajo ‘work,’ 02838 (Chile, control) ........115
Figure 4.2.
Linear predictive coding, [x] trabajo ‘work,’ 02838 (Chile, control) .....116
Figure 4.3.
VCV [ofi] (ofidio ‘ophidia’) with three points of F2 measurement
(1 offset glottal pulse of VC transition, 2 V resting state,
3 onset glottal pulse of CV transition) ....................................................118
Figure 4.4.
F2 difference, [fa] familia ‘family,’ 17137 (Chile, experimental) ..........118
Figure 4.5.
F2 stable, [fo] telefónica ‘telephone,’ 17137 (Chile, experimental) .......119
Figure 4.6.
Average COG (kHz) by Chilean groups in /f/ and /x/ conditions............121
Figure 4.7.
Average COG (kHz) by Mexican groups in /f/ and /x/ conditions ..........121
Figure 4.8.
Average [x] (/x/) LPC by group ...............................................................125
Figure 4.9.
Individual [x] (/f/) LPC by token of experimental participants ...............127
Figure 4.10.
[ɸ] LPC spectra .......................................................................................130
Figure 4.11.
[f] LPC spectra ........................................................................................130
Figure 4.12.
Average [ɸ] (/f/) LPC by experimental speaker ......................................131
xx
Figure 4.13.
Chilean experimental participant 19727 (male):
bilabial [ɸ], labiodental [f] ......................................................................132
Figure 4.14.
Chilean experimental participant 17137 (male):
bilabial [ɸ], labiodental [f] ......................................................................132
Figure 4.15.
Comparison of /f/ allophone ratios by frequency ....................................135
Figure 4.16.
Formality and allophonic rates across production tasks ..........................138
Figure 4.17.
Comparison of /f/ allophone ratios by task ..............................................138
Figure 4.18.
Comparison of /f/ allophone ratios ir/ser by task .....................................141
Figure 4.19.
Comparison of /f/ allophone ratios by following vowel .........................144
Figure 4.20.
Comparison of /f/ allophone ratios by preceding vowel .........................145
Figure 4.21.
Comparison of /x/ allophone ratios by following vowel ........................147
Figure 4.22.
Comparison of /x/ allophone ratios by preceding vowel ........................148
Figure 4.23.
Comparison of /f/ allophone ratios by stress ...........................................152
Figure 4.24.
Comparison of /x/ allophone ratios by stress ..........................................153
Figure 4.25.
Comparison of /f/ allophone ratios by position .......................................155
Figure 4.26.
Comparison of /x/ allophone ratios by position ......................................156
Figure 4.27.
Locus equations for Chilean assimilator 19727: /fu/ versus /fa/ .............160
Figure 4.28.
Locus equations for Chilean assimilator 19727: /xi/ versus /xa/ ............160
Figure 4.29.
Maximum, minimum and average /f/ COG values by participant ..........168
Figure 4.30.
Maximum, minimum and average /x/ COG values by participant .........169
Figure 4.31.
Acoustic similarity among various obstruents ........................................172
Figure 4.32.
Average perceptual rating by allophonic comparison ............................173
Figure 4.33.
Vowel-specific /f/ discrimination within [f]~[x] tokens .........................176
Figure 4.34.
Vowel-specific /x/ discrimination within [x]~[ç] tokens ........................177
xxi
Figure 4.35.
Average perceptual rating by manner .....................................................179
Figure 5.1.
Average /f/ locus equations (slope)
and center of gravity (kHz) comparisons ................................................188
Figure 5.2.
UFT geometric representations for place in consonants and vowels ......210
Figure 5.3.
Spectral shape of place element singletons
|U|, |A| and complex |U A| .......................................................................213
Figure 5.4.
PSM geometric representations ..............................................................215
Figure 5.5.
Lateral view of vocal tract with AP tract variables .................................219
Figure 5.6.
Gestural score for spam [spæ̃ m] .............................................................220
Figure 5.7.
Consonants superimposed over nucleic vowel within the syllable .........221
Figure 5.8.
Intrusive vowels in Spanish and Scots Gaelic ........................................222
Figure 5.9.
Articulatory Phonology landmarks in Optimality Theory ......................223
Figure 5.10.
Three distinct CV alignments .................................................................225
Figure 5.11.
Interaction between maximize contrast and minimize effort ..................247
Figure 5.12.
Articulatory distance: [f], [i], [e], [o], [u], [x] .........................................248
Figure 5.13.
Ranking summary for Latin American Spanish fricative allophony ......252
Figure 5.14.
Constraint reranking and factorial typology of CV/VC constraints .......258
Figure 5.15.
Three possibilities for the speaker-listener exchange redux ...................260
Figure 5.16.
Average /xi, xe/ locus equations (slope) and rate of [ç] allophone (%) ...262
Figure 5.17.
Average /fu, fo/ locus equations
and rate of [ɸ]/[x] allophones (assimilators) ...........................................264
Figure 5.18.
Average /fu, fo/ LEs, rate of [ɸ]/[x] allophones (%)
(experimental/control) .............................................................................266
Figure 5.19.
Years of formal education and rate of [ɸ]/[x] allophones
by participant ..........................................................................................267
Figure 6.1.
Classic OT fixed constraint rankings, C1 » C2 » C3 ..............................273
xxii
Figure 6.2.
Stochastic OT variable constraint rankings, C2 » C3 .............................274
Figure 6.3.
Stochastic OT variable constraint rankings, C3 » C2 .............................275
Figure 6.4.
Chile experimental /f/ bilabialization:
~55% [f], ~45% [ɸ] (see Table 4.6) ........................................................275
Figure 6.5.
Mexico experimental /f/ bilabialization:
~85% [ɸ], ~15% [f] (see Table 4.6) ........................................................275
Figure 6.6.
Experimental /f/ velarization: ~95% [f], ~5% [x] (see Table 4.6) ..........276
Figure 6.7.
Chilean /x/ palatalization: ~85% [ç], ~15% [x] (see Table 5.20) ...........276
Figure 6.8.
Mexican /x/ palatalization: ~75% [x], ~25% [ç] (see Table 5.20) ..........276
xxiii
LIST OF ABBREVIATIONS
ALIGN
ALIGNMENT constraint
AP
Articulatory Phonology
C
consonant
Con
Constraint inventory
CV
consonant-vowel
Eval
Evaluator
Gen
Generator
ID
IDENT constraint
OCP
Obligatory Contour Principle
OT
Optimality Theory
SPE
The Sound Pattern of English
UFT
Unified Feature Theory
UG
Universal Grammar
V
vowel
xxiv
LIST OF SYMBOLS
CAPS
constraint title or Latin orthography
attested and theoretically optimal candidate in a grammatical hierarchy
attested but theoretically non-optimal candidate in an ungrammatical
hierarchy
>
etymological source
»
OT hierarchical dominance
<…>
orthography, grapheme or digraph
/…/
phonemic representation
[…]
phonetic representation or phonetic feature
σ
syllable
unattested but theoretically optimal candidate in an ungrammatical
hierarchy
xxv
1
CHAPTER 1
CONSONANT-VOWEL AFFINITIES
AND THE CASE OF SPANISH
Generative phonology seeks to explain the mental representation (i.e., grammar)
of a speaker’s linguistic sound system, the way in which this representation accounts for
a speaker’s productions, how such a representation came to be acquired in a particular
speaker and how phonological differences between speakers (of the same and different
dialects and languages) can be accounted for with a universal set of theoretical
components. This dissertation contributes in a novel way to the above goals by
empirically and theoretically analyzing three phonological processes of Latin American
Spanish: labiodental velarization, labiodental bilabialization and velar palatalization,
exemplified in basic terms in (1) through (3).
(1)
(2)
(3)
fumo
foto
gente
‘I smoke’
‘photo’
‘people’
[fumo]
[fot̪ o]
[xen̪ t̪ e]
~
~
~
[xumo]
[ɸot̪ o]
[çen̪ t̪ e]
In (1), the voiceless labiodental fricative [f] alternates with the voiceless velar fricative
[x]. (2) demonstrates a similar alternation between [f] and the voiceless bilabial fricative
[ɸ]. The final process under consideration in this thesis appears in (3) in which the
voiceless velar fricative [x] alternates with the voiceless palatal fricative [ç]. While the
process exhibited in (3) is well known among Hispanic linguists (Canfield 1981, Lipski
1994), few studies have documented (1) and (2) (Perissinotto 1975, Resnick 1975,
Sanicky 1988, Lipski 1995, Calvo Shadid 1996, Mazzaro 2005, 2011); those studies that
do exist suffer from a lack of phonetic analysis and imprecision in reporting the results. I
2
endeavor to supplement the above sources by providing detailed phonetic analyses in
order to more accurately describe the processes.
Once the generalizations are better understood, I present a formal analysis framed
within Optimality Theory (Prince and Smolensky 1993/2004, McCarthy and Prince
1993/2001) that builds on a growing body of literature incorporating articulatory
(Browman and Goldstein 1986, 1989, 1990, 1991, 1992, 1995; Byrd 1996; Gafos 2002;
Kochetov 2002; Davidson 2003; Bateman 2007; inter alia) and acoustic (Kohler 1990;
Steriade 1993; Byrd 1994; Jun 1994, 1995; Herrick 1999; inter alia) considerations.
Beyond the grammatical representation proposed, I diachronically trace the
properties from their origins to present day dialectal differentiation and synchronic
allophony. Recent work by Baker, Archangeli and Mielke (2011) and Garrett and
Johnson (2013) has explored how phonological processes originate, why they originate
where they do (as opposed to elsewhere) and how they expand from their point(s) of
origin throughout a speech community (and perhaps farther). With these questions, I
combine articulatory and acoustic phonetics and sociolinguistic factors in order to shed
light on their interdependency in comprehensively accounting for and understanding
phonological phenomena (cf. Mielke 2008).
I present in the next section the data as described in the currently available
literature. Section 1.2 is a crosslinguistic typological look at similar processes. Based on
the Spanish and crosslinguistic data, section 1.3 problematizes the issues that they raise
and presents the nine research questions that guide the dissertation. Section 1.4 motivates
other issues addressed in and lays out the linguistic contributions of the dissertation. The
final section of this introductory chapter outlines the organization of the thesis.
3
1.1. Spanish data and generalizations
In various Latin American dialects of Spanish–namely, Argentina, Bolivia, Chile,
Colombia, Costa Rica, the Dominican Republic, Ecuador, El Salvador, Guatemala,
Mexico, Panama, Peru and Puerto Rico–alternations affecting the two voiceless, nonstrident fricatives of Spanish–labiodental /f/ and velar /x/ (Resnick 1975, Canfield 1981,
Lipski 1994, inter alia)–have been reported, introduced briefly in (1) through (3). In the
cases of (1) and (2), velarization and labialization, there exists a small set of descriptive
accounts (Perissinotto 1975, Resnick 1975, Sanicky 1988, Lipski 1995, Calvo Shadid
1996, Mazzaro 2005, 2011). Two possible reasons for the dearth of information might be
due to the fact that they are non-standard and involve two of the less frequently occurring
phonemes in the Spanish phonemic inventory. Of the eighteen phonemes in Latin
American Spanish, /f/ (13) and /x/ (15) are exceeded in infrequency of occurrence only
by /ʎ/1 (14), /j/2 (16), /tʃ/ (17) and /ɲ/ (18) (Piñeros 2009). For this reason, one goal of this
dissertation is to improve and update the understanding of these underreported processes.
A second goal is to unify the /f/ alternations with an analysis of /x/ palatalization
in Spanish (3). While there is no shortage of literature on palatalization in general (see
Bateman 2007 and the sources cited therein for an extensive survey), Spanish is
frequently excluded from this discussion. Furthermore, to my knowledge, the obvious
similarity in all three processes has not been pointed out. The fact that the targets /f/ and
/x/ form the exclusive natural class of voiceless non-strident fricatives in Latin American
1
The palatal lateral /ʎ/ is limited in its distribution due to its merger with the palatal glide /j/ in
the vast majority of dialects. Notable exceptions include Bolivia, Paraguay, northwest Argentina,
Peru, southern Ecuador and rural Spain (Canfield 1981: 12, Piñeros 2009: 236).
2
See, however, Hualde and colleagues (1997, 2004, 2008) on the controversial (phonemic) status
of the vowel [i]~semivowel [i̯ ]~glide [j] alternation.
4
Spanish and that the output allophones seem to be determined by the following vowel
(see Chapter 4 for further analysis of this claim) suggests that a uniform analysis is not
only possible but preferable (i.e., economical) from a theoretical perspective.
Table 1.1. Grammatical /f/ velarization data
Stressed
Initial
Unstressed
Stressed
Non-initial
Unstressed
Orthography
Output
Category
Gloss
Source
fumo
[fu] [xu]
verb
‘I smoke’
fuerte
[fu̯ e] [xu̯ e]
adjective
‘strong’
forma
[fo] [xo]
noun
‘form’
futuro
[fu] [xu]
noun
‘future’
fumando
[fu] [xu]
gerund
‘smoking
fotógrafo
[fo] [xo]
noun
photographer
difunto
[fu] [xu]
adjective
‘deceased’
afuera
[fu̯ e] [xu̯ e]
adverb
‘outside’
sofoco
[fo] [xo]
noun
‘suffocation’
perfumería
[fu] [xu]
noun
‘perfumery’
M (188)
afuerino
[fu̯ e] [xu̯ e]
noun
‘outsider’
novel
teléfono
[fo] [xo]
noun
‘telephone’
M (188)
P (283)
R (24)
M (188)
P (283)
Sources: Sources are listed by first initial of last name and are adapted from Resnick
(1975: 24), Piñeros (2009: 283), Mazzaro (2011: 188) and one novel form verified by a
native Spanish-speaking Chilean informant to complete the paradigm.
Table 1.1 shows that labiodental [f] may alternate with velar [x] when parsed as the onset
of a syllable whose nucleus is a back vowel /u/ (diphthong included) or /o/. While the
data, when compiled from multiple sources and viewed globally, seem to suggest that
position in the word and stress are irrelevant, permitting the process in any permutation
of those variables, this has been tested empirically by only one source (Mazzaro 2011).
To that end, I aim to examine these factors in the experimental study that I conducted,
detailed in Chapters 3 and 4. Also absent from the data are forms such as (4).
5
(4)
bofetada
/bofet̪ ad̪ a/→[f], *[x]
noun ‘slap’
It is not sufficient that /f/ simply be adjacent to a back vowel. The suggested
ungrammaticality of the velar allophone in (4) shows that the triggering vowel is
tautosyllabic with the target fricative; directionality seems to be uniquely regressive. This
is another unexplored issue that my experiment seeks to address. With respect to the
quality of the following vowel, consider Table 1.2.
Table 1.2. Ungrammatical /f/ velarization data
Stressed
Initial
Unstressed
Stressed
Non-initial
Unstressed
Orthography
Output
Category
Gloss
Source
fíjate
[fi] *[xi]
verb
‘Notice!’
P (55)
feria
[fe] *[xe]
noun
‘festival’
M2 (188)
falla
[fa] *[xa]
noun
‘flaw’
P (55)
fideo
[fi] *[xi]
noun
noodle
M2 (188)
festejar
[fe] *[xe]
verb
‘to party’
familia
[fa] *[xa]
noun
‘family’
desfile
[fi] *[xi]
noun
‘parade’
M2 (188)
café
[fe] *[xe]
noun
‘coffee’
R (24)
estofado
[fa] *[xa]
noun
‘stew’
M2 (188)
sacrificar
[fi] *[xi]
verb
‘to sacrifice’
M1 (60)
cafetera
[fe] *[xe]
noun
‘coffee pot’
alfajor
[fa] *[xa]
noun
pastry
M1 (60)
M2 (188)
Sources: Sources are listed by first initial of last name and are adapted from Perissinotto
(1975: 55), Resnick (1975: 24) and Mazzaro (2005: 60, 2011: 188).
Unlike in Table 1.1, velarization is unattested when /f/ precedes a non-back vowel /i/, /e/
or /a/, as is the case above; once again, neither stress nor position within the word seems
to be relevant. Both data sets, when viewed together, suggest that a following [u] or [o]
conditions the change. (5) summarizes the facts thus far in a simplified linear rule.
6
(5)
Descriptive generalization of (restricted) /f/ velarization
/f/→[x] / σ[ __ [u, o]
The rule in (5) states that an underlying voiceless labiodental fricative /f/ is realized as a
voiceless velar fricative [x] when followed by a back vowel. The facts are not, however,
that straightforward, as (5) is only indicative of the simplest case. For reference I will
refer to dialects that attest (5) as restricted (velarizing) dialects; i.e., they limit /f/
velarization to the back vowels [u] and [o]. On the other hand, there are dialects in which
the speakers have extended velarization to all contexts; i.e., the [f]~[x] alternations from
Table 1.2 are instead grammatical. (6) is a simplified version of the process of such a
grammar.
(6)
Descriptive generalization of (free) /f/ velarization
/f/→[x]
In (6), /f/ velarization is not limited to back vowels, but is instead optional in any context.
While this fact seems to contradict the previous suggestion that /f/ velarization is an
assimilatory process, two points might be raised to maintain the position that, even if
freely velarizing dialects exist, /f/ velarization is still a context-dependent assimilation (in
most cases). First, it is interesting to note that, while there are restricting dialects that
limit /f/ velarization to [u, o] and free dialects that permit it before all five Spanish
vowels [i, e, a, o, u], there are no dialects that allow the process only in the context of the
subsets [i, e], [i, e, a] or [a]. If /f/ velarization is not assimilatory and dialects that freely
velarize are taken as the default, the restricted dialects would then need an explanation as
to why [u, o] permit the process while [i, e, a] (in restricting dialects) do not. Based on
this observation, I will argue in Chapter 6 that freely velarizing dialects represent
7
diachronic change from restricted dialects and that /f/ velarization in restricted dialects, in
turn, is best analyzed as a context-dependent assimilation (Chapters 4 and 5). This
argument is corroborated by the fact that the restrictive dialects are in the vast majority
(at least at the time of the last comprehensive dialectal survey). Of the 84 dialects
reported to allow /f/ velarization, 62 of them (approximately 73.8%) limit the process to
back vowels, while 22 (26.2%) make no distinction and permit the process before any
vowel (Resnick 1975).
At this point I have detailed one of the three phonological processes to be
analyzed in this dissertation. The first process, /f/ velarization, was shown to attest two
varieties: those grammars that restrict the assimilation to back vowels (5), and those that
permit the velar realization [x] of /f/ in any context (6). In either grammar, while the
possibility for velarization differs, the faithful output [f] is always an option. That is, [x]
even in the restricted dialects is never obligatory. I next examine the second property
under consideration: /f/ bilabialization.
The same sources cited previously (Perissinotto 1975, Resnick 1975, Sanicky
1988, Lipski 1995) also report on /f/ bilabialization, a process in which /f/ is realized as
the voiceless bilabial fricative [ɸ] in grammatical and ungrammatical contexts identical to
those in Tables 1.1 and 1.2. Like /f/ velarization, word position and stress do not seem to
affect bilabialization, but this has not been verified.
It also seems to be the case that bilabialization shows the same restricted versus
free dichotomy in terms of where the process is permitted: only before back vowels or
possible anywhere. While Perissinotto (1975) and Sanicky (1988) suggest that the
process occurs at relatively higher rates before back vowels but possible before other
8
vowels, Lipski (1994, 1995) and Piñeros (2009) argue that bilabialization is categorically
restricted to back-vowel environments.
Also like /f/ velarization, /f/ bilabialization is regressive and non-obligatory in
both dialects: the output candidates [f] and [ɸ] are equally possible outputs. (7) and (8)
below summarize these generalizations.
(7) Descriptive generalization of (restricted) /f/ bilabialization
/f/→[ɸ] / σ[ __ [u, o]
(8) Descriptive generalization of (free) /f/ bilabialization
/f/→[ɸ]
In (7), an input voiceless labiodental fricative /f/ is realized as a voiceless bilabial
fricative [ɸ] only when followed by the back vowels [u, o]. In dialects that exhibit (8), the
bilabial [ɸ] realization of /f/ is not restricted by the following vowel.
The final process to be considered is /x/ palatalization. Preceding front vowels [i]
and [e], velar /x/ variably fronts to palatal [ç] in Argentina, Chile, Mexico and Uruguay
(Resnick 1975, Lipski 1994).
Table 1.3. Grammatical /x/ palatalization data
Initial
Non-initial
Orthography
Output
Category
Gloss
Stressed
jefe
[xe] [çe]
noun
‘boss’
Unstressed
gitana
[xi] [çi]
noun
‘gypsy’
Stressed
mejilla
[xi] [çi]
noun
‘cheek’
Unstressed
hereje
[xe] [çe]
noun
‘heresy’
Source: Piñeros (2009: 284).
9
As Table 1.3 shows, velar /x/, which corresponds orthographically to <gi>, <ge>, and
<j>, is realized as the voiceless palatal fricative [ç] when followed by [i] or [e]. Unlike /f/
velarization and bilabialization, free /x/ palatalization before any other vowel (/u, o, a/)
besides the front vowels is rare, reported only for San Luis, Argentina, and Yucatán,
Mexico, by Resnick (1975: 214, 215). Consequently, the outputs *[çu] and *[ço] from
underlying /xu/ and /xo/ (e.g., jugar ‘to play’ and joven ‘young’) are predicted to be
ungrammatical. Yet like the two previously discussed processes, /x/ palatalization is
never obligatory: i.e., [xi] and [çi] may freely alternate for a given lexical item within a
single speaker. (9) contains a linear description of the facts discussed above: an input
voiceless velar fricative /x/ is realized as a voiceless palatal fricative [ç] when followed
by the front vowels [i, e].
(9) Descriptive generalization of /x/ palatalization3
/x/→[ç] / σ[ __ [i, e]
Returning to the frequency of the processes, as mentioned previously, /x/ palatalization is
an accepted and well documented property of particular dialects of Latin American
Spanish; /f/ bilabialization and velarization, on the other hand, are less researched and,
consequently, less understood by the Spanish linguistic community. This is not to say,
however, that native speakers are similarly naïve to the processes, particularly
velarization. The public’s consciousness of this phonological property might be due to
the fact that it creates a neutralization with the faithful output of the velar phoneme /x/.
3
Some Latin American dialects have phonemically replaced velar /x/ with glottal /h/ (Canfield
1981, Lipski 1994). This fact, however, does not drastically affect the theoretical analysis. If
palatalization yields output forms [çi] and [çe], Richness of the Base (Prince and Smolensky
1993/2004) would argue that the corresponding input form is irrelevant since palatalization
results from constraint interaction. Since palatalization is ungrammatical before [a, o, u], whether
or not this output context includes [x] or [h] is outside the scope of this thesis.
10
(10) Neutralization of /f/ and /x/ as a consequence of /f/ velarization
/f/
/x/
eiei
[f]
[x]
…
Two consequences of speakers’ awareness are relevant at the moment. The first is that,
because speakers recognize this phenomenon, they frequently utilize non-standard
orthography to represent its application. The following is a headline from an online
newspaper La Extra Grupo Diario de Morelia, ‘The Extra Daily Group of Morelia,’ from
Morelia, Michoacán, Mexico.
(11)
Tiran
cadáver muerto ya
fallecido de un dijunto
throw-3pl.pres cadaver dead already deceased of a deceased
‘Dead body of previously deceased victim thrown’
Disregarding the redundancy and grim nature of the content, the final word dijunto
‘deceased, dead’ is a non-standard spelling of the expected difunto. Here, the grapheme
<j> associated with the velar fricative [x] has replaced <f>, the labiodental [f], preceding
the back vowel /u/ <u>, precisely the environment and expected outcome of /f/
velarization.
Authors are also able to manipulate the orthography to convey that a subject is a
speaker of such a dialect. In Por qué me quité del vicio, ‘Why I gave up my bad habits,’ a
poem by Manuel Bernal, the author creates a narrator who displays many non-standard
and stigmatized phonological processes in Spanish, one of which is velarization.
(12)
30
31
32
33
Dicen que jue el corazón.
¡Yo no sé lo que haiga sido!
Pero sento en la concencia
que jue mi vicio cochino
They say that it was my heart.
I don’t know what it must have been!
But I feel it in my conscience
that it was my bad habits, pitiful one
11
Again <j> has been used to represent [x] instead of the expected <f>, as the third singular
past tense of ser ‘to be’ is fue ‘was.’ These types of spelling deviations document
speakers’ sensitivity to the process.
In the next section, similar data from various languages are compared to the
Spanish properties as the first step toward a theoretical analysis. The typological
implications revealed factor greatly into the proposal put forth in this dissertation.
1.2. Crosslinguistic consonant-vowel interactions
Place assimilations between adjacent segments, frequently consonants, have been
extensively studied in the phonological literature (see, e.g., Jun 1995, Cho 1997).
Arguably less common are place interactions between consonants and vowels (see
Clements 1989, Sagey 1990, Hume 1992, Ní Chiosáin and Padgett 1993, Clements and
Hume 1995, Cho 1997, inter alia), specifically cases in which a consonant assumes the
place of articulation of an adjacent vowel compared to the inverse case in which a vowel
is affected by a neighboring consonant.
Ní Chiosáin and Padgett (1993: 25) (henceforth, NC&P) even claim that cases in
which consonants adopt the place of a following vowel are unattested. I argue in Chapter
4 that this is precisely the explanation necessary for the Spanish data in question.
However, such processes have been shown to be less crosslinguistically rare as
previously thought in light of the following data from various languages and language
families throughout the world.
Table 1.4 presents a limited view of CV assimilations comparable to the Spanish
data presented above (see, for an extensive discussion of additional examples, NC&P
1993, Cho 1997, Lee 1997, Bateman 2007). For our present purposes, this brief overview
12
serves to show that, crosslinguistically, consonant-vowel interactions are well attested
with various manners of articulations triggered by a variety of vowels in both directions
of assimilation.
Table 1.4. Crosslinguistic CV assimilations
Manner
Language
Bemba
Stop
Fricative
Affricate
Nasal
(Niger-Congo)
Bulgarian
(Balto-Salvic)
Ewe
(Niger-Congo)
Japanese
(Altaic)
Lezgian
(NE Caucasian)
Mina
(Chadic)
Kurpian
(Balto-Slavic)
Sentani
(Papuan)
Target
Output
Context
/t, d, k, ɡ/
[f]
__ /u/
/k, ɡ/
[c, ɟ]
__ /i/
/s, z/
[ʃ, ʒ]
__ /i/
/h/
[ç] [ɸ]
__ /i/ __ /u/
/t͡ s/
[t͡ ʃ]
__ /i/
/t͡ s, d͡ z/
[t͡ ʃ, d͡ ʒ]
__ /i/ or /i/ __
/m/
[ɲ]
__ /i, ɛ/
/n/
[ɲ]
__ /i, j/
Source
Guthrie
(1967-1971)
Scatton
(1984)
Duthie
(1996)
Hirayama
(2005)
Haspelmath
(1993)
Frajzyngier
et al. (2005)
Rubach
(2014)
Cowan
(1965)
Given the typological possibilities predicted by combining these three factors (target
manner, vocalic trigger and direction), the theoretical analysis proposed for the properties
in Latin American Spanish must necessarily take into consideration all the above in order
to avoid permitting assimilations that are ungrammatical in Spanish, yet that the theory
must allow to exhaustively account for attested assimilations crosslinguistically. I
formulate my research questions in the next section with this caveat in mind.
1.3. Research questions
In this brief section, I set forth the nine research questions that inform this
dissertation. In light of the multifaceted goals of phonology, a comprehensive analysis is
13
necessary. To that end, I divide (with much obvious overlap) the research questions into
three main categories: phonetic and descriptive, phonological and theoretical and
sociolinguistic and diachronic.
1.3.1. Phonetic/descriptive
1. Are the fricative phonemes /f/ and /x/ synchronically involved in allophonic
alternations in Latin American Spanish? If so, what are the allophones?
2. Are the phonological processes productive (i.e., not merely lexicalized)?
1.3.2. Phonological/theoretical
3. Can the processes be conceived as assimilations conditioned by the
surrounding vowel(s)? If so, are the assimilations regressive or progressive?
4. Are the assimilations conditioned by stress or position in the word?
5. How can a formal model of phonology–Optimality Theory–account for the
phenomena?
6. How can this theoretical analysis limit the assimilations to the subset of
fricatives {/f/, /x/} to the exclusion of all other phonemes in Spanish?
7. What role, if any, do articulatory and acoustic phonetics play in the theoretical
model proposed?
1.3.3. Sociolinguistic and diachronic
8. What crosslinguistic typological implications does 7 entail?; i.e., why do
certain linguistic systems attest certain properties while other systems do not?
9. How can these questions be viewed from a diachronic perspective of Spanish?
I justify each of these questions, motivate the dissertation and introduce the relevant
background literature in the next section.
14
1.4. Motivation, theoretical relevance
and contributions of the thesis
The primary purpose of this dissertation is to fill two gaps in the Spanish
linguistics literature: to provide updated phonetic analyses on the properties in question
and to account for them in a unified theoretical manner, both of which are novel
contributions to the field.
Concerning the first goal, there are three limitations to the data that I present in
section 1.1. The majority of the resources consulted are relatively outdated, and the more
recent sources mostly cite the earlier sources. This thesis updates the literature by posing
and answering research question 1, which addresses the problem suggested by relying on
outdated sources: do the processes remain productive?
The second limitation is a consequence of the first: the processes have not been
thoroughly analyzed phonetically. In Chapter 4, I provide two phonetic measurements
(center of gravity and spectral peaks) to supplement the data and better describe the
processes (in line with research question 2).
The third criticism is a consequence of the second: because spectral analysis had
not been conducted, the data from the original sources mostly reflect the investigators’
impressionistic auditory determinations. Resnick (1975: 7) specifically states that some
reported output forms were “heard” as indicated. As a result of this imprecision, different
transcription conventions are used in referring to these processes, not always conforming
to modern standards of the International Phonetic Alphabet (IPA). Examples appear
below. Note that all transcriptions here are faithful to the source. In Chapter 4, I render
my results in conventional IPA transcription.
15
Table 1.5. Transcription conventions in /f/ allophony literature
Symbol
Example
Gloss
Explanation
Source
[ẍ]
fuerte [ẍwerte]
‘strong’
“rounded bilabiovelar fricative”
Perissinotto
(1975: 56)
‘coffee’
“voiceless bilabial
fricative,
generally weaker
and of lower pitch
than [f]”
Resnick
(1975: 25)
Resnick
(1975: 25)
Sanicky
(1988: 274)
[ҏ]
café [kaҏe]
[fҏ]
café [kafҏe]
‘coffee’
“voiceless mixed
bilabiallabiodental
articulation”
[fɸ]
afuera [afɸwe]
‘outside’
not described
(but assumedly,
same as above)
Given the high degree of variability in similar processes from Table 1.4, I further explore
the extent to which languages vary. Since it has been documented that grammars may
elect to limit consonant-vowel interactions to regressive CV assimilation, progressive VC
assimilation or permit both, question 3 is necessary since the possibility of VC
conditioning (see (4)) has not been systematically explored in Spanish. Furthermore, no
study exists to determine if stress or position of the fricative in the word affects the output
(research question 4). The answers to these first four research questions inform the
theoretical proposal, which is research question 5 and the second of the two main goals of
the thesis. This is itself a noteworthy contribution given that an adequate theoretical
account that unifies the three processes has not been proposed.
Yet beyond the Spanish data and OT analysis are more global phonological
questions that the dissertation addresses. Approaching Table 1.4 from a different
16
perspective, instead of asking how to account for the differences formally, one might
wonder what determines which type of assimilation will be exhibited in which language
if so many possibilities exist (cf. Weinreich, Labov and Herzog’s (1968) actuation
problem). Baker, Archangeli and Mielke (2011: 347), on whose work this thesis depends
greatly, frame the question in the following manner: “if sound change is conceived as the
accumulation of coarticulation, and coarticulation is widespread, how can some speech
communities resist phonetic pressure to change?” This issue deals directly with dialectal
differentiation (see also Garrett and Johnson 2013) and is the motivation behind research
questions 8 and 9.
In response to this issue, Baker and colleagues propose three criteria.
(13)
a.
b.
c.
phonetic motivation (coarticulation)
inter- and intra-speaker variability
association with sufficient sociolinguistic factors
(13) is, in essence, a formula for phonologization–i.e., the grammaticalization of a
phonetic process and the answer to the question: why do we have phonological
processes? To account for the origin of /f/ and /x/ allophony in Spanish, I apply (13) to
the data under consideration in Chapters 3, 4 and 5. For the moment, I briefly explain
each prerequisite in turn.
(13a) claims that one path to a phonological process is to begin with a
coarticulated sequence. In the speech stream, segments are not articulated individually,
but rather overlap in terms of the gestures necessary to form each sound. Two sounds
whose gestures intersect are prime candidates for asymmetrical influence in that one set
of gestures obscures another. The result of such a situation is an error (if one chooses to
17
call it that), or minimally, output that is target deviant given the failure of one of the
segments to fully surface. This coarticulated output forms the catalyst for a phonological
process.
However, coarticulation in speech is rampant, yet sound change is rare and slow
to occur. These two facts seem to be at odds. That is, if coarticulation is inevitable and
constant, its result–so called output errors–should likewise be inevitable and constant,
predicting that sound change is, additionally, inevitable and constant (i.e., the logic
behind accumulation of error theories; see Baker et al. 2011: 347, 348 for discussion).
However, such is not the case; the vast majority of coarticulated productions fail to
grammaticalize, and even after grammaticalization begins, it is slow to spread. For this
reason, Baker et al. argue that two additional components are necessary.
The first is (13b), variability in terms of output within and between speakers.
Work by Ohala (1981), Blevins (2004) and Beddor (2012) has shown that listeners are
cognizant of the coarticulatory possibilities of speech. Consequently, they are able to
undo the result of coarticulation in order to derive the speaker’s intended target
(corrective (Ohala) or conservative (Beddor) listeners). Thus, while it is necessary to
have coarticulated speech for sound change to occur (i.e., its motivation), it is not
sufficient given the capabilities of listeners to filter out coarticulatory effects. Baker,
Archangeli and Mielke’s variability in output (13b) interferes with this filtering
mechanism, thus presenting, in part, the missing variable for grammaticalization. When
listeners do not perceive the coarticulation that they expect, their ability to apply any rule
or pattern of undoing the coarticulation is disrupted. Ohala (1981) describes three
18
possible situations in the speaker-listener interaction. For clarity, Figure 1.1 concretely
applies these scenarios to possible outcomes for /f/ bilabialization in Spanish.
Figure 1.1. Three possibilities for the speaker-listener exchange
(a) Little to no effect
(b) Correction
(c) Hypocorrection
/fo/
/fo/
/fo/
/fo/
/fo/
/ɸo/
[fo]
[fo]
[fo] [fɸo]
[ɸfo] [ɸo]
[fo]
[fo] [fɸo]
[ɸfo] [ɸo]
[ɸo]
In Figure 1.1 (a), the speaker intends to produce a lexical item with the sequence /fo/. In
realizing this intention, she articulates [fo] with little coarticulation. Consequently, the
listener hears [fo] and interprets the signal as the speaker intended, /fo/. In interaction (b),
on the other hand, the speaker realizes /fo/ with any one of varying degrees of bilabial
coarticulation with the following [o], resulting in a partially bilabialized [fɸ], partially
labiodental [ɸfo] or completely bilabial [ɸ]. Crucially, this speaker may produce any of
these for any given /fo/ utterance. The listener in this scenario, aware of the coarticulatory
possibilities between [f] and [o], corrects for the gestural effect and interprets the (targetdeviant) output as the speaker intended, /fo/. Situation (c) includes the same degree of
output variability on the part of the speaker, yet in this instance, the listener fails to
correct for coarticulation, and the output is (mistakenly) taken at face value and
(re)interpreted by the listener as the speaker’s intended target (hypocorrection (Ohala
1981), innovation (Beddor 2012)). Once this listener becomes a speaker herself, the new
target is instead /ɸ/→[ɸ]. (For a similar approach, see Zampaulo 2013.)
19
In this way, Baker et al. (2011) show why accumulation of error theories fail. Of
the three situations above, 66% of errors ((a) and (b)) result in a failure to grammaticalize
given the claim that speakers are able to reverse engineer the signal (Ohala 1981, Beddor
2012). It is only the rare occasion from scenario (b) that leads to phonologization of a
coarticulatory effect. The conclusion is that, while speaker variability in the output is the
catalyst for phonological processes, the listener actuates the change.
Even with all the above, the landscape is still insufficient for propagation of the
property. As Baker et al. note, one question is the origin of a change; a separate (but not
entirely unrelated) issue is its dissemination beyond one, two or a few speakers.
Compounding the rarity of sound change is the final requirement (13c) that the types of
listeners who will fail to correct the signal (Figure 1.1 (c)) be in a social situation such
that the newly grammaticalized process spreads throughout a speech community, geolect,
sociolect or language variety. Without such disseminating social variables, the property
remains in the domain of idiolectal variation.
Bringing the discussion back to the data at hand, that the sources I have cited
claim /f/ bilabialization, /f/ velarization and /x/ palatalization to be properties of
particular Spanish dialects in Latin America is evidence that they, at some point, went
through the proposed phases in (13). Research questions 8 and 9 apply the criteria from
(13) to these properties in order to account for their origins and expansion.
The remaining research questions to be justified–6 and 7–concern an observation
that, to my knowledge, has not been made previously. It is curious that Latin American
Spanish exhibits three processes with numerous similarities, including the target,
triggering context and output. Compare all three in Table 1.6.
20
Table 1.6. Commonalities among the processes
Property
Target
Trigger
Output
velarization
/f/
__ /u, o/
[x]
bilabialization
/f/
__ /u, o/
[ɸ]
palatalization
/x/
__ /i, e/
[ç]
Commonality
non-strident fricative
following vowel
preserved: frication
altered: place
Focusing on the shared properties of the three processes, all target a non-strident
fricative, are triggered by a following (back or front) vowel and result in an output that
preserves the frication (and voicing) of the fricative yet which adopts the place of the
vocoid. It stands to reason, given these similarities, that a uniform analysis is warranted.
As no such uniform analysis exists, Chapter 5 approaches the theoretical account with
this goal in mind.
The question that arises from Table 1.6, however, is: why should it be the case
that only the non-strident fricatives are susceptible in Spanish? As Table 1.4 showed, all
manners are possible targets by some language in the world. What determines which
manner is susceptible in which language? Research question 6 poses this very issue.
I argue in Chapter 5 that the answer is found in articulatory and acoustic phonetics
and how each is encoded in the theory. However, before proceeding, a brief discussion
concerning the (non-)stridency of [f] is necessary. Under traditional assumptions
(Chomsky and Halle 1968), /f/ is aligned with [+strident] segments (e.g., /s/) as opposed
to [–strident] /ɸ/ and /x/. As an acoustic feature, stridents (commonly used
interchangeably with sibilants when referring to coronal segments) exhibit “greater
noisiness” (ibid.: 329) than their counterparts, the technical correlate of which is higher
21
energy frication. Recent phonetic, phonological and articulatory work (Ladefoged and
Maddieson 1996: 138, Núñez Cedeño and Morales-Front 1999: 72, Ladefoged 2006: 170,
Martínez Celdrán and Fernández Planas 2007: 105, Mielke 2008: 127), however, has
argued that [strident] is applicable only to coronal segments (rendering strident and
sibilant completely synonymous), and that /f/, along with /ɸ/ and /x/, are all [–strident] (if
binary) or unspecified for stridency (if privative). In line with the sources cited above, I
assume that /f/ is non-strident in this dissertation. This assumption is crucial for the
theoretical component presented in Chapter 5. For this reason, I delay detailed
argumentation for my non-strident treatment of /f/ until then.
Returning to the research questions, research question 7 explores the specific
contributions of articulatory and acoustic phonetics. In terms of acoustic phonetics, a
convincingly large body of literature has argued that the acoustic signal of fricatives is
highly variable (McMurray and Jongman 2011) and, consequently, identification of
fricative place is dependent on transitional cues found on the following vowel (Harris
1958; Manrique and Massone 1981; Soli 1981; Stevens 1985; Zeng and Turner 1990;
Fernández, Feijóo, Balsa and Barros 1998; Deng and O’Shaughnessy 2003; Feijóo and
Fernández 2003; Nittrouer and Lowenstien 2009; McMurray and Jongman 2011). This
differs from other manners of articulation in that place is identified by unique cues in the
articulation of the consonant itself (bootstrapped by the transitional vocalic cues)
(Ladefoged 2006, Deng and O’Shaughnessy 2003). Therefore, the argument is as
follows: while fricatives do exhibit spectral cues to place, they are highly variable (and
hence unreliable), which consequently increases the importance of transitional cues
located on the following vowel.
22
I will argue in Chapters 4 and 5 that this line of reasoning accounts for two facts:
why the (non-strident) fricatives are vulnerable–the lack of strength in acoustic cues to
place relative to other manners (and strident segments)–and why the triggering context is
a following vowel–the locus of fricative place identification is the vocalic transition. I
theoretically formalize these arguments following Optimality Theory (OT, Prince and
Smolensky 1993/2004, McCarthy and Prince 1993/2001), Articulatory Phonology (AP,
Browman and Goldstein 1986, et seq.; Byrd 1996; Gafos 2002; Kochetov 2002;
Davidson 2003; Bateman 2007; inter alia), the Production Hypothesis (Steriade 1993,
Byrd 1994, Jun 1994, 1995) and Dispersion Theory (Flemming 1995).
1.5. Organization of the thesis
In this introductory chapter, I have provided the data on Latin American Spanish
/f/ velarization, /f/ bilabialization and /x/ palatalization that serve as the subject of the
phonological analysis to be proposed. I have further motivated the dissertation by
pointing out the limitations in the data and extending the implications of the analysis to
current theoretical issues concerning diachrony, featural assumptions and the role of
articulatory and acoustic phonetics in dialectal differentiation. Additionally, I set forth the
nine research questions that guide the remainder of the thesis, which is structured as
follows. Chapter 2 summarizes and criticizes the two previous analyses of the properties.
Chapter 3 details the methodological protocol employed in a four-part perception and
production empirical study. Chapter 4 presents the experimental results that form the
basis for the Optimality Theoretic analysis in Chapter 5. Chapter 6 concludes the
dissertation with remaining theoretical issues and directions for future research.
23
CHAPTER 2
SPANISH FRICATIVE ALLOPHONY:
PREVIOUS ACCOUNTS
Within the generative paradigm, Lipski (1995) examines /f/ bilabialization
specifically addressing the Spanish of Iquitos, Peru, while Mazzaro (2005) investigates
the variety of /f/ velarization attested in Corrientes, Argentina.1 Both authors adopt an
autosegmental approach (Goldsmith 1976, Clements 1985, McCarthy 1988) in arguing
that the alternations are motivated by the Obligatory Contour Principle (OCP) (Leben
1973, Goldsmith 1976, McCarthy 1986); in doing so, they predict ungrammatical output.
These analyses overgenerate precisely as a consequence of the putative OCP restrictions.
The present chapter summarizes the two aforementioned proposals and argues that both
suffer from theoretical and empirical inadequacies.
I begin with Mazzaro in section 2.1, followed by Lipski in 2.2. I violate
chronological order because Lipski (although predating Mazzaro) actually accounts for
the theoretical criticisms I levy against the latter. Lipski is, however, not without issues;
in 2.2.2 I address the remaining problems that Lipski faces. In light of the theoretical
challenges that both Lipski and Mazzaro encounter, I consider an external (language
contact) explanation for the data in 2.3, considering both African and American
indigenous languages. Ultimately, I reject the external approach, as well, a fact that
leaves /f/ velarization and bilabialization as of yet unaccounted for, thus motivating the
remainder of the dissertation.
1
Lipski (1995) and Mazzaro (2005)–as far as I could tell–exhaust the list of theoretical accounts
concerning /f/ allophony. Although ubiquitous in dialectal literature from a descriptive
perspective (Resnick 1975: 214, Canfield 1981: 31, Lipski 1994: 201, inter alia), I found no
theoretical analysis of /x/ palatalization in Spanish. For this reason, the present chapter focuses
primarily on /f/ alternations.
24
2.1. Mazzaro (2005)
Mazzaro (2005) seeks to address the apparent case of complementary distribution
among the /f/ allophones–labiodental [f], velar [x], glottal [h] and elided [Ø]–attested in
the sociolinguistic interviews she conducted with three Spanish speakers from Corrientes,
Argentina, as seen in Table 2.1 below.
Table 2.1. Distribution of /f/ allophones in Corrientes, Argentina
[f]
[f]amilia
[f]estejábamos
[Ø]
*[Ø]amilia
[Ø]estejábamos
[h]
[x]
*[h]amilia
*[x]amilia
*[h]estejábamos
*[x]estejábamos
sacri[f]icada
*sacri[Ø]icada
sacri[h]icada
*sacri[x]icada
[f]orma
*[Ø]orma
*[h]orma
[x]orma
[f]ui
*[Ø]ui
*[h]ui
[x]ui
Source: Adapted from Mazzaro (2005: 60).
Note: Spanish orthography and glosses are as follows: familia ‘family,’ festejábamos ‘we
celebrated’ (imperfect), sacrificada ‘sacrificed’ (feminine adjective), forma ‘form’
(noun), fui ‘I went’ (preterit).
Unsurprisingly, the (faithful) labiodental allophone [f] is grammatical irrespective of the
vowel that follows it, as indicated by the grammaticality of every form in the leftmost
column. Mazzaro argues, on the other hand, that elision–seen in the second column from
the left–is only possible in the environment of the mid front vowel [e], at least for this
particular Argentinean dialect (the asterisk ‘*’ and shading indicate ungrammatical output
forms). Debuccalization resulting in glottal [h] is grammatical only if /f/ precedes the
high front vowel [i]. Velarization in the rightmost column is solely attested in the
25
environment of back vowels, high [u] and mid [o]. It is this lattermost fact that proves
crucial for Mazzaro’s theoretical analysis.
As mentioned at the beginning of this chapter, Mazzaro motivates /f/ velarization
in the context of back vowels by proposing an Obligatory Contour Principle (OCP)
restriction (Leben 1973, Goldsmith 1976, McCarthy 1986). Such restrictions formalize
the observation that (in some languages) particular adjacent identical elements (features,
segments, etc.) are prohibited. The result of OCP-driven repair strategies is improved
contrast. The following section expands on the theoretical analysis in more detail,
providing the necessary background information for section 2.1.2 in which I argue that
Mazzaro’s proposal is, ultimately, untenable.
2.1.1. Analysis
Assuming feature geometry (Clements and Hume 1995, inter alia) in which
phonological features are arranged hierarchically (see (1) below), Mazzaro claims that the
sequences /fu/ and /fo/ violate a proposed [labial] OCP restriction that prohibits adjacent
[labial] segments; i.e., a sequence of consecutive segments that contain the feature
[labial] within their geometric representations is prohibited (1a). Following the Uniform
Feature Theory (UFT) (Reighard 1972, Clements 1989, Clements and Hume 1995),
consonants and vowels share the set of place features, allowing for straightforward
explanations for phenomena in which consonants and vowels interact with respect to
place. Under this assumption, /f/, /u/ and /o/ are all specified for [labial] (as opposed to,
traditionally, /u/ and /o/ being characterized as [+back, +round] (SPE, Chomsky and
Halle 1968)) since the lips are required for both labiodental [f] articulation and [u] and
[o] rounding, as well.
26
In order to resolve this [labial] OCP violation, Spanish opts to delink the [labial]
feature of one of the offending segments, in this case, that of the fricative /f/ (1a),
according to Mazzaro. At this point the previously labiodental fricative is devoid of
place, resulting in an intermediate glottal fricative [h] (1b). The derivation, however, does
not end there, as Mazzaro argues that [h] becomes velar [x] via propagation of the
[dorsal] feature of the following back vowel (1c).2
(1)
a. Delinking due to OCP
//f
╪
│
[lab]
u//
│
│ [dor]
[lab]
→
b. Debuccalization
/h
u/
│
[dor]
→
c. Assimilation
[x
u]
│
[dor]
The analysis outlined in (1), however, must confront the empirical and theoretical
problems discussed in the following section.
2.1.2. Criticisms
In such a proposal, [labial] delinking (1a) should be obligatory, as its motivation–
the Obligatory Contour Principle–implies. An initial problem with Mazzaro’s analysis is
that delinking and assimilation are not obligatory, as [xu]/[xo] freely alternate with their
respective unassimilated counterparts [fu]/[fo] as indicated in Chapter 1. Although there
are clear allophonic preferences depending on the following vowel–apparent in Table
2.2–categorical complementary distribution is not achieved, which contrasts with other
cases of clear OCP phenomena: the restriction against multiple [labial] segments in
2
It is worth mentioning that, at the point of assimilation (1c), the back vowels /u/ and /o/–
endowed with both [dorsal] (backness) and [labial] (roundedness)–may propagate either of their
place features. However, assimilation of [labial] merely reintroduces Mazzaro’s [labial] OCP
violation and is therefore not selected.
27
Arabic root morphemes (McCarthy 1988), multiple voiced obstruent prohibitions in
words of patrimonial origin in Japanese (Itô and Mester 1986), etc.
Table 2.2. Percentage of occurrence of /f/ allophones in Corrientes, Argentina
[f]
[Ø]
N
[h]
N
%
/fa/
39
100
/fe/
18
86
/fi/
38
90
/fo/
17
85
3
/fu/
35
30
81
3
%
N
[x]
%
N
Total
%
N
%
39
100
21
100
42
100
15
20
100
70
116
100
14
4
10
Source: Adapted from Mazzaro (2005: 61).
Note: Shading indicates unattested output (zero occurrences).
The free variation between [fu]~[xu] and [fo]~[xo] demonstrated by speakers of these
dialects is left unexplained if the triggering mechanism is a [labial] OCP restriction. In
other words, speakers are able to variably produce [fumo] and [xumo] at rates of thirty
and seventy percent, respectively (Mazzaro 2005: 61), suggesting that the application of
Mazzaro’s [labial] OCP restriction is optional. She does not detail under what
circumstances (1a) would fail to apply in order to arrive at [fumo] as a surface
representation; i.e., when a labial OCP violation is tolerated.
Although the above issue of optionality is unaddressed in the analysis, current
understandings of the OCP as a family of (violable) OT constraints (see also Chapter 6
for additional proposals) are able to account for such free variation. A stronger argument
is to question why the [labial] feature of /f/ necessarily delinks as opposed to the [labial]
feature associated with /u/, a solution which is just as plausible and just as adequately
28
resolves the [labial] OCP violation from a theoretical perspective. This stipulation is
independently unmotivated.
Furthermore, other [labial]-[labial] sequences in the language such as [po] poro
‘pore,’ [bu] burla ‘mockery’ and [mu] música ‘music’ are unaffected by an OCP
restriction against adjacent [labial] segments under the relatively uncontroversial
assumption that /p/, /b/ and /m/ are also specified as [labial]. Mazzaro does not address
why apparent violations of a [labial] OCP such as those above undergo no repair (Place
delinking of the consonant), necessarily implying that the restriction is not generalizable,
only operating with respect to /f/ and not any other [labial] segment in the Spanish
phonological inventory. This stipulation is also unmotivated in her account.
With respect to (1b), Mazzaro relies on the OCP to derive [h] in an intermediate
stage after delinking, a step which is crucial to her analysis for the following reason: she
seeks a uniform account for additional data in which /f/ patterns with /s/ such that both
fricatives are able to debuccalize in various contexts: e.g., sacrificada [sakɾihikað̪ a]
‘sacrificed’ (feminine) and mis amigos [mihamiɣoh] ‘my friends.’ In the former example,
Mazzaro proposes that speakers have extended the OCP prohibition (against adjacent
labials) that results in [h] to a process of /f/ debuccalization in general, even before front
vowels /i/ and /e/ (why debuccalization should not also apply to /a/ is not apparent). If
such is the case, an asymmetry arises.
Following debuccalization in (1b), the [dorsal] specification of /u/ propagates,
resulting in [xu]. The [coronal] specification of /i/ in sacri[hi]cada, on the other hand,
does
not:
hypothetical
coronal-assimilated
outputs
*sacri[si]cada
(alveolar),
*sacri[θi]cada (interdental) and *sacri[çi]cada (palatal) must all be ruled out. Mazzaro,
29
following Rice and Avery (1991), accounts for the lack of assimilation in sacri[hi]cada
by assuming that coronals, in their role as the universally unmarked place of articulation,
are underspecified for place, thus preventing any place assimilation. Such an assumption
is problematic given that coronal consonants trigger assimilation in Spanish.
(2) Nasal place assimilation in Spanish
a. alianza
b. manta
c. ganso
d. ancho
[ali̯ an̟ θa]
[mãn̪ t̪ a]
[ɡanso]
[ãn̠ t͡ ʃo]
‘alliance’
‘blanket’
‘goose’
‘wide’
interdental (north and central Spain)
dental
alveolar (single closure)
palato-alveolar
If nasal place assimilation requires a four place distinction in the coronal region, then the
features [anterior] and [distributed] are arguably necessary, which are, in McCarthy’s
(1988) terms, dependents of [coronal]. If such is the case, [coronal] cannot therefore be
underspecified. Once underspecification becomes untenable, assimilating [coronal] to a
debuccalized /f/ in sacri[hi]cada parallel to (1c) can no longer be prevented, and the
unattested possibilities *sacri[si]cada, *sacri[θi]cada or *sacri[çi]cada are incorrectly
predicted.
On the other hand, one might argue that Mazzaro could claim that [coronal] is
specified for consonants only, accounting for the assimilation facts presented in (2), yet
underspecified in vowels, explaining why [h] in sacri[hi]cada remains unassimilated.
That
is,
positing
a
consonant/vowel
distinction
with
respect
to
[coronal]
(under)specification explains why [coronal] assimilation is attested in consonantconsonant interactions yet is unattested in the context of consonant-vowel. However, the
/x/ palatalization data presented in Chapter 1 suggest that /i/ and /e/ are indeed specified
for [coronal]; without this assumption, the fronting of velar /x/ to palatal [ç] in the
30
particular environment of these vowels would be a curious anomaly as no relevant feature
could explain the apparent case of [coronal] assimilation. This suggests that neither
coronal consonants nor vowels are underspecified for the [coronal] feature, which again
leaves Mazzaro unable to rule out [coronal] assimilation in sacri[hi]cada.
Not critical to the analysis but an oversight nonetheless, Mazzaro also leaves
unaddressed the fact that elision (/f/→[Ø]) is licit only when /f/ precedes the mid front
vowel [e], e.g., [Ø]estejábamos ‘we celebrated’ (imperfect), cf. *[Ø]amilia ‘family.’ Such a
derivation requires a mechanism separate from /f/→[x] since delinking in velarization
targets the Place node (if Mazzaro’s //f//→/h/→[x] process from (1) is assumed), while
elision delinks at the Root node.
Even though assimilation (1c) does seem to be on the right track, I see no reason
to assume that debuccalization due to a [labial] OCP constraint (1a) must precede it, as
(a) Spanish attests no other OCP restrictions, reducing the analysis to an ad hoc solution,
and (b) it makes incorrect predictions for labial stops (oral and nasal) followed by back
vowels (i.e., no repair attested in poro ‘pore,’ burla ‘mockery,’ etc.).
Lipski (1995), summarized in the following section, addresses some of the
criticisms above. Section 2.2.2, however, argues that further analysis of the properties in
question is merited given his own theoretical and empirical challenges.
2.2. Lipski (1995)
Lipski and Mazzaro are similar in their analyses in that they both (a) assume
feature geometry and (b) rely on the Obligatory Contour Principle. They differ crucially
with respect to the feature targeted by the OCP, perhaps a consequence of the fact that
Mazzaro analyzes /f/ velarization while Lipski principally investigates /f/ bilabialization
31
(with occasional reference to /f/ velarization). With this difference, Lipski’s analysis
provides possible responses to the criticisms from the previous section. However, the
additional assumptions he must make, which I explore in the section below, are not
without their own inadequacies.
2.2.1. Analysis
Although very much related to /f/ bilabialization and velarization, Lipski’s data
are specific to a variety of Spanish spoken in Iquitos, Peru (in addition to the coastal
regions of southeastern Mexico, northwestern Ecuador and western Colombia), where
hypercorrection and errors by communities of second language (L2) learners of Spanish
throughout history (Lipski 1995: 297, 298) resulted in the alternations seen in (3) (Lipski
1995: 286, 287).
(3)
a.
juicio
juez
Juan
[hwi] ~ [ɸi]
[hwe] ~ [ɸe]
[hwa] ~ [ɸa]
b.
fin
feo
fácil
[ɸi]
[ɸe]
[ɸa]
c.
fumar
fue
formación
[ɸu] ~ [hu]
[ɸwe] ~ [hwe]
[ɸo] ~ [ho]
~ [hwi]
~ [hwe]
~ [hwa]
‘judgment’
‘judge’ (noun)
(proper name)
‘end’ (noun)
‘ugly’ (masculine)
‘easy’
‘to smoke’ (infinitive)
‘s/he/it went’ (preterit)
‘formation’
In (3a), the grapheme <j> represents an underlying velar fricative /x/ (commonly realized
as glottal [h] in this dialect); when followed by the semivocalic [w]3 manifestation of the
high back vowel /u/ as the non-nucleic member of a diphthong (/ui/, /ue/ and /ua/,
respectively), [hw] alternates with the bilabial fricative [ɸ], a common realization of
3
In (3), I faithfully follow Lipski’s transcription convention; specifically, /uV/ diphthongs are
rendered with a superscript labiovelar on-glide [w], which, for Lipski’s purposes, is treated as part
of a natural class with monophthong /u/ ([u]).
32
labiodental /f/ for these speakers. (Alternative spellings for (3a) representative of
pronunciation include ficio, fez and Fan.)
(3b) reflects the inverse of (3a) in which the bilabial fricative [ɸ] in the
environment of non-back vowels /i/, /e/ and /a/ is realized as glottal [h] with the
labiovelar diphthong [wV], normally the surface realization of underlying /xuV/ (thus,
(3b) is alternatively written juin, jueo and juácil).
Lastly, bilabial [ɸ] in the context of back vowels /u/ (including when part of a
diphthong) and /o/ alternates with glottal [h] in (3c), resulting in possible orthographic
alternatives jumar, jue and jormación. Note that, although they are superficially similar
alternations, (3b) differs from (3c) crucially in that the glottal allophone of (3b) contains
a diphthong [hwV] not present in the bilabial variant [ɸ] (*[ɸwV]), while (3c) attests no
such distinction: either [ɸ] alternates with [h] (with both variants lacking a diphthong–
*[hw]), or [ɸwV] with [hwV] (both containing a diphthong). It is the surface realization of
the diphthong in (3b) that proves crucial for Lipski’s analysis, detailed below.
Lipski, like Mazzaro, assumes that phonological features are organized
geometrically. Although various hierarchical structures have been proposed (see
McCarthy 1988 for review), Lipski (1995: 288) adopts the following (irrelevant
nodes/features ignored).
(4)
a.
[p, b, m, f]
│
Place
│
[labial]
b.
[f, ɸ]
│
Place
│
[round]
c.
[u, o]
│
Place
│
[round]
33
In addition to the structures above, his analysis relies on two critical assumptions. First,
he assumes that the labial consonants [p], [b], [m] and (occasionally) [f] (4a) are specified
with a uniquely consonantal [labial] feature that does not (cannot) interact with vowels.
Second, in dialects like those exemplified in (3) in which bilabial fricative [ɸ] allophonic
alternations exist, [f] and [ɸ] (4b) instead contain the place feature [round] that does (can)
interact with the traditional [round] feature associated with the back rounded vowels [u]
and [o] (4c).
Lipski explains the data in (3) by proposing not a [labial] OCP prohibition (contra
Mazzaro 2005), but rather an OCP restriction on adjacent [round] features. In this way,
Lipski is superior to Mazzaro, as the former does not predict phonological repair in data
such as poro ‘pore,’ burla ‘mockery’ and música ‘music.’ If we accept Lipski’s
distribution of the features [labial] and [round] in (4), [po] in poro, [bu] from burla and
[mu] in música represent sequences of [labial]-[round], not [round]-[round]; thus an OCP
restriction against consecutive [round] features is not violated, and no phonological
process is necessitated. As such, Lipski correctly predicts which labial segments may (or
may not) interact in this variety of Spanish, given his [round] OCP analysis; Table 2.3
illustrates these proposed interactions.
Table 2.3. Labial interaction vis-à-vis [round] OCP in the Spanish of Iquitos, Peru
[labial]
[f]
[ɸ]
[u]
*
*
[o]
*
*
[p]
[round]
[round]
[b]
[m]
[f]
Note: The asterisk ‘*’ and shading indicate ungrammatical output forms assuming the
operation of a [round] OCP restriction.
34
According to Lipski, [f] (in certain dialects in which [ɸ] is an allophone of /f/), [ɸ] and
the high back vowels [u] and [o] all contain [round] specifications. When /f/ appears in
the environment of a round vowel–e.g., fumar ‘to smoke’ (infinitive), fue ‘s/he/it went’
(preterit) and formación ‘formation’ (3c)–the OCP against [round] segments is violated,
necessitating repair. [p], [b], [m] and [f] (when no allophonic alternation exists), on the
other hand, are strictly [labial] and do not trigger an OCP violation when adjacent to a
[round] vowel.
Therefore, once [round] delinks in response to the [round] OCP violation, Lipski
claims that the feature remains unattached–in his terms, becomes a “floating
autosegment” (Lipski 1995: 283)–and can reattach freely to either a consonant or a vowel
to the extent that the resultant structure is well-formed, or delete if reattachment cannot
be accomplished grammatically. Consequently, Lipski avoids a criticism faced by
Mazzaro; while Mazzaro must stipulate without additional motivation that OCP-induced
delinking occurs only on the consonant, Lipski argues that, to the extent that the delinked
structure and the structure created after delinking are both well-formed, delinking may
affect the consonant or the vowel. Let us take fumar ‘to smoke’ (infinitive) as a starting
point.
(5)
[ɸ]
│
Place
╪
[round]
[u]
│
Place
│
[round]
Lipski begins with the premise (without developing the point further) that labiodental /f/
in these dialects has as a possible output allophone [ɸ] (more on this in section 2.2.2),
35
reflected in the [ɸu] representation of fumar in (5). In these dialects where [ɸ] is a
possible /f/ allophone, the operation of a [round] OCP restriction prohibiting consecutive
[round] features will delink the [round] specification of [ɸ]. All else being equal (i.e.,
voicelessness and continuancy), the consonant post-delinking is simply a placeless
fricative, i.e., glottal [h]. In this case, Lipski argues that delinking the [round] feature
associated with [u] is impossible, as the high back unrounded vowel [ɯ] that would result
does not exist in Spanish.
Accounting for (3a, b), however, is not as straightforward. To illustrate, a partial
representation of fin ‘end’ (noun) appears in (6).
(6)
[ɸ]
│
Place
│
[round]
[i]
│
Place
│
[coronal]
Arriving at [hwin] from (6) cannot be due to Lipski’s [round] OCP, as the high front
(unrounded) vowel [i] is not specified as [round] (or [labial], for that matter). Like
Mazzaro, Lipski argues that speakers have extended the OCP restriction to a simple
process of [round] delinking in any environment, including before non-back (specifically,
non-round) vowels, [i] in this case. Once [round] is able to delink freely, Lipski argues
that it remains unassociated in the underlying representation of forms where, e.g.,
[ɸ]~[hw] alternations exist. The output form that results depends on the direction to which
this free [round] feature associates. Both possibilities are exemplified below.
36
Figure 2.1. Direction of free [round] association
(a)
ficio (< juicio)
[ɸ]
│
Place
[V]
│
Place
│
[–back]
[round]
(b)
juin (< fin)
[h]
│
Place
[wV]
│
Place
│
[–back]
[round]
Option (a), representative of an etymological form such as juicio ‘judgment’ seen in (3a),
contains a delinked [round] feature that historically originated from the semivocalic half
[w] of the diphthong [wi] < /ui/. If, as Lipski suggests, [round] came to be unassociated in
underlying forms, this floating feature could either reattach to the vowel to its right,
resulting in expected [hwi], or it could propagate to the consonant to its left. The addition
of [round] to an otherwise placeless fricative explains the [ɸ] alternative, ficio, and the
disappearance of the diphthong, as only one [round] feature is available (*[ɸw]).
Conversely, in (b), an additional diphthong appears that does not derive from an
underlying /u/, but rather from the [round] feature originally associated with [ɸ] < /f/. For
example, the juin [hwin] pronunciation of fin ‘end’ (noun) (3b) is explained by linking the
floating [round] feature to the vowel that follows, thus creating a diphthong, and leaving
[h] (< [ɸ]) placeless. Alternatively, nothing prevents [round] from reattaching to the
consonant, accounting for the simultaneous existence of forms such as fin [ɸin].
Although superior to Mazzaro in the two ways discussed above–(a) avoiding
overgeneralization by predicting ill-formedness in well-formed output in Spanish such as
poro, burla and música, and (b) going beyond mere stipulation concerning the target
37
(consonant or vowel) of OCP delinking–Lipski’s analysis also falls victim to various
theoretical and empirical problems, explored in the next section.
2.2.2. Criticisms
It is first necessary to qualify the discussion that follows. As previously
mentioned, (3) represents pronunciations found in only a few regions of Peru, Mexico
and Colombia, while the data presented in Chapter 1 that form the basis of this thesis–
culled from the numerous sources cited–have been reported in various areas throughout
Latin America. The goal of this dissertation is not to analyze a particular dialect like
Lipski, but rather to uniformly account for three related processes in Spanish by way of
typological analysis. Given our differing scopes and research questions, a few criticisms
that follow might not necessarily be criticisms per se, but rather consequences of our
unique goals; I caution the reader to keep this in mind.
With the above caveat, I will first address the principle, yet most dubious,
component of the analysis, the dissociated [round] autosegment. Lipski argues that once
[round] delinking via OCP is analogized to every instance of [round] irrespective of
adjacent segments, the free [round] feature becomes part of the underlying representation,
a fact that he uses to explain the [ɸ]~[hw] alternations. It is not made clear how speakers
acquire this floating feature as part of their lexical entries other than suggesting that it is
due to the fact that speakers recognize such alternations exist, a circular explanation–i.e.,
a free [round] features exists because of the [ɸ]~[hw] alternation, which is in turn
explained by positing a free [round] feature.
Beyond the notion of a free feature in underlying specifications, the issue of how
/f/ comes to be realized as [ɸ] in (3) is a central analytical goal of this dissertation; Lipski
38
(1995) does not address this question and instead simply assumes it, which is a
consequence of his research goals and not necessarily a fault in the analysis. While he
seeks to explain the [ɸ]~[hw] alternation, I explore /f/→[ɸ] and relate it to /f/→[x] and
/x/→[ç]. To the extent that (/f/ >) [ɸ]~[h] (3c) is similar to the /f/ velarization data that I
wish to explain, the application of a [round] OCP to forms in (3c) in accounting for /f/
allophony is what must be scrutinized, as it is here where the analyses (at least
marginally) overlap.
Part of Lipski’s application of the OCP to these data relies on the assumption that
[f], [ɸ], [u] and [o] form a natural classed defined by the feature [round]. Attributing
[round] in addition to the widely accepted [labial] to /f/ is contradictory, demonstrated
explicitly when Lipski states, on the one hand, that “labiodental /f/ [is] characterized by
an exclusively consonantal feature such as [labial] and [does] not interact with a [round]
specification found on rounded vowels” (Lipski 1995: 285), but also that, “in Spanish
dialects in which /f/ is realized as [ɸ], /f/ is specified as [round]” (Lipski 1995: 289), on
the other. That is, /f/ is [labial] and does not interact with vowels when no alternation
exists, yet [round] and does succumb to vocalic assimilation when bilabial [ɸ] is a
possible allophone. This is not an explanation, but rather a mere description of the data.
Lipski must stipulate this [round] versus [labial] distinction, as he must
differentiate [f] and [ɸ] from the other traditionally [labial] segments [p], [b] and [m] in
order to avoid the overapplication issue that Mazzaro faces, i.e., predicting
ungrammaticality in poro, burla and música. However, as Ní Chiosáin and Padgett
(1993), Clements and Hume (1995), Cho (1997) and Beckman (2000) (inter alia) show,
labial consonants [p], [b], [m], [f] and [ɸ] and the rounded vowels [u] and [o] do form a
39
natural class. Thus, relying on [round] to distinguish {[f], [ɸ], [u], [o]} from {[p], [b],
[m]} is ad hoc. Once [round] can no longer be recruited to group [f] and [ɸ] with the
rounded vowels to the exclusion of [p], [b] and [m], the proposed [round] OCP of Lipski
is no different than Mazzaro’s [labial] OCP, and thus falls victim to the same criticism
presented previously concerning the predicted {[p], [b], [m]} and {[u], [o]} interaction.
I also argued that, theoretically, Mazzaro has no way to limit which segment–the
consonant or the vowel–delinks in response to the OCP. Lipski avoids this issue by
claiming that either segment is a possible target as long as the resultant structure remains
well-formed. In the default scenario (3c), Place is delinked from the underlying fricative,
and glottal [h] is the expected outcome. Delinking [round] from a vocalic segment is only
possible in the case of an underlying diphthong (3a), as delinking [round] from
monophthong /u/ or /o/ results in an unrounded back vowel [ɯ] or [ɤ], respectively.
On this point, Lipski notes that “removal of [round] [from a back vowel] would
create a phonologically nonpermitted (nonlow) back unrounded vowel” (Lipski 1995:
291). As Lipski states, unrounded back vowels are unattested in Spanish,4 yet we cannot
theoretically rule out delinking [round] from /u/ or /o/ simply by arguing that Spanish
lacks the vowel that would result (a non-low unrounded back vowel), as such an
4
To list a few examples regarding unrounded back vowels in other languages, Turkish
phonemically contrasts high back rounded /u/ and unrounded /ɯ/ (Göksel 2005: 9, contra Kornfilt
1990: 623, who argues instead for the central vowel /ɨ/); Korean, on the other hand, contains a
four-way contrast: short rounded /u/ and unrounded /ɯ/, in addition to long rounded /uː/ and
unrounded /ɯː/ (Kim 1990, Lee 1993, Yeon 2001). Concerning the mid back vowels, Scottish
Gaelic phonemically contrasts rounded /o/ and unrounded /ɤ/ (Lamb 2001), while Thai (Hudak
1990, Tingsabadh and Abramson 1993, Smyth 2002)–like Korean–also contrasts for length.
Beyond phonemic contrasts, Turkish attests rounding harmony (Göksel op. cit.) which requires
rounded~unrounded alternations. Because the theory must allow such alternations as a
consequence of phonological processes, Lipski cannot rule out vocalic [round] delinking simply
because the Spanish data lack the unrounded result.
40
argument is circular.5 Additionally, while the respective unrounded counterparts to [u]
and [o] are crosslinguistically infrequent, they are not inarticulable. Put in other terms,
the theoretical mechanism must be constrained by components of the theory, not by the
data it wishes to explain.
Since no adequate (internal) theoretical solution was found in the previous
sections, one might question if (external) sociolinguistic factors could account for the
phenomena. Indeed, Lipski (1995: 287, 297–299) alludes to this point while briefly
discussing various situations of language contact that have shaped Spanish, specifically in
Latin America, throughout its development. Lipski does not propose, however, that the
properties in question have extra-Spanish origins; rather, I am taking language contact in
general and questioning if it could account for the alternations. Lipski, on the other hand,
uses language contact to argue that imperfect L2 learning of Spanish by the African and
indigenous populations can account for his specific [ɸ]~[hw] data set from (3b).
He suggests that it is because the African and indigenous language speakers
lacked [ɸ] that they needed an alternative when acquiring Spanish, which happened to
result in [hw]. Not unlike how, e.g., English borrowed the French word menu [məny] as
[mɛñ juʊ̯] by decomposing the foreign final front rounded vowel [y] into [juʊ̯] with the
glide [j] maintaining frontness and the (high) back rounded vowel [uʊ̯] roundedness,
Lipski claims that the L2 Spanish learners of the time, when confronted with [ɸ],
decomposed the fricative into glottal [h] to preserve frication and vocalic [w] for
labiality/roundedness. He relies on hypercorrection in explaining (3a), in that these
speakers then (mistakenly) reinterpreted (standard) Spanish /xuV/, e.g., Juan, as a case of
5
However, translating the autosegmental proposal (in which the segmental source of [round]delinking or the lack thereof must be stipulated by force) into OT terms is an issue of proposing
either segmental markedness/inventory (e.g., *[+back, –round]) or IDENT([round]) constraints.
41
[hw] < [ɸ], resulting in [ɸan] Fan. His logic is that, if these L2 acquirers of Spanish
already had a representation for [ɸ] from their first language, there would be no need to
replace it with [hw]. For this reason, Lipski in fact concludes that African and indigenous
influence cannot be the source of Spanish [ɸ].
However, the above explanation applies only to the dialects that Lipski addresses,
while this dissertation is broader in scope; furthermore, because it is not the main point of
his article, Lipski does not delve into detail concerning precisely what the African and
indigenous substrata were (see, however, Lipski 2005); this is the goal of the following
subsections. I will ultimately conclude, like Lipski, that language contact cannot be
responsible for the data under consideration, and that an alternative internal approach
remains necessary.
2.3. External explanations
Given the complex sociolinguistic situation of colonial Latin America–in addition
to the above issues of the internal analyses–it is not unreasonable to pose the following:
are the alternations under consideration a consequence of language contact? That the
previously cited literature (Perissinotto 1975, Resnick 1975, Sanicky 1988, Lipski 1995,
Mazzaro 2005, Piñeros 2009) limits the processes to Latin America to the exclusion of
Spain suggests such a solution. Furthermore, Spanish previously experienced a similar
situation under different circumstances; although it is not without debate, Otero (1971,
inter alia) argues that certain properties particular to Spanish (e.g., initial /f/ deletion (see
Chapter 6), sibilant devoicing and apical /s̺ /) not found in other Romance languages can
be derived from contact with Basque following the Muslim conquest of the Iberian
Peninsula in 711 CE. Could language contact account for the fricative alternations under
42
investigation, as well? This research question informs the remainder of this chapter with
the purpose of briefly surveying the phonemic, phonetic and phonological properties
(where relevant) of the African and indigenous languages with which Latin American
Spanish came into contact.
2.3.1. The African substratum
Identifying the African languages spoken by the 9.5 million slaves (Moreno
Fraginals 1984: 5) imported during approximately three and a half centuries of Spanish
colonization is not an easy task, made more difficult by the fact that Spanish colonial
records paid little attention to origin (Rout 1976: 27). However, through historical
documentation and cultural and linguistic analyses (see, e.g., Schwegler 2006), various
degrees of specificity have been achieved concerning the languages and ancestral regions
of the African diaspora. While many details of these complex issues fall outside the scope
of this dissertation, this section summarizes the principal African languages that came
into contact with colonial Latin American Spanish.
Rout (1976) and Thornton (1992) divide Africa, respectively, into four and three
major regions along the western and southern African coasts from which slaves were
imported (the difference in division owes to the fact that Rout includes Portuguese slave
regions (omitted here), while Thornton limits himself to Spain).
Table 2.4. Origins of colonial Latin American slaves by region
Region
Corresponding countries
Upper Guinea
Guinea, Guinea-Bissau, Liberia, Mali, Senegal, Sierra Leone
Lower Guinea
Benin, Cameroon, Côte d’Ivoire, Ghana, Nigeria, Togo
West Central
Angola, Congo, Gabon
Sources: Compiled from Rout (1976) and Thornton (1992).
43
Of these fifteen nation-states, the majority of Spanish colonial slaves came from those of
the Upper and Lower Guinea regions, especially during early colonization, with
expansion into the West Central region as demand increased (Lipski 2005: 47). Within
these countries we find the following primary languages and linguistic groups.
Table 2.5. African languages and linguistic groups by region
Region(s)
Language(s) or linguistic group(s)
Angola, Congo
Kongo (Bantu)
Benin
Ewe, Fon (Kwa)
Côte d’Ivoire
Akan (Kwa)
Ghana
Gã (Kwa)
Nigeria
Edo, Igbo, Yoruba (Niger-Congo)
Senegal
Mandinka (Mande), Wolof
Sierra Leon, Liberia
Mende (Mande)
Sources: Compiled from Thornton (1992: 186–192) and Lipski (2005: 44, 45).
Note: Larger linguistic groups generally spoken in the countries are listed in parentheses.
Admittedly Table 2.5 is non-exhaustive and overlooks numerous African languages that
came into contact with Latin American Spanish, yet the contribution made by the
principal linguistic groups above, according to Thornton and Lipski, suffices for our
present purposes. Given that the fricative inventories and related processes of these
languages are our primary concern, Table 2.6 is limited to these aspects, disregarding
irrelevant consonants and vowels.
If it is the case that Spanish /f/ velarization and bilabialization, in addition to /x/
palatalization, are a product of language contact, the necessary prediction is that the
44
African languages under consideration will have equivalent (or sufficiently similar)
phonemes, allophones or phonological processes.
Table 2.6. Fricatives of African languages
Language
Akan
Labialized C(s)
w
/h /
/f s h/
Edo
Process(es)
/hi, hwi /→[çi, çɥi]
/f v s z x ɣ h/
/ɸ f v s z x ɣ ɦ/
/f v s z x ɣ/
Ewe
w
w
Fon
/x ɣ /
Gã
/ʃw hw/
Igbo
Fricatives
w
w
/si, zi/→[ʃi, ʒi]
/f v s z ʃ h/
w
w a
/k ɡ / (/ɣ h /)
/f (v) s z (ʃ) ɣ (h ɦ)/
Kongo
/f v s z/
Mende
/f v s h/
/f s x () (h)/
Yoruba
/f s ʃ h/
Sources: Akan (Dolphyne 1988: 29), Edo (Elugbe 1989: 77), Ewe (Duthie 1996: 12), Fon
(Lefebvre and Brousseau 2002: 16), Gã (Kropp Dakubu 2002: 66), Igbo (Clark 1990: 14,
Ikekeonwu 1999: 108), Kongo (Welmers 1973: 70), Mende (Tateishi 1990: 99), Wolof
(Ka 1994: 63, Torrence 2005: 12) and Yoruba (Bamgbos̩ e 1966: 6).
Wolof
Note: Shading indicates an African property hypothetically relevant for Spanish.
a
Phonemes appearing in parenthesis are dialectal.
Table 2.6 suggests that the Spanish processes under investigation in this dissertation did
not originate from African influence. While Lipski’s explanation for bilabial [ɸ]
decomposition is supported by the various languages that contain labialized consonants
(thus providing the necessary structure for decomposition), very little evidence
corroborates the hypothesis that Latin American Spanish /f/ velarization, /f/
bilabialization and /x/ palatalization were transferred from an African language. Only one
language, Ewe, attests /ɸ/ phonemically. Of the sources consulted, no language manifests
45
[ɸ] allophonically, for that matter. The only relevant phonological process comes from
Akan in which /h/ and labialized /hw/ palatalize preceding the high front vowel /i/.
Given Tables 2.6 and 2.7 such a dearth of comparable phonemes, allophones and
phonological processes in the African languages cannot account for the vast territory over
which fricative allophony is attested in Latin America.
Table 2.7. Phonological processes by country
Assimilation type(s)
/f/ bilabialization
Country / countries
Cuba
/f/ velarization
/x/ palatalization
Uruguay
/f/ bilabialization and /f/ velarization
Bolivia, Colombia, Costa Rica,
Dominican Republic, Ecuador,
El Salvador, Guatemala, Panama,
Peru, Puerto Rico
/f/ bilabialization and /x/ palatalization
/f/ velarization and /x/ palatalization
all three
Source: Resnick’s (1975) dialectal survey.
Argentina, Chile, Mexico
Note: Resnick reports that Honduras, Nicaragua, Paraguay and Venezuela attested none
of the three processes in question.
Even if Akan contains a similar palatalization process, and Ewe phonemically contrasts
the voiceless bilabial fricative /ɸ/ with /f/ and /x/, it seems a stretch to say that these two
facts are able to account for the widespread distribution of /x/ palatalization and /f/
bilabialization throughout Latin American Spanish as demonstrated by Table 2.7, not to
mention the complete lack of any contributor to /f/ velarization other than the observation
that Edo, Ewe, Fon and Wolof contain the voiceless velar fricative /x/ as a phoneme; this,
46
however, is inconsequential, as Latin American Spanish does (and did during colonial
times), as well–even if some dialects contained glottal /h/ in its place.
Since African linguistic influence must be rejected, the next section explores the
possibility that American indigenous languages contributed to the processes. Ultimately,
this hypothesis is also rejected.
2.3.2. The American indigenous substratum
In the last section of this chapter, I explore the possibility that contact between
Spanish and American indigenous languages accounts for the phonological processes in
question. Not unlike the situation concerning African influence, the multitude of
indigenous languages precludes me from completing an exhaustive survey (see Campbell
1997), yet for our present purposes it suffices to examine the five languages with the
greatest number of reported speakers in the two countries from which I gathered data,
Chile and Mexico (which also happen to attest all three properties; see Table 2.7 above).
Table 2.8. Indigenous languages by population
Country
Chile
Mexico
Language
Number of speakers
1. Mapudungun
200,000
2. Quechua
8,200
3. Rapa Nui
5,840
4. Huilliche
2,000
5. Aymara
900
1. Nahuatl
1,376,026
2. Yucatec Maya
759,000
3. Mixtec
423,216
4. Zapotec
410,901
5. Tzeltal Maya
31,730
Sources: Chile (Ethnologue, Lewis 2009); Mexico (Instituto Nacional de Estadística y
Geografía [National Institute of Statistics and Geography] 2005).
47
In order to determine if the above languages influenced Spanish specifically in the cases
of /f/ bilabialization, /f/ velarization and /x/ palatalization, it is necessary to examine their
fricative inventories and allophonic processes.
Table 2.9. Fricatives of American indigenous languages
Language
Labialized C(s)
Phonemes
Relevant allophones
Aymara
/s ʃ x h/
Huilliche
/f θ s/
Mapudungun
/f (v)a θ (ð) s (ʂ)/
/f/ [ɸ v x h]
Quechua
/(f) s (ʃ) h/
/p/ [ɸ]
Rapa Nui
/v (s, ʂ) h/
w
Mixtec
/k /
/(v) (ð) s ʃ (ʒ) (nʒ) x/
Nahuatl
/kw/
/(f) s ʃ (h)/
Tzeltal Maya
/b/ [, ɸw], /x/ [ç, h]
/s ʃ h/
Yucatec Maya
/s ʃ h/
w
w
Zapotec
(/k ɡ /)
/(f) s z (ʂ ʐ) ʃ ʒ (x)/
Sources: Aymara (Hardman 2001: 12), Huilliche (Álvarez-Santullano Busch 1992: 63),
Mapudungun (Zúñiga 2000: 5), Mixtec (North and Shields 1977: 21, Macaulay 1996:
19), Nahuatl (Launey 2011: 4), Quechua (Cerrón-Palomino 1987: 96–99, Rodríguez
Champi 2006: 1, 5), Rapa Nui (Du Feu 1996: 182), Tzeltal Maya (Gerdel 1974: 101),
Yucatec Maya (Brody 2004: 60) and Zapotec (Munro 2005: 90, Merrill 2008).
Note: Shading indicates an indigenous property hypothetically relevant for Spanish.
a
Phonemes appearing in parenthesis are dialectal; segments appearing in brackets are
allophones.
Before proceeding to any detailed discussion, it must be mentioned that Quechua,
Mixtec, Maya and Zapotec, although treated monolithically here, are instead language
families consisting of many individual languages, not always mutually intelligible. They
are represented in the table above with as many dialectal variants as possible.
48
Referring specifically to Mixtec, Nahuatl and Zapotec, once again there is
evidence of labialization, lending support to Lipski’s decomposition hypothesis.
Concerning, however, the three processes relevant for this dissertation–/f/ bilabialization,
/f/ velarization and /x/ palatalization–there is again little evidence to suggest that they
have their origins in American indigenous languages. Mapudungun attests [ɸ, v, x, h] all
as /f/ allophones, yet Zúñiga (2000) does not specify if they are in free variation or
context-dependent (complementary distribution). Quechua realizes /p/ as [ɸ] in coda
position (Rodríguez Champi 2006), yet this appears to be a completely distinct process
from those considered here.
According to Macaulay (1996: 19, fn. 3), /b/ is normally realized as the voiced
bilabial fricative (approximate) [] in intervocalic position in Mixtec, yet a labialized
voiceless bilabial fricative [ɸw] is possible in one native word: [ʒuɸwʔe] žubeʔe ‘door.’ It
is doubtful that the limited distribution of this allophone had any significant impact. To
further the point, Macaulay (1996: 25) states that Mixtec has borrowed fuerza ‘strength’
from Spanish as [ɸweɾsa] (her transcription convention), which is either the result of [f]
decomposition along the lines of Lipski, or is indeed faithful to the Spanish input to
which these Mixtec speakers are/were exposed, which, by definition, means that the
process originated in Spanish and was instead transferred to Mixtec.
Concerning velar /x/ properties, Macaulay (1996: 20) does note that palatal [ç] is
a possible allophone before the high front (oral and nasal) vowels /i, ı/̃ of Mixtec,
reminiscent of the /x/ palatalization data in Spanish. However, if one wishes to argue that
(synchronic) Spanish /x/ palatalization has its origins in Mixtec /x/ palatalization, one
would need to account for the prevalence of palatalization in Argentina and Chile, far
49
outside the attested region of Mixtec speakers. For these reasons, an indigenous influence
hypothesis must also be rejected.
Additionally, in the languages for which /f/ and /x/ are listed dialectally (in the
parenthesis in Table 2.9), the respective sources note that the phonemes are non-native,
and appear only in Spanish borrowings. In this case, it cannot be that Spanish /f/ and /x/
phonological processes originated with the indigenous languages, as these languages
lacked the requisite phonemes, but rather that Spanish influenced the substrata.
Furthermore, when viewed in light of the development of Spanish, reference to
external factors such as African or indigenous origin might be redundant, as /f/, [f], /ɸ/
and [ɸ] have a long, interconnected history (while Penny 2002 does appeal to external
French influence, it is not the catalyst for /f/ bilabialization, assimilation is). According to
the narrative presented in Penny (2002: 90), bilabial [ɸ] was a common output realization
of /f/ in the late spoken Latin of Cantabria. With [ɸ]’s increased usage, Penny argues that
/f/ was eventually replaced by phonemic /ɸ/, and progressed through (roughly) five stages
of development.
Table 2.10. Five stages of Latin F
Stage 1
Stage 2
Stage 3
Stage 4
Stage 5
/ɸ/→[ʍ, ɸ]
/ɸ/→[ʍ, h, ɸ]
/ɸ/→[ʍ, h, ɸ]
/h/→[f, h]
/f/ versus /Ø/
[ʍ] preceding
the glide [w]
[ʍ] preceding
the glide [w]
phonemicization
of [f]
dissimilation,
[h] preceding
syllabic [u, o]
e.g., forno
‘oven’
generalization
of [h]
preceding all
vowels
French
influence,
[f] reintroduced
preceding [w]
and [ɾ]
e.g., fuerte,
frío
[ɸ] elsewhere
[ɸ] elsewhere
[h] elsewhere
assimilation,
[ʍ] preceding
the glide [w]
e.g., fuerte
‘strong’
[ɸ] elsewhere
[h]-dropping,
[Ø] elsewhere
e.g., horno
50
While I leave the details of this process for Chapter 6 when I frame the synchronic
properties in a diachronic perspective, Table 2.10 suffices to serve as evidence that, at
one point in the development of Spanish, bilabial [ɸ] existed allophonically, perhaps
rendering unnecessary an external hypothesis.
2.4. Conclusion
This chapter has reviewed two previous theoretical analyses in Lipski (1995) and
Mazzaro (2005) and has argued that their proposed OCP restrictions in accounting for /f/
bilabialization and /f/ velarization are inadequate. Additionally, I have ruled out external
accounts that attempt to explain the data by arguing for African and/or American
indigenous influence/transfer.
The following two chapters are devoted to clarifying and updating the previously
reported data based on the criticisms from Chapter 1. Returning to the bilabial [ɸ]~glottal
[h] alternation from (3c)–the data set from Lipski (1995) closest to the basis for this
dissertation–while it is the case that, in certain Latin American dialects, glottal [h] is
indeed the output of etymological /x/ (and probably phonologized as /h/) (Canfield 1981,
Lipski 1994), most of the sources cited thus far (Perissinotto 1975, Resnick 1975,
Mazzaro 2005, Piñeros 2009) argue that velar [x] is the output of /f/ preceding a back
vowel (at least in the dialects they consider), and not simply [h] via debuccalization, as
such would leave unexplained the clear preference for /f/ unfaithfulness specifically in
the context of rounded back vowels (/f/→[f] faithfulness elsewhere) claimed by the above
sources. Although this difference might be due to the fact that Lipski is addressing output
from a specific dialect, not necessarily to the exclusion of velar [x] in other dialects, this
51
dissertation endeavors to better examine the nature of the possible /f/ allophones via
acoustic analysis in Chapter 4.
Attesting to the complicated nature of the data, Sanicky (1988) reports on seven
possible /f/ allophones in (occasionally) overlapping contexts in the Spanish of 129
speakers from Misiones, Argentina. Although sociolinguistic in nature (and consequently
statistically analyzed by age and socioeconomic level, the details of which are irrelevant
here), my point in presenting her data is that Lipski (1995) and Mazzaro (2005) are not
mutually exclusive; i.e., even though Lipski analyzes bilabialization /f/→[ɸ]~[hw] while
Mazzaro accounts for velarization /f/→[x], both processes may exist in a single dialect,
as Table 2.11 demonstrates (see also Table 2.7).
Table 2.11. Distribution of /f/ allophones in Misiones, Argentina
normal speech rate
[f]
rapid or intervocalic
[v]
[ɸ]~[fɸ]
v
[ ]
||/f/
[||f]uimos
/fC/
[fɾ]ío
[ɸl]or
/fa/
[f]amilia
[ɸ]ábrica
/fe/
ca[f]é
per[ɸ]ectamente
/fi/
clasi[f]icar
di[ɸ]icultoso
/fo/
uni[f]ormado
a[ɸ]ónica
/fu/
[f]unciona
[ɸ]unda
[x]~[h]
[ɦ]
[x]amilia (n=1)
[x]uerte
Source: Adapted from Sanicky (1988).
Note: Spanish glosses (left to right, top to bottom) are as follows: fuimos ‘we went’
(preterit), frío ‘cold,’ flor ‘flower,’ familia ‘family,’ fábrica ‘factory,’ café ‘coffee,’
perfectamente ‘perfectly,’ clasificiar ‘to classify’ (infinitive), dificultoso ‘difficult,’
uniformado ‘standardized’ (masculine adjective), afónica ‘lost one’s voice’ (feminine
adjective), funciona ‘s/he/it functions’ (present indicative), funda ‘s/he/it founds’ (present
indicative), fuerte ‘strong.’
52
Exemplified in the first row from Table 2.11 above, Sanicky reports that, in post-pausal
position, [f] is the only attested allophone. Followed by a consonant, both [f] and [ɸ] are
possible (in addition to their rapid speech counterparts), as seen in the second row. This is
also the case when followed by the vowels [a, e, i, o]. In the final row, Sanicky reports
that the context /fu/ permits the greatest range of allophony, allowing [f], [ɸ] and [x]. In
this respect Sanicky differs from Perissinotto (1975), Resnick (1975), Lipski (1995),
Mazzaro (2005) and Piñeros (2009), who all state that velarization is possible in the
context of /o/, as well. Of course dialects differ, and given the large geographic area
covered under the term ‘Latin American Spanish,’ the amount of dialectal difference
possible (and attested) is not unexpected. Even so, accurate phonetic reporting is
paramount, and it is with this goal in mind that I supplement the data from Chapter 1 with
additional experimental data in Chapter 4, outlining its corresponding methodology in
Chapter 3.
In this chapter, I ruled out a (sociolinguistic) external account and the Obligatory
Contour Principle (with respect to both [labial] and [round] features) as possible
explanations for the alternations. I agree with the fundamental idea of Lipski (1995) and
Mazzaro (2005) that assimilation to the following vowel accounts for the processes
(Piñeros (2009: 283) comes to the same conclusion, yet his purpose is simply descriptive
and not theoretical). The issues now become clarifying the relevant data and formalizing
the observations and generalizations from the data in Optimality Theoretic terms.
Concerning the first goal, Chapter 3 outlines the methodology followed in gathering data,
while Chapter 4 presents the experimental results. To address the second goal, the OT
analysis is presented in Chapter 5, based on Chapter 4’s results.
53
CHAPTER 3
EXPERIMENTAL METHODOLOGY
In Chapter 1, I introduced the basic facts of /f/ bilabialization, /f/ velarization and
/x/ palatalization in Latin American Spanish. The cursory nature of this overview was a
result of the available literature on these processes specific to Spanish. Independent of the
ubiquity of phonological assimilation in general, the data for the three properties under
investigation are found primarily in outdated sociolinguistic corpora (Perissinotto 1975,
Resnick 1975, Sanicky 1988, Calvo Shadid 1996) for which the research questions and
experimental variables (e.g., socioeconomic status, education and age) are not necessarily
relevant here. To that end, the detailed research questions that follow (repeated from
Chapter 1) inform the descriptive/empirical (1 and 2), theoretical (3 through 7) and
sociolinguistic (8 and 9) components of this dissertation.
1. Are the fricative phonemes /f/ and /x/ synchronically involved in allophonic
alternations in Latin American Spanish? If so, what are the allophones?
2. Are the phonological processes productive (i.e., not merely lexicalized)?
3. Can the processes be conceived as assimilations conditioned by the
surrounding vowel(s)? If so, are the assimilations regressive or progressive?
4. Are the assimilations conditioned by stress or position in the word?
5. How can a formal model of phonology–Optimality Theory–account for the
phenomena?
6. How can this theoretical analysis limit the assimilations to the subset of
fricatives {/f/, /x/} to the exclusion of all other phonemes in Spanish?
54
7. What role, if any, do articulatory and acoustic phonetics play in the theoretical
model proposed?
8. What crosslinguistic typological implications does 7 entail?; i.e., why do
certain linguistic systems attest certain properties while other systems do not?
9. How can these questions be viewed from a diachronic perspective of Spanish?
To pursue answers to the above questions, data that go beyond what is currently available
is required. The primary sources (Perissinotto 1975, et seq.; see Piñeros 2009 for brief
descriptions of all three processes) on which Lipski (1995) and Mazzaro (2005) rely date
from the 1970s and 1980s and, consequently, fail to conform to modern linguistic
descriptive standards–specifically, detailed phonetic analyses. Furthermore, the
methodology outlined in this chapter will consider other relevant phonological factors–in
addition to the quality of the adjacent vowel(s)–that have not been addressed previously
at a systematic level concerning these alternations: i.e., word frequency (Bybee 2010) and
the influence of (primary) prosodic stress and position within the word (Beckman 1998).
However, perhaps more important than those precise contextual details is the fact that
sources disagree regarding the nature of the output allophones themselves, specifically
with /f/ alternations. As mentioned in the previous chapter, Sanicky (1988) alone lists
eight possibilities (see Table 2.7), ignoring the instances in which she differs from
Perissinotto (1975) and Resnick (1975) in her descriptions. Consequently, the goal of the
first research question is to better describe the segments involved in the alternations,
followed by determining the additional factors that govern the processes.
The above criticisms demonstrate that the current descriptive data are lacking; the
purpose of Chapters 3 and 4 is to fill this empirical gap in order to propose a better
55
informed theoretical analysis in Chapter 5. In section 3.1, I discuss the demographics and
background information of the participants. Sections 3.2 through 3.5 describe and provide
sample tokens from the four principal tasks of the experiment: a sociolinguistic interview,
an AX perception task, a sentence reading task and a picture description task (also
referred to as tasks 1, 2, 3 and 4, respectively). Section 3.6 summarizes the chapter.
3.1. Participants
A total of 37 native speakers of various Spanish dialects participated in at least
one of the four primary components of the study. Eight of these participants took part in
at least three of the four phases; these participants received the equivalent of US$20 as
compensation. Seven participants who were not compensated took part in a small scale
verification study designed to analyze the perceived nativeness of the CV tokens used in
the main perception task. No participant self-reported any hearing or specific language
impairments. For the sociolinguistic interview, sentence reading and picture description
tasks, I recruited four Chilean (two male, two female) and four Mexican (two male, two
female) speakers.
Before continuing, a brief comment on why Chile and Mexico were selected for
fieldwork is in order. Recall from Table 2.7 that Chile, Mexico and Argentina exhibit all
three phonological processes in question (Resnick 1975). Furthermore, /x/ palatalization
is robustly pervasive in Chile, attested in all speakers at rates nearing one hundred
percent (Resnick 1975). To be able to draw larger conclusions, however, I sought to
supplement the Chilean data with one additional dialect. Argentina would have been the
logical choice in order to control as many variables as possible, as Chile and Argentina
are relatively socioculturally and linguistically similar. However, within the set of
56
Spanish fricatives, Argentina variably attests the palatoalveolar sibilants /ʃ/ and /ʒ/
(Resnick 1975, Canfield 1981, Lipski 1994), while Chile does not. Given that this
dissertation specifically addresses fricative phenomena, comparability in that regard was
crucial. I instead selected Mexico, which shares with Chile the more restricted fricative
inventory /f, s, x/.
Regarding group membership in the three production tasks (1, 3 and 4), I divided
the aforementioned Chilean and Mexican participants post hoc into four groups based on
the properties (/f/ bilabialization, /f/ velarization and /x/ palatalization) that they exhibited
after phonetic analysis. The Chilean and Mexican experimental groups attested at least
one of the /f/ properties in question in addition to /x/ palatalization, while the Chilean and
Mexican control groups provided evidence only of /x/ palatalization.
Concerning the two perception tasks, the first is a small scale judgment task (see
§3.3.2.1) designed to verify the perceived nativeness of the tokens used in the main
perception AX discrimination task. The nativeness rating task consists of one group of
seven listeners (three Chileans, one Colombian, one Cuban, two Spaniards) who did not
take part in any of the production tasks or the main AX discrimination task in order to
avoid biases, specifically during the primary perception task. Various dialects were used
in order to globally determine the abstract nativeness of each syllable.
If a significant difference is reached on individual tokens, a dialectal comparison
will be necessary to determine if such is able to account for the discrepancy. That is, even
though the token [çe] might be perceived as perfectly native-like to a Chilean, a Spaniard
or a Colombian who lacks palatalization might conceivably mark it low regarding its
nativeness, this judgment having nothing to do with its quality as a sample.
57
In the AX discrimination task, ten participants from various dialects (two
Colombians, one Puerto Rican, six Spaniard, one Venezuelan) who did not participate in
any other task comprise the control group. The experimental group consists of speakers
who attest any of the three phonological processes (or a combination thereof) under
investigation: a group of six palatalizing Chileans (who also took part in no other task),
the four Mexican production participants and six additional Mexican listeners (four
females, two males) who likewise took part in the perception task alone.
In the remaining subsections of 3.1, I provide detailed background information for
the participants and groups discussed above, after which I explain the procedures, tokens
and justifications for each task in sections 3.2 through 3.5.
3.1.1. Chilean production participants
The four native speakers of Chilean Spanish (two male, two female) who
participated in the three production tasks were recorded at the Linguistics Laboratory at
the University of La Frontera in Temuco, Chile, from 17 July through 1 August 2012. I
provide their biographical information including participant number, gender, age, highest
level of education completed1 and occupation in Table 3.1.2 Post hoc group data
organized according to property attested are available in Chapter 4.
1
Education indicates the highest level of education attained (or in the case of the students, the
major in which they were enrolled at time of testing). Chilean primary education (educación
básica) consists of eight grades spanning ages six through 13. Secondary education (educación
media) is divided into four grades, ages 14 to 17 (“technical” refers to one of the four possible
career or educational paths that students elect during their final two years of secondary
education). As of 1965, all twelve primary and secondary grades have been compulsory; prior to
1965, only primary education was compulsory. Post-secondary university education is optional.
2
To maintain anonymity, no names or other identifying information (addresses, phone or
governmental numbers, etc.) were collected. Participants were instructed to select a random five
digit number at the beginning of the testing session for coding, task matching and analytical
purposes only. In Table 3.1 and all subsequent tables concerning demographics, participants are
listed in chronological, not numerical, order of testing.
58
Table 3.1. Individual biographical information for Chilean production group
Group
Experimental
Control
Participant
Gender
Age
Education
Occupation
19727
M
40
8th primary
janitor
17137
M
34
4th secondary
janitor
15246
F
28
pedagogy
student
02838
F
40
technical
secretary
Note: The participants are grouped by occupation for clarity in presentation only. Mean
and standard deviation for age are 35.5 and 5.74, respectively.
In addition to the above, participants were asked if they spoke or had any contact with
languages other than Spanish, and if so, what type of contact and to what extent. Given
the presence of indigenous languages (see Table 2.8), it was necessary to determine that
Spanish was their first and primary language, which turned out to be true for all
participants; it was also the case that no participant reported speaking any language other
than Spanish. Consequently, no Chilean participant was excluded from analysis. The
following subsection describes in the same manner as this subsection the separate group
of Chileans who took part in the perception task only.
3.1.2. Chilean perception participants
Both the Chilean and Mexican groups are further subdivided into production and
perception participants. The four Chilean participants discussed in section 3.1.1 only
participated in the three production tasks: the sociolinguistic interview and the sentence
reading and picture description tasks. A separate group of Chileans to be detailed in the
current subsection took part solely in the AX discrimination task. This differs from the
group of the four Mexicans who participated in the interview, the picture description task
and the AX discrimination perception task, but not the sentence reading task.
59
This asymmetry is due to a combination of time constraints and the relative
literacy of the two participant sets. While my Chilean participants were literate and thus
able to complete the sentence reading task, my Mexican participants reported difficulty
with reading and writing. In order to keep the time required to take the experiments
uniform between the two groups, I replaced the sentence reading task with the perception
task for my Mexican participants and compensated by recruiting a separate group of
Chileans to complete the perception component. Table 3.2 below provides a group-bytask summary.
Table 3.2. Group to task correspondences, Chilean and Mexican participants
Interview
Chile1
Mexico2
Reading
Picture
Chile2
Mexico1
Perception
A second group of six native speakers of Chilean Spanish (all females, referred to as
Chile2 above) serves as the experimental group for the AX discrimination task. This test
was administered via Microsoft PowerPoint version 2010 on 8 March 2013. Participants
were tested online using their own personal computers and were instructed to use
headphones.
Although the Chilean perception participants did not take part in any production
task (to empirically corroborate that they do, in fact, palatalize), as Resnick (1975) and
many others note, /x/ palatalization is attested in virtually one hundred percent of Chilean
speakers and is as close to diagnostic as possible. For this reason using two separate
60
groups of Chilean participants presents no confounding issue, as the perception group
assuredly palatalizes, as well. In Table 3.3, I provide their biographical information.
Table 3.3. Individual biographical information for Chilean perception group
Participant
Gender
Age
Education
Occupation
00007
F
60
Master’s
professor
00008
F
33
bachelor’s
architect
84210
F
32
bachelor’s
teacher
33469
F
20
bachelor’s
student
00070
F
30
bachelor’s
X-ray technician
08692
F
20
bachelor’s
model
Note: Mean and standard deviation for age are 32.5 and 14.67, respectively.
Three participants (84210, 33469 and 08692) claimed knowledge and meaningful
exposure to no language other than Spanish; the remaining three listed exposure only to
English, the details of which appear in Table 3.4.
Table 3.4. Chilean listeners’ exposure to English (in years)
Participant
Age
Age of Onset
Length
00007
60
12
48
00008
33
5
12
00070
30
5
25
Mean
41
7.33
28.33
SD
16.52
4.04
18.23
In the case of participant 00007, age of onset is relatively late while the length of
exposure is prolonged (this participant was tested in the United States). For participant
61
00008 the facts are reversed with early, but brief, exposure. In the case of participant
00070, exposure is both early and prolonged, yet all three participants list Spanish as their
first and primary language. Furthermore, 00007 and 00070 reside in Chile; consequently,
no participant was excluded. In the next subsection, I detail the Mexican production
participants.
3.1.3. Mexican production participants
A total of four speakers of Mexican Spanish originally took part in the interview
and picture description tasks, in addition to the perception task (which was supplemented
by another six Mexican participants (§3.1.4)). Participants 26911 and 35792 were tested
between 24 and 31 October 2012 in the security room at a marketplace in Cholula; the
remaining two were tested in Coralville, Iowa, on 17 August 2013. Table 3.5 contains the
results of the participants’ biographical questionnaire.3
Table 3.5. Individual biographical information for the Mexican production group
Group
Experimental
Control
Participant
Gender
Age
Education
Occupation
26911
M
35
3rd primary
artisan
35792
F
38
6th primary
performer
11111
F
54
high school
homemaker
71017
M
31
Master’s
sports coach
Note: Mean and standard deviation for age are 39.5 and 10.08, respectively.
3
The Mexican educational system is divided into three levels: primary (educación primaria,
grades one through six, corresponding to ages six through 12), secondary, the U.S. equivalent of
middle school (educación secundaria, grades seven through nine, ages 12 through 15), and the
equivalent of high school (bachillerato/educación preparatoria, grades ten through twelve, ages
15 through 18).
62
Two participants list familiarity with Nahuatl. As was the case for the Chilean
participants, age of onset and length of exposure are the criteria I considered for possible
exclusion, although neither factor reached such a level for any participant to warrant
removal from the study.
Table 3.6. Mexican production participants’ exposure to Nahuatl (in years)
Participant
Age
Age of Onset
Length
26911
35
10
25
35792
38
0
38
Mean
37.75
3.75
34
SD
5.25
4.79
6.68
The participant for whom age of onset is indicated as 0 reported that she is a
simultaneous bilingual. For participant 26911, age of onset for Nahuatl is also relatively
low; additionally, participant 71017 listed 25 years of exposure to English beginning at
age three. However, all three claim Spanish as their primary language. For these reasons,
no participant was eliminated based on linguistic background.
3.1.4. Mexican perception participants
For the AX discrimination task, six Mexican participants (four female, two male)
supplemented the four production participants for a total of ten listeners in the Mexican
perception group (these six listeners participated in no other task from the experiment).
All perception participants were tested between 24 and 31 October 2012 with the
exception of 98765, who took part in the study on 4 September 2013 at the Second
Language Acquisition lab in Iowa City, Iowa. The childcare worker and housekeeper
completed the task in the personal home office of a colleague originally from Puebla. The
63
secretary participated in the library at her Puebla elementary school. The tortilla vendor
was interviewed in the backroom of his tortilla shop and, similarly, the produce vendor
participated from her roadside tent, both in Puebla. Table 3.7 summarizes their
biographical information, excluding the four production participants from Table 3.5.
Table 3.7. Individual biographical information for the Mexican perception group
Participant
Gender
Age
Education
Occupation
53209
F
29
6th primary
childcare
00004
F
57
2nd primary
housekeeper
18743
F
45
biology
secretary
00005
M
58
6th primary
tortilla vendor
00006
F
48
none
produce vendor
98765
M
40
PhD
professor
Note: Mean and standard deviation for age are 46.17 and 10.9, respectively.
Two speakers disclosed knowledge of indigenous languages acquired as simultaneous
child bilinguals: Zapotec (participant 53209) and Mixtec (00004). Participant 98765
listed English as an adult second language with formal study beginning at age 28. In his
case, age of onset was well past the critical period. For the simultaneous bilinguals, both
claimed to be Spanish-dominate. Like all other groups, no participants were excluded.
3.1.5. Control groups
The final two groups of participants serve as controls in the nativeness rating and
AX discrimination tasks. In the case of the former, the task consists of this single group,
and any significant differences with individual tokens will be addressed independently.
Concerning the latter, the control group will be compared to the two perception groups
64
detailed above: the Chilean and Mexican perception groups. Both control groups are
composed of speakers from disparate Spanish dialects for the following reasons.
The goal of the nativeness rating task is to determine how native-like the tokens
produced by a native-English speaking phonetician are perceived to be. To do so it was
necessary to collect judgments from a variety of dialects in order to ensure that uniformly
low ratings on particular tokens are indeed indicative of an issue with the sample, and not
due to any given speaker’s unfamiliarity with allophones possible in other dialects. The
following section details the group of listeners who participated in this preliminary task.
3.1.5.1. Nativeness rating participants
Seven native speakers of Spanish (three males, four females) from differing
dialects comprise the group for the preliminary nativeness rating task. Listeners
participated via Microsoft PowerPoint version 2010 using their own personal computers
and headphones from 25 February through 2 March 2013. Table 3.8 contains relevant
biographical information.
Table 3.8. Individual biographical information for the nativeness rating group
Participant
Dialect
Gender
Age
Education
Occupation
00009
Chile
M
37
doctorate
professor
00010
Chile
F
36
doctorate
professor
00011
Chile
F
55
university
medical tech.
55532
Colombia
F
26
Master’s
student
66603
Cuba
M
66
doctorate
professor
12883
Spain
M
29
Master’s
student
66985
Spain
F
27
Master’s
childcare
Note: The participants are grouped by dialect for clarity in presentation only. Mean and
standard deviation for age are 39.43 and 15.33, respectively.
65
As is the case with the previous groups discussed, knowledge and familiarity with
languages other Spanish must be examined in order to determine if either are sufficiently
high to warrant exclusion. Participants listed contact with the following languages other
than Spanish.
Table 3.9. Nativeness controls’ exposure to English (in years)
Participant
Age
Age of Onset
Length
00009
37
20
17
00010
36
6
30
55532
26
15
11
66603
66
25
41
12883
29
8
21
66985
27
6
8
Mean
36.83
13.33
21.33
SD
15.01
7.99
12.37
Table 3.10. Nativeness controls’ exposure to French (in years)
Participant
Age
Age of Onset
Length
55532
26
16
3
66985
27
15
1
Mean
26.5
15.5
2
SD
0.71
0.71
1.41
Even though overall contact with English is extensive for this group, the preliminary task
was merely for token verification. Consequently, any potential influence is not as great a
concern had the task been part of the main experimental protocol. Furthermore, with the
exception of the two Spaniards (who reside in Spain), all participants were tested in the
66
United States; the high English influence is thus not unexpected given this fact.4 (The
French influence, with high age of onset and low years of exposure, is negligible.) The
following subsection explores the final group of participants.
3.1.5.2. Task 2: AX discrimination control participants
The last group of participants consists of the control group for task 2, the AX
discrimination task. Ten native speakers of Spanish (six males, four females) from
various dialects listened to the recordings via Microsoft PowerPoint version 2010 using
their own personal computers and headphones from 28 January through 28 February
2013. I provide their biographical information in Table 3.11.
Table 3.11. Individual biographical information for the AX discrimination controls
Participant
Dialect
Gender
Age
Education
Occupation
46969
Colombia
M
27
Master’s
student
05187
Colombia
M
35
Master’s
student
32456
Puerto Rico
F
33
Master’s
student
25310
Spain
M
31
Master’s
student
15300
Spain
F
38
Master’s
student
24538
Spain
M
27
Master’s
student
99976
Spain
F
31
Master’s
student
91823
Spain
M
35
Master’s
student
75931
Spain
F
27
Master’s
student
28043
Venezuela
M
50
Master’s
student
Note: The participants are grouped by dialect for clarity in presentation only. Mean and
standard deviation for age are 33.4 and 6.96, respectively.
4
Participant 66603 reports that he is a simultaneous Ladino-Spanish bilingual, in addition to
having studied Portuguese, French, Italian, American Sign Language, German, Russian, Latin
and Greek. He does not list age of onset or years of exposure for languages other than Ladino,
Spanish and English.
67
Given Table 3.11, two differences–education and occupation–between the perception
control group and the other experimental groups are apparent. While the control group for
the AX discrimination task is highly educated, its members are being compared to
speakers of disproportionate socioeconomic and educational status. Ideally these factors
would be equivalent or at least similar, and thus controlled. However, given the nature of
the three allophonic properties–which, according to previously cited sources (Perissinotto
1975, et seq.) are most prevalent among the lower classes–these differences are
necessary, as the goal of the task is to compare perception of the allophones between two
groups of speakers: those who use the allophones and those who do not. For this reason
the asymmetry is a necessary (possible) confound.
As is the case for all other groups, exposure to languages other than Spanish
represents another possible confounding variable. Those languages for the perception
control group are listed in the following tables.5
Table 3.12. Discrimination controls’ exposure to French (in years)
5
Participant
Age
Age of Onset
Length
46969
27
22
3
05187
35
23
12
25310
31
12
6
15300
38
16
2
24538
27
10
8
99976
31
15
3
91823
35
9
6
Mean
32
15.29
5.71
SD
4.2
5.53
3.5
One participant, 75931, is a simultaneous Basque-Spanish bilingual.
68
Table 3.13. Discrimination controls’ exposure to English (in years)
Participant
Age
Age of Onset
Length
46969
27
16
11
05187
35
12
13
32456
33
10
23
25310
31
5
26
15300
38
9
29
24538
27
12
15
99976
31
5
26
91823
35
14
17
75931
27
7
20
28043
50
20
30
Mean
33.4
11
21
SD
6.96
4.83
6.8
Table 3.14. Discrimination controls’ exposure to German (in years)
Participant
Age
Age of Onset
Length
25310
31
23
4
99976
31
18
3
75931
27
11
4
Mean
29.67
17.33
3.67
SD
2.31
6.03
0.58
Table 3.15. Discrimination controls’ exposure to Italian (in years)
Participant
Age
Age of Onset
Length
46969
27
22
1
75931
27
18
2
Mean
27
20
1.5
SD
0
2.83
0.71
69
Table 3.16. Discrimination controls’ exposure to Portuguese (in years)
Participant
Age
Age of Onset
Length
05187
35
32
3
32456
33
31
1
28043
50
40
10
Mean
39.33
34.33
4.67
SD
9.29
4.93
4.73
Although exposure to and knowledge of languages other than Spanish are extensive in
this group, such is not unexpected given their education and the fact that, with the
exception of 75931 (who took the test in Spain), all were tested in the United States.
However, it is the case that, with few exceptions, the additional languages were acquired
relatively late with low lengths of exposure (the obvious exception being English), and
for all participants, Spanish was listed as their primary (and first) language. For these
reasons no control participant was excluded from the analysis.
Table 3.17 condenses the current subsection and expands on Table 3.2 by
providing a complete list of which tasks were performed by which groups.
Table 3.17. Group to task correspondences, all participants
N
Interview
Chile1
4
Chile2
6
Mexico1
4
Mexico2
6
Control1
7
Control2
10
Total
37
Perception
Sentences
Pictures
Native
8
26
4
8
7
70
Having thoroughly detailed the participants and group membership in section 3.1, the
goals of sections 3.2 through 3.5 are to (i) document the protocol for each task of the
experiment, (ii) justify the task by relating it to the research questions posed at the
beginning of this chapter and (iii) provide sample tokens and fillers.
3.2. Task 1: sociolinguistic interview
There are four main tasks to the empirical component of this dissertation:
sociolinguistic interview (task 1), AX discrimination (task 2), sentence reading (task 3)
and picture description (task 4). Task order was randomized for each participant; I use
task 1 through 4 for ease of reference only.
The testing sessions in Chile and Mexico, requiring two consecutive hours per
participant, were conducted entirely in Spanish. Task 1, the sociolinguistic interview, is
an informal dialogue between researcher and participant with the goal of eliciting the
target sounds in spontaneous and natural speech in order to respond to research question
1: determining if the phonological processes remain productive and, if so, what are the
output allophones.
The first part of the interview consisted of a written questionnaire (see Appendix
A for both Spanish and English versions of the document) in which participants answered
questions on biographical information (age, gender, city of birth, current city of
residence, education and occupation) and linguistic background (if they know, are
familiar with or have studied languages other than Spanish, if they have or have had
contact with languages other than Spanish and if so, to what degree is/was the contact).
Upon completing the questionnaire, the participants were asked various openended discussion questions in order to elicit the target sounds in a conversational register.
71
While the exact wording of the ten questions can be found in Appendix A, the general
topics of the questions included childhood memories and stories, questions about family
and friends, familial and cultural traditions, leisure activities, food and the regional
climate. The participants were instructed that, if uncomfortable (or for whatever reason),
they could elect not to answer any question. Two questions in particular, Can you tell me
a funny story from your childhood? and Can you tell me an embarrassing story from your
childhood?, caused the most difficulty, as the majority of participants claimed that they
simply could not remember any.
Chilean production participants (Chile1) were recorded at the Linguistics
Laboratory at the University of La Frontera in Temuco, Chile, from 17 July through 1
August 2012 using a Sony DCR-SR47 digital video camera recording in high definition.
Audio responses were recorded on a Fostex FR-2LE 24 bit 48 Hz recorder using an
Audix HT5 small condenser headset microphone powered by an Audix APS-911
phantom power supply. The audio files were transferred to an HP Mini 210 Netbook and
analyzed using Praat phonetic analysis software version 5.3.41 (Boersma and Weenink
2013). The visual files were transferred in the same manner, and screen captures were
created via the print screen key while viewing the videos in Windows Media Player
version 12.0. The same equipment and analytical process apply to the Chilean production
group for tasks 3 (sentence reading) and 4 (picture description).
The Mexican production participants were tested at five locations in Cholula and
Puebla, Mexico, between 24 and 31 October 2012. Unlike Chile, no video recording was
available as the experiment was not conducted in a laboratory but rather throughout
various sites in the aforementioned cities: the security room at a marketplace, the
72
roadside tent of a produce vendor, etc. (see §3.1.3 for further details). All tasks requiring
audio recording (the interview and picture description tasks) were conducted with a Shure
Beta 58A microphone directly connected to an HP Mini 210 Netbook via an HDE XLR
to USB converter cable. The computer was running Audacity audio editing and recording
software, version 1.3. Phonetic analysis was carried out in Praat.
The following subsection details the methodology followed for task 2, the
perception task, and includes details concerning the nativeness rating task, as well.
3.3. Task 2: perception
Task 2 is an AX discrimination task in which participants determine on a scale of
1 to 5 if two CV syllables are the same (1) or different (5). Four types of CV syllables
appear. There are two combinations for which the expected answer is ‘different’: AB and
BA; there are also two ‘same’ conditions: AA and BB. This task specifically targets
research questions 6 and 7: how can the analysis within Optimality Theory limit
assimilation to just the fricatives /f/ and /x/, and how can acoustic phonetics be
incorporated into the analysis.
Recall that, in Chapter 2, I summarized proposals by Lipski (1995) and Mazzaro
(2005) that claim that the processes are motivated by Obligatory Contour Principle
restrictions that rule out adjacent [labial] segments. I argued that such proposals
overgeneralize and consequently predict unattested phonological repair in forms such as
poro ‘pore,’ burla ‘mockery’ and música ‘music’ with non-fricative [labial]-[labial]
sequences. The issue that such data pose concerns how the assimilations might be limited
to just the labiodental /f/ and velar /x/ (non-sibilant) phonemes.
73
In order to test the prediction from Chapter 1 that perception in the form of the
unavailability and relative weakness of acoustic cues to fricative place (i.e., the lack of
invariance in the fricative signal, the lack of invariance in fricative production and the
high reliance on the vowel due to fricative-vowel coarticulation) plays a role in
explaining why only /f/ and /x/ participate in assimilation, task 2 is an AX discrimination
task that gauges the perceived similarity between pairs of sounds. The weakness of place
cues in the non-sibilant fricatives compared to the strength of cues to place in the
remaining sounds of Spanish (stops, affricates, sibilants, nasals and liquids; see the
literature and discussion in Chapters 1, 4 and 5) predict perceived similarity/place
confusion in the cases of [f], [x] and their respective allophonic alternations [ɸ] and [ç].
Let us take as representative the pairs [fu] [ɸu], [xi] [çi] and [pu] [ku]. On the assumption
that non-sibilants {[f], [x], [ɸ], [ç]} contain weak cues to place in light of the reasons
above, while stops {[p], [k]} attest stronger cues, misperception (perceived similarity) is
expected in the former set and unexpected in the latter. Such would suggest a perceptual
motivation in limiting assimilation to /f/ and /x/ in response to research questions 7 and 8.
The 70 unique syllables used in the AX discrimination task were recorded by a
native English speaking, 40-year-old male originally from Pittsburgh, Pennsylvania. He
has been Professor of Linguistics at the University of La Frontera in Temuco, Chile, for
two years. I required a trained phonetician familiar with the dialects in question in order
to articulate the crosslinguistically rare allophones (e.g., [ɸi], [ça]) necessary for the task.
He has traveled, lived, researched and taught in Chile for the past twenty years, in
addition to being married to a Chilean for the previous five. So although he is not a native
speaker of Spanish, the combination of his personal background and professional training
74
more than qualifies him to produce the syllables used in the perception task. Yet even
given the above profile, the purpose of the nativeness rating task (§3.3.2.1) is to verify
the perceived accuracy of these syllables.
The tokens for the discrimination task were recorded at the Linguistics Laboratory
at the University of La Frontera in Temuco, Chile, on 10 June 2012 using a Fostex FR2LE 24 bit 48 Hz recorder and an Audix HT5 small condenser headset microphone
powered by an Audix APS-911 phantom power supply, the same equipment used to
record the Chilean production participants. I used Audacity (version 1.3) audio editing
and recording software to isolate each CV syllable, which I then combined into the test
stimuli following the format [CV CV]. Sixty-five ‘different’ pairs serve as target stimuli,
15 pairs of approximants were filler while 20 ‘same’ pairs were added as a qualifying
measurement to ensure that participants do not absentmindedly mark ‘different’ (no
participant was disqualified), resulting in 100 total stimuli.6 A short 500 millisecond (ms)
interstimulus interval (ISI) was used between each pair in order to target abstract
representations, and the order for each ‘different’ pair (AB or BA) was randomized
among the participants.
The pairs were presented in a PowerPoint presentation on an HP Mini 210
Netbook running Microsoft Office version 2010. Each page of the presentation contained
one embedded [CV CV] pair that played automatically after each page advanced.
Participants were instructed to press either the ‘enter’ or ‘space bar’ keys to continue to
the next page.
6
While this is admittedly a low ratio of ‘different’ to ‘same’ tokens, this was simply one of a
four-part study in which practicality constituted an overriding factor.
75
Appendix B provides the instructions as they appeared to the participants in
addition to the English translation. Participants were asked to rate how they perceived
each pair given the following five point Likert scale.
(1)
1
2
3
4
5
‘identical’
‘similar’
‘you are unsure’
‘a little different’
‘completely different’
iguales
semejantes
Ud. no está seguro/a
un poco diferentes
completamente diferentes
Participants in Mexico listened to the task using Sennheiser PXC 450 NoiseGard
headphones, and stated aloud their answers while I documented them. I followed this
process due to the fact that the first two participants (whose perception answers I
eliminated) had difficulty registering their answers on paper. The control participants
kept track of their own answers on a separate Microsoft Word document.
In the next section I describe the selection and organization of the tokens and
fillers used in this task.
3.3.1. Justification for tokens and conditions
Table 3.18 lists the conditions involved in each pair.
Table 3.18. AX discrimination task consonant pairs
Target
Stop
Nasal
Fricative
Approximant
Same
[f] [x]
[p] [k]
[m] [n]
[f] [ç]
[] [ð]
[p] [p]
[f] [ɸ]
[b] [ɡ]
[x] [ɸ]
[] [ɣ]
[b] [b]
[ɸ] [ç]
[ð] [ɣ]
[k] [k]
[x] [ç]
[f] [s]
[x] [s]
[ɸ] [s]
[ç] [s]
[ɡ] [ɡ]
76
The conditions are divided by manner of articulation. The first condition is the target
condition, and compares the allophones involved in the three fricative assimilations: /f/
velarization, /f/ bilabialization and /x/ palatalization. To determine the relative similarity
among the other sounds of Spanish in response to the criticism of Lipski’s and Mazarro’s
analyses, approximants, stops and nasals are also considered. The filler conditions
permutated the fricative allophones from the first condition, in addition to the alveolar
sibilant [s], and included the voiced spirantized fricatives/approximant series. I also
included four ‘same’ conditions in which the voiced and voiceless bilabial and velar stops
are repeated within the same pair (the AA and BB tokens). For inclusion in the study,
participants had to achieve at least 80% accuracy on the ‘same’ tokens (at least 16/20
‘same’ tokens rated 1 or 2) to ensure that they were responding based on the stimuli and
not uniformly responding 5 ‘different.’ Chilean listener 33469 responded 2 to a single
token, [ɡu] [ɡu]; otherwise, all participants rated every ‘same’ token 1. Consequently, no
perception participant was disqualified.
Three manners of articulation attested in Spanish are missing from the above:
affricate, lateral and rhotic. As there is only one phonemic affricate in Spanish, the
palatoalveolar [t͡ ʃ], there is no other place of articulation with which it could form a pair.
The same holds true for the alveolar lateral [l]7 and rhotics [ɾ, r]. I therefore excluded
them from the task.
Every consonant pair was repeated five times, once for each canonical vowel of
Spanish: [i, e, a, u, o]. Thus, there are five target [f] [x] token pairs, for example: [fi] [xi],
7
The palatal lateral /ʎ/ (e.g., llamar [ʎamaɾ] ‘to call’) is phonemic in an increasingly limited
number of dialects in northern Spain, Paraguay and the Andes (Peru, Bolivia and Ecuador)
(Resnick 1975, Canfield 1981, Lipski 1994, Hualde 2005, Piñeros 2009, inter alia). Given its
reduced appearance in the Hispanophone world and absence in the specific dialects in question, I
do not consider it here.
77
[fe] [xe], and so on. The 100 total tokens are the result of the sixteen ‘different’ pairs in
addition to the four ‘same’ filler pairs repeated five times each. Order was randomized
such that the same stimuli–[fi] [xi]–for example, appeared as [xi] [fi] for any given
participant.
Furthermore, the 100 tokens were randomized such that no two consecutive
tokens shared either consonants or vowels in any of the four syllables. That is, [fi] [xi]
followed by [sa] [fa] was disallowed given that one syllable in both pairs contains [f].
[bo] [ɡo] followed by [no] [mo] was similarly disallowed given the shared vowel [o].
In the preceding section I have explained the procedure for the AX perceptual
discrimination task (task 2). The following subsections analyze the phonetic details of the
token syllables used in the perception task and describe the nativeness rating task
designed to test the reliability of the samples.
3.3.2. Token analysis
Given that the producer of the 70 syllables used in the perception task is a nonnative speaker of Spanish, I wanted to ensure that any judgments derive only from the
perceived difference between the two consonants in each pair, and not from any other
factor, including fricative and vowel length (analyzed in §§3.3.2.3 and 3.3.2.5), vowel
quality (§3.3.2.4) and, specifically, any general unfamiliarity or discomfort with the
syllables themselves if they are perceived as foreign. The goal of the nativeness rating
task (§3.3.2.1) is to control for the lattermost variable by asking a separate group of seven
Spanish speakers to rate how native-like they perceive the individual tokens on a scale of
1 ‘native-like’ to 5 ‘not native-like’ (i.e., completely foreign sounding). The results are
presented below.
78
3.3.2.1. Native rating
In the nativeness rating task, participants heard the 70 CV syllables used in the
AX discrimination task in isolated contexts as embedded sound files in a Microsoft
PowerPoint (version 2010) presentation. The tokens consisted of five repetitions (once
per vowel) paired with the 14 consonants necessary for the main perception task
described above divided into five conditions: approximants (A, [], [ð], [ɣ]), fricatives (F,
[f], [x], [s]), fricative allophones (FA, [ɸ], [ç]), nasals (N, [m], [n]) and stops (S, [p], [k],
[b], [ɡ]).
Like the discrimination task, the syllables were randomized such that no
consecutive tokens shared either the consonant (e.g., [fi] did not follow [fa]) or the vowel
([no] did not follow [mo]). Participants controlled the progress of each page, yet the
sound files played automatically following advancement.
Appendix B contains the original instructions in Spanish and their English
translation. Given the Likert scale in (2) (different from that of the main perception task
from (1)), participants were asked to rate the degree to which they considered each
syllable native-like.
(2)
1
2
3
4
5
nativo al español
un poco nativo
Ud. no está seguro/a
un poco extraño para el español
no existe en español
‘native to Spanish’
‘somewhat native’
‘you are unsure’
‘a little strange for Spanish’
‘nonexistent in Spanish’
79
Figure 3.1. Mean nativeness rating by condition
The average rating per condition (approximant A (n=105, SD=1.52), fricative F (n=105,
SD=1.24), fricative allophone FA (n=70, SD=1.38), nasal N (n=70, SD=1.05) and stop S
(n=140, SD=1.00)) based on the scale in (2) is presented in Figure 3.1 above. A repeated
measures ANOVA revealed a significant overall effect (assuming standard significance
p<0.05, which will be assumed throughout this dissertation) for condition (F(4,
24)=7.427; p=0.000). Pair-wise post hoc comparisons with Bonferroni adjustment
showed that approximants differed significantly from fricatives (p=0.027) and stops
(p=0.012); no other comparisons between conditions were significant.
Although it is traditionally accepted that parametric analyses such as ANOVAs
require that certain assumptions be met (e.g., normal distribution, interval
measurements), Norman (2010) argues that ordinal data with smaller sample sizes are
still possible with parametric statistics. Concerning the data above, using an ANOVA
even with such a small sample size (a possible violation of normal distribution) and
Likert scale data (ordinal as opposed to interval) yielded the same results as a non-
80
parametric Friedman test: an overall significant effect (χ2(4)=16.01; p=0.003) with
approximants differing significantly from fricatives (p=0.01) and stops (p=0.018) in
Bonferroni adjusted post hoc pair-wise comparisons.
Consequently, given that Norman (2010: 628) argues that, even with ordinal
Likert data and groups as small as five, parametric tests remain accurate, I will present
parametric analyses in this dissertation (principally, Chapter 4).
That the approximants and fricative allophones were rated least native-like is
unsurprising. Spanish phonotactics relegates the approximants [], [ð] and [ɣ] to postcontinuant position; it is thus highly unexpected to hear a continuant in phrase-initial
position, explaining the participants’ reluctance to rate them native-like, which is not a
comment on the segments themselves, but rather on the positional expectation.
It is worth noting that, although significantly different from two other categories
(fricatives and stops) and rated the least native-like of the five, the approximant category
average did not exceed 3. In fact, individually speaking, there were only four syllables
whose average was above that threshold, which suggests that the vast majority of the
syllables are indeed native-like (at least perceived as such by these participants).
Table 3.19. Individual syllable mean nativeness ratings greater than 3 (‘unsure’)
Syllable
Mean rating (n=7)
SD
[ɣi]
4.14
1.21
[ɣu]
3.86
1.46
[ɣo]
3.71
1.25
[çu]
3.71
1.5
81
The three highest syllables (i.e., least native-like) all contain velar approximants. The
only other syllable above 3 includes the palatal fricative preceding a back vowel [çu],
which does not conform to expectations of palatalization in the environment of front
vowels.
Additionally, the fricative allophones [ɸ] and [ç]–although not significantly
different from the fricatives, nasals or stops–were the other condition to exceed 2 on the
rating scale. Except in the cases of [çi] and [çe] for the three Chilean speakers, these
seven participants do not produce these segments and probably hear them rather
infrequently given that they are most commonly attested in lower class dialects, while the
participants come from educated backgrounds. However, it is also possible that the
participants categorized [ɸ] as [f] and [ç] as [x] due to their acoustic similarity (Mielke
2012), thus explaining their relatively low ratings.
Therefore, it was necessary not just to determine if the syllables were perceived as
native-like, but to verify that the stimuli were indeed the segments that I claimed them to
be. For this reason I analyzed via Praat the spectral properties of the 25 fricatives used as
tokens in the perception task (five iterations–once per vowel–of [f], [ɸ], [x], [ç] and [s]).
3.3.2.2. Fricative spectral analyses
To verify the identity of the segments used in the perception task, I calculated the
spectral means for the target fricatives in order to compare them with known values for
the corresponding segments. Fricatives are characterized by the turbulent airflow
(frication) created by a narrow passage between articulators in the vocal tract, resulting in
spectra consisting of random aperiodic noise that lacks the canonical and unique formant
82
structures of, e.g., vowels and other sonorants (Ladefoged and Maddieson 1996,
Ladefoged 2001, Johnson 2012).
For this reason, additional methods of analysis are necessary. Jongman, Wayland
and Wong (2000) recognize nine cues for fricative identification (expanded by
McMurray and Jongman (2011) to 24). Of those, the spectral mean was associated with
the least amount of speaker-/context-dependent variance, making it the best predictor for
uniquely identifying fricatives according to place.
Spectral mean is the average frequency (in Hz) as weighted by intensity (dB)
present in a defined spectrum and is commonly referred to as the first spectral moment
(M1), the center of gravity (COG) or the centroid frequency. I argue that such a
measurement is a valid predictor to place as it serves as the principal indicator in
numerous fricative studies (e.g., Padgett and Żygis 2007, Nawrocki 2008, Baker et al.
2011). Additional cues (fricative duration, skewness (third spectral moment), kurtosis
(fourth spectral moment), fricative-vowel transitional measurements, etc.) are unreliable
in uniquely identifying fricative place of articulation, i.e., vary greatly according to the
speaker and/or phonetic context (Ladefoged and Maddieson 1996: 173, McMurray and
Jongman 2011: 227). Consequently, I present no analysis of these cues here (yet see
Chapters 1, 4 and 5 for a detailed discussion on how their lack of predictability factors
into the proposal put forth in this dissertation).
Traditionally, spectral moments such as spectral mean for any given fricative are
computed as averages of at least three spectral slices extracted from initial, middle and
final positions (if not computed over the entire fricative). A fourth transitional window
that straddles the fricative offset and vowel onset is also possible. Figure 3.2 diagrams
83
such divisions on a condensed waveform and spectrogram of the token [fi] as produced
by the native English speaker recruited to record the CV syllables for the nativeness
rating and AX discrimination tasks.
Figure 3.2. CV syllable [fi] with four fricative spectral divisions (1 initial, 2 medial, 3
final, 4 transitional) and fricative and vocalic onsets and offsets
Onset
1
Fricative
2
Offset Onset
3
Vowel
Offset
4
Since my current goal is to identify spectral properties unique to the fricatives
themselves, I will not take transitional measurements at locus 4 (see, however, Chapter 4
for coarticulatory measurements at this window, and Chapter 5 for a discussion on their
relevance to the theoretical analysis). Jongman et al. (2000) found that spectral mean was
most reliable in medial position 2; for these reasons, I calculated M1 values at a 40 ms
window in the middle of frication.
All M1 values were taken from fast Fourier transform (FFT) spectral slices
extracted in Praat phonetic analysis software using a Hamming window to reduce the
effects of clipping at the edges of each sample. I maintained Praat’s standard amplitude
84
weighting of 2 and used the default frequency range of 22.05 kHz. Table 3.20 presents
the average spectral mean (center of gravity) measurements for the 25 fricative tokens
used in the perception tasks; included for reference are the corresponding values from
Jassem’s (1968) comprehensive acoustic analysis of fourteen fricatives from various
languages.
Table 3.20. Average spectral mean (Hz) of token fricatives (n=5 per fricative) in
nativeness rating/AX discrimination tasks
[ɸ]
[f]
[s]
[ç]
[x]
Mean
2898.27
3729.58
5080.87
3369.15
1821.07
SD
493.16
407.31
1167.86
344.17
293.39
Jassem
4100
4200
5200
3500
1600
Before comparing the centers of gravity, notice that, in general, there is an inverse
relationship between energy concentration and place of articulation concerning
fricatives–the further back the point of articulation in the vocal tract (left to right in Table
3.20), the lower the energy concentration (Ladefoged and Maddieson 1996). Stridents or
sibilants (e.g., [s]) represent the exception, identified by relatively high energy
concentrations as there are two sources of frication inherent in their articulation: the locus
of the narrowing at the articulators (dental, alveolar or postalveolar) and a secondary
obstruction as the air passes between the teeth (Ladefoged 2001: 56).
The averages of each spectral mean for the tokens from task 2 align quite
similarly with those reported by Jassem with one exception, the bilabial fricative [ɸ].
However, note that [ɸ] also contains the second highest standard deviation. The
maximum center of gravity for the bilabial fricative is 3551 Hz from the [ɸo] syllable, a
85
number that more closely approaches Jassem. Furthermore, in terms of ordinal
measurements, the highest to lowest average center of gravity is equivalent for both my
tokens and Jassem (with the [ɸ] caveat): [s] > [f] > [ɸ] > [ç] > [x] (which, unsurprisingly,
follows the inverse relationship in terms of frequency and place of articulation discussed
above).
Thus, it seems to be the case that, even though the tokens were produced by a
native English speaker, their spectral properties conform to what is expected of each
fricative. The responses to the perception task, I argue, are therefore reflective of
differences between the pairs, and not due to irregularities in the individual fricatives.
3.3.2.3. Duration
Having verified in the previous subsection the identify of each fricative used in
the perception task, without controlling for additional variables it cannot be argued that
‘different’ responses in the AX discrimination task are necessarily due to the distinct
places of articulation perceived in the fricatives. Two additional factors to which
participants may have responded include differences in length (of both the fricatives and
vowels) and vowel quality. In the current subsection, I analyze the duration of the
fricatives and vowels that comprise the 25 fricative target CV tokens of the perceptual
discrimination task. In the subsection that follows, I measure the formant structure of the
vowels of the same syllables to control for vowel quality.
For each fricative, duration was manually measured in Praat from the onset to the
offset of frication using the corresponding waveform and spectrogram (see Figure 3.2).
Unlike the spectral properties of fricatives, which more reliably indicate place of
articulation, duration is instead correlated with voicing (Jongman et al. 2000; Gordon,
86
Barthmaier and Sands 2002; Silbert and de Jong 2008; McMurray and Jongman 2011),
voiceless fricatives tending to be longer than their voiced counterparts (Ohala 2010). For
present purposes, duration is measured to ensure that answers of ‘different’ during the
AX discrimination task are not based on temporal properties, but rather on perceived
differences in place. This is true for the length of each vowel segment, as well.
Following the measurement protocol described in McMurray and Jongman (2011:
245), vocalic duration was measured from the onset of periodic voicing and dark energy
bands indicative of formant structure to the termination of both (Figure 3.2).
Table 3.21. Fricative duration of target CV tokens (in seconds)
[ɸ]
[f]
[s]
[ç]
[x]
[i]
0.199
0.194
0.155
0.211
0.189
[e]
0.187
0.166
0.22
0.179
0.154
[u]
0.16
0.168
0.151
0.215
0.172
[o]
0.172
0.153
0.167
0.222
0.162
[a]
0.166
0.15
0.173
0.186
0.164
Mean
0.177
0.166
0.173
0.203
0.168
SD
0.016
0.017
0.028
0.019
0.013
In Table 3.21, the length of each fricative for every fricative vowel pair, in addition to
means and standard deviations by category, are provided. A non-parametric Friedman
test was conducted on the data (and on the two data sets that follow) as they reflect the
results of a single subject–in this case, the producer of the stimuli introduced in 3.3–
rendering a repeated measures ANOVA inappropriate. Concerning the average length by
condition ([ɸ], [f], [s], [ç], [x]) of the token fricatives, no statistically significant
87
differences result (χ2(4)=8.8; p=0.066). Thus, overall, the durations of the fricatives are of
comparable length.
Table 3.22. Vocalic duration (in seconds) of fricative CV tokens
[i]
[e]
[u]
[o]
[a]
[ɸ]
0.254
0.267
0.273
0.247
0.263
[f]
0.250
0.252
0.251
0.274
0.267
[s]
0.278
0.277
0.256
0.255
0.269
[ç]
0.251
0.240
0.269
0.222
0.259
[x]
0.251
0.279
0.255
0.251
0.281
Mean
0.257
0.263
0.261
0.250
0.268
SD
0.012
0.017
0.009
0.019
0.008
The durations of each vowel component for every fricative vowel pair appear in Table
3.22; the Friedman test also revealed no significant differences (χ2(4)=4.889; p=0.299),
indicating comparable duration concerning the vowels. Given the above it is unsurprising
that there were also no significant differences in overall syllable duration (χ2(4)=5.12;
p=0.275).
Table 3.23. Syllable duration (in seconds) of target CV tokens
[ɸ]
[f]
[s]
[ç]
[x]
[i]
0.453
0.461
0.428
0.458
0.452
[e]
0.437
0.418
0.471
0.453
0.421
[u]
0.438
0.445
0.407
0.47
0.441
[o]
0.423
0.393
0.436
0.444
0.421
[a]
0.417
0.429
0.428
0.437
0.445
Mean
0.434
0.429
0.434
0.452
0.436
SD
0.014
0.026
0.023
0.013
0.014
88
Having demonstrated that duration was controlled and thus ruling out duration as one
possible source of perceived difference during the discrimination task, the next
subsection examines formant structure to rule out vowel quality as another possible
source of difference.
3.3.2.4. Vowel quality
Using the same 25 CV syllables as the preceding section, F1 and F2 values were
measured in Praat in order to determine if variances in vowel qualities (differences in the
resonant properties of, e.g., [e] in [fe] versus [xe]) might lead to answers of ‘different.’
Table 3.24 lists the average formant values from a 20 ms window in the middle of each
vowel8 in order to avoid coarticulation effects from the preceding fricative at vowel onset
and possible utterance final creaky voice (Ladefoged 2006: 121) at vowel offset.
Table 3.24. F1 and F2 values (Hz) of vowels in target CV tokens
[i]
8
[e]
[u]
[o]
[a]
F1
F2
F1
F2
F1
F2
F1
F2
F1
F2
[ɸ]
284
2075
379
1835
315
789
394
826
595
1056
[f]
267
2078
390
1797
320
738
403
813
590
1090
[s]
251
2053
359
1831
322
761
382
849
584
1054
[ç]
251
2088
354
1837
321
751
399
862
568
1089
[x]
271
2051
392
1859
328
727
416
812
582
1049
Mean
264.8
2069
374.8
1832
321.2
753.2
398.8
832.4
583.8
1068
SD
14.08
16.26
17.51
22.29
4.66
23.8
12.44
22.28
10.20
20.15
For [çju], [çjo] and [çja], a discernible palatal approximant [j] is present in the transition between
the front fricatives and the following non-front vowels. This articulatorily facilitative off-glide
was left unspliced in order to maintain as much naturalness as possible in the speech tokens. For
this reason, the F2 values for [u], [o] and [a] in these syllables were taken from a 20 ms window
near the end of the vowel in order to avoid influence from [j] (whose F2 values began at 2097,
2074 and 1807 Hz, respectively). Furthermore, the palatal components comprised 0.077, 0.092
and 0.083 seconds of total vowel duration (included in the measurements from Table 3.22).
89
Since each syllable within the perception task pairs utilizes the same vowel (e.g., [o] in
[fo] [xo]), it is not necessary to make inter-categorical comparisons. In other words,
whether or not the group [o] as a collective is significantly different from the group [u] as
a collective is irrelevant. It is, however, necessary to determine how [o]-like each
individual [o] is, since differences in the two [o] articulations must be ruled out as a
possible source for an answer of ‘different’ given the pair [fo] [xo].
To determine how closely the vowels are clustered, I calculated the range
encompassing two standard deviations both above and below the means for all categories,
a range which statistically subsumes 95.45% of all values for a normally distributed data
set, or the equivalent of p>0.05 in between group comparisons.
Comparing Tables 3.24 and 3.25, every vowel falls within the range necessary for
its respective category such that no vowel is a statistical outlier (p>0.05) relative to the
other four vowels within its cluster; i.e., the vowels in the two articulations of, e.g., [o] in
the pair [fo] [xo] are statistically equivalent in terms of F1 and F2 values, ruling out
vowel quality as a source of perceived difference in the AX discrimination task. Figure
3.3 represents in vowel chart format the F1 and F2 values from Table 3.24 in order to
demonstrate how little variability exists within each vowel quality.
Table 3.25. Ranges (two SDs) of F1 and F2 values (Hz) of vowels in target CV tokens
[i]
[e]
[u]
[o]
[a]
F1
F2
F1
F2
F1
F2
F1
F2
F1
F2
Mean
264.8
2069
374.8
1832
321.2
753.2
398.8
832.4
583.8
1068
SD
14.08
16.26
17.51
22.29
4.66
23.8
12.44
22.28
10.20
20.15
High
293
2102
409.8
1877
330.5
800.8
423.7
877
604.2
1108
Low
236.6
2036
339.8
1787
311.9
705.6
373.9
787.8
563.4
1028
90
Figure 3.3. Vowel distribution by F1 and F2 values (Hz) of target CV tokens
F2
2500
2000
1500
1000
500
0
0
200
[i]
F1
[u]
400
[e]
[o]
600
[a]
800
In terms of the nativeness of the vowels, Table 3.26 provides the ranges reported by
Martínez Celdrán and Fernández Planas (2007: 175, 181).
The vowels produced by my speaker fall within range relative to vowels of native
Spanish with the following eight exceptions with the relevant value for comparison
shaded below: F1 of [se] (359 Hz) and [çe] (354 Hz); F1 of [so] (382 Hz); and every F2
of [a] ([ɸa] 1056 Hz, [fa] 1090 Hz, [sa] 1054 Hz, [ça] 1089 Hz and [xa] 1049 Hz). In
general, the articulation of the low central [a] in this speaker’s Spanish is farther back
than a native [a] by an average of 228.4 Hz, while the differences with the two instances
of non-native [e] and one of [o] are negligible (20, 25 and 3 Hz, respectively).
Table 3.26. Native ranges of F1 and F2 values (Hz) of Spanish vowels
[i]
[e]
[u]
[o]
[a]
F1
F2
F1
F2
F1
F2
F1
F2
F1
F2
High
414
2523
587
2212
449
1175
656
1313
1002
1635
Low
241
1832
379
1676
275
622
385
793
536
1296
91
3.3.3. Interim summary
In the preceding subsections, I analyzed in detail the individual CV tokens
pertinent to the fricative properties under investigation. Crucially, the syllables constitute
the experimental tokens in the AX discrimination task to determine if perceptual
similarity plays a role in the assimilations. To control for fricative identity, I acoustically
analyze the 25 syllables containing the four fricative allophones [f], [ɸ], [x] and [ç]
relevant to the discussion–plus [s]–and show that the fricatives are indeed what they are
claimed to be. Duration was controlled by measuring each fricative and vowel component
and demonstrating no significant differences, while vowel quality was measured by
plotting F1 and F2 values and determining that no statistical outliers exist within the
vowel clusters. The two sections that follow detail the protocol followed in the
production tasks: sentence reading (§3.4) and picture description (§3.5).
3.4. Task 3: sentence reading
The final two tasks are production tasks of a more controlled nature than the
sociolinguistic interview designed to elicit the environments appropriate for the
processes. To that end, they respond to research questions 1 (what are the allophonic
alternations of /f/ and /x/, if any?), 2 (are they productive?), 3 (are they conditioned by
the surrounding vowel(s) in a regressive and/or progressive direction?) and 4 (is stress
deterministic?).
In the first production task, participants read 110 which were recorded in
Audacity (and subsequently analyzed in Praat) using a Fostex FR-2LE 24 bit 48 Hz
recorder using an Audix HT5 small condenser headset microphone powered by an Audix
APS-911 phantom power supply. Participants read the sentences on an HP Mini 210
92
Netbook running Microsoft Office PowerPoint version 2010. Each page of the
presentation contained one sentence that the participants were instructed to read at a
normal, conversational rate. There was no time limit imposed, and the participants
progressed to each subsequent sentence by clicking the mouse and advancing the page.
Productions were recorded and analyzed using Praat software.
The sentences were randomized such that no two consecutive sentences contained
target words that shared either the fricative in question (/f/ or /x/) or the vowel that
followed it. I constructed sentences instead of simple word lists for three reasons, the first
being the resultant naturalness of the speech, as speakers tend to self-monitor and thus
provide less naturalistic speech when reading isolated words in list format (Bayley 2002,
Foulkes, Scobbie and Watt 2010). Second, embedding the target words in sentence-length
contexts allowed me to avoid fillers since the participants would have no reasonable way
of determining which word (or words) were indeed the desired targets in each sentence.
Such was necessary to keep at a minimum the number of sentences in the task given the
vast range of conditions, which, if combined with fillers, would render the test
impractical and begin to introduce the possibility of participant fatigue. Third, sentences
allow me to manipulate the surrounding phonetic context in order to primarily explore the
effects of preceding and following vowels and prosody.
During the experiment, each sentence occupied two lines of text on the computer
screen. To control for length, the sentences were measured in terms of characters,
including spaces (as opposed to morphemes, syllables or words, measurements which
involve variation), given that the task is a reading task. The descriptive statistics
concerning length appear below.
93
Table 3.27. Length of sentences (n=110) by character (including spaces) in task 3
Characters
Mean
SD
High
Low
65.41
9.59
84.60
46.21
Note: ‘High’ represents mean plus two standard deviations; ‘low,’ mean less two
standard deviations; sentences below the minimum or above the maximum are
significantly different (p<0.05).
None of the 110 sentences fall outside this range (see Appendix C).
In addition to sentential length, word position within the sentence, segmental
position within the word and frequency of the target word were also considered.
Regarding position within the sentence, the literature on domain-initial strengthening
(Fougeron and Keating 1997; Keating, Cho, Fougeron and Hsu 1997; Cho and Keating
2009; additionally, see Beckman 1998) shows that consonants at the left (utteranceinitial) edge of prosodic domains resist reduction and are instead articulated with more
extreme gestures (greater and longer articulatory contact). Assuming Prosodic Phonology
(Nespor and Vogel 1986, Selkirk 1986), the prosodic hierarchy is organized into the
following constituents.
(4)
Utt (utterance)
|
IP (intonational phrase)
|
PhP (phonological phrase)
|
PWd (prosodic word)
|
Ft (foot)
|
σ (syllable)
94
Specifically, Fougeron and Keating (1997) show a positive correlation between degree of
contact between articulators and height in the prosodic hierarchy; i.e., the effect of
strengthening on a consonant that begins an utterance is greater than that on one that
begins an intonational phrase, and so on. To that end, target words consistently appeared
in sentence-medial (or, minimally, not in absolute initial or absolute final) position in
order to avoid such effects. Stress, however, was manipulated as an independent variable
in order to determine if word-level prosody is a conditioning factor, following the logic
that domain-initial strengthening conflicts with reduction processes. The position of /f/
and /x/ within the word (initial versus non-initial) was also manipulated given the
domain-initial strengthening facts above.
Furthermore, it was necessary to manipulate word-level frequency.9 As Sanicky
(1988: 275) notes, (at least concerning /f/ velarization) the [x] allophone occurs most
frequently in the irregular preterit forms of ser ‘to be’ and ir ‘to go’: fui [xu̯ i] ‘I
was/went,’ fuiste [xu̯ ih.t̪ e] ‘you were/went,’ etc. In the relevant dialects, the process may
have been lexicalized in these highly frequent forms due to their near-uniform production
with the velar [x] allophone. As Bybee (2010) and Coetzee and Kawahara (2013)
demonstrate, (variable) phonological processes are more likely to occur in high frequency
words than in low frequency words (and, as a corollary, diachronic change is likely to
occur more quickly in the former). Given these facts, frequency acted as an additional
dependent variable.
9
Defining “frequency” is not as straightforward as one would assume. That corpora and lexical
databases can be accepted as indicative of individuals’ experiences and lexicons is usually taken
for granted; however, the threshold between “frequent” versus “infrequent” is not always made
explicit. Framing the issue solely in terms of lexical frequency also ignores frequency (or, in
more general terms, probabilities) at other levels of analysis, e.g., segmental combinations,
morphemes or lexical neighbors (see discussions in Bybee 2010 and Coetzee and Kawahara
2013).
95
Concerning research question 2, frequency is one aspect to consider when
examining the productivity of the processes. Were /f/ velarization, /f/ bilabialization and
/x/ palatalization only to occur in (highly) frequent words, one might argue that such a
scenario suggests lexicalization and not active grammatical processes. On the other hand,
demonstrating the processes in less frequent words suggests a systematic/grammatical
property that is not lexicalized, but rather applies (even if variably so) in the appropriate
phonological contexts irrespective of frequency. For this reason, word-level frequency
was manipulated in task 3.
The target words were taken from the LEXESP Spanish corpus (Sebastián-Gallés,
Cuetos, Martí and Carreiras 2000). Token frequencies were computed by totaling the
individual frequencies of the inflected forms for each lexeme (derivational forms
excluded). For example, the overall frequency of profund- (adjective, ‘deep, profound’;
699 occurrences) represents the sum of the frequencies reported for profund-a (feminine,
singular; 275), profund-o (masculine, singular; 261), profund-a-s (feminine, plural; 103)
and profund-o-s (masculine, plural; 60); not included were derived forms (even though
they share the same root): profund-a-mente (adverb, ‘profoundly’), profund-idad (noun,
‘profundity’), profund-izar and its conjugations (verb, ‘to deepen, to detail’), profund-ización (noun, ‘deepening’), etc.10
For each CV combination within each condition (e.g., /fu/ within the /f/
assimilating condition; see §§3.4.2 through 3.4.6), I ranked the qualifying words from
most to least frequent, selecting the two words with the highest and lowest frequencies.
Borrowings, abbreviations and interjections were disqualified.
10
The LEXESP corpus also contains common idiomatic phrases. Inflected forms of relevant
tokens that appear separately in these phrases were also included (derivatives excluded).
96
After selection, I followed the procedure in Marshall and van der Lely (2006). I
performed an independent samples t-test (standard α level at 0.05) with frequency
(converted to natural log values for ease of comparison) as the relevant variable in order
to ascertain that the ‘frequent’ tokens were statistically different from the ‘infrequent’ set.
Consequently, for any given vowel (/i, e, u, o, a/) within each condition, there are
maximally eight possible tokens: two positions (initial, non-initial) two stress patterns
(stressed, unstressed) two frequencies (frequent, infrequent). A sample series for the /f/
assimilating condition, /fu/, appears in Table 3.28 (see Appendix C for the complete
frequency analysis).
Table 3.28. Sample token list for /f/ assimilating condition, task 3 /fu/ tokens
Frequent
Frequency
Natural log
Infrequent
Frequency
Natural log
a.
fuga
131
4.88
fuco
2
0.69
b.
futuro
905
6.81
funesto
24
3.18
c.
profundo
699
6.55
difuso
62
4.13
d.
difusión
148
5.00
efusivo
7
1.95
Mean
470.75
5.81
23.75
2.49
SD
391.69
1.011
27.18
1.49
p
0.0103 (statistically significant difference between frequencies)
a. /fu/ initial, stressed; ‘escape’ (noun) and ‘fucus’ (noun, type of algae)
b. /fu/ initial, unstressed; ‘future’ (noun) and ‘disastrous’ (adjective)
c. /fu/ non-initial, stressed; ‘profound’ (adjective) and ‘vague’ (adjective)
d. /fu/ non-initial, unstressed; ‘diffusion’ (noun) and ‘effusive’ (adjective)
97
In addition to the sentences specifically designed for each token, it is possible that the
equivalent context appears elsewhere in the task in non-target words used in the
sentences of separate tokens. For example, the target word designed for the (frequent)
non-initial, unstressed, non-assimilating /fe/ context is diferente ‘different’ (5). However,
the same set of variables (non-initial, unstressed, non-assimilating /fe/) appears in the
word profesiones ‘professions, careers,’ which happens to be part of the sentence
designed for the token word ilegible ‘breath, stamina,’ which instead targets the
(infrequent) non-initial, stressed assimilating /xi/ condition (6).
Such words are not listed independently in the appendix, nor are they considered
when calculating frequency by condition, but they do form part of the data analysis
present in Chapter 4.
(5)
Intended target:
diferente, /fe/, non-initial, unstressed, frequent
Juana busca un estilo personal diferente después de años de lo mismo.
Juana is looking for a different personal style after years of the same (style).
(6)
Intended target:
ilegible, /xi/, non-initial, stressed, infrequent
Additional context: profesiones, /fe/, non-initial, unstressed, frequent (cf. (5))
En ciertas profesiones, la escritura ilegible puede causar problemas graves.
In certain professions, illegible handwriting can cause serious problems.
While a complete, non-randomized version of every sentence from task 3 (and the
corresponding frequency analysis) is available in Appendix C, the following subsections
explain in detail the twelve conditions from the task and provide sample sentences for
each.
98
3.4.1. Justification for tokens and conditions
Due to the discrepancies explored in Chapter 1 of the available descriptions of the
processes, the purpose of the production tasks is to elicit the fricative allophones in as
many different phonological contexts as possible. Beyond prosodic structure and the
position of the fricative within the word, the surrounding segmental context has also been
argued to be deterministic. It is this latter fact that gives rise to the following
experimental conditions: /f/ assimilating, /f/ non-assimilating, /f/ progressive assimilation
and /f/ with ir ‘to go’ and ser ‘to be’ paradigms (§3.4.2); and /x/ assimilating, /x/ nonassimilating and /x/ progressive assimilation (§3.4.3).
3.4.2. /f/ conditions
According to the literature, bilabialization and velarization of /f/ tend to occur in
the presence of the back vowels /u/ and /o/ (Lipski 1995, Mazzaro 2005, Piñeros 2009);
however, some have argued against such a categorical treatment, claiming that the
processes are possible irrespective of the following vowel (Perissinotto 1975, Resnick
1975, Sanicky 1988). With these preliminaries in mind, the following conditions were
considered in selecting tokens for the sentence reading task.
3.4.2.1. Assimilating
Twenty words containing the sequences /fu/, /fuV/ ([u̯ V] diphthong) and /fo/
comprise the /f/ assimilating condition. As described above, each CV(V) combination
includes eight tokens given the three independent variables with two factors each:
position, stress and frequency. This is demonstrated by Table 3.29.
99
Table 3.29. Task 3 /f/ assimilating tokens (n=20)
Initial
/fu/
/fuV/
/fo/
Non-initial
Stressed
Unstressed
Stressed
Unstressed
frequent
fuga
futuro
profundo
difusión
infrequent
fuco
funesto
difuso
efusivo
frequent
fuente
11
afuera
11
infrequent
fuelle
frequent
fondo
fortuna
uniforme
metáfora
infrequent
fobia
fogoso
aforo
aforismo
refuerzo
Note: Glosses are as follows (left to right, top to bottom): ‘escape’ (noun), ‘future’
(noun), ‘profound’ (adjective), ‘diffusion’ (noun); ‘fucus’ (noun, type of alga),
‘disastrous’ (adjective), ‘vague’ (adjective), ‘effusive’ (adjective); ‘fountain, resource’
(noun), ‘outside’ (adverb); ‘stamina’ (noun), ‘reinforcement, support’ (noun); ‘fund’
(noun), ‘fortune’ (noun), ‘uniform’ (noun and adjective), ‘metaphor’ (noun); ‘phobia’
(noun), ‘fiery’ (adjective), ‘capacity’ (noun), ‘aphorism’ (noun).
The tokens appeared embedded in sentences edited by two native Spanish speakers–one
Chilean (male, age 37) and one Mexican (female, 33)–for grammatical, pragmatic and
cultural clarity. Sample sentences from the /f/ assimilating condition are provided in (7)
and (8).
(7)
/fo/, initial, stressed, frequent
Según los jefes no hay fondos suficientes para un aumento de sueldo este año.
According to the bosses there are not enough funds for raises this year.
(8)
/fo/, initial, stressed, infrequent
A la vecina que le tiene fobia a los animales no le gusta ir de caminata.
The neighbor who has a fear of animals does not like to go hiking.
11
Unstressed tokens for the /fuV/ diphthongs are not provided as such forms are relatively rare in
Spanish given that the Latin source of the vast majority of /uV/ diphthongs is stressed short O [ɔ]:
PORTA > puerta ‘door,’ BONUS > bueno ‘good.’ A few exceptions are found in infinitival verbal
forms with final stress, notably: secuestrar, *secostrar ‘to kidnap’ (cf., secuestro ‘kidnapping’
noun), frecuentar, *frecontar ‘to frequent’ (cf. frecuente ‘frequent’ adjective) and amueblar,
*amoblar ‘to furnish’ (cf. mueble ‘furniture’ noun). Derived environments also circumvent
expectation demonstrated by the following penultimately stressed forms: puerta > puertita,
*portita ‘little door’ (diminutive), bueno > buenísimo, *bonísimo ‘very good’ (superlative).
100
3.4.2.2. Non-assimilating
In the non-assimilating /f/ condition, the same position stress frequency
paradigms are applied to words in which /f/ appears with the allegedly non-assimilating
vowels /i, e, a/. Twenty-four tokens (three factors, two variables per factor three
vowels), provided in Table 3.30, result; corresponding sample sentences are given in (9)
and (10).
Table 3.30. Task 3 /f/ non-assimilating tokens (n=24)
Initial
/fi/
/fe/
/fa/
Non-initial
Stressed
Unstressed
Stressed
Unstressed
frequent
fino
final
beneficio
dificultad
infrequent
físico
filete
grafito
sáfico
frequent
fecha
fenómeno
efecto
diferente
infrequent
ferro
festón
prefecto
plumífero
frequent
falta
familiar
artefacto
tarifa
infrequent
fario
farsante
calafate
pífano
Note: Glosses are as follows (left to right, top to bottom): ‘fine’ (adjective), ‘final’ (noun
and adjective), ‘profit’ (noun), ‘difficulty’ (noun); ‘physical’ (adjective), ‘fillet’ (noun),
‘graphite’ (noun), ‘Sapphic’ (adjective); ‘date’ (noun), ‘phenomenal’ (adjective), ‘effect’
(noun), ‘different’ (adjective); ‘iron’ (noun), ‘festoon, scallop’ (noun, adornment),
‘prefect’ (noun), ‘feathery’ (adjective); ‘lack’ (noun), ‘familiar’ (adjective), ‘artifact’
(noun), ‘tariff’ (noun); ‘luck’ (noun), ‘fraud’ (noun or adjective), ‘caulker’ (noun), ‘fife,
fifer’ (noun).
(9)
/fi/, initial, stressed, frequent
A la novia no le gusta el pelo muy fino que tiene.
The bride/girlfriend does not like the really fine hair that she has.
(10)
/fi/, initial, stressed, infrequent
Es mejor tener clase en un aula física en vez de electrónica.
It is better to have class in a physical classroom than in an electronic one.
101
3.4.2.3. Progressive assimilation
Until this point, every /f/-vowel combination has been tautosyllabic, necessitating
a regressive/anticipatory assimilation analysis. The previous literature is silent concerning
the possibility of heterosyllabic progressive/perseveratory assimilation. In pursuing this
unexplored question, it was necessary to include forms containing an assimilating vowel
{/u/, /o/}, followed by /f/, followed by a non-assimilating vowel {/i/, /e/, /a/}. /f/ initial
words were discarded such that all three segments in the /VfV/ sequence were internal to
the prosodic word. This resulted in eight tokens (/u, o/ stressed/unstressed
frequent/infrequent), provided in Table 3.31. (11) and (12) present sample sentences.
Table 3.31. Task 3 /f/ progressive assimilation tokens (n=8)
Non-initial
/ufV/
/ofV/
Stressed
Unstressed
frequent
bufanda
suficiente
infrequent
bufete
acúfeno
frequent
profano
oficial
infrequent
ofidio
estrofa
Note: Glosses are as follows (left to right, top to bottom): ‘scarf’ (noun), ‘sufficient’
(adjective); ‘office’ (noun), ‘tinnitus’ (noun, medical condition); ‘layperson’ (noun),
‘official’ (noun and adjective); ‘ophidia(n)’ (noun and adjective, zoological class of
serpents), ‘verse, stanza’ (noun).
(11)
/ufV/, stressed, frequent
La madre de Sebastián le tejió una bufanda roja, su color favorito.
Sebastian’s mother wove for him a red scarf, his favorite color.
(12)
/ufV/, stressed, infrequent
Paula fue al bufete del abogado para discutir su próxima demanda.
Paula went to the lawyer’s office to discuss her next lawsuit.
102
3.4.2.4. Lexicalized velarization with ir and ser
Given the frequency with which the /f/-initial past tense forms (in both indicative
and subjunctive moods) of ir ‘to go’ and ser ‘to be’ are reported to velarize–approaching
one hundred percent according to Sanicky (1988)–in the grammars of speakers who allow
velarization, these verbal paradigms were separated into their own condition in
determining lexicalization of the processes (frequency being irrelevant given that the
verbs and corresponding conjugations are highly frequent). The relevant forms appear in
Table 3.32.
Table 3.32. Task 3 verbal paradigms for ir ‘to go’ and ser ‘to be’ (n=8)
Preterit
a
Past subjunctive 1
b
Past subjunctive 2
yo
fui
fuera
fuese
tú
fuiste
fueras
fueses
él, ella, Usted
fue
fuera
fuese
nosotros
fuimos
fuéramos
fuésemos
vosotros
fuisteis
fuerais
fueseis
ellos, ellas, Ustedes
fueron
fueran
fuesen
Note: From top to bottom by column, the personal pronouns are 1 singular ‘I,’ 2 singular
‘you’ (informal), 3 singular ‘he, she, you (formal), it,’ 1 plural ‘we,’ 2 plural ‘you’
(informal, used primarily in Spain), and 3 plural ‘them (masculine, feminine), you’
(formal and informal, used primarily in Latin America).
a
Preterit forms are the simple past tense: yo fui ‘I went/I was,’ tú fuiste ‘you went/you
were,’ etc.
b
Past subjunctive 1 forms are more common than the alternative 2 forms and are roughly
translated as ‘to go’ or ‘to be’ with past tense matrix clauses that take subjunctive
complements, e.g., Quería que fueras ‘I wanted you to go (past subjunctive).’
103
I excluded from the task the shaded forms: fuéramos did not appear in the LEXESP
corpus, the vosotros ‘you plural’ forms are not part of the participants’ Latin American
grammar and the –se past subjunctive alternative forms appear less frequently in general
than the –ra forms. This resulted in eight unique forms for the ir and ser condition (1 and
3 singular in the past subjunctive–fuera and fuera–are indistinct). (13) and (14) provide
sample sentences.
(13)
ir, ‘to go,’ preterit, 1 singular
Cuando estaba en Perú fui a Lima, pero no tuve tiempo para ir a Cuzco.
When I was in Peru I went to Lima, but I did not have time to go to Cusco.
(14)
ir, ‘to go,’ past subjunctive, 3 singular
La madre quería que su hijo fuera al doctor ya que su dolor ha persistido.
The mother wanted her son to go to the doctor because his pain has persisted.
The preceding sections explained the labiodental /f/ conditions in the sentence reading
task. The same conditions are detailed in the following sections as they apply to velar /x/.
3.4.3. /x/ conditions
Unlike /f/ allophony, velar /x/ palatalization is conditioned by the front vowels /i,
e/. However, the /x/ tokens followed the same protocol with respect to position within the
word, stress and frequency in assimilating, non-assimilating and progressively
assimilating conditions, although there are no lexicalized /x/ tokens comparable to ir and
ser.
3.4.3.1. Assimilating
Eighteen words containing the sequences /xi/, /xiV/ ([i̯ V] diphthong) and /xe/
constitute the set of /x/ assimilating tokens. Table 3.33 provides the relevant forms for
which two sample sentences appear in (15) and (16) below.
104
Table 3.33. Task 3 /x/ assimilating tokens (n=18)
Initial
/xi/
/xiV/
/xe/
Non-initial
Stressed
Unstressed
Stressed
Unstressed
frequent
giro
gigantesco
registro
lógico
infrequent
gila
jineta
ilegible
frígido
frequent
12
13
religión
13
infrequent
elogioso
frequent
gente
general
ejemplo
imagen
infrequent
geno
geógrafo
manijero
hereje
Note: Glosses are as follows (left to right, top to bottom): ‘turn, spin’ (noun), ‘gigantic’
(adjective), ‘register’ (noun), ‘logical’ (adjective); ‘stupid, silly’ (noun and adjective),
‘manner of riding a horse’ (noun), ‘illegible’ (adjective), ‘frigid’ (adjective); ‘religion’
(noun); ‘praiseworthy’ (adjective); ‘people’ (noun), ‘general’ (noun and adjective),
‘example’ (noun), ‘image’ (noun); ‘lineage’ (noun), ‘geographer’ (noun), ‘manager or
foreman of a farmstead, prone to inciting problems’ (noun and adjective), ‘heretic’
(noun).
(15)
/xi/, non-initial, stressed, infrequent
En ciertas profesiones la escritura ilegible puede causar problemas graves.
In certain professions, illegible handwriting can cause grave problems.
(16)
/xe/, initial, stressed, frequent
Mis amigos no se divirtieron porque había demasiada gente en la playa.
My friends did not have fun because there were too many people on the beach.
3.4.3.2. Non-assimilating
Preceding the back vowels /u/, /o/ and /a/, velar /x/ fails to palatalize. The twentyfour tokens (eight per vowel) that appear in Table 3.34 also differ with respect to the
position of the fricative, stress and frequency.
12
13
Stressed, word-initial /xiV/ sequences were absent from the LEXESP corpus.
Similar to unstressed /fuV/ diphthongs, unstressed /xiV/ diphthongs are just as rare in Spanish
for the same historical reason–the Latin source of most /iV/ diphthongs is stressed short E [ɛ]:
PETRA > piedra ‘rock,’ BENE > bien ‘well.’ Exceptions are also found in infinitival verbal forms
with final stress: arriesgar, *arresgar (cf. riesgo ‘risk’ noun). Unsurprisingly, derived
environments are additional exceptions: piedra > piedrita, *pedrita ‘little rock’ (diminutive).
105
Table 3.34. Task 3 /x/ non-assimilating tokens (n=24)
Initial
/xu/
/xo/
/xa/
Non-initial
Stressed
Unstressed
Stressed
Unstressed
frequent
justo
juventud
ajuste
lujurioso
infrequent
judo
jurel
bajura
sojuzgado
frequent
joven
jornada
enojoso
trabajo
infrequent
jota
joroba
piojoso
navajo
frequent
jarro/a
jamás
tajante
hoja
infrequent
jade
jayán
majada
sonaja
Note: Glosses are as follows (left to right, top to bottom): ‘just, fair’ (adjective), ‘youth’
(noun), ‘adjustment’ (noun), ‘lustful’ (adjective); ‘judo’ (noun, martial arts), ‘scad, horse
mackerel’ (noun, fish), ‘shortness’ (noun, lack of height), ‘subjugated, oppressed’ (noun
and adjective); ‘young (one)’ (noun and adjective), ‘workday’ (noun), ‘angry’ (adjective),
‘work’ (noun); ‘(the letter) j’ (noun), ‘hump’ (noun, of a camel), ‘lice-ridden, nitpicky’
(adjective), ‘puddle’ (noun); ‘pitcher, jug’ (noun), ‘never’ (adverb), ‘emphatic,
unwavering’ (adjective), ‘leaf’ (noun); ‘jade’ (noun), ‘strong man’ (noun and adjective),
‘pen, corral’ (noun, for animals), ‘rattle’ (noun, for a child).
(17)
/xu/, initial, stressed, frequent
El niño se quejó de que no era justo su castigo.
The boy complained that his punishment was not just.
(18)
/xo/, initial, stressed, frequent
Cuando era joven, Ramón solía causar muchos problemas para sus padres.
When he was young, Ramon tended to cause a lot of problems for his parents.
3.4.3.3. Progressive assimilation
The same tautosyllabic, regressive observations hold for velar palatalization that
were introduced previously in 3.4.2.3 regarding /f/ allophony; no one has examined the
possibility of heterosyllabic, progressive /x/ assimilation in any systematic way. Again
excluding the word-initial /x/ series in this particular context, there are eight tokens that
vary according to prosody and frequency.
106
Table 3.35. Task 3 /x/ progressive assimilation tokens (n=8)
Non-initial
/ixV/
/exV/
Stressed
Unstressed
frequent
quijote
hijo
infrequent
rijoso
prefijo
frequent
prejuicio
lejos
infrequent
viejuco
hollejo
Note: Glosses are as follows (left to right, top to bottom): ‘Quijote’ (proper noun, the
main protagonist of the classic Cervantes novel Don Quijote), ‘son’ (noun);
‘quarrelsome’ (adjective), ‘prefix’ (noun); ‘prejudice’ (noun), ‘far’ (adverb); ‘elderly’
(noun and adjective), ‘fruit skin’ (noun).
(19)
/ixV/, stressed, frequent
Los enemigos rijosos se gritaron por horas después de la pelea.
The quarrelsome enemies shouted for hours after the fight.
(20)
/exV/, stressed, infrequent
Cuando Braulio va de vacaciones, viaja muy lejos para escapar del estrés.
When Braulio goes on vacation, he travels far to escape the stress.
3.5. Task 4: picture description
The third and final production task is a picture description task in which I provide
the participants with images whose contents contain the targeted words and ask them to
name and describe the items depicted. The picture description task is a production task of
a more natural speech-style than sentence reading. While I was able to create the
sentences used in task 3, I could not directly control how the participants elected to
describe the pictures in task 4, no matter how specific the images might be. The tradeoff
is that while the sentence reading task is more controlled in that I am able to target the
desired contexts, it is a less conversational/natural style of speech, which consequently
reduces the probability of phonological processes that occur outside monitored speech.
107
The two production tasks (other than the sociolinguistic interview) complement each
other by compensating for the possible drawback inherent in the other.
In task 4, participants orally described 50 pictures that appeared on an HP Mini
210 Netbook running Microsoft Office PowerPoint version 2010. Each page of the
presentation contained one image. Participants were given fifteen seconds to describe
each picture in as much detail as possible before the program automatically proceeded to
the next page. While participants were allowed to describe the pictures however they
wished, during the task they were also permitted to refer to a document with the
following suggestions regarding aspects to discuss about the object(s) in question.
(35)
a.
su nombre
its name
b.
cualquier otro sinónimo
any other synonym
c.
sus atributos físicos
its physical attributes
d.
quién(es) suele(n) usarla (si se aplica)
who tends to use it (if applicable)
e.
para qué sirve o para qué se usa (si se aplica)
for what it is used (if applicable)
f.
dónde se encuentra o dónde se usa (si se aplica)
where it is found or used (if applicable)
g.
cualquier otro aspecto que le parece relevante o importante si hay tiempo
any other aspect that seems relevant to you if there is time
The images were randomized such that no two consecutive images targeted words that
shared either the fricative in question (/f/ or /x/) or the vowel that followed it (with the
understanding that participants might not have necessarily used the word intended,
108
anyway). The participants’ descriptions were recorded in Audacity (and subsequently
analyzed in Praat) using the same equipment as described previously during the
sociolinguistic interview and sentence reading task, i.e., a Fostex FR-2LE 24 bit 48 Hz
recorder with an Audix HT5 small condenser headset microphone powered by an Audix
APS-911 phantom power supply. Appendix D provides a complete set of instructions in
both the original Spanish and translated English, in addition to the entire set of images.
3.5.1. Justification for tokens and conditions
In order to best replicate the sentence reading task in a less controlled register, the
same conditions from task 3 repeat in task 4, with the following exception: frequency is
no longer relevant, as the images are all relatively frequent (see Appendix D) in the
LEXESP database (a necessary condition in order to find a recognizable picture for each
word). Otherwise, the same conditions from task 3 (§3.4) recur: stress, word-position,
vowel quality and directionality. The conditions for each are as follows: assimilating
conditions (including /fuV/ and /xiV/ diphthong), progressive assimilation and nonassimilating contexts. In the next subsection, I list for task 4 sample tokens from each of
the above conditions.
3.5.2. /f/ and /x/ conditions
The picture description task is divided into six conditions (three per phoneme)
totaling 50 target words. Tables 3.36 and 3.37 list each word by condition with sample
images following the relevant table (images appeared in full color for the participants
during the experiment, yet are rendered in black and white for the present section and
Appendix D).
109
Table 3.36. Task 4 real word /f/ conditions (n=26)
Initial
Assimilating
Nonassimilating
Progressive
assimilation
Non-initial
Stressed
Unstressed
Stressed
Unstressed
/fu/
fútbol
fumar
perfume
confundido
/fuV/
fuego
/fo/
foto
fotocopiar
alfombra
teléfono
/fi/
fila
edificio
figura
científico
/fe/
feria
feliz
enfermo
cafetería
/fa/
falda
familia
elefante
gafas
/ufV/
bufanda
sofá
/ofV/
estufa
profesor
Note: Glosses are as follows (left to right, top to bottom): ‘football’ (noun), ‘to smoke’
(verb), ‘perfume’ (noun), ‘confused’ (adjective); ‘fire’ (noun); ‘photo’ (noun),
‘photocopy’ (verb), ‘rug, carpet’ (noun), ‘telephone’ (noun); ‘line’ (noun), ‘building’
(noun), ‘figure’ (noun), ‘scientist’ (noun); ‘festival, market’ (noun), ‘happy’ (adjective),
‘sick, ill’ (adjective), ‘cafeteria’ (noun); ‘skirt’ (noun), ‘family’ (noun), ‘elephant’ (noun),
‘(sun)glasses’ (noun); ‘scarf’ (noun), ‘couch, sofa’ (noun); ‘stove’ (noun), ‘professor’
(noun).
Figure 3.4. Task 4 sample images (initial, stressed) /f/ condition, (a) assimilating fútbol
‘football’ and (b) non-assimilating fila ‘line’
a.
b.
110
Table 3.37. Task 4 real word /x/ conditions (n=25)
Initial
Assimilating
Nonassimilating
Progressive
assimilation
Non-initial
Stressed
Unstressed
Stressed
Unstressed
/xi/
giba
jirafa
mejilla
página
/xiV/
región
/xe/
gema
gemelos
mujer
imagen
/xu/
jugo
juguete
juez
brújula
/xo/
joya
joroba
callejón
ojo
/xa/
jaula
jabón
pijamas
pájaro
/ixV/
frijoles
oreja
/exV/
hijo
conejo
Note: Glosses are as follows (left to right, top to bottom): ‘hump’ (noun, camel), ‘giraffe’
(noun), ‘cheek’ (noun), ‘page’ (noun); ‘region’ (noun); ‘jewel’ (noun), ‘twins’ (noun),
‘woman’ (noun), ‘image’ (noun); ‘juice’ (noun), ‘toy’ (noun), ‘judge’ (noun), ‘compass’
(noun); ‘jewelry’ (noun), ‘hump’ (noun, camel), ‘alley(way)’ (noun), ‘eye’ (noun);
‘birdcage’ (noun), ‘soap’ (noun), ‘pajamas’ (noun), ‘bird’ (noun); ‘beans’ (noun), ‘ear’
(noun); ‘son’ (noun), ‘rabbit’ (noun).
Figure 3.5. Task 4 sample images (initial, unstressed) /x/ condition, (a) assimilating jirafa
‘giraffe’ and (b) non-assimilating juguete ‘toy’
a.
b.
Although there are 50 total tokens in Tables 3.36 and 3.37 above, there are only 49
images in the actual task. This is due to the fact that two of the images were selected to
allow for two possible target words depending on how the participants chose to describe
111
them: giba, joroba ‘camel’s hump’ and figura, imagen ‘figure,’ ‘image.’ In many cases,
participants avoided the target words entirely; however, any instance of /f/ or /x/ as part
of the targeted lexical item or not was included in the analysis. Furthermore, two images
represented the same target word. For región, maps of Chile and Mexico were used
depending on the participant (see Appendix D).
3.6. Conclusion
In this chapter, I described in the detail the 37 Spanish speakers who participated
in the production- and perception-based methodology designed to improve on the
presently available descriptions of the processes. To that end, I additionally explained the
protocol of the four tasks (available in detail in the appendices), justified them by
exploring their usefulness vis-à-vis how they address the research questions and provided
sample tokens and filler conditions. The following chapter reports on the results of each
task. In Chapter 5, I put forth the Optimality Theoretic analysis based on those results.
112
CHAPTER 4
STATISTICAL ANALYSIS:
RESULTS, DISCUSSION AND IMPLICATIONS
The preceding chapter detailed the methodological procedures of the empirical
portion of this dissertation; this chapter presents the results of these experimental tasks
and addresses their implications for the theoretical analysis of the data under
consideration. Recall that there were three main production tasks in which four Chileans
(experimental=2, control=2) responded to ten interview questions, read 110 sentences
and described 50 pictures for the purpose of eliciting allophones of the labiodental /f/ and
velar /x/ fricatives. With the exception of the sentence reading task, four Mexican
speakers (experimental=2, control=2) also participated in the above tasks. The same eight
speakers–in addition to six other Chilean, ten Mexican and ten control participants from
various dialects of Spanish–took part in the perceptual discrimination task.
It is important to note that the tasks are collapsed in the sections that follow as the
relevant statistical comparisons are within and between groups (Chile versus Mexico and
experimental versus control)–not comparisons among the tasks–given that I am not
interested in examining task effects. The one exception, however, is section 4.3.2 in
which I discuss the productivity of the processes. The production tasks were designed to
differ slightly in formality to determine if the unfaithful allophones appear in all registers
(from the open-ended interview answers and picture descriptions to the more formal,
monitored speech of reading); such would suggest that the processes are indeed
productive (i.e., an active part of the grammar). For this reason I compare the individual
tasks only in this subsection; for all other comparisons the data are compiled. Table 4.1
summarizes the data by task, group and dialect totals.
113
Table 4.1. Number of fricatives analyzed (n=1667)
experimental
Chile
control
experimental
Mexico
control
Task 1
Task 3
Task 4
/f/
67
185
94
/x/
42
174
74
/f/
36
191
98
/x/
45
180
97
/f/
9
50
/x/
9
54
/f/
9
134
/x/
6
113
Group
Dialect
636
1283
647
122
384
262
I begin this chapter with an explanation of the analytic tools used in the statistical and
phonetic analyses. The remainder of the chapter addresses each research question in turn,
concluding with a discussion of what such results suggest for the theoretical contribution
of the thesis in Chapter 5. Before proceeding to the methods and measurements necessary
for any meaningful presentation of the results, I repeat below the subset of research
questions relevant to the result sections of this chapter (with the remaining research
questions addressed by the theoretical analysis and concluding chapters).
1. Are the fricative phonemes /f/ and /x/ synchronically involved in allophonic
alternations in Latin American Spanish? If so, what are the allophones? (§4.2)
2. Are the phonological processes productive (i.e., not merely lexicalized)?
(§§4.3 and 4.4)
3. Can the processes be conceived as assimilations conditioned by the
surrounding vowel(s)? If so, are the assimilations regressive or progressive?
(§4.5)
4. Are the assimilations conditioned by stress or position in the word? (§§4.6, 4.7)
114
6. How can this theoretical analysis limit the assimilations to the subset of
fricatives {/f/, /x/} to the exclusion of all other phonemes in Spanish? (§4.8.3)
7. What role, if any, do articulatory and acoustic phonetics play in the theoretical
model proposed? (§§4.8.1 and 4.8.2)
With these questions in mind, I proceed to a description of how and with what
measurements the data were analyzed.
4.1. Analytic tools
I rely on a variation of fast Fourier transforms (FFT) known as linear predictive
coding (LPC)–in conjunction with the center of gravity (COG) measurements briefly
introduced in Chapter 3–to determine the identity of the fricatives in question. To
determine the degree of coarticulation between the fricatives and following vowels, locus
equations were computed to plot the relative difference in second formant values.
4.1.1. Center of gravity
Mentioned previously in Chapter 3, the center of gravity (COG) is the mean
frequency expressed in Hertz of a given spectral window–for present analyses, 40 ms
extracted from the midpoint of the fricative (Figure 3.2). It has been argued in a large
body of literature for various languages (English: Baker et al. 2011, McMurray and
Jongman 2011; Polish: Padgett and Żygis 2007, Nawrocki 2008; Spanish: Mazzaro 2011)
that, although fricatives are inherently highly variable (Ladefoged and Maddieson 1996,
McMurray and Jongman 2011), certain spectral cues are more reliable in uniquely
identifying place of articulation in fricatives relative to other cues. Thus, I use center of
gravity to determine the identity of the output allophones in section 4.2, and I argue that
reliance on such a measurement is valid based on the studies above. Furthermore, center
115
of gravity was necessary in order to make direct comparisons with Mazzaro (2011), as
she relies on COG values in her analyses, as well.
4.1.2. Fast Fourier transform and linear predictive coding
The 40 ms window from which COG values are computed is known as a fast
Fourier transform (FFT), an extraction from a spectrum that converts the signal to its
representative sine wave (Johnson 2012: 60). Figure 4.1 below represents an FFT sample
of [x] from the token trabajo ‘work’ (noun) as produced by participant 02838 (Chile,
control).
Figure 4.1. Fast Fourier transform, [x] trabajo ‘work,’ 02838 (Chile, control)
20
10
dB
0
–10
0
5
kHz
10
15
In their crosslinguistic survey of fricatives, Gordon, Barthmaier and Sands (2002) use
FFT spectra as an additional diagnostic cue for place of articulation. Yet it is not
necessarily the spectrum as a whole that is relevant, but rather the spectral peaks in terms
of amplitude (dB), which, theoretically, occur at predictable frequencies depending on the
fricative (in the case of [x] from Figure 4.1, the first two main spectral peaks occur
116
around 1015 and 4200 Hz, consistent with Jassem (1968: 204), Manrique and Massone
(1981: 1145) and Martínez Celdrán and Fernández Planas (2007: 107)).1
As Johnson (2012: 71) notes, such peaks are not always salient in FFT spectra,
obscured by the fact that FFTs rely on individual harmonics. Linear predictive coding
(LPC; also commonly known as LPC smoothing) reduces FFT waves to the broadest
peaks; compare Figure 4.1 with 4.2–the LPC equivalent of 4.1–below.
Figure 4.2. Linear predictive coding, [x] trabajo ‘work,’ 02838 (Chile, control)
20
10
dB
0
–10
0
5
kHz
10
15
Due to the clearly defined peaks in the LPC rendering, I rely on LPC spectra–to the
extent that they facilitate identification of spectral speaks–as an additional resource for
fricative identification when such is not possible via COG values alone. In fact, as
McMurray and Jongman (2011: 227) show, the spectral peak with the highest amplitude
(the maximum peak) is a slightly better predictor of place (r2=0.493) than center of
1
Sources differ with respect to the number of peaks reported: Jassem relies on the first three to
four peaks depending on the fricative, Martínez Celdrán and Fernández Planas list two to three,
while Manrique and Massone provide two.
117
gravity (r2=0.425). However, as mentioned above, I present the spectral peaks only when
center of gravity is insufficient for distinction.
4.1.3. Locus equations and second formants
In testing Baker and colleagues’ (2011) proposal that coarticulation is a
prerequisite for grammaticalization, it was necessary to determine the degree to which the
fricatives under investigation and the following vowels attest gestural overlap. To this
end, section 4.8.1 relies on locus equations (LE) (Lindblom 1963; Sussman, McCaffrey
and Matthews 1991; Sussman and Shore 1996), linear regressions that determine the
relative difference in second formant (F2) frequencies at two points in the spectrum: CV
transition and vocalic resting state. Following the protocol in Nirgianaki (2011), the F2
value for CV transition was taken at the first glottal pulse at fricative offset/vowel onset.
In the case of VC coarticulation in progressive tokens, the F2 transition value was taken
at the last vocalic glottal pulse at vowel offset/fricative onset. Resting state F2 values
were manually estimated at the vowel’s midpoint (Figure 4.3).
Locus equations quantify the extent to which two segments are coarticulated by
plotting F2 transitions along the y-axis and F2 vowel midpoints along the x-axis. A
difference in the two values for a single CV (or VC) token suggests lesser degrees of
coarticulation given that the F2 transition value is independent from that of the vowel
(Figure 4.4), resulting in a flatter regression line. Contrastingly, no (to little) change in
the F2 measurement from transition to the middle of the vowel suggests that the F2 value
at the point of consonant offset is instead dependent on that of the vowel, suggesting a
greater degree of coarticulation (Figure 4.5) and exhibiting a steep regression line.
118
Figure 4.3. VCV [ofi] (ofidio ‘ophidia’) with three points of F2 measurement (1 offset
glottal pulse of VC transition, 2 V resting state, 3 onset glottal pulse of CV transition)
2
[
1
3
o
2
f
i
]
Figure 4.4. F2 difference, [fa] familia ‘family,’ 17137 (Chile, experimental)
F2 CV
transition
1593.21
Hz
F2 mid-V
1069.75
Hz
[
f
a
]
119
Figure 4.5. F2 stable, [fo] telefónica ‘telephone,’ 17137 (Chile, experimental)
F2 CV
transition
835.4 Hz
F2 mid-V
1031.17
Hz
[
f
o
]
Locus equations take the form y=kx+c in which y represents F2 values at the VC or CV
transition, x the F2 at vowel midpoints, k the slope and c the y-intercept of the regression
line. The prediction of Figure 4.4 is that non-coarticulated segments will result in flat,
horizontal slopes approaching 0 as the F2 transitions (y) will vary independently of F2
mid-vowel (x) values. Figure 4.5, however, represents coarticulated segments in which
slopes approach 1 due to the fact that changes in the F2 transition (y) directly correlate
with changes in the F2 of the vowel (x).
Also reported are the r2 values of the regression lines. An r2 approaching or equal
to 1 suggests that x is an accurate predictor of y–i.e., that the two values are correlated
(that of coarticulated contexts); conversely, an r2 approaching or equal to 0 suggests that
x is not an accurate predictor of y (that of non-coarticulated utterances). I return to this
discussion in section 4.8.1.
Having explained the methods of analysis, I respond to research question 1 in the
next section, which asks if there are differences in the output realizations of /f/ and /x/.
120
This question serves two purposes: to determine if phonological processes are even
present, as the absence of alternations obviates any need for a grammatical account; and,
if it is determined that there are alternations, to improve the data presented in Chapter 1
by providing phonetic analyses that go beyond impressionistic determinations in
identifying the allophones.
4.2. Allophone identity
Recall from Chapter 1 that the data are relatively underreported. The few
descriptive sources available (Perissinotto 1975, Resnick 1975, Sanicky 1988, Calvo
Shadid 1996) are outdated and consequently lack phonetic specificity (or asked research
questions for which phonetic details were irrelevant). I also offered various criticisms in
Chapter 1, principle among them are the impressionistic nature and imprecise reporting
of the data. Here, I supplement the data with phonetic measurements (center of gravity
values and spectral peaks) to determine the actual identity of the output allophones.
A mixed-model ANOVA with the independent variables group (experimental,
control) and country (Chile, Mexico) was performed to determine how each affects the
center of gravity value of the various allophones. For /f/, country was insignificant (F(1,
4.880)=4.937; p=0.078), indicating that Chilean and Mexican speakers do not differ in
terms of COG measurements associated with each /f/ segment. There was a main effect,
however, for group (F(1, 4.879)=24.933; p=0.004). The group effect suggests that the
four experimental speakers differed significantly in terms of COG measurements for each
allophone from the four control speakers. Regarding /x/, both group (F(1, 4.947)=18.998;
p=0.007) and country (F(1, 4.984)=8.198; p=0.035) were significant. In this case,
experimental participants differ from control participants and Chilean speakers differ
121
from Mexican speakers concerning the COG values associated with the /x/ allophones.
Figures 4.6 and 4.7 summarize the average center of gravity measurements for each
allophone in the Chilean and Mexican dialects, respectively; Table 4.2 provides token
counts for each condition.
Figure 4.6. Average COG (kHz) by Chilean groups in /f/ and /x/ conditions
5
4
kHz
3
2
1
0
[f]
[ɸ]
[x]
[ç]
/f/
/x/
Experimental
2800.91
802.22
513.58
750.78
2008.15
/f/
/x/
Control
4280.38
1149.11
3427.27
Figure 4.7. Average COG (kHz) by Mexican groups in /f/ and /x/ conditions
5
4
kHz
3
2
1
0
[f]
[ɸ]
[x]
[ç]
/f/
/x/
Experimental
2691.12
459.32
213.40
556.07
2294.85
/f/
/x/
Control
3409.88
1238.22
3714.36
122
Table 4.2. Number of tokens by condition
Chile
Mexico
Total
Experimental
Control
Experimental
Control
[f]
186
325
8
143
[ɸ]
157
[x]
208
211
64
106
589
[ç]
85
111
1
13
210
Total
636
647
122
262
1667
49
662
206
The conclusions to be drawn from Figures 4.6 and 4.7 are that, based on center of gravity
means, Chilean and Mexican speakers are not distinct in their production of /f/ allophones
as determined by COG; the four speakers that I have identified as experimental
participants, however, do make /f/ distinctions not found in the four control participants.
These differences are indicated by two unique COG clusters within the experimental
group: high COG [f] versus low COG [ɸ] and [x] (to be disambiguated below).
Regarding /x/, the conclusions are twofold: Chilean speakers differ from Mexican
speakers, and experimental participants differ from control participants. However, unlike
/f/, every country group combination makes the same allophonic distinction in the case
of /x/: high COG [ç] versus low COG [x].
At this point I have labeled the allophones labiodental [f], bilabial [ɸ], velar [x]
and palatal [ç] based on expectations from previous literature; I have not yet argued for
their precise places of articulation. I have only shown that allophonic variation does exist
(without which no analysis would be necessary). The next logical question–indeed, the
second clause of research question 1–is this: if there is allophonic variation, what are the
allophones? For this issue, it is not sufficient to merely demonstrate that there are distinct
123
center of gravity clusters (high versus low); the center of gravity values themselves must
be taken into account. To that end, consider Table 4.3.
Table 4.3. Group COG averages (Hz) by allophonic conditions
[f]
[ɸ]
[x]
[ç]
Experimental
2796.38
720.65
699.83
2011.48
Control
4014.40
1178.90
3457.37
Mean
3657.46
720.65
957.67
2865.24
SD
1887.51
384.44
500.31
973.63
As I previously argued based on numerous studies, center of gravity measurements have
been shown to (relatively) uniquely characterize fricative places of articulation (Jongman
et al. 2000, McMurray and Jongman 2011). When we compare Table 4.3 with known
center of gravity values in Table 4.4, we can determine the identity of the allophones in
question.
Table 4.4. Crosslinguistic COG (Hz) averages
Author
Language(s)
[f]
[ɸ]
[x]
[ç]
Jassem
various
4100
4200
3500
1600
JWW
English
5108
GBS
various
4415
4276
4532
NCF
Greek
5484
3905
4471
Mazzaro
Spanish
3588
2237
Mean
4539
4200
3480
3534
SD
763
0
887
1675
Sources: Jassem (1968), Jongman et al. (JWW 2000), Gordon et al. (GBS 2002),
Nirgianaki et al. (NCF 2010), Mazzaro (2011).
124
At first glance, the two simplest cases appear to be the labiodental [f] (faithful realization
of /f/) and palatal [ç] (palatalized /x/) allophones, whose relatively high average COG
values of 3657.46 and 2865.24 Hz, respectively, align closely with those from Table 4.4.
Less apparent is why [x] as the faithful allophone of /x/ (957.67 Hz) is well below the
cited sources (3480 Hz) (bilabial [ɸ] is also addressed below). Notice, however, that
Mazzaro’s COG value for Spanish velar [x] is lowest among the languages, which lends
support for the low values in the Spanish data reported herein.
Two facts must be considered at this point. First, although COG has been argued
to be a relatively contextually and sociolinguistically independent marker of place
(McMurray and Jongman 2011: 227), fricatives are notoriously variable compared to
other manners of articulation (Ladefoged and Maddieson 1996: 173, McMurray and
Jongman 2011: 227), as evidenced by the discrepancies among the languages in Table
4.4. McMurray and Jongman, in fact, argue that, while COG is a relatively better cue to
place than the other cues that they test, it is not itself infallible, as they report its r2 value
at only 0.425.
Second, a closer look at Mazzaro’s data (2011: 132) reveals interesting nuances.
She divides [x] COG values into fourteen groups: education {high, low} the quality of
the following vowel {[a, e, i, o, u, u̯ e, u̯ i]}. Lowest among these is the [xu̯ i] context for
the low education group with a mean COG at 1758 Hz. [xu] in the low education group is
the most variable: 2093 Hz (average) with a high standard deviation of 1155 Hz. In light
of Mazzaro’s contextual differences and variation in the data, I argue that the low COG
average for the faithful velar allophone [x] by my participants is not so unexpected.
125
However, given the indeterminacy of the velar allophone’s center of gravity,
Figure 4.8 provides the average LPC spectra of [x] by group in order to determine the
spectral peaks. (Recall from Figure 4.2 above that LPC analyses are FFT spectra reduced
to their broadest peaks.)
Figure 4.8. Average [x] (/x/) LPC by group
40
20
dB
CE
0
CC
ME
-20
MC
-40
0
5
10
15
20
25
kHz
Note: CE, Chile experimental (n(participants)=2, n(fricatives)=208); CC, Chile control
(n(p)=2, n(f)=210); ME, Mexico experimental (n(p)=2, n(f)=64); MC, Mexico control
(n(p)=2, n(f)=105).
Irrespective of the /x/ center of gravity values for my participants, it is clear in Figure 4.8
that the first two LPC spectral speaks occur around 1000 and 4000 Hz for all groups,
which are consistent with the expected [x] peaks reported by Jassem (1968: 204) and,
specifically for Spanish, Manrique and Massone (1981: 1145) and Martínez Celdrán and
Fernández Planas (2007: 107). In Table 4.5 below, I provide the specific peak values
from the velar LPC spectra above as measured by Praat.
126
Table 4.5. Average [x] (/x/) LPC spectral peaks by group
First peak (Hz)
Second peak (Hz)
CE
820.31
4054.69
CC
1265.62
3937.5
ME
1054.69
3164.06
MC
1289.06
3960.94
Mean
1107.42
3779.3
SD
218.5
413.27
Identification of two of the four allophones now remains: velar [x] not as a faithful output
of /x/ but rather as an assimilated realization of /f/, and the bilabial fricative [ɸ] as a
separate /f/ allophone (i.e., disentangling the low COG /f/ category from Figures 4.6 and
4.7). Table 4.6 summarizes the ratios of each /f/ allophone produced by the experimental
participants.
Table 4.6. Ratio of /f/ allophones by experimental participants
[f]
Chile
Mexico
[ɸ]
[x]
Total /f/
N
%
N
%
N
%
N
%
19727
107
54.87
85
43.59
3
1.54
195
100
17137
79
52.32
72
47.68
0
0
151
100
26911
8
22.22
27
75
1
2.78
36
100
35792
0
0
22
95.65
1
4.35
23
100
Although my participants produced a significantly greater number of bilabial [ɸ]
allophones as compared to velar [x] (p=0.019), Figures 4.6 and 4.7 presented at the
beginning of this subsection indicated that there are indeed three distinct /f/ allophones
(Table 4.7) with consistent relative COG values. The same protocol employed above to
127
determine [x] as the faithful output of /x/ can be applied to the third /f/ allophone
tentatively labeled [x] as well. Figure 4.9 provides the LPC spectra of the five [x] outputs
of /f/ as produced by my experimental participants.
Table 4.7. COG (Hz) of /f/ allophones by experimental participants
[f]
[ɸ]
[x]
Chile
2800.91
802.22
513.58
Mexico
2691.12
459.32
213.40
Mean
2746.02
630.77
363.49
SD
77.63
242.47
212.26
Figure 4.9. Individual [x] (/f/) LPC by token of experimental participants
40
20
afuera
dB
0
fue
fuego
-20
perfume
afuera
-40
0
5
10 15
kHz
20
25
Note: afuera ‘outside’ (Chile 19727); fue ‘he went’ (Chile 19727); fuego ‘fire’ (Chile
19727); perfume ‘perfume’ (Mexico 26911); afuera (Mexico 35792).
The above velar [x] allophones of /f/ in Figure 4.9 are comparable to the faithful velar [x]
allophones of /x/ from Figure 4.8 in that the LPC spectra all peak around the expected
frequencies: 1000 and 4000 Hz. Table 4.8 lists the specific [x] (/f/) peak frequencies.
128
Table 4.8. Individual [x] (/f/) LPC spectral peaks by token of experimental participants
First peak (Hz)
Second peak (Hz)
afuera
796.86
3750
fue
1031.25
3820.31
fuego
867.19
3421.88
perfume
1148.44
5226.56
afuera
1875
4710.94
Mean
1143.75
4185.94
SD
431.34
752.67
While the velar allophones of /f/ exhibit lower COGs comparable to those of /x/, both [x]
groups’ LPC spectra and spectral peaks align with the phonetic literature for [x]. Viewing
these findings in light of Mazzaro (2011), who examined similar phenomena, rate of
velar /f/ allophony and the requisite phonetic context coincide. Mazzaro (2011: 124)
analyzed 467 labiodental fricatives from 22 speakers in a picture description production
task similar to task 4 described in this dissertation. Of these, 423 (or 90.6%) were realized
faithfully as [f]; the remaining 44 fricatives (9.4%) were assimilated [x]. In my case, 405
instances of /f/ were analyzed in my experimental groups. Discounting the contribution of
the bilabial allophone (as Mazzaro reports no instances of [ɸ] with which to compare), [f]
accounts for 47.9% (194 tokens) of my data, while the five instances of [x] total 1.23%;
the difference in ratios between the two studies is not significant (p=0.642).
In terms of the vocalic context most conducive for the velar realization of /f/,
Mazzaro (2011: 69, 70) provides the following percentages of [x] allophones for the 1811
instances of /f/ analyzed from her sociolinguistic interview: 6.13% [u̯ e, u̯ i] (n=111),
1.05% [u] (n=19), 0.06% for both [i] and [e] (or one token) and 0% for [o] and [a];
combined, 7.23% (n=132) of all /f/ realizations are velar [x], 98.48% (or 130 instances)
129
of which are conditioned by /u/. Comparing Mazzaro’s [x] to /f/ ratio with my overall [x]
to /f/ ratio (five tokens out of 405, or 1.23%), the difference is not significant (p=1). In
terms of the triggering vowel, 100% of my velarized /f/ tokens are triggered by /u/,
which, when compared to Mazzaro, is also not significant (p=1).
The conclusions from the preceding discussion are bipartite. First, I argued that
the identity of one of two low COG /f/ allophones is indeed velar [x] given the results
from LPC spectra and spectral peak analyses. Second, I demonstrated that its relatively
low rate of occurrence is not uncharacteristic (cf. Mazzaro 2011). The final discussion
concerns the other low center of gravity /f/ allophone, which I will argue is bilabial [ɸ].
The challenge in identifying [ɸ] with any sort of phonetic certainty is that one
must be able to distinguish it from labiodental [f]. The two are easily misperceived
bidirectionally (Mielke 2012) since they share similar phonetic properties; for this reason,
contrasting /f/ and /ɸ/ phonemically is relatively marked, the most widely known
exception being Ewe (which also contrasts the voiced counterparts /v/ and /∫/; see Utman
and Blumstein 1994).2
As Jassem (1968) and Ladefoged and Maddieson (1996) note, both [f] and [ɸ]
display relatively flat spectra (Figures 4.10 and 4.11; adapted from Jassem 1968: 197, his
Figures 15 and 16). Additionally, Jassem (1968: 204) reports little difference in the
spectral peaks of each segment (if they peak at all): [ɸ], 600 and 1300 Hz; [f], 500 and
1400 Hz (labeled F1 and F2 by Jassem).
2
Given the high likelihood of misperception, Utman and Blumstein (1994: 231) show that Ewe
speakers enhance the distinction by increasing the amplitude with which [ɸ] is routinely
articulated. Interestingly, they also found differences in amplitude between the English and Ewe
articulations of labiodental [f], with English [f] being higher in amplitude. Thus, by lowering the
amplitude of [f] while increasing that of [ɸ], Ewe speakers maximize the contrast of an unrelated
cue to aid in distinction (cf. English roundedness of palatoalveolar [ʃ] vis-à-vis alveolar [s]).
130
Figure 4.10. [ɸ] LPC spectra
20
dB 10
0
0
1
2
3
4
5
6
7
4
5
6
7
kHz
Figure 4.11. [f] LPC spectra
20
dB 10
0
0
1
2
3
kHz
Given the ease with which [f] and [ɸ] are confused due to their many phonetic/acoustic
similarities, visual cues represent a possible source of distinction (cf. the [f]/[θ]
distinction in English; McGuire and Babel 2012). In what follows, I provide the LPC
spectra and spectral peaks for the bilabial realizations of /f/ for both Chilean and Mexican
experimental groups, and screen captures of video recordings for the Chilean participants
to confirm the presence or absence of labiodental contact. As our point of departure,
observe Figure 4.12 and Table 4.9 which detail the average LPC [ɸ] spectra and spectral
peaks for the four speakers that comprise the experimental groups.
131
Figure 4.12. Average [ɸ] (/f/) LPC by experimental speaker
40
20
dB
CE1
0
CE2
ME1
-20
ME2
-40
0
5
10
15
20
25
kHz
Note: CE1, Chile experimental 19727 (n=85); CE2, Chile experimental 17137 (n=72);
ME1, Mexico experimental 26911 (n=27); ME2, Mexico experimental 35792 (n=22).
Table 4.9. Average [ɸ] (/f/) LPC spectral peaks by experimental speaker
First peak (Hz)
Second peak (Hz)
CE1
843.75
2273.44
CE2
773.44
3234.36
ME1
679.69
3140.63
ME2
937.5
2203.13
Mean
808.60
2712.89
SD
109.09
550.11
With the exception of Mexican participant 26911, the shapes of the bilabial fricatives’
spectra in Figure 4.12 are relatively flat. The peak values for [ɸ] in Table 4.9, subtle as
they may be, relatively align with the expected peak values reported by Jassem (1968).
Crucially, both peaks are significantly lower than those for [x] (p=0.05 and p=0.021,
respectively). The conclusion is that the allophone that I have labeled bilabial [ɸ]: (i) is
132
phonetically distinct from velar [x] in my speakers, (ii) exhibits flat spectra and (iii)
includes spectral peaks at frequencies consistent with the bilabial fricatives analyzed in
previous studies. To distinguish [ɸ] from [f]–for which spectral analysis is insufficient
given the overlap in acoustic properties–compare screen captures (a) and (b) from Figures
4.13 and 4.14 which contrast bilabial [ɸ] and labiodental [f] productions within the two
Chilean experimental participants. (Recall that visual recordings were not possible with
the Mexican speakers.)
Figure 4.13. Chilean experimental participant 19727 (male): bilabial [ɸ], labiodental [f]
a.
futura ‘future’ (f. adj.), task 3
b.
café ‘coffee,’ task 1
Figure 4.14. Chilean experimental participant 17137 (male): bilabial [ɸ], labiodental [f]
a.
aforismo ‘aphorism,’ task 3
b.
beneficios ‘benefits,’ task 3
133
In (a) from Figures 4.13 and 4.14, no labiodental contact is present; instead, the speakers
articulate bilabial [ɸ] with labial narrowing and slight lip protrusion (Ladefoged and
Maddieson 1996: 141). I visually verified the lack of labiodental contact in the two
Chilean experimental participants’ bilabial tokens. In (b), the retraction of the lower lip
toward a visible portion of the upper teeth in articulating labiodental [f] is apparent.
The goal of this subsection was to respond to research question 1. I have argued
that speakers do make distinctions in the realizations of labiodental /f/ and velar /x/.
Regarding /f/, three allophones–faithful [f], velar [x] and bilabial [ɸ]–were identified by
their centers of gravity, spectral shape, spectral peaks and visual cues. Concerning the
velar phoneme /x/, these speakers produce two allophones–faithful [x] and palatal [ç]–as
suggested by the same acoustic measurements. The descriptive contribution is twofold: I
have supplemented previous studies on the phenomena (which relied on impressionistic
determinations and non-standard IPA) by analyzing the sounds phonetically and have
corroborated other research (Mazzaro 2011) in terms of rates of allophonic variation.
Now that I have shown that there are phonetic differences in the [f], [ɸ], [x] and
[ç] output realizations of /f/ and /x/, research question 2 asks if the appearance of these
allophones is productive or the result of lexicalization. I address this in the next section.
4.3. Productivity
The mere presence of variation is insufficient evidence for a synchronic analysis
since an alternation does not necessarily mean that it is productive. Recall from Chapter 3
that Sanicky (1988) notes that the past tense forms of ir ‘to go’ and ser ‘to be,’ all of
which contain a stressed initial /f/, are velarized at rates approaching 100% in dialects
that have labiodental velarization. Such a fact calls into question the productivity of the
134
process. Compounding the controversy is the low rate of velarization outside
conjugations of those two verbs that both Mazzaro (2011) and I (in the previous section)
report, not to mention the fact that Mazzaro lists no appearances of the bilabial allophone.
This fact, viewed in light of the dates of the sources that claim that bilabialization
exists (Perissinotto 1975, Resnick 1975, Sanicky 1988), in addition to Penny’s (2002)
diachronic approach to [ɸ], all conspire to suggest that /f/ allophony might not be
productive–a conclusion also reached by Calvo Shadid (1996), who attributes current
instances of [ɸ] as relics from Penny’s proposed evolutions (see §6.3). However, I
ultimately argue that the appearance of the unfaithful /f/ allophones is not predictable by
register by showing that assimilation rates do not vary across tasks;3 I also claim that it is
not lexically predictable (with the two ir and ser exceptions detailed in §4.4). These
conclusions suggest that the allophony identified in the previous section is truly the result
of a synchronic, productive grammatical process.
4.3.1. Frequency effects in the reading task
I begin by looking at the rates at which speakers produce unfaithful /f/ allophones
depending on frequency of the target words. Briefly discussed in Chapter 3, Bybee
(2010) and Coetzee and Kawahara (2013) argue that phonological processes are more
likely to appear in high frequency words than in low frequency ones. This general
tendency straightforwardly explains why very high frequency words (e.g., ir ‘to go’ and
ser ‘to be’) might attest lexicalization of applicable phonological processes. What such a
3
I leave aside the productivity of velar palatalization as there is no such controversy comparable
to that of the /f/ processes. It is accepted in the Spanish linguistic literature and community that
/x/ palatalization is highly productive and in fact notoriously characteristic of the majority of
Chilean speakers of all social classes (Canfield 1981: 33, Lipski 1994: 201, Sadowsky and
Salamanca 2011). The productivity of Mexican /x/ palatalization in particular regions (namely the
highlands), although not as widespread within the country as the Chilean case, is also
uncontroversial (Resnick 1975: 215, Canfield 1981: 62, Huskey 2010).
135
tendency implies is that, if a phonological process can be demonstrated in low frequency
words, such results suggest that the process remains productive in the grammar.
Under this assumption, I manipulated word-level frequency as a variable in the
sentences that the eight Chilean participants read as part of task 3 (the Mexican speakers
did not participate in the reading task). Given that /f/ allophony is the only concern, this
section analyzes the two experimental speakers and leaves aside the control participants
who uniformly produced /f/ faithfully as [f]. The prediction is that, all else being equal
(specifically, the following vowel, as comparing the rate of, e.g., frequent words in the [a]
(non-trigger) condition to infrequent words in the [u] (trigger) condition is a confounding
variable), /f/ allophones should appear at nearly equivalent rates in both frequent and
infrequent word conditions if velarization and bilabialization are productive. If frequent
words show significantly higher rates of allophony compared to infrequent words, then
one might instead argue that the processes have been lexicalized. Figure 4.15 compares
the ratio of /f/ allophones between frequency conditions (F frequent, I infrequent) in task
3 by the two Chilean experimental participants.
Figure 4.15. Comparison of /f/ allophone ratios by frequency
100%
80%
60%
[ɸ]
40%
[f]
20%
0%
F
I
[a]
F
I
[e]
F
I
[i]
F
I
[o]
F
I
[u]
136
Table 4.10. Comparison of /f/ allophone ratios by frequency: descriptive statistics
[f]
[a]
[e]
[i]
[o]
[u]
[ɸ]
Total /f/
N
%
N
%
N
%
F
8
66.67
4
33.33
12
100
I
8
80
2
20
10
100
F
7
87.5
1
12.5
8
100
I
8
80
2
20
10
100
F
9
69.23
4
30.77
13
100
I
8
88.89
1
11.11
9
100
F
2
25
6
75
8
100
I
3
33.33
6
66.67
9
100
F
5
21.74
18
78.26
23
100
I
7
35
13
65
20
100
A generalized linear mixed model (GLMM) resulted in no significant differences from
the comparison of the ratios of bilabial versus labiodental allophones (see Table 4.10)
between each word-level frequency (F(1, 115)=0.126; p=0.724). This comparison
suggests that at least bilabialization applies equally to frequent and infrequent words alike
(no velar allophones were produced in the reading task).
Given the previous discussion concerning the tendency of phonological processes
to most readily apply to words of higher frequency, these results would suggest that the
process is productive (i.e., not lexicalized) as it occurs in infrequent words at rates equal
to frequent words.
In the following subsection, I continue my argument for productivity by
comparing the ratio of allophones over different task types–sociolinguistic interview,
sentence reading and picture description–on the assumption that the formality/register of
the task correlates with different rates of assimilation.
137
4.3.2. Assimilation rates across all production tasks
In this section, rates of allophony by the Chilean and Mexican experimental
participants (n=2, n=2, respectively) are compared across production tasks to determine if
register/formality affects the frequency of the processes. My methodology employed
three production tasks: a personal interview (task 1), sentence reading task (task 3) and a
picture description task (task 4).
The tasks differed in terms of formality to provide a second measure of
productivity in addition to frequency. Task 3 (reading) is the most formal and,
consequently, most susceptible to relatively unnatural, monitored speech (Bayley 2002,
Foulkes, Scobbie and Watt 2010), yet was necessary to control the phonological
environments under investigation. Task 1, the sociolinguistic interview, is less formal
than the reading task; however, I would argue that the participants still maintained a
certain degree of guarded speech given the extralinguistic circumstances: an interview
with an unfamiliar, non-native speaker in a laboratory setting. The picture description
task is the most open-ended and informal of the three. Participants were given
suggestions for topics to address when describing the pictures although they were free to
describe them however they wished and were instructed to speak in a natural,
conversational style.
If the processes are lexicalized, the prediction is an inverse correlation between
the rate of assimilation and formality of the task; i.e., the rate of assimilation (greater
numbers of unfaithful allophones with a concomitant decrease in the faithful allophone
[f]) should increase as the formality of the task decreases. Figure 4.16 summarizes the
above relations and predictions.
138
Figure 4.16. Formality and allophonic rates across production tasks
4 picture
1 interview
3 reading
–formal
+formal
fewer [f]
more [f]
more [ɸ]/[x]
fewer [ɸ]/[x]
Figure 4.17. Comparison of /f/ allophone ratios by task
100%
80%
60%
[x]
40%
[ɸ]
20%
[f]
0%
4 1 3 4 1 3 4 1 3 4 1 3 4 1 3
[a]
[e]
[i]
[o]
[u]
139
A GLMM analysis on the three tasks resulted in no significant overall effect (F(2,
353)=0.838; p=0.434). Recall that it was predicted that [f] should increase while [ɸ]/[x]
should decrease as register/formality increases if register/formality affects the process.
With the exception of [a] and [e], this prediction does not hold in Table 4.11 when
progressing from task 4 to 1 to 3 (least to most formal).
Table 4.11. Comparison of /f/ allophone ratios by task: descriptive statistics
[f]
[a]
[e]
[i]
[o]
[u]
[ɸ]
[x]
Total /f/
N
%
N
%
N
%
N
%
4
8
28.57
20
71.43
0
0
28
100
1
8
72.73
3
27.27
0
0
11
100
3
18
64.29
10
35.71
0
0
28
100
4
11
44
14
56
0
0
25
100
1
7
50
7
50
0
0
14
100
3
26
76.47
8
23.53
0
0
34
100
4
16
80
4
20
0
0
20
100
1
7
87.5
1
12.5
0
0
8
100
3
38
79.17
10
20.83
0
0
48
100
4
6
19.35
25
80.65
0
0
31
100
1
0
0
4
100
0
0
4
100
3
6
26.09
17
73.91
0
0
23
100
4
3
12.5
18
75
3
12.5
24
100
1
7
26.92
17
65.38
2
7.69
26
100
3
12
27.27
32
72.73
0
0
44
100
The conclusion from this is that task type has no statistical effect on the amount of
assimilated allophones [ɸ]/[x] produced, suggesting that the processes are indeed
productive rather than lexicalized, given their appearance in formal tasks at rates equal to
140
those in informal ones. These results, paired with those from the analysis of frequency in
the previous section, which reached the same conclusion, support an argument for
productivity. This is not the case, however, for a particular subset of the data to which I
have alluded previously–the highly frequent verbs ser ‘to be’ and ir ‘to go.’ I divide out
these token in the next section to support Sanicky’s (1988) claim that assimilation rates in
the past tense forms of these verbs approach one hundred percent.
4.4. Ir and ser lexicalization
While I argued above that the processes are productive, the literature contradicts
such a conclusion for ir ‘to go’ and ser ‘to be’ in the past tense (Sanicky 1988). To begin,
review Table 4.12 (an expanded version of Table 3.32).
Table 4.12. Verbal paradigms for ser ‘to be’ and ir ‘to go’4
Preterit
indicative
Imperfect
subjunctive
4
Person, number
Pronoun
Verb
Translation
1 sg.
yo
fui
I was/went
2 sg.
tú
fuiste
you were/went
3 sg. / 2 frml.
él, ella / Usted
fue
s/he/it was/went
1 pl.
nosotros
fuimos
we were/went
2 pl.
vosotros
fuisteis
you were/went
3 pl. / 2 pl.
ellos, ellas / Uds.
fueron
they were/went
1 sg.
yo
fuera
me to be/go
2 sg.
tú
fueras
you to be/go
3 sg. / 2 frml.
él, ella / Usted
fuera
her/him/it to be/go
1 pl.
nosotros
fuéramos
us to be/go
2 pl.
vosotros
fuerais
you to be/go
3 pl. / 2 pl.
ellos, ellas / Uds.
fueran
them to be/go
I exclude the shaded quadrants as vosotros is most prevalent in Castilian Spanish of central and
northern Spain–dialects outside those under consideration–and fuéramos was absent from the
lexical database consulted.
141
Every form begins with a (stressed) initial /f/ followed by /u/; consequently, all
contextual factors are inherently controlled by the data. The question is if the allophone
([f], [ɸ] or [x]) produced in these forms by the four experimental participants is
dependent upon task. The predictions are identical to those of 4.3.2 above: the more
formal the task, the fewer assimilated allophones are predicted.
Figure 4.18. Comparison of /f/ allophone ratios ir/ser by task
100%
80%
60%
[x]
40%
[ɸ]
20%
[f]
0%
IS
E
IS
4
E
IS
1
E
3
Table 4.13. Comparison of /f/ allophone ratios ir/ser by task: descriptive statistics
[f]
4
1
3
[ɸ]
[x]
Total /f/
N
%
N
%
N
%
N
%
IS
0
0
1
100
0
0
1
100
E
49
34.27
91
63.64
3
2.1
143
100
IS
4
25
11
68.75
1
6.25
16
100
E
34
56.67
25
41.67
1
1.67
60
100
IS
2
10.53
17
89.47
0
0
19
100
E
105
63.25
61
36.75
0
0
166
100
142
Given the low number of ir/ser tokens (labeled IS in Table 4.13; n=36) compared to all
other conditions (E, elsewhere; n=369), statistical analysis would not be equitable given
the substantial difference in sample size. However, descriptively comparing the ir/ser
condition to the elsewhere condition within tasks, it is clear that, generally, ir and ser
tend to be assimilated at higher rates than other tokens. Specifically, assimilated ir and
ser account for 100% of task 4 (although this is only one token), 75% (12/16) in task 3
and 89.47% (17/19) in task 1.
When these ir/ser assimilated rates are compared to the rates of assimilation
elsewhere (65.74%, 43.34% and 36.75% by task), it is apparent that assimilation is
preferred in the common verbs ‘to be’ and ‘to go’ at rates higher than in all other
contexts. Furthermore, these results contradict the prediction of task effects, with the
most formal task (task 3) exhibiting the highest number of assimilated ir/ser tokens.
Thus, while I have argued that the processes are an active part of the grammar, for the
experimental participants it also appears that they have been lexicalized in the frequent
verbs ir and ser, supporting Sanicky’s (1988) results.
In the next three sections, I compare the dependent variables of preceding and
following vowel, stress and position within the word to identify which factor (or
combination thereof) determines the output allophones. I begin by analyzing the
surrounding vocalic context.
4.5. Regressive versus progressive assimilation
Recall from Chapter 1 that the literature has identified two types of grammars
concerning the /f/ processes: those that are limited to the back vowels /u, o/ and those that
permit the unfaithful allophones irrespective of the following vowel (Resnick 1975).
143
Such an observation raises questions for an analysis that assumes assimilation; permitting
the unfaithful allophone to surface with no regard for the vowel that follows contradicts a
proposal that attributes the unfaithful allophone to the identity of the vowel. However,
other sources suggest that [ɸ] and [x] are impossible before non-back vowels /a, e, i/
(Perissinotto 1975, Piñeros 2009).
Furthermore, the contribution (or irrelevance) of the preceding vowel has not been
examined for these processes. This is not an inconsequential issue given the
crosslinguistic data, also from Chapter 1, that show that progressive assimilation
comparable to the Spanish processes in question is just as possible. This issue raises two
questions: does Spanish in fact only include regressive assimilation, and, if so, why,
given that progressive assimilation is (cross)linguistically permissible?
With respect to velar palatalization, the literature is relatively in agreement
concerning its restriction to the front vowels /i, e/ (Resnick 1975, Piñeros 2009,
Sadowsky and Salamanca 2011); this common process is seen in many languages (see
Bateman 2007) which I also reviewed in Chapter 1. However, the possibility (or not) of
progressive conditioning for palatalization in Spanish has not been addressed. The goal of
this section is to disentangle these two issues: do the surrounding vowels (preceding
and/or following) condition assimilation?
I begin with the case of /f/. To answer the above questions, I first compare the
ratios of the allophones among following vowels (CV). Remember that, for /f/, I limit the
analysis to the four experimental participants, as the four control participants exhibit no
unfaithful /f/ allophones. Also recall that I collapse the three production tasks.
144
Figure 4.19. Comparison of /f/ allophone ratios by following vowel
100%
80%
60%
[x]
[ɸ]
40%
[f]
20%
0%
[a]
[e]
[i]
[o]
[u]
Table 4.14. Comparison of /f/ allophone ratios by following vowel: descriptive statistics
[f]
[ɸ]
[x]
Total /f/
N
%
N
%
N
%
N
%
[a]
34
50.75
33
49.25
0
0
67
100
[e]
44
60.27
29
39.73
0
0
73
100
[i]
61
80.26
15
19.74
0
0
76
100
[o]
12
20.69
46
79.31
0
0
58
100
[u]
22
23.4
67
71.28
5
5.32
94
100
A GLMM analysis determined that there are significant differences between the rates of
the allophones within each vowel context (F(4, 362)=16.247; p<0.001). Post hoc tests
with Bonferroni adjustments revealed that [i] and [e] differ significantly from [u] and [o]
(p<0.001), and [u] and [o] similarly differ from [e] and [i] (p<0.001); [u] and [o] do not
differ from each other (p=0.767). In terms of probability of assimilation, it is clear that
faithful [f] is preferred in the non-back vowel contexts /i, e/, while the back vowels /u, o/
favor the unfaithful bilabial [ɸ] (and velar [x] to a much lesser extent).
145
These results support the descriptive accounts that claim that the following vowel
is deterministic (Perissinotto 1975, Piñeros 2009) and straightforwardly lend themselves
to theoretical analysis in that that roundedness of /u, o/ accounts for bilabial [ɸ] while the
backness explains [x] (see Chapter 5 for the seemingly neutral effect of [a]).
Even though the above suggests that that identity of the allophone depends on the
following vowel, this does not necessarily exclude VC progressive assimilation. As I
demonstrate in Chapter 1, languages may opt to assimilate regressively, progressively or
both within the same grammar. To examine this possibility, observe Figure 4.20 and
Table 4.15.
Figure 4.20. Comparison of /f/ allophone ratios by preceding vowel
100%
80%
60%
[x]
[ɸ]
40%
[f]
20%
0%
[a]
[e]
[i]
[o]
[u]
A GLMM revealed no effect for the preceding vowel (F(4, 157)=0.466; p=0.76) in the
VC context; i.e., there were no differences between [f] and [ɸ] rates depending on the
prior vowel. When the percentage of each allophone is compared within each vowel
individually, every rate is around 50% (Table 4.15).
146
Table 4.15. Comparison of /f/ allophone ratios by preceding vowel: descriptive statistics
[f]
[ɸ]
[x]
Total /f/
N
%
N
%
N
%
N
%
[a]
18
42.86
22
52.38
2
4.76
42
100
[e]
22
56.41
17
43.59
0
0
39
100
[i]
20
46.51
23
53.49
0
0
43
100
[o]
11
42.31
15
57.69
0
0
26
100
[u]
8
53.33
7
46.67
0
0
15
100
The conclusion from the above is that the preceding vowel has no influence on which
allophone appears, suggesting that progressive assimilation is not a factor in these
processes. This is not unexpected considering that Ghavami (2002) and Lee (1997, 1999),
from a phonetic perspective (for a theoretical approach, see also Fujimura 1979, 1988,
1990 and Clements 1990), show greater coarticulatory overlap in CV sequences than in
VC sequences (even though VC assimilation is also attested crosslinguistically; see
Chapter 1 and the sources cited therein).
With the above, I have responded to the research questions that guide this
subsection with respect to /f/: bilabialization and velarization are triggered by following
back vowels /u, o/, preceding vowels being irrelevant. The data do not, however,
categorically exclude the possibility of bilabial [ɸ] appearing before the non-back vowels
/a, e, i/, characteristic of the freely assimilating dialects; rather, the data support a greater
probability for [ɸ] in the context of the back vowels. This detail will be addressed in
Chapters 5 and 6.
For the moment, I apply the same manner of analysis to the palatalization data.
Figure 4.21 and Table 4.16 compare the allophone rates within the contexts of each
147
following vowel. In this case, since all groups produced palatal [ç] allophones for /x/
(Figures 4.6 and 4.7), every participant is included.
Figure 4.21. Comparison of /x/ allophone ratios by following vowel
100%
80%
60%
[ç]
40%
[x]
20%
0%
[e]
[i]
[a]
[o]
[u]
Table 4.16. Comparison of /x/ allophone ratios by following vowel: descriptive statistics
[x]
[ç]
Total /x/
N
%
N
%
N
%
[e]
55
30.56
125
69.44
180
100
[i]
24
22.02
85
77.98
109
100
[a]
164
100
0
0
164
100
[o]
192
100
0
0
192
100
[u]
149
100
0
0
149
100
The GLMM analysis in this case resulted in significance with respect to the following
vowel (F(4, 18.913)=20.579; p<0.001). Bonferroni adjusted post hoc tests revealed a high
distinction for palatal [ç] in the contexts of /i/ and /e/ (p<0.001 in both cases). /a, o, u/ are
not distinct from each other in the context of [x] (p=1). The conclusion is that, like /f/, the
148
following vowel plays a significant role in assimilation, uniquely conditioned by /i/ and
/e/ to the exclusion of /a, u, o/. However, faithful [x] is also possible with following /i, e/,
yet at much lower rates. It must now be determined if the preceding vowel is predictive.
Figure 4.22. Comparison of /x/ allophone ratios by preceding vowel
100%
80%
60%
[ç]
40%
[x]
20%
0%
[e]
[i]
[a]
[o]
[u]
Table 4.17. Comparison of /x/ allophone ratios by preceding vowel: descriptive statistics
[x]
[ç]
Total /x/
N
%
N
%
N
%
[e]
87
65.91
45
34.09
132
100
[i]
56
65.12
30
34.88
86
100
[a]
93
86.11
15
13.89
108
100
[o]
72
84.71
13
15.29
85
100
[u]
20
54.05
17
45.95
37
100
The GLMM on the preceding vowel did result in significant differences (F(4,
443)=7.465; p<0.001). However, Bonferroni adjusted post hoc tests determined that no
vowel is significantly different from another in terms of predicting which allophone will
149
occur ([a]~[e] and [e]~[o], p=0.107; [a]~[u] and [o]~[u], p=0.131; [a]~[i] and [i]~[o],
p=0.132; [a]~[o], [e]~[i], [e]~[u] and [i]~[u], p=1). That is, no preceding vowel differs
from any other in affecting (or not) the following consonant.
The reason for this discrepancy (overall significance with no adjusted post hoc
significance) might be due to the fact that a chi-square test determined that, given my
particular data, the preceding vowel was not independent from the following vowel
(χ2(16, 448)=232.608; p<0.001). In other words, due to the freedom afforded in two of
the production tasks (the sociolinguistic interview and the picture description task),
tokens such as dije [d̪ ixe] ‘I said’ included ambiguous contexts; i.e., if the speaker
produced the palatalized output [d̪ içe], determining if palatalization is the effect of the
preceding [i] or the following [e] is not possible in isolation. I was only able to control for
such a confounding factor in the sentence reading task in which the (targeted) preceding
vowel tokens were accompanied by only non-assimilating following vowels (e.g., lejos
‘far’).
If the post hoc analyses are considered, the conclusion is that the preceding vowel
is not a predictor of assimilation; Table 4.16, however, indicated that this was not the
case for the following vowel. Concerning /x/, the data suggest that, like /f/, palatalization
is also conditioned by the following vowel, but limited to the front vowels /i/ and /e/.
Again, the process is optional, as [x] is possible in these very contexts even though the
preference is to palatalize. The preceding vowel remains irrelevant.
Now that it has been determined that the crucial environment for /f/ is a following
back vowel while /x/ requires a following front vowel, various questions fall out from
such a discovery. Assuming that place homorganicity is optimal for a particular language
150
in a given CV context, two possibilities exist to satisfy an assimilating constraint: the
consonant adopts the place of the vowel or the vowel the place of the consonant. While
Spanish happens to opt for consonantal place unfaithfulness in these contexts, the
alternative is also possible. Below I respond to why such an alternative is ruled out. In
Chapter 5, I account for how this exclusion is encoded theoretically.
In explaining why regressive CV assimilation is preferred and progressive VC
assimilation avoided, Ghavami (2002) and Lee (1997, 1999) used locus equations to
show higher degrees of coarticulation in CV sequences than in VC sequences.
Theoretically this is logical when one considers syllabic structure; CV sequences are
heterosyllabic while VC sequences usually cross a syllable boundary. In the particular
CV case of fricative-vowel, numerous acoustic studies have shown that the CV transition
is crucial in identifying the place of the fricative relative to other manners of articulation
(Harris 1958; Manrique and Massone 1981; Soli 1981; Zeng and Turner 1990; Stevens
1985; Fernández, Feijóo, Balsa and Barros 1998; Deng and O’Shaughnessy 2003; Feijóo
and Fernández 2003; Nittrouer and Lowenstien 2009; McMurray and Jongman 2011).
Furthermore, cues to fricative place in non-stridents are relatively weak inherently
(Ladefoged 2001, Deng and O’Shaughnessy 2003, McMurray and Jongman 2011).
On the other hand, vocalic formants that cue place are stronger and more stable
than, e.g., center of gravity or spectral speak for fricatives. Stevens (1989) argues that, in
a gestural (coarticulated) blend, the more stable the gesture, the more resistant the
segment is to influence from the competing gesture. This concept applied to the case of
(weakly cued) fricative overlapped with (strongly and stably cued) vowel accounts for the
unfaithfulness in the fricative as opposed to the vocoid. While it is true that there is
151
measurable consonantal influence on the vowel, as well (Stevens and House 1963;
Strange, Verbrugge, Shankweiler and Edman 1976; Hillenbrand, Clark and Nearey
2001), the vocalic influence on the consonant is more robust.
To take a canonical case for example, the velar stop /k/ in [k̟ i] has a slightly more
anterior articulation than [ku] due to the fronting influence of /i/. A small phonetic shift
such as this, however, does not explain the complete assimilation of place that is at stake
in /f/→[ɸ]/[x] and /x/→[ç], nor does it account for the propagation of the process
throughout a speech community. Indeed, I argue that /f/ and /x/ allophony began as subtle
coarticulation in intermediate steps toward [ɸ]/[x] and [ç], yet this cannot be the only
explanation. As Baker et al. (2011) note, coarticulation is universal, yet not every
overlapped sequence results in a productive process that also happens to disseminate
throughout a lect. I provide an account of how the processes evolved and propagated in
section 4.8 after I have explored the remaining descriptive issues in the next two sections.
In this section, I argued that the following vowels account for the realization of
the unfaithful allophones. Sections 4.6 and 4.7 determine if stress and position within the
word also influence rates of assimilation.
4.6. Prosodic effect
It was determined in the previous section that the following vowel conditions the
processes. Another possible source of influence concerns prominence, both prosodic and
positional. Stressed syllables and word-initial information (among other domains) seem
to be phonologically prioritized given their psycholinguistic role in lexical storage,
access, retrieval and processing (Beckman 1998 and the sources cited therein).
Consequently, Beckman argues that languages tend to avoid unfaithfulness to input form
152
in these prominent positions. Concerning the Spanish processes under consideration here,
the theory would therefore predict fewer instances of unfaithfulness (i.e., a higher
probability of [f] (/f/) and [x] (/x/)) in stressed and word-initial syllables and a greater
number of unfaithful allophones ([ɸ]/[x] (/f/) and [ç] (/x/)) elsewhere. This contradicts
Mazzaro (2011), who found an effect for both stress and position with the highest rate of
occurrence of the unfaithful velar [x] allophone of /f/ in stressed, word-initial position
(she did not investigate bilabial [ɸ] or palatal [ç]).
Given this disagreement, I sought to shed additional light on this issue by
exploring the influence of stress (and, in the next section, position of the fricative in the
word) on /f/ and to add to the description of the process, as–to my knowledge–neither
variable has been examined in the case of palatalization.
I begin with labiodental /f/. As the control participants exhibited no /f/
alternations, they are excluded here, yet included in the examination of velar /x/ below
since all groups displayed the same [x]~[ç] alternation.
Figure 4.23. Comparison of /f/ allophone ratios by stress
153
Table 4.18. Comparison of /f/ allophone ratios by stress: descriptive statistics
[f]
[a]
[e]
[i]
[o]
[u]
[ɸ]
[x]
Total /f/
N
%
N
%
N
%
N
%
S
40
81.63
9
18.37
0
0
49
100
U
77
76.24
24
23.76
0
0
101
100
S
54
91.53
5
8.47
0
0
59
100
U
66
73.33
24
26.67
0
0
90
100
S
86
92.47
7
7.53
0
0
93
100
U
60
88.24
8
11.76
0
0
68
100
S
77
77.78
22
22.22
0
0
99
100
U
33
57.89
24
42.11
0
0
57
100
S
86
59.31
54
37.24
5
3.45
145
100
U
17
56.67
13
43.33
0
0
30
100
To determine if stress is predictive, a GLMM compared stressed (S) versus unstressed
(U) tokens; the results determined that stress is significant (F(1, 358)=13.830; p<0.001);
stress decreases the odds of assimilation by 68.4%. I now apply the same analysis to
palatalization below.
Figure 4.24. Comparison of /x/ allophone ratios by stress
154
A GLMM on the /xi, xe/ conditions (since /xa, xu, xo/ contained no variability) revealed
no significant differences for stress (S stressed versus U unstressed) (F(1, 285)=1.237;
p=0.267), indicating that stress in not an allophonic predictor; whether [x] or [ç], both
occur at equal rates regardless of stress when the vowel is controlled. Unlike /f/, the
conclusion is that prosody is irrelevant for /x/.
Table 4.19. Comparison of /x/ allophone ratios by stress: descriptive statistics
[x]
[e]
[i]
[a]
[o]
[u]
[ç]
Total /x/
N
%
N
%
N
%
S
26
24.76
79
75.24
105
100
U
29
38.67
46
61.33
75
100
S
10
16.39
51
83.61
61
100
U
14
29.17
34
70.83
48
100
S
53
100
0
0
53
100
U
111
100
0
0
111
100
S
80
100
0
0
80
100
U
112
100
0
0
112
100
S
61
100
0
0
61
100
U
88
100
0
0
88
100
A second manifestation of prominence is word-initiality, the final variable addressed in
the next section.
4.7. Position within the word
Extending the same prosodic predictions to position in the word, we would expect
the faithful allophones [f] and [x] to occur at higher rates than the unfaithful allophones
[ɸ]/[x] and [ç] in word-initial position. Mazzaro (2011), on the other hand, shows the
155
unfaithful [x] (/f/) more frequently than [f] in initial position. Figure 4.25 and Table 4.20
address these predictions for /f/.
Figure 4.25. Comparison of /f/ allophone ratios by position
Table 4.20. Comparison of /f/ allophone ratios by position: descriptive statistics
[f]
[a]
[e]
[i]
[o]
[u]
[ɸ]
[x]
Total /f/
N
%
N
%
N
%
N
%
I
61
77.22
18
22.78
0
0
79
100
N
56
78.87
15
21.13
0
0
71
100
I
31
91.18
3
8.82
0
0
34
100
N
89
77.39
26
22.61
0
0
115
100
I
69
93.24
5
6.76
0
0
74
100
N
77
88.51
10
11.49
0
0
87
100
I
77
77
23
23
0
0
100
100
N
33
58.93
23
41.07
0
0
56
100
I
71
58.2
49
40.16
2
1.64
122
100
N
32
60.38
18
33.96
3
5.66
53
100
156
A GLMM revealed no significant allophonic difference concerning position in the word
(F(1, 358)=2.776; p=0.097); i.e., for any given allophone, its rate of appearance is
unaffected by initial or non-initial position. The same analysis is applied to /x/ below.
Figure 4.26. Comparison of /x/ allophone ratios by position
Table 4.21. Comparison of /x/ allophone ratios by position: descriptive statistics
[x]
[e]
[i]
[a]
[o]
[u]
[ç]
Total /x/
N
%
N
%
N
%
I
24
30.38
55
69.62
79
100
N
31
30.69
70
69.31
101
100
I
3
9.09
30
90.91
33
100
N
21
27.63
55
72.37
76
100
I
41
100
0
0
41
100
N
123
100
0
0
123
100
I
46
100
0
0
46
100
N
146
100
0
0
146
100
I
120
100
0
0
120
100
N
29
100
0
0
29
100
157
The GLMM revealed no significant difference between allophones based on position
(F(1, 285)=3.302; p=0.07); i.e., the rate of appearance for [x] is equal irrespective of
position (I initial or N non-initial). Such data suggest that position, like stress, is
irrelevant for /x/.
As an interim summary, I have at this point provided evidence that suggests the
following conclusions with respect to /f/ bilabialization, /f/ velarization and /x/
palatalization.
(1)
a.
b.
c.
d.
e.
There are four allophones involved in the processes: [f], [ɸ], [x] and [ç].
All three processes are productive.
Ir ‘to go’ and ser ‘to be’ seem to be lexicalized.
Assimilation is exclusively regressive and vowel dependent.
Stress and position are mostly irrelevant (the exception being stressed /f/).
With (1), I have responded to research questions 1 through 4 concerning the descriptive
and theoretical aspects of the dissertation. The conclusions from (1) are sufficient for the
Optimality Theoretic analysis, the formal account of the synchronic grammatical
representation of the processes in the minds of the speakers. However, the tools available
in OT cannot (and do not necessarily claim to) account for why and how a particular set
of speakers came to acquire such properties while other sets of speakers did not. As
Mielke (2008) argues, any account for any given property that strives for completeness
should address the origin, dissemination and mental representation of said property.
With the origins and dissemination of the property in mind, the final section of
this chapter responds to research questions 6 and 7 before proceeding to the OT account
in Chapter 5: how can (and why should) the analysis be limited to /f/ and /x/, and what
role does perception play in the properties’ genesis and propagation?
158
4.8. Toward the grammatical analysis
As mentioned in Chapter 1, Baker et al. (2011) and Garrett and Johnson (2013)
outline possible paths toward the grammaticalization of phonetic processes. One such path
begins with a phonetic motivation. In the case of Baker et al. (2011), that motivation is
coarticulation, not unlike the slightly fronted /ki/→[k̟ i] example. This articulatory effect,
however, must be combined with an acoustic component.
Ohala (1981, 1989, 1993) has shown that listeners are sensitive to coarticulated
sequences and are able to reconstruct the signal filtering out the coarticulation to arrive at
the underlying form intended by the speaker (see Figure 1.1, Chapter 1). In order to
disrupt this process, Baker and colleagues argue that variability in the signal is necessary
such that the filtering rule that listeners might apply is disrupted. Once listeners are
unable to reconstruct the signal, the coarticulated sequence is taken at face value. When
listeners become speakers and produce the coarticulated variant to greater and greater
degrees of coarticulation, what began as a phonetic overlap is now a phonological variant
for that speaker.
Yet not every process recently grammaticalized by a few speakers is uniformly
spread throughout a speech community. Beddor (2012) shows that listeners diverge
concerning the extent to which they rely on coarticulated cues in parsing speech. She
terms innovative listeners those that are more likely to take the signal at face value and
conservative listeners those that prefer to reconstruct the signal. Assuming that an
innovative listener has grammaticalized a coarticulated sequence, the third component in
Baker et al.’s analysis concerns the sociolinguistic factors associated with the speaker. In
order for a sound change to spread throughout a dialect/speech community, the
159
phonological process must coincide with social factors that favor its dispersion. Without
this third component to expand the use of the property, the grammaticalization will end
with the few speakers who reached this point. This complex confluence of events, Baker
et al. argue, accounts for why sound change is relatively infrequent and slow compared to
the predicted frequency and speed with which sound change should occur under
accumulation of error theories.
In this section, I apply the proposal in Baker et al. to the Spanish properties under
investigation to determine if the three criteria that they posit–coarticulation, variability
and association with sufficient social factors–can account for the origin and spread of /f/
and /x/ allophony throughout Latin America. I begin with coarticulation.
4.8.1. Coarticulatory effects
To determine if the /f/ and /x/ phonological processes under investigation are
motivated by coarticulation, locus equations are necessary. Recall from section 4.1.3 that
locus equations (LE) (Lindblom 1963; Sussman, McCaffrey and Matthews 1991;
Sussman and Shore 1996) are linear regressions that report coarticulation based on the
difference in second formant (F2) frequencies measured at the CV transition (y-axis) and
vocalic midpoint (x-axis). A (large) difference in the two values suggests little to no
coarticulation given that the F2 transition value is independent from that of the vowel; a
flatter regression line with low slope and low r2 values are expected. On the other hand, a
minor difference in the F2 values from transition to the middle of the vowel suggests that
the F2 at the point of transition depends on that of the vowel, suggesting a greater degree
of coarticulation; greater r2 values and a steep regression line with higher slope are
predicted.
160
Figure 4.27. Locus equations for Chilean assimilator 19727: /fu/ versus /fa/
2000
y=0.8954x+91.669
r²=0.9208
F2
transition
(Hz)
1500
[fu]
1000
[fa]
y=0.1565x+985.73
r²=0.03
500
500
1000
1500
F2 mid-vowel (Hz)
2000
Figure 4.28. Locus equations for Chilean assimilator 19727: /xi/ versus /xa/
2500
y=0.8494x+329.81
r²=0.8147
F2
transition
(Hz)
2000
[xi]
1500
[xa]
y=0.2961x+929.33
r²=0.0647
1000
1000
1500
2000
F2 mid-vowel (Hz)
2500
Crucially, coarticulation is not predicted across all vocalic contexts equally. As section
4.5 shows, the back vowels /u, o/ condition /f/ allophony while the front vowels /i, e/
trigger palatalization. Therefore, greater evidence of coarticulation is expected in the
limited subset /fu, fo, xi, xe/ with less coarticulation expected for all other combinations
(/fi, fe, fa, xu, xo, xa/). With these predictions in mind, observe Figures 4.27 and 4.28
161
below, which report the LEs for Chilean experimental participant 19727 (male) in the /fu/
versus /fa/ and /xi/ versus /xa/ contexts.
As the above figures comparing an assimilating versus non-assimilating CV
context show, the predictions are borne out. The slope and r2 values in the /fu/ and /xi/
contexts are relatively high (/fu/: slope=0.895, r2=0.92; /xi/: slope=0.849, r2=0.81) while
the same values in the /fa/ and /xa/ conditions are considerably lower (/fa/: slope=0.157,
r2=0.03; /xa/: slope=0.296, r2=0.06).
What this suggests is that Chilean speaker 19727 exhibits a high degree of
coarticulation when producing [fu] and [xi], which results from significant gestural
overlap between the consonant and vowel. [fa] and [xa], on the other hand, attest less
coarticulation, the source of which is less gestural overlap when articulating the
consonant and vowel. This point will prove crucial for the theoretical analysis in the next
chapter.
Before proceeding to the complete results, note that slopes were calculated over
raw F2 values in Hz and not normalized across speakers. While normalization is usually
standard procedure when comparing, e.g., males and females given the variation in the
size of the vocal tract inherent in this comparison (see Rosner and Pickering 1994,
Thomas 2002), two facts obviate the need for normalization. Ronquest (2012) argues that
normalization is not necessary when gender is not an independent variable, which
happens to be the case here. Furthermore, the relative comparisons are between
fricative/vocalic conditions, assimilating/non-assimilating speakers and Chilean/Mexican
dialects. Since the slope values do not involve collapsing male and female speakers, each
slope is instead the result of F2 comparisons within each speaker.
162
For the moment, Table 4.22 provides the values for the remaining LEs for all
participants in the /fV/ condition to verify that the expected results from participant
19727 hold for the other speakers.
Table 4.22. /f/ locus equations for all participants
/u/
/o/
/i/
/e/
/a/
slope
r2
slope
r2
slope
r2
slope
r2
slope
r2
19727
0.90
0.92
0.84
0.74
0.59
0.27
0.17
0.03
0.16
0.03
17137
0.79
0.59
0.70
0.52
0.22
0.01
0.08
0.00
-0.95
0.16
02838
0.86
0.74
0.98
0.82
0.08
0.01
-0.01
0.00
0.06
0.00
15246
0.68
0.44
0.68
0.37
0.21
0.06
-0.18
0.02
0.03
0.00
Mexico 26911
exp.
35792
1.03
0.99
0.86
0.92
-0.08
0.02
0.31
0.72
0.05
0.09
0.89
0.87
0.99
1.00
-0.23
1.00
0.52
0.69
-0.18
0.03
Mexico 11111
contr. 71017
0.58
0.37
0.54
0.92
0.44
0.13
0.15
0.02
0.05
0.00
0.68
0.53
0.58
0.28
0.30
0.06
0.25
0.13
0.27
0.41
Mean
0.80
0.68
0.77
0.70
0.19
0.20
0.16
0.20
-0.06
0.09
SD
0.15
0.23
0.17
0.27
0.27
0.34
0.21
0.31
0.38
0.14
Chile
exp.
Chile
contr.
The first relevant comparison is between the /fu, fo/ versus /fi, fe, fa/ contexts. A mixedmodel ANOVA revealed that, overall, the average slope (F(1, 69.706)=92.887; p<0.001)
and r2 values (F(1, 65.818)=59.336; p<0.001) are significantly higher in the shaded
quadrants than the unshaded ones. The conclusion from this is that the sequences [fu, fo]
exhibit higher coarticulation than all other [fV] sequences. This is not surprising given
that I argued previously that /u, o/ trigger /f/ allophony, and it is now clear why: the
gestures of [f] overlap at higher degrees with [u, o] than with [i, e, a], which allows the
vocalic features of the back vowels to condition assimilation. Applying this conclusion to
the proposal by Baker et al. (2011), the bilabial properties do have a phonetic motivation.
163
The next comparison explores dialectal differences. Table 4.23 compares the averages
between the Chilean and Mexican groups irrespective of experimental/control
membership.
Table 4.23. /f/ locus equations by geolect
/u/
Chile
Mexico
/o/
/i/
/e/
/a/
slope
r2
slope
r2
slope
r2
slope
r2
slope
r2
Mean
0.81
0.67
0.80
0.61
0.28
0.09
0.02
0.01
-0.18
0.05
SD
0.10
0.21
0.14
0.21
0.22
0.12
0.15
0.02
0.52
0.08
Mean
0.80
0.69
0.74
0.78
0.11
0.30
0.31
0.39
0.05
0.13
SD
0.20
0.29
0.22
0.34
0.31
0.47
0.16
0.37
0.18
0.19
Between the two countries, the difference in slope across all vowel contexts did not reach
significance (F(1, 74)=0.057; p=0.811) (neither did r2 (F(1, 4.467)=2.793; p=0.162)),
which suggests that the degree of coarticulation between each respective /fV/ sequence is
similar in both Chilean and Mexican speakers. The last labiodental LE analysis compares
the experimental and control groups with regions now collapsed.
Table 4.24. /f/ locus equations by group
/u/
Exp.
Control
/o/
/i/
/e/
/a/
slope
r2
slope
r2
slope
r2
slope
r2
slope
r2
Mean
0.90
0.84
0.85
0.80
0.13
0.33
0.27
0.36
-0.23
0.08
SD
0.10
0.18
0.12
0.21
0.36
0.47
0.19
0.40
0.50
0.06
Mean
0.70
0.52
0.70
0.60
0.26
0.07
0.05
0.04
0.10
0.10
SD
0.12
0.16
0.20
0.32
0.15
0.05
0.19
0.06
0.11
0.21
164
The group comparison reveals no significance difference in slope (F(1, 6.22)=1.421;
p=0.277). It is worth noting, however, that, while the /u, o/ slopes for the controls tend
toward coarticulation, they do so to a lesser degree than the experimental group. Baker et
al. (2011) argue that non-assimilators who tend toward coarticulation but not to the extent
of missing the intended faithful segment (like the assimilators) are a crucial component in
the grammaticalization process, as it is this tendency toward coarticulation by the nonassimilators that creates the environment in which eventual assimilators misperceive the
signal.
From the above, I conclude the following: that, overall, /fu, fo/ exhibit greater
gestural overlap than /fi, fe, fa/; that Chileans and Mexicans coarticulate to comparable
degrees; and that assimilators coarticulate at slightly greater degrees than nonassimilators.
I now apply the same analyses to /x/, beginning with individual results below
(note that the blank regions for participant 35792 are due to the fact that she produced
only one /xi/ token, meaning that no regression line–which requires at least two data
points–could be computed).
As was the case with /f/ above, the relevant triggering vowels (/i, e/ for
palatalization) demonstrate significantly higher slope (F(1, 73.180)=9.114; p=0.003) and
r2 values (F(1, 79)=51.444; p<0.001) in Table 4.25 below, suggesting greater gestural
overlap. Conversely, /u, o, a/ exhibit low values for both measures, indicating little
coarticulation. In Table 4.26, averages across Chilean and Mexican speakers are
compared.
165
Table 4.25. /x/ locus equations for all participants
/i/
/e/
slope
r
0.81
0.90
0.92
0.88
02838
1.06
15246
Mexico 26911
exp.
35792
Chile
contr.
r
19727
0.85
17137
/u/
2
Chile
exp.
slope
2
/o/
slope
2
r
0.92
0.15
0.90
0.79
0.96
1.07
0.92
0.91
0.53
0.96
/a/
slope
r
2
slope
r2
0.03
0.28
0.02
0.30
0.06
0.19
0.01
0.10
0.00
-0.05
0.00
0.95
0.19
0.09
0.20
0.12
0.50
0.29
0.81
0.82
-0.18
0.01
0.40
0.33
0.30
0.07
0.44
0.68
0.08
0.25
0.48
0.72
0.26
0.05
0.44
0.73
0.22
0.06
0.07
0.01
-0.02
0.00
Mexico 11111 -0.02
contr. 71017 0.86
0.00
0.36
0.47
0.15
0.10
0.39
0.15
0.59
0.60
0.73
0.34
0.05
0.20
0.08
-0.08
0.03
0.45
0.18
Mean
0.73
0.75
0.66
0.68
0.12
0.08
0.23
0.17
0.29
0.16
SD
0.37
0.34
0.29
0.30
0.13
0.08
0.19
0.25
0.23
0.21
Table 4.26. /x/ locus equations by geolect
/i/
Chile
Mexico
/e/
/u/
/o/
/a/
slope
r2
slope
r2
slope
r2
slope
r2
slope
r2
Mean
0.94
0.89
0.92
0.87
0.09
0.04
0.25
0.12
0.26
0.11
SD
0.09
0.06
0.11
0.08
0.18
0.04
0.13
0.15
0.23
0.13
Mean
0.46
0.56
0.40
0.48
0.16
0.12
0.22
0.23
0.32
0.21
SD
0.44
0.50
0.05
0.31
0.06
0.09
0.26
0.33
0.26
0.27
Productions by the Chileans, although articulated with descriptively higher gestural
overlap than the Mexicans in the assimilating /xi, xe/ contexts, do not reach significance
(F(1, 4.058)=5.904; p=0.071).
The final velar comparison concerns differences between the experimental
(assimilating) and control (non-assimilating) speakers. Table 4.27 collapses regional
differences into the two groups above.
166
Table 4.27. /x/ locus equations by group
/i/
Exp.
Control
/e/
slope
r
2
Mean
0.77
SD
/u/
slope
2
r
0.88
0.67
0.21
0.08
Mean
0.71
SD
0.49
/o/
slope
2
r
0.78
0.16
0.27
0.10
0.65
0.65
0.44
0.36
/a/
slope
r
2
slope
r2
0.09
0.23
0.19
0.12
0.03
0.06
0.11
0.19
0.36
0.18
0.03
0.57
0.09
0.07
0.23
0.16
0.46
0.29
0.40
0.18
0.04
0.22
0.13
0.12
0.23
Although the experimental group coarticulates to a slightly higher degree than the control
participants within the subset of triggering vowels /i, e/, this difference is not significant
(F(1, 5.228)=0.018; p=0.898). As was the case with /f/ coarticulation, slope for the
control speakers, although lower than the assimilating group, still shows gravitation
toward coarticulation in the front vowel contexts as compared to equally low slopes for
both groups in the non-front vowel contexts.
The confounding variable here, however, is that both the experimental and control
groups contain Chilean participants for whom palatalization is ubiquitous. If group
comparison is limited solely to the Mexican speakers, the difference is slightly clearer:
/xi/ slope=0.53, r2=0.96 experimental versus /xi/ slope=0.33, r2=0.37 control; /xe/
slope=0.44, r2=0.71 experimental versus /xe/ slope=0.35, r2=0.26 control. Recall that the
argument here is that the tendency to slightly coarticulate /x/ in the context of /i, e/
creates the impetus for eventual assimilators to misperceive and exaggerate the
coarticulation, leading to grammaticalization of the process.
This section has tested the predictions of Baker et al. (2011), who argued that
phonological properties might be explained by three factors, the first of which is an
underlying phonetic motivation. I have shown that Spanish /f/ and /x/ unfaithfulness is
167
triggered by coarticulation, but only by back vowels in the case of /f/ and front vowels in
the case of /x/. Both properties showed greater coarticulation by the assimilating group,
yet with the non-assimilating group exhibiting patterns toward assimilation. Regional
differences were only identified with /x/; Chileans exhibited high degrees of
coarticulation, offering a possible explanation for why palatalization is pervasive by these
speakers.
However, as Baker and colleagues note, while coarticulation is universal, not
every instance is grammaticalized (the majority, in fact, are not). A second necessary
component is variability in the signal so that Ohala’s (1993) hypocorrectors or Beddor’s
(2012) innovative listeners cannot successfully undo the phonetic effects, resulting in a
failure to reconstruct the intended target.
The next subsection explores the degree of variability in /f/ and /x/ produced by
these speakers.
4.8.2. Variance in center of gravity values
As Ladefoged and Maddieson (1996: 173, 181) and Silbert and de Jong (2008:
2778) note, variability among different instances of the same segment (between speakers
and within single speakers) is characteristic of fricatives, making the task of determining
unique cues to place for individual fricatives a difficult phonetic endeavor. While this
was precisely the goal of McMurray and Jongman (2011), they found that some cues
predict fricative place more accurately than others. The complete range of r2 values for all
24 of their fricative cues spans from 0.011 to 0.678 (M=0.284, SD=0.219), suggesting
that phonetic cues to fricative place are quite variable. Following Baker, Archangeli and
Mielke (2011), variation is measured by center of gravity in the present section.
168
Given that fricatives are inherently variable, they are especially susceptible to the
type of grammaticalization described by Baker et al. (2011). Without such variability, the
coarticulation will simply go unnoticed, or will fail to disrupt the phonetic corrections
that Ohala (1981) shows that listeners are able to make. For this reason, Baker,
Archangeli and Mielke argue that phonological change is both rare and slow given the
complex series of events that must coincide.
To determine variability, Figures 4.29 and 4.30 below report the maximum,
minimum and average center of gravity values for each phoneme, /f/ and /x/ respectively,
for all eight production participants (listed 1 through 4, experimental; 5 through 8 control;
1, 2, 5, 6, Chile; 3, 4, 7, 8, Mexico).
Figure 4.29. Maximum, minimum and average /f/ COG values by participant
Note: 1 and 2, Chile experimental participants 19727 and 71737; 3 and 4, Mexico
experimental 26911 and 35792; 5 and 6, Chile control 15246 and 02838; 7 and 8, Mexico
control 11111 and 71017.
169
Figure 4.30. Maximum, minimum and average /x/ COG values by participant
Note: 1 and 2, Chile experimental participants 19727 and 71737; 3 and 4, Mexico
experimental 26911 and 35792; 5 and 6, Chile control 15246 and 02838; 7 and 8, Mexico
control 11111 and 71017.
As shown in Figure 4.29, both inter- and intra-speaker variability obtain given the large
COG ranges, differences in those ranges between speakers and SDs for /f/ in every
participant excluding Mexican experimental participant 26911 (referenced as 3 above).
Figure 4.30 presents the same calculations for /x/.
The expected variability expressed by the range between the maximum and
minimum COG value for /x/ is exhibited by the experimental participants (1 through 4),
yet the control participants show comparable ranges, suggesting that the velar fricative–at
least for this phonetic measurement for these particular speakers–is relatively more stable
than /f/. However, note the subtle variation in the control group’s averages, indicating
some degree of variability, even if minor.
170
Given the degree of variability both between and within speakers in the ranges
and SDs produced for /f/ and to a lesser extent /x/, the second condition put forth by
Baker et al. (2011)–generating contexts in which reconstruction of the signal is disrupted
via variability–is met.
In Tables 4.28 and 4.29, the maximum, minimum and mean COG values by
allophone are reported to determine if the ranges for each output overlap.
Table 4.28. Maximum, minimum and average COG values for /f/ allophones by group
[x]
[ɸ]
Experimental
[f]
Control
Experimental
Max
704
2992
11964
8809
Mean
394
721
4014
2796
Min
146
125
1105
1025
SD
218
384
1940
1427
In Table 4.28, if the lower end of the control groups’ [f] continuum overlaps with the
upper end of the experimental participants’ [ɸ] and [x], this shared phonetic space of
(target-deviant) [ɸ]-like [f] productions by the control speakers lends itself to
misperception by the experimental group, who, when acting as speakers themselves,
reproduce the surface variant [ɸ] as opposed to the intended [f].
Indeed, the maximum of [ɸ] is greater than the minimum of the control [f],
meaning that these low COG [f] articulations by the controls are susceptible to
misperception as [ɸ]. No overlap was attested for the velar allophone even though the gap
is relatively minor (401 Hz). Table 4.29 applies the same rationale to /x/.
171
Table 4.29. Maximum, minimum and average COG values for /x/ allophones by group
[x]
[ç]
[x]
Experimental
[ç]
Control
Max
1873
3466
3132
4821
Mean
706
2011
1179
3457
Min
95
729
162
1858
SD
346
592
506
711
With respect to /x/, the assimilated variant [ç] by the experimental group highly overlaps
with the control group’s faithful variant /x/. Crucially, observe the difference in both
groups’ [x] allophones; that is, the higher COG [ç]-like articulations of [x] by the control
speakers extend beyond the phonetic space permitted by the experimental group’s [x],
creating a situation in which reconstruction to the intended [x] target by prospective
innovative listeners is impeded, and the perceived surface form [ç] is taken at face value.
This section applied Baker et al.’s (2011) variability requirement for
grammaticalization to the Spanish fricative data. By examining the range of center of
gravity values for each participant, I argued that /f/ exhibited high degrees of inter- and
intra-speaker variability; the same was true for /x/ only in the case of the four
experimental participants. I furthermore demonstrated how this variability could result in
hypocorrection (Ohala 1981), the failure of listeners to correct for phonetic effects. In the
next section, I present an additional perceptual argument before concluding the chapter.
4.8.3. Perception
At this point, I have demonstrated that two of the three criteria posed by Baker et
al. (2011) hold for the Spanish processes: phonetic motivation and variability in the
signal. I address the third component–sociolinguistic factors to enable expansion of the
172
property–in Chapter 5, yet for the moment it is necessary to address perception, as it
plays a critical role in component two.
While variability might facilitate hypocorrection by disrupting the hearer’s ability
to reconstruct the intended signal, it might simply be the case that two segments are
inherently similar. Mielke (2012) tested for the acoustic similarity among 58 segments
from various languages. Figure 4.31 contains the relevant portion of his results (Mielke
2012: 152) (gray boxes are my emphasis).
The critical region spans only six segments in which we find bilabial [ɸ] and
labiodental [f] as adjacent neighbors, velar [x] separated from them by only a single
segment. Furthermore, while the palatal fricative [ç] was not part of the inventory tested
by Mielke, the palatal stop [c] is only two segments above [x]–from this, I would argue
that assuming that [ç] would not be too far removed from [c]’s position is reasonable.
Figure 4.31. Acoustic similarity among various obstruents
173
A logical prediction from Mielke’s similarity results is that simple misperception
between the members of the pairs [f]~[ɸ], [f]~[x] and [x]~[ç] is the impetus for the
phonological processes instead of the coarticulation facts presented above. In fact,
Mazzaro (2011: 113) reaches this conclusion for the [f]~[x] velarization data that she
analyzes, as participants misperceived [f] as [x] (and vice versa) in the context of back
vowels 64.2% of the time. To determine if misperception accounts for the processes, I
present below the results from my AX discrimination task (task 2).
Figure 4.32. Average perceptual rating by allophonic comparison
Recall that participants heard 100 [CV CV] pairs with alternating initial consonants (e.g.,
[fo] [xo]) and were instructed to rate on a scale of 1 ‘same’ to 5 ‘different’ how similar
(or not) they perceived the difference within each pair to be. Figure 4.32 compares the
average perceptual rating of each allophonic pair by group. A GLMM revealed a
significant effect for allophonic contrast (F(10, 345)=3.303; p<0.001). The post hoc tests
determined that [f]~[ɸ] differs from both [f]~[x] (p<0.001) and [x]~[ç] (p<0.001); [f]~[x]
174
and [x]~[ç] do not differ significantly (p=0.972). What this indicates is that the [f]~[ɸ]
distinction is misperceived more readily than the other two contrasts. This holds true for
each group above, as group was not significant (F(10, 345)=0.124; p>0.999), nor was
there a group allophonic type interaction (F(20, 345)=0.503; p=0.965).
If perception is predictive of the processes, in the [f]~[ɸ] and [f]~[x] conditions,
Mexican participants should have rated the pairs the lowest numerically (i.e., perceived
as most similar) of the three groups given that they attest the properties most frequently.
By that same logic, Chilean participants should have rated [x]~[ç] closer to 1, and should
have differed from the other two groups. These predictions only held for the [f]~[ɸ]
condition, but it was not differentiated by country or group; in other words,
(mis)perception by itself does not seem to correlate with exhibiting the property in
production. These results seem to contradict Mazzaro’s (2011) findings, yet support work
by Fernández, Feijóo, Balsa and Barros (1998) who show that [f] in the context of [u] is
actually easier to perceive than in the context of all other vowels. This is surprising if
misperception is assumed given that it is precisely the /fu/ sequence that conditions both
bilabialization and velarization most frequently.
Furthermore, Feijóo and Fernández (2003) find high rates of discrimination in
both C and CV contexts for [f] and [x] by Galician speakers who identified any one of the
five Galician fricatives /f, θ, s, ʃ, x/ or one affricate /t͡ ʃ/ in 600 real stimuli of the shape
FVCV (6 consonants 5 vowels 20 speakers). Their results (Feijóo and Fernández
2003: 5) for the fricatives relevant to Spanish (/f, x/) are reported below (averages do not
total 100% as the excluded consonants account for the remaining percentages).
175
Table 4.30. Fricative misperception in Galician
F condition (isolated)
perceived as /f/
perceived as /x/
/f/ stimulus
78.4
1.5
/x/ stimulus
0.4
99.5
FV condition
perceived as /f/
perceived as /x/
/f/ stimulus
83.9
1.4
/x/ stimulus
0
100
It is clear that [f] and [x] are able to be accurately discriminated (at least by these 30
speakers of Galician). (The results also support the important role of transitional vocalic
cues given that accuracy increased slightly across the board in the FV condition.)
Accuracy in discriminating [f] and [ɸ] by my participants was significantly lower,
however; unsurprising since, in Spanish, the contrast is allophonic. Yet perception of the
contrast is not impossible. As previously discussed, Ewe and other African languages
contrast both segments phonemically (with the caveat that the distinction is enhanced
through amplitude and/or roundedness maximization (Utman and Blumstein 1994,
Ladefoged and Maddieson 1996)).
I further argue that misperception cannot be the only explanation, as such would
also predict bidirectional alternations: i.e., /f/→[x] (e.g., fumo ‘I smoke’ [xumo]) in
addition to its inverse /x/→[f] (jota ‘the letter j’ *[fot̪ a]). While such data are possible in
Latin American Spanish, /x/→[f] appears (i) in grammars that also attest /f/ velarization,
(ii) at much lower rates of frequency and (iii) in all vocalic contexts. These facts lead
176
Lipski (1995) to conclude that /x/→[f] is the result of hypercorrection (sociolinguistic,
not Ohala’s (1981) terminology) by speakers of /f/→[x] dialects.
In the absence of bidirectionality, one-to-one segmental misperception must be
questioned. The notion remains viable if misperception varies with respect to the
following vowel, however. Given the vowel-dependent conditioning facts for which I
argue above, if the allophones are easier to confuse in the contexts of the specific vowels
that trigger the processes, they might be due to asymmetrical misperception. Consider
Figures 4.33 and 4.34.
Figure 4.33. Vowel-specific /f/ discrimination within [f]~[x] tokens
If /f/ allophony is solely the result of misperceiving [f] as [x] in the context of [u, o], the
prediction is that those speakers who attest the property should have more difficulty in
discriminating [f] and [x] precisely in front of those vowels as compared to those who do
not have the property as part of their grammar. As Figure 4.33 shows, this is not the case.
While the /f/ assimilating Mexicans are slightly less accurate in [f]~[x] discrimination for
177
/u, o/ compared to (i) the Chilean speakers and (ii) to the /i, e, a/ context within the
Mexican group, neither difference is significant; i.e., no significant effect was found for
group (F(4, 114)=.094; p=.984), vocalic condition (F(4, 114) =0.159, p=0.958) or group
condition interaction (F(4, 114)=0.268, p=0.898).
In other words, Mexicans who assimilate /f/ are just as accurate in discriminating
[f] from [x] as speakers from other dialects (without the phonological process), and have
no difficulties discriminating the consonants in any vocalic context. The Mexican group
is, in fact, more accurate than the Chilean and Control participants with respect to the
non-back-vowel condition.
The same logic can be applied to /x/ palatalization. The prediction is that Chilean
listeners–who attest palatalization at significantly high rates–will be less accurate in
discriminating [x] and [ç] in two comparisons: relative to the Mexican and Control
listeners who lack such a process, and relative to the conditioning vowels /i/ and /e/.
Figure 4.34. Vowel-specific /x/ discrimination within [x]~[ç] tokens
178
As can be seen in the figure above, while the perceived difference between [x] and [ç] in
the /i, e/ context within the Chilean group might be less than /u, o, a/, this minor
difference did not reach significance (F(5, 110)=0.040; p=0.999). Nor is the difference
between groups significant (F(5, 110)=0.072; p=0.996) (the group vocalic context
interaction is also insignificant (F(5,110)=0.737; p=0.598)). Furthermore, the Chilean
listeners outperformed the Mexican and Control groups in general.
These results argue against an account that attributes the processes to
misperception alone; the conclusion is that it is not the case that the perceived similarity
among the allophones can solely account for the origin of the processes. If so, [f] and [ç]
are only four segments away in the acoustic similarity tree from Figure 4.31. Assuming
misperception, how do we prevent an output such as /f/→[ç]: figura ‘figure,’ *[çiɣuɾa]
(or its inverse)?
I would argue that the missing component is indeed (co)articulation. An [f] in the
context of [u], compelled to retract by assimilating forces (see Chapter 5), becomes
gradiently more [x]-like in this particular environment given the backness of the vowel. It
is this coarticulated utterance that is then subject to misperception, not the inherent
similarity between any [fo] and [xo] output. The proximal cause remains the
coarticulation, of which (the increased likelihood of) misperception is the result.
At this point we are now able to answer a longstanding question throughout this
dissertation: why are fricatives susceptible while other manners resist assimilation (in
Spanish, at least)? Beyond the previous discussion concerning the inherent variability in
the signal of fricatives and the external locus of its identification of place (the vowel), it
is also the case that other manners better resist misperception given the strength of their
179
acoustic cues (Deng and O’Shaughnessy 2003, Ladefoged 2006). Figure 4.35 reports on
the perceived similarity by manner for all listener groups.
Figure 4.35. Average perceptual rating by manner
The GLMM found that, while group was not significant (F(10, 1630)=0.368; p=0.961),
consonantal manner was (F(15, 1630)=7.903; p<0.001). The target, non-strident
fricatives (the /f/, /x/ category) were perceived as most similar at ratings significantly
different from all other manners (p<0.001 for all comparisons). What this suggests, in
line with previous research (Kohler 1990, Steriade 1993, Byrd 1994, Jun 1994, 1995,
Herrick 1999), is that acoustically weak segments (e.g., [f] and [x]) are more likely
phonological targets than acoustically strong ones (e.g., stridents, stops and nasals).
Conversely, acoustically strong segments more readily resist phonological targethood
than acoustically weak segments (see Chapter 5 for expanded theoretical discussion and
formalization).
180
By this reasoning we are able to limit the grammar to only the most vulnerable
segments–here, the two non-strident fricatives. In this way I respond to the criticism from
Chapter 2 against the overproductive analyses by Mazzaro (2005) and Lipski (1995),
whose [labial] OCP analyses incorrectly predict assimilation in música ‘music’ and burla
‘mockery.’ It is the strength of the phonetic cues to place that protect nasal [m] and stop
[b] from assimilation.
4.9. Conclusion
This chapter responded to six of the nine research questions that set forth the
descriptive, theoretical and sociolinguistic contributions of this thesis. I repeat the
relevant questions below, after which I provide an answer based on the conclusions
supported by the data and results presented above.
1. Are the fricative phonemes /f/ and /x/ synchronically involved in allophonic
alternations in Latin American Spanish? If so, what are the allophones?
By analyzing centers of gravity and comparing them to known values for various
fricative segments, I argued in section 4.2 that there are four identifiable allophones
involved in the processes described in this dissertation: labiodental [f], bilabial [ɸ], velar
[x] and palatal [ç]. These conclusions were further supported by spectral shape, spectral
peak and visual cues, all of which coincide with previous phonetic analyses of similar
properties (namely, Mazzaro 2011).
2. Are the phonological processes productive (i.e., not merely lexicalized)?
I argued for the productivity of the processes in section 4.3. No significant differences in
rates of assimilation were attested in the frequent versus infrequent contexts and among
production task comparisons. Section 4.4, on the other hand, offered two possible
181
exceptions in line with previous research (Sanicky 1988): ir ‘to go’ and ser ‘to be,’ whose
rates of assimilation suggest lexicalization (at least for the past tense forms) for the four
experimental participants.
3. Can the processes be conceived as assimilations conditioned by the
surrounding vowel(s)? If so, are the assimilations regressive or progressive?
Given the freely assimilating data in Chapter 1, it was worth questioning what factor
actually conditions the allophony. In section 4.5, I argued that it is the following vowel
that triggers the unfaithful allophones in all three processes; I additionally excluded the
preceding vowel as deterministic. This is not surprising in light of the numerous literature
on the importance of the following vowel in fricative identification (see the sources cited
above), in addition to the studies that demonstrate an articulatory affinity in CV
sequences greater than that in VC sequences (Lee 1997, 1999, Ghavami 2002). The
properties are therefore assimilations exclusively in the regressive direction. This
conclusion leaves the freely assimilating dialects (in which the unfaithful allophone may
appear before any vowel) unexplained, an issue I address in Chapter 6.
4. Are the assimilations conditioned by stress or position in the word?
In addition to the surrounding vowels, stress and position of the fricative within the word
were also considered. Although Mazzaro (2011) found an effect for both, my results
showed a trend but no significant effect for either (other than the case of stressed /f/) in
sections 4.6 and 4.7, respectively.
6. How can this theoretical analysis limit the assimilations to the subset of
fricatives {/f/, /x/} to the exclusion of all other phonemes in Spanish?
182
I provided a general overview of consonant-vowel assimilations in Chapter 1, which
demonstrated that all manners of articulation are susceptible to assimilation. The question
that this fact raises is: what limits assimilation to a subset of all possible manners in a
given language, and how can this be accounted for crosslinguistically? I suggested in
section 4.8.3 that perception contributes to limiting the process to {/f, x/} in Spanish, as
participants were least successful in discriminating place in the only two non-sibilant
fricatives in the phonological inventory of Spanish.
7. What role, if any, do articulatory and acoustic phonetics play in the theoretical
model proposed?
Beyond accounting for the data from a descriptive and explanatory perspective–the goals
of the above questions–another aspect of this dissertation concerns the dialectal
differences themselves; specifically, how did the dialects of Spanish come to be
differentiated concerning these three properties? That is, why do these particular speakers
from these particular dialects have these properties as part of their grammars, while other
speakers from other dialects do not? The three criteria proposed by Baker and colleagues
(2011) include coarticulation as phonetic motivation, variability in the signal and
association with sufficient sociolinguistic factors to disseminate the property. I addressed
two of the three in sections 4.8.1 and 4.8.2, showing via locus equations that
coarticulation exists (providing a motivation and genesis for the processes) and that
variability is rampant (not unexpected in fricatives; see Ladefoged and Maddieson 1996
and Silbert and de Jong 2008). I explore the social aspect in the next chapter in which I
encode the above findings in the Optimality Theoretic proposal that comprises the formal
account of the analysis.
183
CHAPTER 5
OPTIMALITY THEORETIC ANALYSIS
There were three principal goals of this dissertation: to update the descriptive
literature of the phonological properties in question through detailed phonetic analysis; to
account for the origin and expansion of these properties following the predictions of
Baker, Archangeli and Mielke (2011); and to account for the alternations within
Optimality Theory (Prince and Smolensky 1993/2004, McCarthy and Prince 1993/2001).
This chapter presents the theoretical proposal with this final goal in mind, bringing
together the conclusions from the first two goals in a formal phonological account.
Specifically, recall the detailed research questions from Chapter 1.
1. Are the fricative phonemes /f/ and /x/ synchronically involved in allophonic
alternations in Latin American Spanish? If so, what are the allophones?
2. Are the phonological processes productive (i.e., not merely lexicalized)?
3. Can the processes be conceived as assimilations conditioned by the
surrounding vowel(s)? If so, are the assimilations regressive or progressive?
4. Are the assimilations conditioned by stress or position in the word?
5. How can a formal model of phonology–Optimality Theory–account for the
phenomena?
6. How can this theoretical analysis limit the assimilations to the subset of
fricatives {/f/, /x/} to the exclusion of all other phonemes in Spanish?
7. What role, if any, do articulatory and acoustic phonetics play in the theoretical
model proposed?
184
8. What crosslinguistic typological implications does 7 entail?; i.e., why do
certain linguistic systems attest certain properties while other systems do not?
9. How can these questions be viewed from a diachronic perspective of Spanish?
At this point, I have responded to questions 1 through 4. The present chapter (and
Chapter 6) expands on questions 5 through 9.
I structure the chapter as follows. Section 5.1 defends the validity of an
Optimality Theoretic account by arguing against previous proposals that claim that such
an account is untenable. In section 5.2, I lay out the fundamentals of the theoretical
components assumed. Section 5.3 presents the OT analysis for each process–/f/
bilabialization, /f/ velarization and /x/ palatalization–and concludes by offering the
factorial typology for possible grammars predicted by the constraints necessary in the
analysis. The third component of Baker et al.’s (2011) proposal that I set aside in Chapter
4–the sociolinguistic situation surrounding the processes that accounts for their
dissemination–is resolved in section 5.4 based on the preceding discussion from this
chapter. I conclude in section 5.5.
5.1. Arguments for a synchronic analysis
I claimed in the previous chapter that /f/ and /x/ allophony in Latin American
Spanish are explained by two indispensable components: coarticulation and vocalic
context sensitivity. I argued that, while perception plays a role in the account, simple
segment-to-segment confusion cannot be the underlying cause for the processes. I instead
demonstrated that vowel-specific coarticulation (i.e., gestural overlap) is the motivation
for the alternations. While such an explanation is in line with previous accounts of /x/
palatalization, my coarticulated data are in direct conflict with the theoretical prediction
185
from Bateman (2007) concerning the extent to which /f/ may assimilate. I present and
ultimately argue against her prediction in section 5.1.2. First, however, I refute Mazzaro’s
(2011) perceptual account in 5.1.1. Section 5.1.3 offers an additional piece of evidence in
favor of a synchronic explanation: crosslinguistic variation.
5.1.1. Perception
The first alternative to my coarticulatory approach is that of perception: do [f] and
[x] alternate because they are inherently acoustically similar? As Mielke (2012) shows
(see Figure 4.31), this indeed seems possible given their adjacency is his phonetic
neighbor analysis. This approach, however, is untenable for two reasons. The first is that,
assuming the segment-to-segment misperception, there are two predictions that fail: that
perceptual confusion and consequently the phonological process should be bidirectional
(I have argued previously that it is primarily unidirectional /f/→[x]); and that perceptual
confusion and consequently the phonological process should be vowel-insensitive (I have
argued previously that it is vowel dependent). The second reason that straightforward
perceptual replacement fails is that Feijóo and Fernández (2003) show that /f/ and /x/
remain (highly) discernible in both isolated and fricative-vowel conditions. For these
reasons, an explanation that relies on one-to-one misperception is inadequate.
However, it is possible to maintain a perceptual account if vowel-specific
differences are included. Mazzaro (2011) ultimately claims that [f] alternates with [x] in
the specific contexts of [u] and [o] because [fu, fo] are perceptually closer to [xu, xo] than
[f] is to [x] elsewhere. There are also two ways in which this explanation fails. As I
demonstrated in Figure 4.33, all three listener groups (Chilean, Mexican and control)
exhibited insignificant differences in perception when [f]~[x] discrimination is broken
186
down by vocalic context. Furthermore, Fernández, Feijóo, Balsa and Barros (1998) argue
that [f] is easier to perceive in the context of [u] given that [u]’s roundedness amplifies
the frication produced by the ladiodental obstruction. This conclusion directly conflicts
with Mazzaro (2011).
On the other hand, my participants demonstrated significant differences in the
degrees of coarticulation of each fricative depending on the following vowel (see Figures
4.21 and 4.24). Consequently, I argued that an articulatory motivation is preferable to an
underlying perceptual one, even though perception plays a crucial role in the
coarticulation’s grammaticalization into a phonological process. As Baker, Archangeli
and Mielke (2011) argue, and as I show for /f/ and /x/ in Spanish in Tables 4.27 and 4.28,
slightly coarticulated (i.e., target-deviant) productions of [f] (/f/) and [x] (/x/) that
approach the phonetic spaces for [x]/[ɸ] (/f/) and [ç] (/x/) are those that are most
susceptible to misperception. Thus, while it is the perceptual failure to correct for
coarticulation that phonologizes the process, the underlying cause is the gestural overlap.
As Ohala (1989: 183) notes, “what looks similar to the eye in these displays
[spectrograms] will sound similar to the ear and thus be subject to confusion.”
Mazzaro (2011: 10) cites and uses this quote to argue for her perceptual account.
However, what one assumes to be the point of departure for the sound change
(coarticulation or perception) determines how Ohala is interpreted. Essentially, what
results is the classic chicken and egg causality dilemma: do [f] and [x] within the context
of [u] and [o] resemble each other acoustically because of their (co)articulation, or do
speakers (co)articulate [f] and [x] in the context of [u] and [o] because of their perceived
187
similarity? I argue for the former in that coarticulated [f] results in a relatively [x]-like
output that speakers may (or may not) interpret as an intended [x].
There is an inverse relationship between the slope calculated from the locus
equations of each fricative-vowel combination and the resultant center of gravity seen in
Figure 5.1 and Table 5.1. As coarticulation (slope) increases (more or less moving from
front to back in terms of vocalic location), center of gravity tends to decrease. Given that
[f] is associated with a higher COG relative to [x], the inverse relationship just described
suggests that the more [f]-like /f/ productions are associated with the front vowels and
less coarticulation while the more [x]-like /f/ productions appear with back vowels and
more coarticulation (see the shaded regions below).
In other words, as the vowel associated with /fV/ moves from front to back (i, e »
a » o, u), coarticulation increases with concomitant lowering of the COG value for /f/, in
fact approaching that of /x/. Thus, I argue that the perceptual facts associated with /f/ are
a consequence of the (co)articulatory facts, and that Mazzaro (2011) has the causal
relationship inverted. Furthermore, such a conclusion lends itself to an OT explanation in
which recent work in Articulatory Phonology (AP; Browman and Goldstein 1986, et seq.)
has formally accounted for these types of variable and gradient coarticulations.
As a result, I conclude that the driving force behind the processes is not
perception, but rather (co)articulation. Without a synchronic analysis for the /f/ data in
this dissertation, it is highly coincidental that the phonetic contexts with the highest
degree of coarticulation result in place-predictable effects on the fricative.
188
Figure 5.1. Average /f/ locus equations (slope) and center of gravity (kHz) comparisons
4.5
/fa/ /fe/ /fi/
/fu/ /fo/
4
COG 3.5
(kHz) 3
Renaud
Mazzaro
2.5
2
0
0.5
1
LE (slope)
Table 5.1. Average /f/ locus equations and center of gravity (Hz) comparisons: data
Chapter 4
Mazzaro (2011)
LE
COG
COG
/fo/
0.963
2529
3284
/fu/
0.902
2208
2713
/fi/
0.296
3473
4196
/fe/
0.124
3476
3973
/fa/
0.029
3186
3533
Average /x/ COG (Mazzaro 2011)
2185
On the other hand, this conclusion contradicts the theoretical predictions of Bateman
(2007), whose AP-based OT account of palatalization categorically rules out for /f/ the
type of assimilatory proposal that I pursue. In the next section I address and ultimately
reject those predictions, as well.
189
5.1.2. Coarticulation
Based on the empirical results presented in Chapter 4 and the argumentation
above, I concluded that the Latin American Spanish /f/ (and /x/) alternations are due to
fricative-vowel assimilations motivated by coarticulation. This is problematic in light of
Bateman (2007), who argues that, in the case of /f/, such an explanation is theoretically
impossible. Her extensive survey of palatalization and (secondarily) palatalized segments
(both phonemic and derived) resulted in only two reported cases of (fully) palatalized
labials, which she dismisses as the final step of a diachronically multifaceted process, as
opposed to a productive synchronic alternation.
In the case of Moldavian (a Romanian dialect), /p, b, f/ alternate with [kj, ɡj, ʃ] in
the context of /i/ (Bateman 2007: 104). When assumed to be a direct alternation (the
shaded portions below), it seems like a case of labial palatalization. Yet crosslinguistic
analysis led her to posit intermediate steps such that the palatalization derives not from
/i/, but from a historic yod /j/ that triggered consonantal hardening. Observe the steps in
Table 5.2 (idem., her Table 3.1).
Table 5.2. Palatalized labials in Moldavian
Gliding
/albinə/
/lupi/
/fier/
albjinə
lup
Hardening
Deletion
Current
Gloss
albʝinə
albɡjinə
alɡjinə
alɡjinɨ
‘bee’
lupɕ
lupkj
N/A
lupkj
‘wolf’ (pl.)
fɕer
fʃer
ʃer
ʃer
‘iron’
j
lupj
fjer
190
Other than one additional language, Tswana (Bateman 2007: 131),1 her 117-language
survey revealed no other instances of (full) labial palatalization. Due to the fact that the
palatalization of labials seems to be highly marked, she seeks to prevent its generation.
To rule out the possibility of (complete) labial palatalization, Bateman (2007: 223, her
(18)) proposes the following constraint.
(1)
IDENT-Articulator IO
“An input oral gesture of a particular major articulator, tongue or lips, must be
associated with an output segment which is associated with the same major
articulator.”
The theoretical formalization in (1) (more on OT constraints in §5.2.1) is designed to
prevent alternations that convert segments articulated with the lips (e.g., [f]ier from
Bateman’s Moldavian data) directly to segments articulated with the tongue ([ʃ]er), and
vice versa. This way, she can elegantly encode a restriction that the theory seems to place
on languages. As a consequence, (1) predicts the impossibility of directly accounting for
the Spanish /f/→[x] data addressed throughout this dissertation, as such a process would
violate (1) given the disagreement in major articulators from input labial /f/ to output
dorsal/velar (tongue body) [x].
At this juncture, one of two possibilities must be correct: either /f/→[x] is not a
direct, synchronic alternation and instead requires an alternative analysis, or Bateman’s
constraint is too restrictive. The rationale that follows argues for the latter: that
Bateman’s (1) makes incorrect predictions and must instead be reconsidered, maintaining
a synchronic analysis for /f/ velarization.
1
The phonological descriptions here are relatively complex and morpheme specific. For a
complete discussion, I direct the interested reader to Bateman (2007: 131, §3.4). For present
purposes, it is sufficient to recognize that she concludes for Tswana just as she concluded for
Moldavian, that the apparent cases of labial palatalization are instead historic vestiges.
191
There are three arguments to support this line of reasoning; the first is conceptual.
Bateman explores palatalization across many languages, yet proposes a constraint that
limits data that go beyond palatalizing processes. Thus, in attempting to account for a
marked process in her data, she overgeneralizes and restricts the theory where such a
restriction might prove problematic elsewhere. Note that I do not dispute her
palatalization results, yet the (theoretical) explanation that she puts forth in limiting
palatalization is too broad.
Additionally, in encoding her constraint, she simplifies the components of
Articulatory Phonology (AP). While I will expand on the theory in section 5.2.3 below,
for the moment it is sufficient to note that, in AP, three major articulators are assumed:
lips, tongue tip and tongue body. Even though the latter two are both subdivisions of the
tongue, the seminal work for AP (Browman and Goldstein 1986, 1989, 1990), technically
speaking, separates the tongue into two major articulators given the high degree of
flexibility and productivity of processes involving the tongue blade/coronal region. In this
way, the authors justify endowing the tongue tip with its own unique (theoretical) status
separate from the tongue body.
If the appropriate tripartite division is then applied to (1)–“an input oral gesture of
a particular major articulator [lips, tongue tip or tongue body] must be associated with an
output segment which is associated with the same major articulator”–the unintended
consequence is that particular palatalization processes for which she accounts are also
ruled out as they require the conversion from the tongue body to (separate major
articulator) tongue tip. According to (1), this should not be permitted. Relevant examples
appear in (2) (Bateman 2007: 230 (Mwera, among many others), 250 (Bulgarian)).
192
(2)
a. /k/→[t͡ ʃ]
b. /k/→[c]
[k] Tongue Body (TB)
[k] Tongue Body (TB)
[t͡ ʃ] Tongue Tip (TT)
[c] Tongue Tip (TT)2
For these reasons, (1) is theoretically inadequate and cannot be considered a
counterargument to the synchronic analysis of Spanish /f/→[x].
The second reason that (1) makes improper predictions is that the data for which it
was designed involve (mostly) categorical, allomorphic alternations in derived
environments; in contrast, the Spanish processes are productive, root-internal and nonobligatory. A synchronic analysis must be a viable option as two instances of the same
word might be produced differently (one with the faithful allophone, the other with the
assimilated variant) by the same speaker, as was the case for many words in the data from
Chapter 4. Thus, the historical analysis that led Bateman to her rejection of synchronic
labial palatalization does not apply here. In fact, recent developments in Optimality
Theory (for Stochastic OT (Boersma and Hayes 2001), see Chapter 6) are designed to
handle such variable/optional processes with which classic OT encountered difficulties.
The final argument contra Bateman’s prediction is that (1) suffers from empirical
challenges. As the results of my production experiments show, a high degree of
coarticulation with /f/ was found only in the /u, o/ contexts. Bateman argues that the
articulatory justification for her constraint in (1) is that gestural overlap between the lips
and the tongue is impossible. Deng and O’Shaughnessy (2003: 228), however, explicitly
refute such a claim: “Often, not all articulators are directly involved in a phoneme’s
2
With respect to (2b), the place features for palatals (e.g., [c]) are not without controversy. The
debate concerns whether palatal segments should be considered strictly [coronal], or [coronal]
with a secondary [dorsal] specification, assuming the standard [coronal]-[dorsal]-[labial] placefeature division. In Articulatory Phonology (§5.2.3), however, palatals are specified with a unique
[palatal] place feature, which is itself a subordinate of the Tongue Tip Constriction Location
(TTCL) tract variable (see Table 5.9).
193
constriction, leaving other articulators free to coarticulate. For example, phonemes
with a labial constriction /p, b, m, f, v/ allow the tongue to coarticulate” (emphases
mine). The conclusion is, therefore, that the prediction derived from Bateman’s constraint
in (1) is incorrect, and that a grammatical account can be maintained. The third argument
in favor of a representational approach is provided in the next section.
5.1.3. Systematic crosslinguistic variation
The final argument in support of a grammatical analysis comes from
crosslinguistic comparisons. As I show in Chapter 1, there are numerous permutations
that any language may select among the manners targeted by this type of assimilation, the
vowels identified as triggers and the outputs of the processes. Optimality Theory (see the
following section) is uniquely capable of capturing such a varied set of facts with a
limited number of theoretical components. As I will argue, Spanish is no exception to
these comparable processes, all of which must be explained in some way even if one is
able to alternatively account for a few of them. That is, if it turns out to be the case that I
incorrectly disregarded Mazzaro (2011) and Bateman (2007) concerning the specific data
for Spanish, the analysis can just as easily be applied to these similar properties in other
languages precisely because OT is designed to operate in such a fashion.
Specifically, the fact that Spanish /f/→[x] is unidirectional and solely regressive is
curious since, as previously discussed, bidirectional [f]↔[x] and progressive assimilation
are just as logically possible. As I will demonstrate in the remainder of the chapter, OT is
simultaneously able to account for such cases of language specificity and crosslinguistic
variability. The next section begins the analysis by presenting the necessary theoretical
mechanisms.
194
5.2. Theoretical components
The formal analysis relies on two main theories–Optimality Theory (OT, Prince
and Smolensky 1993/2004, McCarthy and Prince 1993/2001) and Articulatory Phonology
(AP, Browman and Goldstein 1986, et seq.)–and borrows from and slightly modifies two
others–the Production Hypothesis (Steriade 1993, Byrd 1994, Jun 1995) and Dispersion
Theory (Flemming 1995). The contribution of each is as follows. The representational
account for each property is framed within OT (§5.2.1). The featural and assimilatory
nature of the properties derives from AP (§5.2.3). The Production Hypothesis is adapted
to explain the manner restrictions that Spanish places on the processes (§5.3.4.1), and
Dispersion Theory informs the typological implications when the three processes are
considered collectively (§5.3.4.2). In the following section, I outline the core tenets of the
primary theoretical component, OT.
5.2.1. A general overview of Optimality Theory
Previous rule-based theories of phonology found in, for example, The Sound
Pattern of English (SPE, Chomsky and Halle 1968) account for phonological phenomena
by way of linear, inviolable and (at times) language-specific rules.
Current phonological theory, on the other hand, adopts a different approach.
Optimality Theory assumes a universal set of violable and (mostly) independently
rankable constraints belonging to one of three categories: markedness, faithfulness and
alignment. Markedness constraints, grounded in articulatory and acoustic/perceptual
facts, regulate the well-formedness of output structures. That is, markedness constraints
reflect the economy of certain articulatory configurations by imposing such
configurations on output forms. As a different means to the same end, markedness
195
constraints also proscribe articulatory inefficiency. Without motivating the constraints,
grounding them in articulatory and acoustic facts or demonstrating their universality at
some level, they are reduced to language-specific, ad hoc rules that serve no explanatory
purpose and lend themselves to the same criticism SPE analyses faced.
One such markedness constraint whose underlying motivation accounts for the
Spanish data at hand is AGREE (Lombardi 1999: 272). Lombardi formalizes AGREE as
follows.
(3)
AGREE
Obstruent clusters agree in voicing.
Extrapolating away from the details, the markedness constraint in (3) captures the fact
that sharing features (here, [voice]; see, e.g., Piñeros 2007 for AGREE(Place)) between
adjacent segments is less marked (more articulatorily facilitative) than a situation in
which each segment is associated with separate features. The effect of (3) is that output
forms in violation of AGREE will be less preferred than forms that comply by sharing the
relevant feature specification.
Divergence from input, however, also comes at a cost. While languages seek
articulatory economy, they simultaneously require contrast. These two goals conflict
when contrast is sacrificed for the sake of economy, and vice versa. To maintain
linguistic
distinctions,
faithfulness
constraints
penalize
disagreement
between
corresponding input and output forms (McCarthy and Prince 1995).
(4)
IDENT(F)
Corresponding input and output segments are identical with respect to feature F.
196
Output candidates run afoul of (4) when they contain features that differ from the input.
Thus, IDENT constraints preserve input contrast in output forms.
The final constraint category, alignment, will be addressed below. For the
moment, the discussion of markedness and faithfulness suffices to illustrate how OT
operates in the simplest case. In linear phonological theories, the goal was to get from
point A (the input) to point B (the output) by describing and correctly ordering the
necessary rules. While these rules did regulate well-formedness (i.e., markedness), they
and their relative orders were largely language-specific, i.e., ad hoc.
OT incorporates the universality of the rules, reconceptualized as constraints, by
positing that every language contains every constraint; languages may differ in the degree
to which they tolerate marked structures, or, if not, how they correct them. In other
words, an attested output form in language A is (universally) imperfect in that it violates
some constraint whose violation is not tolerated in another language. Optimality Theory
does not seek the perfect output, as such does not exist given the inherent conflict
illustrated above; it simply seeks the optimal form for the language in question.
Optimality can only be evaluated through an understanding of the three main
components of OT: Gen, Con and Eval. Gen generates an infinite array of possible output
candidates based on an input string. Con is the set of universal constraints ranked by
priority in a given grammar. Dominant (highly-ranked) constraints must be obeyed, while
violation of subordinate (lowly ranked) constraints is permitted. The optimal candidate–
the output form selected by the language from the infinite set–is determined by Eval.
Eval is a function that evaluates the candidate array and determines the optimal candidate
197
based on minimal violation; i.e., the candidate that violates the least prioritized constraint
in the grammar while satisfying all others (relative to the losing candidates).
Evaluations are reported in OT tableaux, exemplified by (5), utilizing the two
constraints introduced thus far. Input forms consistently appear at top-left, followed by
the relevant constraints ranked left to right in order of dominance (indicated by »).
Pertinent output candidates are provided in rows at left below the input; the symbol
designates the optimal candidate based on constraint violations indicated by an asterisk *.
(5)
Constraint interaction, AGREE(Place) » IDENT(Place)
/np/
a.
[np]
b.
[mp]
AGREE(Place)
IDENT(Place)
*!
*
Candidate (5b) [mp] is optimal even though output bilabial [m] differs from input
alveolar [n], violating IDENT(Place) due to the place discord between corresponding
segments. However, because of the subordinated status of IDENT(Place), this violation is
minimal and tolerated. Candidate (5a) [np] is ruled out by this grammar. While (5a) is
superior with respect to IDENT(Place) since the input [coronal] feature of /n/ is preserved
in output [n], this satisfaction comes at the expense of an AGREE(Place) violation, which
is fatal (indicated by the exclamation mark) since AGREE(Place) dominates
IDENT(Place). Once fatality has been determined, constraints irrelevant to the evaluation
are shaded. Now compare the grammar from (5) with (6).
If the hierarchy is reversed, candidate (6a) is now selected since its IDENT(Place)
satisfaction is optimal and AGREE(Place) violation minimal, whereas (6b) fatally
violates IDENT(Place).
198
(6)
Constraint interaction, IDENT(Place) » AGREE(Place)
/np/
a.
[np]
b.
[mp]
IDENT(Place)
AGREE(Place)
*
*!
With ranking permutation, the same set of constraints predicts other possible grammars
(dialects/languages). This factorial typology, as it is known, will be a crucial argument in
section 5.3.5.
For the moment, tableaux (7) through (9) provide the final three scenarios of basic
constraint interaction. As the direct conflict between AGREE and IDENT is unable to
illustrate them, hypothetical candidates and constraints are used in their place.
(7)
Constraint interaction, S, T
/input/
a.
[candidate a]
b.
[candidate b]
S
T
*!
Candidate (7a) is optimal as it violates neither constraint; candidate (7b)’s only violation
is fatal. The idea here is that no argument is established for a fixed ranking between S and
T. A dashed line in the tableau and a comma in the constraint hierarchy indicate such a
scenario. If dominance were reversed–T, S–candidate (7a) remains the winner. The logic
still applies in the event that candidate (7b) violates constraint S instead.
(8)
Constraint interaction, W, X
/input/
W
a.
[candidate a]
*
b.
[candidate b]
*
X
*!
199
(8) is another case in which a fixed ranking cannot be established. Instead of neither
candidate violating one of the two constraints, in this tie, both candidates equally violate
the same constraint–here, constraint W. Determining the optimal candidate falls to
constraint X. Since candidate (8b) incurs one X violation relative to candidate (8a)’s X
satisfaction, (8a) is optimal. Again, the dominance between W and X cannot be
established since the order X, W results in the same output (constraint W is simply
irrelevant in this case since X suffices to determine optimality). Additionally, if the locus
of the tie is constraint X and candidate (8b) instead violates W, order is still
indeterminable, and candidate (8a) remains optimal.
(9)
Constraint interaction, Y, Z
/input/
Y
a.
[candidate a]
*
b.
[candidate b]
**!
Z
(9) illustrates the final example of variable dominance. Candidates (9a) and (9b) violate
constraint Y once, with candidate (9b) incurring a second Y violation. This additional
violation is fatal, rendering Z irrelevant; the order Z, Y still selects (9a). The same is true
if only constraint Z is violated in the same unequal manner.
Now that the role of markedness and faithfulness constraints in mediating the
relationship between input-output segments has been explored, I contrast this with
alignment, the third constraint category available in Con. Alignment constraints are
unique in that they do not regulate input-output correspondences, but rather the relative
collocation of edges within phonological and morphological domains as they pertain to
output forms (McCarthy and Prince 1993). (10) provides the alignment template.
200
(10)
ALIGN(Category1, Edge1, Category2, Edge2)
Category1Category2 such that Edge1 of Category1 and Edge2 of Category2
coincide
‘For every category1 there exists some category2 such that the {L, R} edge of
category1 and the {L, R} edge of category2 coincide.’
‘Edge’ may refer to either the left or right margin of a given domain; ‘category’ specifies
the domains: e.g., affixes and feet (McCarthy and Prince 1993) relative to prosodic
words; segments relative to syllables and syllables to segments (Itô and Mester 1993,
1999). For the analysis of Spanish /f/ and /x/ in section 5.3, the family of markedness
constraints that compels fricative-vowel assimilation involves the alignment of particular
consonantal and vocalic gestural landmarks (Gafos 2002, Davidson 2003) within
Articulatory Phonology (Browman and Goldstein 1986, et seq.). Adopting AP is
necessary given the inadequacy of the alternative analyses considered in the next section.
5.2.2. Alternative analyses and featural assumptions
Beyond the basic principles of Optimality Theory outlined above, two additional
aspects must be considered to complete the analysis for Spanish: (i) motivating and
encoding the particular markedness constraints necessary for these processes; and (ii)
restricting the application of these constraints to the set of non-strident fricatives /f, x/. I
begin with restricting assimilation from point (ii).
Various proposals have been put forth to limit the application of markedness
constraints given that, as indicated above, languages diverge concerning which marked
structures are tolerated in differing phonological conditions. The conclusion from this is
that OT must have at its disposal faithfulness constraints that leave susceptible those
201
segments that succumb to markedness reduction while simultaneously protecting those
segments that resist it.
In what follows, I briefly review four proposals of such faithfulness constraints
and comment on their tenability concerning the Spanish data. The first to be addressed is
Positional Faithfulness.
5.2.2.1. Positional Faithfulness
Positional Faithfulness (Beckman 1998) prioritizes privileged positions, notably
the onset. Assuming a place assimilatory imperative between heterosyllabic adjacent
consonants [VC1.C2V], there are two possible outcomes: C1 adopts the place of C2, or C2
the place of C1. To account for the frequency with which languages prefer unfaithfulness
in the coda (C1), Beckman proposes IDENT-ONSET(Place) to preserve place features in
the onset (C2) at the expense of place features in the coda given the additional cue to
place afforded by the vocalic transition between onset/C2 and V. However, such a
constraint is in fact violated in assimilations of the type exhibited by Spanish due to two
crucial differences: the context is CV and not CC, and it is precisely the onset consonant
that assumes the place of the following vowel. Consequently, the context for application
of an IDENT-ONSET analysis is unfulfilled.
Other proposals by Lombardi (1999), Petrova, Plapp, Ringen and Szentgyörgyi
(2006) and Rubach (2008) (inter alia) specifically apply Positional Faithfulness to CV
contexts by positing an IDENT-PRESONORANT(x) constraint that preserves x features
(e.g., voice, place) in presonorant position. For present purposes, IDENT-PRESON has
ostensibly the same effect as IDENT-ONSET in protecting the consonant’s features in
prevocalic (CV) position. Consequently, it also cannot be responsible for the
202
asymmetrical application of assimilation in Spanish since unfaithfulness in the consonant
occurs precisely in presonorant position.
5.2.2.2. Preservation constraints
The second of the four alternative approaches is found in Jun (1994, 1995). While
I ultimately incorporate the fundamental ideas behind Jun in section 5.3.4.1, I briefly
introduce the concepts here. Like Positional Faithfulness, Jun is concerned only with CC
contexts. However, beyond coda-onset asymmetries, Jun explores differences between
targets and triggers with respect to place and manner. The proposal is as follows.
Table 5.3. Differences in place/manner concerning likelihood of target/trigger status
Target [VC1.
Most likely
.C2V] Trigger
Least likely
Manner
nasal
stop
fricative
Place
coronal
labial
dorsal
Most likely
Least likely
stop
nasal
dorsal
labial
coronal
Jun conducts an extensive crosslinguistic survey and, with the generalizations from Table
5.3, captures the probabilities of triggering and targeted places and manners. The
underlying principle derives from the Production Hypothesis (Kohler 1990, Steriade
1993, Byrd 1994): acoustically weak segments are more likely targets (and less probable
triggers) than acoustically strong segments. Conversely, acoustically strong segments
more readily resist reduction (and more frequently trigger assimilation) than acoustically
weak segments.
Given a sequence C1C2, the likelihood of C1 assuming the place of C2 is higher
when C1 is, e.g., a nasal than when C1 is a fricative (i.e., a nasal is a weaker coda than a
203
fricative and is consequently more likely to succumb to assimilation); stated inversely, C1
fricatives resist place assimilation more readily than C1 nasals given the strength of
fricative cues to place in coda position (crucially, note the positional statement given
previous discussions in this dissertation concerning cues to fricative place in CV
position). Regarding place, C1 coronals most readily assimilate while dorsals resist; C2
dorsals, similarly, are the strongest triggers while coronals the weakest.
Jun encodes these empirical observations in a family of PRESERVATION
constraints in which place preservation of the least likely targets and most likely triggers
(shaded above) is prioritized.
(11) Target PRESERVATION
a.
PRES [fric]-C » PRES [stop]-C » PRES [nas]-C
b.
PRES [dor]-C » PRES [lab]-C » PRES [cor]-C
Trigger PRESERVATION
c.
PRES C-[nas] » PRES C-[stop]
d.
PRES C-[cor] » PRES C-[lab] » PRES C-[dor]
The family of faithfulness constraints in (11a) penalizes place disagreement between
input and output fricatives above stops, and between input and output stops above nasals.
Since nasals are the likeliest C1 target, their preservation in this position is least
prioritized. (11b) applies the same principle to place instead of manner, preserving first
dorsals above labials, then coronals.
(11c), on the other hand, approaches preservation from the perspective of the
trigger C2. As nasals are weaker triggers to assimilation than stops, preserving the place
of C1 when followed by a nasal is prioritized over preserving the place of C1 when
followed by a stop. (11d) encodes the same for differing place in the C2 consonant: C1
204
place preservation is given priority when C2 is a coronal given coronal consonants’
relative weakness as a trigger. At the other end of the spectrum, C1 place preservation is
least preferable when C2 is dorsal given dorsal segments’ greater ability to trigger
assimilation.
Depending on where in each hierarchy the assimilation-driving markedness
constraint is collocated, place and/or manner will behave asymmetrically given the
segmental subset to which the faithfulness constraints apply. This is the very idea
necessary to explain why /f/ and /x/ (non-strident fricatives) in CV position are
vulnerable assimilatory targets while all other manners in Spanish are realized faithfully.
The precise formalization from (11), however, must be reconsidered, as Jun (1994, 1995)
was only concerned with CC contexts.
The notion of manner-specific faithfulness constraints, on the other hand, is Jun’s
crucial contribution, and what I ultimately propose for the asymmetrical effect of
assimilation with non-strident fricatives versus all other manners in CV contexts in
Spanish (see §5.3.4.1).
5.2.2.3. Perceptual Faithfulness
One antecedent for manner-specific faithfulness outside CC contexts is Herrick
(1999). To account for the differential effect of place assimilation relative to manner in
Catalan, Herrick adapts the basic idea behind Jun by disassociating cue strength from the
CC context and generalizing perception to all contexts in the theoretical mechanism.
Herrick (1999: 30) argues that each manner contains x number of cues to place depending
on phonetic conditions.
205
Table 5.4. Variable cues to place depending on condition by manner
Intervocalic V_V
Pre-sonorant (C)_V
Coda V_(C)
x cues
Manner
x–1
Manner
x–1
Manner
4
Fricative
3
Fricative
3
Fricative
3
Stop
2
Stop
2
Stop
3
Lateral
2
Lateral
2
Lateral
2
Nasal
1
Nasal
1
Nasal
As shown by Table 5.4, Herrick’s argument is that, phonetically, fricatives contain the
greatest number of cues in all conditions: noise (i.e., frication/spectral cues), duration,
vowel onset transition and vowel offset transition. In pre-sonorant position, fricatives
only lose the vowel offset transition leading into the C as there is no longer a preceding
vowel; similarly for the final condition, fricatives only lose the vowel onset transition
given the absence of a following vowel. Stops only contain noise (assumedly, release
burst energy) and vocalic onset and offset cues (lost in a similar fashion as fricatives
depending on context). In addition to the same vowel offset/onset cues, laterals contain a
unique cue in that laterality implies coronal given the marked structure of non-coronal
laterals (see Herrick 1999: 30 for further justification). Lastly, Herrick argues that nasals
are cued only by vocalic transitions from both directions, a situation which results in one
total cue in the final two contexts if one of the surrounding vowels is removed.
These claims are encoded by IDENT faithfulness constraints in an implicational
hierarchy that seek to preserve place depending on how well-cued a given segment is in
each context.
(12)
ID(Place) 4 CUES » ID(Place) 3 CUES » ID(Place) 2 CUES » ID(Place) 1 CUE
206
No matter where the markedness constraint is located, if (12) is applied to Spanish, the
prediction is not borne out. To force CV assimilation in fricatives, the markedness
constraint that compels assimilation at the expense of place faithfulness must dominate
IDENT(Place) 4 CUES, as Herrick argues that fricatives contain four place cues in the
relevant VCV position. Given the implicational hierarchy from (12), this necessarily
entails assimilation in all other manners as they are less cued than fricatives in all other
positions. This prediction was previously discarded given the discussion concerning the
unattested outcome of OCP proposals for Spanish (Lipski 1995, Mazzaro 2005), in
addition to the crosslinguistic data from Chapter 1 and the results of Bateman’s (2007)
survey that show that languages may select particular manners for CV assimilation
outside of an implicational hierarchy as Herrick proposes.
While perceptual faithfulness is one way to apply Jun’s idea of manner-specific
faithfulness constraints, Herrick errs in his justification for each cue. To begin, I argued
in Chapter 3 that a number of sources (Jongman et al. 2000; Gordon, Barthmaier and
Sands 2002; Silbert and de Jong 2008; McMurray and Jongman 2011) have demonstrated
that duration in fricatives correlates with voicing, not place. Furthermore, I assume that
“noise” in Herrick’s terms concerns frication and/or spectral cues, which I have argued,
based on numerous studies cited previously, are inherently variable and consequently
unreliable. If duration is removed as a cue and noise weighted as 0.5 instead of 1 given its
inaccuracies, fricatives are left with 2.5 cues in intervocalic position, 1.5 elsewhere.
Turning to nasals, Herrick claims that only vocalic transitions cue place.
However, the acoustic literature has argued that listeners are able to take advantage of
formants and antiformants to cue nasal place (Ohala and Ohala 1993, Ladefoged 2001,
207
Wright 2004). Although place cues in nasal formants are weak relative to other cues in
other manners of articulation (Wright 2001, Ladefoged 2006: 193), the crucial point here
is that they are not nonexistent, as Herrick argues. If (anti)formant cues are weighted as
0.5 given their relative weakness, nasals now contain 2.5 cues in intervocalic position,
equivalent to fricatives (assuming all fricatives are equal by ignoring the robust effect for
stridency on sibilants (Wright 2001, 2004)). Applying this adjusted hierarchy to Table 5.4
results in the following: stops, laterals (3 or 2 cues depending on position) » fricatives,
nasals (2.5 or 1.5 cues). Such an account–treating (all) fricatives and nasals equally–still
does not explain the susceptibility of fricatives and resistance of nasals to CV
assimilation in Spanish, as both are predicted to be similarly vulnerable (or resistant).
The conclusion from the above discussion is that OT must be able to encode
asymmetric differences in terms of priority of place preservation relative to manner of
articulation. Positional Faithfulness (Beckman 1998, inter alia) and PRESERVATION
(Jun 1994, 1995) constraints principally protect onset/presonorant consonants, which do
not apply to the Spanish assimilations. Manner-specific Perceptual Faithfulness
constraints, following Herrick (1999), fail empirically as the hierarchical implication is
untenable for Spanish. Disassociating the precise number of available cues from the
constraints is one possible solution, as Wright (2001: 271) argues that specific cues must
be weighted differently (rather than being entirely absent or present) in different positions
under different noise conditions. I ultimately pursue this avenue in section 5.3.4.1 in
limiting the markedness constraint(s) applicable to Spanish.
For the moment, I set aside faithfulness and return to point (i) from the beginning
of section 5.2.2: motivating and encoding the markedness constraints. As I argue in
208
Chapter 4, an assimilatory imperative between a consonant and a following vowel best
accounts for the data. Irrespective of the precise formalization, the constraints must
necessarily operate over a set of features, given the current understanding of OT. The set
of features assumed, however, is not inconsequential since the constraint and its effect are
dependent on what each feature actually defines. While I explore, in section 5.2.3,
Articulatory Phonology (Browman and Goldstein 1986, et seq.)–the theory behind the
markedness constraints that I adopt (which happens to be accompanied by its own
assumptions about the nature of features)–I address possible alternatives to these in the
following four subsections, explaining why they fail and, consequently, why AP must be
assumed.
5.2.2.4. The Sound Pattern of English
The Sound Pattern of English (SPE, Chomsky and Halle 1968) represents one of
the seminal works in generative phonology (see also Jakobson, Fant and Halle 1952).
Take as a representative case for this discussion /f/ bilabialization. Given the rarity of
output [ɸ], it is arguably the most difficult of the three processes to characterize
theoretically.
SPE relies on linear rule application, taking an input and converting it into the
corresponding output under specifically defined conditions. The SPE place features
relevant for /f/ bilabialization in Spanish are as follows.
Table 5.5. Place features in SPE /f/ bilabialization
Input /f/
Output [ɸ]
Trigger condition /u, o/
[+anterior, –coronal, –back]
[+anterior, –coronal, –back]
[+back, +round]
209
The SPE consonantal place features for /f/ are [+anterior, –coronal, –back]; this feature
matrix is equivalent to output [ɸ]. The context in which this process occurs is preceding a
[+back, +round] vowel. Assuming SPE features is problematic for two reasons. The first
is that, featurally, nothing has changed from the input to the output. Concerning the
principal place features, SPE has no method of distinguishing /f/ and [ɸ]. Chomsky and
Halle (1968: 312) do posit a “special feature” (their term), [±distributed], to phonemically
contrast [–distributed] /f/ and [+distributed] /ɸ/. However, the current assumption
(McCarthy 1998) is that [distributed], like [strident] (see discussion below), is a coronal
dependent, distinguishing (namely) [–distributed] alveolars and retroflexes from all other
coronally-articulated
[+distributed]
segments
((inter)dentals,
palatoalveolars,
alveopalatals and palatals). Even so, [distributed] was not defined on vocalic segments in
SPE, leaving unexplained why /u/ and /o/ seem to trigger a change from [–distributed] /f/
to [+distributed] [ɸ].
The second reason that SPE fails is that such a rule does not capture the
assimilatory nature of the process; there is nothing in the feature matrix for [ɸ] that
clearly originates on /u/ or /o/. A rule such as this would better exemplify assimilation if a
specific feature from the trigger context would be shared by the output consonant. This is
not the case here. Consequently, no assimilation is motivated, and no change theoretically
occurs.
One reason that SPE fails to capture CV assimilations in general is that it assumes
separate sets of place features for consonants and vowels. If it wishes to account for such
processes, additional theoretical machinery is necessary in the form of consonant-vowel
featural equivalency/conversion rules: e.g., x consonantal feature↔y vocalic feature.
210
Applying Occam’s Razor, these rules unnecessarily complicate the theory. For this
reason, Unified Feature Theory assumes one set of features for both segmental types.
5.2.2.5. Unified Feature Theory
To straightforwardly account for direct CV interactions, Unified Feature Theory
(UFT, Clements 1989, Clements and Hume 1995) does not distinguish between
consonantal and vocalic place features. It does, however, distribute them across distinct
tiers: C(onsonant)-place and V(owel)-place, retaining the hierarchical geometric structure
of Autosegmental Phonology (Goldsmith 1976). Although separate, interaction between
the two tiers is permitted given the shared C-place node (as a theoretical restriction, only
‘like’ tiers may interact).
Figure 5.2. UFT geometric representations for place in consonants and vowels
(a) Consonants
(b) Vowels
Root
|
C-place
Root
|
C-place
|
vocalic
|
V-place
[labial]
[labial]
[coronal]
[coronal]
[dorsal]
[dorsal]
Under UFT, consonants and vowels now share place features and the C-place node that
ultimately dominates place. In this way, direct interaction is encoded in the spreading of
211
equivalent tiers. Ignoring the geometric structure for the moment, bilabialization in
Spanish is now characterized by the interaction of the features in Table 5.6.
Table 5.6. Place features in UFT /f/ bilabialization
Input /f/
Output [ɸ]
Trigger condition /u, o/
[labial]
[labial]
[labial]
UFT is highly reductive in the sense that all three components–input, output and context–
are defined by the privative feature [labial]. This uniform treatment of place features is an
improvement over SPE, as assimilation is now captured by feature sharing. However, this
assimilation is vacuous in the sense that, while the trigger and the output do share a
feature, it is not distinct from the input, which again suggests a lack of motivation for any
process to apply at all. If the imperative is to share a [labial] feature between consonant
and following vowel, [fo] and [fu] are already in compliance with such a rule/constraint,
removing any justification for modification, rendering unfaithfulness to input extraneous.
To avoid this lack of distinction, some proposals recruit the feature [±strident] (or
simply [strident], if privative) to contrast /f/ and /ɸ/ (see Utman and Blumstein’s (1994)
discussion of Ewe). This is still problematic given that Nartey (1982), Clements and
Hume (1995) and Ladefoged and Maddieson (1996) argue that such a feature should be
reserved for sibilants and relegated to coronal-dependent status. Furthermore, [strident] is
undefined/irrelevant on vocalic segments, and the loss of a strident feature from /f/ in the
context of /u, o/ is not justified.
At this point, the issue of (non-)stridency concerning /f/ is worth addressing.
There are three lines of argumentation that support the treatment of /f/ as non-strident.
212
The first is acoustic. If stridency is defined as noisy high-frequency, high-amplitude
frication, Ladefoged and Maddieson (1996: 138), Ladefoged (2006: 170) and Martínez
Celdrán and Fernández Planas (2007: 105) all demonstrate that uncontroversial stridents
([s], e.g.) attest higher concentrations of frication (at a greater intensity as measured by
dB) than [f].
The second point is articulatory in nature. If stridency is instead defined as the
result of an obstruction to the air flow (such as the teeth) in addition to the original
constriction point (e.g., place of articulation) (Ladefoged and Maddieson op. cit.), then
labiodental [f] does not qualify, since only one point of constriction is present in [f]: the
original labiodental place of articulation. This contrasts with the canonical sibilant [s] in
which the stream is both obstructed at the alveolar region then additionally by the teeth as
the air exits the oral cavity. [f] lacks this second obstruction since the teeth participate in
the primary constriction point.
The third argument is phonological. While UFT treats /f/ as [–strident], Halle
(1992) and Rubach (1994) place [strident] as a [coronal] dependent (not unlike
[distributed]). Under this assumption, stridency is undefined on /f/. Furthermore, Mielke
(2008) presents evidence that /f/ does not form a natural class with the sibilants,
indirectly implying that /f/ is either [–strident] or lacks a [strident] specification entirely.
For the purpose of the Spanish proposal that follows, I consequently assume that /f/ is
non-strident. Whether that is formalized as [–strident] or a lack of specification for
stridency is not crucial for the analysis.
Thus, while UFT accounts for CV interaction and avoids conversion rules, it is
still inadequate. In the next section, I explore the possibility of Element Theory.
213
5.2.2.6. Element Theory
Element Theory (ET, Harris 1994, Harris and Lindsey 1995) assumes that a set of
atomic elements that reflect basic spectral properties can be used (and combined) to
define segments. Each element indicates a particular spectral shape present in the
acoustics of the segment that it purports to define. To characterize /f/ bilabialization, only
two elements are necessary: |U| and |A| in Figure 5.3 (Backley 2011: 23, 24, 27).
Figure 5.3. Spectral shape of place element singletons |U|, |A| and complex |U A|
(a)
|U|
(b)
|A|
(c) |U A|
Two arguments in ET are relevant here. First, ET contends that elements can be
combined in complex segments (c) and that segments are always headed. Headedness
applies to the element that contributes the greatest spectral prominence to the segment. In
the case of segments defined by a single element ((a) and (b)), there is only one possible
element to bear prominence.
In complex segments (c), only one element of the two is allowed to head the
segment. Element (a), when headed (ET convention underlines the element: |U|), defines
both the vowel /u/ and the bilabial fricative /ɸ/, given shared spectral properties (the
214
obvious voicing and manner differences are accounted for by additional elements that are
not relevant here). The complex segment |U A| in which |A| is a dependent of headed |U|
defines both /f/ and /o/ (with the same caveat as above).
Assuming these elemental matrices for the necessary segments to describe /f/
bilabialization, observe Table 5.7.
Table 5.7. Place features in ET /f/ bilabialization
Input /f/
Output [ɸ]
Trigger condition /u, o/
|U A|
|U|
|U|, |U A|
When the ET features are viewed in light of Spanish /f/ bilabialization, the criticisms
from the two previous theories recur. First, assimilation between /f/ and /u, o/ is
unmotivated, as both /f/ and the triggering vowels contain |U|, the only element necessary
to result in [ɸ]. In other words, any place-feature sharing constraint is vacuously satisfied
as /f/ and the vowels already share the only feature that can account for the change.
Furthermore, bilabialization in the context of /o/ is uniquely curious; ET would
(mis)characterize such a process as a dissimilation given that /f/ and /o/ completely share
|U A|, while output [ɸ] necessitates a loss of |A|. This contradicts the coarticulatory facts
of Chapter 4. With SPE, UFT and ET discarded, the final feature theory under
consideration is the Parallel Structures Model.
5.2.2.7. Parallel Structures Model
Parallel Structures Model (PSM, Morén 2003, 2006, inter alia) is an updated
incarnation of Unified Feature Theory. It differs, however, in a few key aspects. UFT’s
contribution of shared place features between consonants and vowels (in order to directly
215
account for consonant-vowel interactions) is extended to manner and laryngeal settings.
Furthermore, the hierarchy is reduced by removing the superfluous intervening vocalic
node for vowel segments (see Figure 5.2). What results is a parallel template for
laryngeal, manner and place features for both consonants and vowels.
Figure 5.4. PSM geometric representations
C-{laryngeal, manner, place} node
[feature x]
[feature z]
[feature y]
V-{laryngeal, manner, place} node
[feature x]
[feature z]
[feature y]
There are two issues with PSM. The first is theoretical. The geometric hierarchy of
Figure 5.4 was a crucial component in Autosegmental Phonology (Goldsmith 1976),
designed to account for phonological features that recurrently participate as natural
classes. Consequently, nodes could be manipulated in order to explain why one subset of
features is consistently active in phonological processes while another subset is not. The
rationale was that the active natural class formed an exhaustive unit under a single
dominating node.
OT constraints, however, need not impose an internal structure on segments (nor
do they conclusively rule it out). Feature Class Theory (FCT, Padgett 1995) argues that
constraints perform the work once accomplished by autosegments, rendering internal
216
structure redundant. FCT re-envisions geometric class nodes as set labels to which
constraints may refer. Any feature within a set is now relevant to constraints in two ways.
An OT constraint may either (i) reference the feature itself (e.g., [labial]) or (ii) reference
the set label to which that feature belongs–in this case, Place, comprising the set of
features {[dor], [lab], [cor]}. The putative node dominating those features is irrelevant to
the constraint manipulating them since the constraint can bypass higher structure by
specifying the class label directly.
PSM, however, requires internal structure to function theoretically. It is stipulated
that vowel (root) segments cannot have a C-class node filled by any consonantal features,
but the C-class node must remain in the structure to dominate the V-class node (and its
vocalic features) for the sake of parallelism and to permit interaction with consonants
(Youssef 2013: 12). This requirement is unmotivated outside its own theoretical ends.
The second issue is empirical. While PSM restructured the organizational
hierarchy of Unified Feature Theory, the (shared, privative) place features themselves are
unchanged.
Table 5.8. Place features in PSM /f/ bilabialization
Input /f/
Output [ɸ]
Trigger condition /u, o/
[labial] (C-place)
[labial] (C-place)
[labial] (V-place)
As a consequence, the same UFT criticisms apply here: /f/→[ɸ] in the context of /u, o/ is
unmotivated assuming a place-feature sharing constraint since /f/ and /u, o/ are both
already [labial].
217
To summarize, the feature sets of The Sound Pattern of English, Unified Feature
Theory, Element Theory and Parallel Structures Model cannot capture the facts for
Spanish /f/ bilabialization. While both [strident] and [distributed] have previously been
utilized to distinguish /f/ and /ɸ/, both have recently been argued to be coronal
dependents (i.e., irrelevant to labials), and neither feature is defined on the triggering
vowels /u, o/ to explain the assimilation. Consequently, the features to which the relevant
OT constraints refer cannot be based solely on the above theories; i.e., an alternative
approach to phonological features is necessary. As mentioned above, this approach is
Articulatory Phonology (AP). In the next section I review in detail the fundamental
components of Articulatory Phonology.
5.2.3. A general overview of Articulatory Phonology
Given the issues discussed above, I follow work in Articulatory Phonology (AP)
by Browman and Goldstein (1986, et seq.), Byrd (1996), Gafos (2002), Kochetov (2002),
Davidson (2003), Bateman (2007) and Bradley (2014, forthcoming) for the featural
representations and markedness constraints adopted in the analysis of Spanish /f/ and /x/
assimilations. AP argues that the grammar abstractly encodes not only linguistic spatial
relations (e.g., place of articulation, a notion shared with the models of the previous
section), but also linguistic timing. As argued in Chapter 4, significantly greater degrees
of coarticulation–the temporal overlap of gestures–were found in the precise fricativevowel conditions that exhibit allophony. Furthermore, variation in production expressed
as a COG continuum from most [f]-like/least [x]-like to most [x]-like/least [f]-like (on a
scale of /f/ velarization, for example) was also present in the output. AP formally captures
both aspects: gestural overlap and gradience in production.
218
In AP, gestures are contrasted via a set of tract variables, the abstract
representations consisting of constriction degree (CD, cf. consonantal manner and vocalic
height in other feature sets) and constriction location (CL, cf. place) for each main
articulator (lip, tongue tip and tongue body). Two other tract variables not directly related
to the analysis at hand–velic aperture (nasality) and glottal aperture (laryngeal contrasts
such as aspiration, voicing, etc.)–complete the set of eight possible AP tract variables
(Browman and Goldstein 1989: 207).
Table 5.9. Articulatory Phonology tract variables
Tract variable
Articulators
LP
lip protrusion
(location)
upper and lower lips, jaw
LA
lip aperture
(degree)
upper and lower lips, jaw
TTCL
tongue tip constriction location
tongue tip, tongue body, jaw
TTCD
tongue tip constriction degree
tongue tip, tongue body, jaw
TBCL
tongue body constriction location
tongue body, jaw
TBCD
tongue body constriction degree
tongue body, jaw
VEL
velic aperture
(degree)
velum
GLO
glottal aperture (degree)
glottis
Specifically, constriction location (CL) refers to place of articulation and is described by
the set of features [labial], [dental], [alveolar], [palatoalveolar], [palatal], [velar], [uvular]
and [pharyngeal] (Gafos 2002, citing Ladefoged 1989), essentially expanded from the
traditional [labial], [coronal] and [dorsal] distinction in UFT and PSM. I argued that a
finer-grained theoretical notion of place was necessary given the inability of [labial]
alone to describe /f/ bilabialization in Spanish (or to successfully contrast /f/ and /ɸ/ in
languages like Ewe). Below, I will demonstrate how [labial] and [dental] accomplish
219
precisely this under AP’s gestural assumptions. Figure 5.5 (adapted from Browman and
Goldstein 1990: 343) locates each variable from Table 5.9 on the vocal tract.
Constriction degree (CD) contrasts seven features (Browman and Goldstein
1990): [closed], [critical], [narrow], [narrow, mid], [mid], [mid, wide] and [wide].
[closed] and [critical] specify stops and fricatives, respectively (cf. [±continuant] (binary)
or Ø versus [continuant] (privative) in standard featural assumptions). [narrow] through
[wide] primarily distinguish vowel height (high to low) but do not factor crucially into
the analysis.
Figure 5.5. Lateral view of vocal tract with AP tract variables
TBCL
TTCL
VEL
LP
LA
TTCD
TBCD
GLO
One representation for the temporal relationship between gestures is the gestural score,
exemplified in Figure 5.6 (adapted from Browman and Goldstein 1991: 316). Tract
variables are arrayed vertically (y-axis) (cf. autosegmental tiers), time is expressed
220
horizontally (x-axis) and contrasting features (gestures) are specified in boxes whose
amount of overlap can be varied depending on degree of coarticulation.
As is seen in Figure 5.6 (below), at the beginning of the utterance, a closed velum
is necessary to articulate the oral gesture for the voiceless alveolar fricative [s] with the
following tract variables: major articulator, tongue tip (TT); constriction location,
alveolar; constriction degree, critical.
Figure 5.6. Gestural score for spam [spæ̃ m]
VEL
closed
LIPS
wide
lab. closed
TT
lab. closed
alv. critical
TB
pharyngeal wide
GLO
wide
[
s
narrow
p
æ̃
m
]
Superficially, it is curious that the articulation of the low front vowel [æ] (tongue body;
CL pharyngeal; CD wide) also begins at the syllable’s left edge (and completely extends
to the right edge) if it is assumed that, even allowing for slight coarticulatory overlap,
segments are for the most part produced consecutively. I will return to this point below.
For the moment, consider the articulation necessary for [p]: a bilabial closure with a
narrow specification for glottal contribution upon release (voiceless unaspirated).
Crucially, note that gestural scores in AP are able to account for coarticulation, as the
221
timing boxes for [s] and [p] overlap. So too do the boxes for [p] and [æ], as well as [æ]
and the velic opening necessary for [m], resulting in nasalized [æ̃ ] immediately preceding
the bilabial closure for [m]. It is by way of gestural scores that AP accounts for
coarticulated phenomena, a crucial component of the theoretical analysis to follow.
Yet before proceeding to discuss how gestural scores have been encoded in OT, I
return to the issue of [æ]. Many phonetic studies have argued that consonantal gestures
are superimposed over vocalic gestures that span the length of the syllable (Kozhevnikov
and Chistovich 1965; Öhman 1966; Perkell 1969; Fowler 1980, 1983; Coleman 1992,
2001). Even though such vocalic gestures might not be perceptible at the extreme left
edge of the syllable, Gafos (1999) argues that this is not due to their absence; instead, the
onset consonant is articulated with its greatest constriction at this point, and the vocalic
gestures are consequently obscured in the acoustic signal. To illustrate, a simplified view
of Figure 5.6 is provided below.
Figure 5.7. Consonants superimposed over nucleic vowel within the syllable
σ
qp
C1 (C2)
C3 (C4)
V
This structure is further corroborated by intrusive vowel data. An intrusive–or
svarabhakti–vowel is a phonetic vocalic gesture perceived between consecutive
consonants which happen to be articulated with little overlap (otherwise CC overlap
would block the surfacing of the vowel). In these contexts, an [ə]-like vowel is possible if
222
the articulators are in a neutral position or if the interval between the consonants is
minimal.
A second, more enlightening possibility is that the intrusive vowel heard between
the consonants is of the quality of the nucleic vowel (Hall 2006), evidence in support of
the structure from Figure 5.7. These intrusive ‘copy’ vowels are possible between both
onset and coda clusters, indicating that consonants are overlaid upon vowels at either
margin.
Table 5.10. Intrusive vowels within onset and coda clusters
Position
Language
Word
Gloss
Standard
Intrusive V
Onset
Spanish
crónica
‘chronic(le)’
[kɾonika]
[koɾonika]
Coda
Scots Gaelic
tarv
‘bull’
[tarv]
[tarav]
Source: Spanish, García-Bellido (1999); Scots Gaelic, Hall (2006).
To better demonstrate the point, Figure 5.8 applies the structure from 5.7 to the relevant
syllables in the Spanish and Scots Gaelic data.
Figure 5.8. Intrusive vowels in Spanish and Scots Gaelic
(a) Spanish, c(o)rónica
k
(b) Scots Gaelic, tar(a)v
t
ɾ
o
r
v
a
As is made explicit in these figures, it is the minute gap between the articulations of the
consonants that permits the underlying vowel to surface as intrusive, evidence that
vowels serve as the syllabic substratum on top of which consonants are overlaid.
223
The point relevant to present purposes is that such arguments further corroborate
the validity of the regressive CV analysis, as the consonant in the crosslinguistically
possible but less frequently attested progressive VC analysis (e.g., b[o.f]etada,
*b[o.x]etada ‘slap’) is actually superimposed over the vowel that serves as its nucleus
([fe]) (Gafos 2002), and not the preceding vowel ([o]), accounting for the reduced
influence of vowels in this position relative to the consonant in question.
At this point, I have reviewed the main arguments in Articulatory Phonology and
have demonstrated how they respond to the issues raised by Spanish /f/ and /x/ processes
for which the alternative theories presented could not account: (i) features that truly
motivate the processes and (ii) a theoretical mechanism that addresses the coarticulatory
(temporal) facts in addition to the spatial (place of articulation) ones. The next section
explores how current OT approaches have incorporated the fundamental tenets of AP.
5.2.4. Articulatory Phonology-based constraints
in Optimality Theory
Gafos (2002: 276) translates Browman and Goldstein’s gestural score into OT
constraints that operate over five (spatial (y-axis) and temporal (x-axis)) landmarks
inherent in the articulation of each segment.
Figure 5.9. Articulatory Phonology landmarks in Optimality Theory
target
onset
center
release
offset
224
Each landmark corresponds to an identifiable point in articulation (as applied to a single
segment), the onset being the point at which the articulators begin to move. The target
denotes the point of contact between the speaker-intended active and passive articulators.
The center is the midpoint in the length of time devoted to maintaining constriction at the
target until release, when articulators start their movement away from the target. Offset is
the release of control of the active articulator and is, in the default scenario, coordinated
with the onset of the following segment.
The OT markedness constraints that motivate assimilation rely on these APinfluenced landmarks and, in this way, capture the vowel-dependent coarticulation facts
that proved crucial for description of the properties in Chapter 4. This is accomplished by
permitting OT alignment constraints to refer to gestural landmarks and, in doing so,
capture how language permits gestures to coordinate. (13) repeats (10), the standard OT
formalization of alignment.
(13)
ALIGN(Category1, Edge1, Category2, Edge2)
Category1Category2 such that Edge1 of Category1 and Edge2 of Category2
coincide
‘For every category1 there exists some category2 such that the {L, R} edge of
category1 and the {L, R} edge of category2 coincide.’
Applied to Gafos (2002), ‘category’ corresponds to a segmental value {C consonant, V
vowel} while ‘edge’ refers to the landmarks from the set {onset, target, center, release,
offset} to be coordinated. While all possible combinations of {C, V} with the five
landmarks result in 20 unique phonetic configurations (CC, CV, VC, VV, five distinct
alignments each), only three main cases that describe phonological phenomena are
necessary here.
225
Within the CV context relevant for the Spanish data, if the vocalic onset remains
constant, it may increasingly encroach on the consonantal landmarks in any one of five
specified positions (Figure 5.9). This is not to say that gradient intermediate positions do
not exist with a phonetic consequence (as measured by locus equations). For the purposes
of demarcating how one configuration is reported phonologically compared to another,
however, the landmarks serve as divisions. Observe Figure 5.10.
Figure 5.10. Three distinct CV alignments
(a) Little to no overlap [CV]
C
V
(b) Secondary articulation [CVV]
C
V
(c) Assimilation [C͜ V]
C
V
In the case of (a), negligible coarticulation is attested when C offset and V onset are
aligned (Bradley 2014) (furthermore, if scenario (a) were instead between consecutive
consonants, it is precisely the configuration necessary for intrusive vowels given the
sizeable gap between gestures (Gafos 2002, Hall 2006)). Scenario (b) overlaps vocalic
with consonantal articulations one landmark to the left, aligning C release (instead of
offset) and V onset, the result of which is a derived secondary articulation on C (Bateman
2007).
Overlapping the vowel by one more landmark leftward, (c) aligns C center and V
onset, at which point the beginning of the vocalic gesture completely encroaches on the
target duration of the consonant, resulting in complete place assimilation of the consonant
to the vowel (Bateman 2007, citing Gafos 2002).
226
Further overlapping the vocalic gestures two more landmarks to the left is the
logical extension of the series, yet phonologically there is no difference, as complete
assimilation to place is achieved by (c); the effect of such utterances is revealed in locus
equation calculations as extreme coarticulation approaching 1.
(14) encodes the AP landmark structures from Figure 5.10 into OT alignment
constraints.
(14)
a.
ALIGN(C, offset, V, onset) (Bradley 2014)
In a CV sequence, align the offset of C with the onset of V.
b.
ALIGN(C, release, V, onset) (Bateman 2007)
In a CV sequence, align the release of C with the onset of V.
c.
ALIGN(C, center, V, onset) (Bateman 2007, citing Gafos 2002)
In a CV sequence, align the center of C with the onset of V.
If constraints (14a), (14b) and (14c) dominate the necessary faithfulness constraints, the
resultant output will be articulated, respectively, in gestures corresponding to (a) (little
overlap), (b) (secondary articulation) and (c) (C place assimilation) from Figure 5.10.
However, the constraints in (14)3 as stated above–particularly (14b) and (14c)–are
much too general, affecting any CV sequence irrespective of all other factors. As
Bateman (2007) and my small crosslinguistic sample from Chapter 1 show, languages are
able to select which vowels trigger CV place assimilation and which consonants are
susceptible to/protected from its effects. With the template from (14) in mind, the specific
constraints necessary for the three Spanish properties are presented in the following
section.
3
See also CV LINK (Zubritskaya 1995), OVERLAP CV (Lee 1997) and PLACE ATTRACTION
(Beckman 1999, 2000). The advantage of Gafos’s (2002) alignment account is the ability to
capture the gradient nature of coarticulation.
227
5.3. Optimality Theoretic analysis
of Spanish fricative-vowel assimilations
All the components are finally in place to begin the OT analysis of Latin
American Spanish /f/ and /x/ allophony. I address each process in turn: section 5.3.1
analyzes /f/ bilabialization, followed by /f/ velarization and /x/ palatalization in sections
5.3.2 and 5.3.3. After the basic assimilatory scenarios are in place, section 5.3.4 restrains
the grammar. The section concludes with a typology of Spanish grammars (with
crosslinguistic implications for CV assimilations in general) in section 5.3.5.
5.3.1. /f/ bilabialization
As the point of departure, consider what is minimally necessary to motivate
/f/→[ɸ] in the context of /u, o/ in OT. Table 5.11 summarizes the AP features that
describe the process (irrelevant tract variables VEL and GLO ignored).
Table 5.11. AP features for /f/ bilabialization
Target /f/
Trigger /u, o/
Output [ɸ]
lip protrusion
[labial, dental]
[labial]
[labial]
lip aperture
[critical]
[narrow], [mid]
[critical]
tongue body loc.
[velar]
tongue body degree
[narrow], [mid]
With the featural assumptions of AP (shared features are highlighted), two previously
unanswered issues can now be addressed simultaneously. The theory is able to contrast /f/
and /ɸ/ (in Ewe, e.g.) given the [dental] gesture of labiodental /f/ absent in bilabial /ɸ/.
Furthermore, since it is the gesture that is relevant in coarticulation, the assimilation in
Spanish is now motivated as a purely [labial] overlap, the result of which is a loss of /f/’s
228
[dental] gesture. There is no effect on the manner of articulation, as output [ɸ] is faithful
to input /f/’s [critical] gesture/feature, of which the phonological categorization is
‘fricative’ and phonetic manifestation is frication. Manner is maintained by virtue of the
markedness constraint to be proposed, which targets only place.
Limiting the conditioning context is also necessary given the discussion in the
previous section concerning the general nature of the alignment constraints and the
vowel-specific nature of the processes for which I argued in Chapter 4. I follow Bateman
(2007) and Bradley (forthcoming), who incorporate specific conditions such as triggering
context into the markedness constraint itself.
(15)
ALIGN-LP(C, center, V, onset)
In a CV context, align the LP onset gesture of V with the center of C.
‘Consonants share lip protrusion gestures with following vowels.’
(15) compels lip gesture sharing between a consonant and a following vowel
(specifically, at the central point of the consonant), the result of which is consonantal
place assimilation. Lip protrusion (location) must be specified, as languages differ in
terms of which vowels are allowed to participate in CV assimilations (see Lee 1997,
1999, Bateman 2007, Chapter 1 of this dissertation and the sources cited therein).
(15) additionally addresses the issue of research question 3, progressive (VC)
versus regressive (CV) assimilation. I have previously explored phonetic reasons why VC
interaction is less frequent; there is now a method of encoding this fact theoretically. A
separate ALIGN-VC constraint exists in the grammar (in all grammars), yet is
subordinated to a point such that its effect does not surface in Spanish. (This is also true
229
of the additional alignment constraints from (14); consequently, they will not be included
in the tableaux that follow.)
As OT seeks an optimal candidate for the language in question and not the
universally perfect one, satisfaction of (15) must come at a cost. Indeed, to be compliant
with ALIGN-LP, /f/ must relinquish its input labiodental gesture, specifically the dental
contribution. Input-output faithfulness is preserved via the IDENT family of constraints;
the relevant version appears in (16).
(16)
IDENT(LP)
Input lip protrusion gestures are preserved on output correspondents.
(16) requires that input segments maintain their lip protrusion gestures in output forms.4
To compel unfaithfulness, markedness must dominate faithfulness, as any
grammar that attests such a process prioritizes economy of articulation (sharing/overlap
of gestures; in this case, the purely (bi)labial gesture of the vowel) over input
maintenance (/f/ as [f]). This interaction is achieved in (17).
(17)
ALIGN-LP » IDENT(LP), IDENT
/foɾmo/ formo ‘I form’
4
a.
[foɾmo]
b.
[ɸoɾmo]
c.
[xoɾmo]
d.
[moɾmo]
ALIGN-LP
IDENT(LP)
IDENT
*!
*
*!
*
*
*!*
In complete form, (16) is technically IDENT(LP)-IO, indicating that the direction of evaluation
is unidirectional, input to output (Pater 1999), and not necessarily bidirectional (input↔output) as
commonly assumed. For simplicity, I omit IO here and in what follows under the understanding
that subsequent IDENT constraints are similarly IO.
230
The assimilated candidate (17b)5 is selected as optimal given its satisfaction of ALIGNLP resulting in bilabial [ɸ] (all else, notably frication and manner, being equal) due to the
strictly [labial] gesture of the following /o/. Its IDENT(LP) violation is permitted because
of the fatal ALIGN-LP violation of the faithful candidate (17a). Candidate (17c)
demonstrates that place unfaithfulness for unfaithfulness’s sake is ruled out, as velar [x]
not only violates IDENT(LP), but does so not to satisfy markedness, as it additionally
violates ALIGN-LP. Although /u, o/ are indeed specified as velar given that the back
vowels involve movement of the tongue body toward the velum/dorsal region of the
vocal tract, assimilating to the [velar] feature/gesture of /u, o/ is not motivated since
ALIGN-LP in this grammar does not make reference to velar overlap (see, however, the
next section). Candidate (17d), while satisfying ALIGN-LP since [m] is bilabial, is in
violation of IDENT, included in the general sense to protect manner and voice. That is,
any category (voice or manner) of unfaithfulness not motivated by ALIGN-LP is in
violation of general IDENT and is also ruled out, as candidate (17a) is (comparatively)
minimally unfaithful to the input by diverging in place alone. There is no ranking
argument here as (17d) still loses under the order IDENT, IDENT(LP). I will assume that
candidates like (17d) are similarly ruled out in the sections that follow and will not
include them (or the general IDENT constraint) elsewhere.
Subordinate to both IDENT constraints are the additional alignment constraints
previously discussed, as their satisfaction is outranked by IDENT. I exclude them for
5
Additionally possible but unlisted candidates that tie with (17b) in terms of optimality are
[f oɾmo], [ɸfoɾmo] and [xɸoɾmo]. To rule out derived secondarily articulated consonants, I
assume that a constraint like that of *SUPERIMPOSE (“segments must not be associated with
both a primary and a secondary oral gesture”; Bateman 2007: 235) is undominated in Spanish.
ɸ
231
clarity here and in all additional tableaux, but the OT assumption that they exist in the
hierarchy yet are not high enough to matter remains.
With the crucial order ALIGN-LP » IDENT(LP), I have accounted for /f/
bilabialization in Spanish: the result of coarticulation between a consonant and a
following [labial] vowel. The next section addresses the alternative result of /f/ allophony
in Spanish: velar [x].
5.3.2. /f/ velarization
In the same /u, o/ context, /f/ assimilation may also result in velar [x]. Table 5.12
compares the AP features relevant for velarization.
Table 5.12. AP features for /f/ velarization
/f/
/u, o/
lip protrusion
[labial, dental]
[labial]
lip aperture
[critical]
[narrow], [mid]
[x]
tongue body loc.
[velar]
[velar]
tongue body degree
[narrow], [mid]
[critical]
Comparing the shared features (also highlighted) in this case, it is clear that, while
ALIGN-LP targets the labial aspect of the back vowels, velarization is motivated by the
fact that /u, o/ are also specified as [velar]. When this gesture is overlapped with /f/, the
result is the velar fricative [x] (manner and voice unchanged). (18) captures these facts.
(18)
ALIGN-TBCL(C, center, V, onset)
In a CV context, align the TBCL onset gesture of V with the center of C.
‘Consonants share tongue body constriction location gestures with following
vowels.’
232
(18) is similar in form to (17) with the exception of the triggering aspect of the vowel. In
this case, the constraint allows for gestural overlap of the dorsality of the vowel by
targeting not its labial gestures, but tongue body gestures. Both constraints are necessary
as, crosslinguistically, languages may allow for either one, both or neither. This is
precisely the typological issue to be addressed in 5.3.5.
Unfaithfulness must also be penalized. Although the output is different given the
different overlapping gesture from the vowel, it is still the case that /f/ is sacrificing its
labiodental constriction (LP gesture). Thus, IDENT(LP) remains the relevant faithfulness
constraint. If markedness again dominates faithfulness, /f/ velarization is compelled,
demonstrated in (19).
(19)
ALIGN-TBCL » IDENT(LP)
/foɾmo/ formo ‘I form’
ALIGN-TBCL
a.
[foɾmo]
*!
b.
[ɸoɾmo]
*!
c.
[xoɾmo]
IDENT(LP)
*
*
A grammar that contains the hierarchy ALIGN-TBCL » IDENT(LP) selects the
assimilated candidate [xo] (19c) as optimal as it satisfies the dominant markedness
constraint, which is fatally violated by the faithful [fo] (19a) and the mis-assimilated [ɸo]
(19b) whose place unfaithfulness is unmotivated.
Both /f/ processes have now been accounted for theoretically. The final property
that must be addressed is velar palatalization.
5.3.3. /x/ palatalization
The necessary features to motivate /x/ palatalization are provided below.
233
Table 5.13. AP features for /x/ palatalization
Target /x/
Trigger /i, e/
Output [ç]
tongue tip location
[palatal]
[palatal]
tongue tip degree
[narrow], [mid]
[critical]
tongue body loc.
[velar]
tongue body degree
[critical]
Again, AP features capture the assimilatory nature of the process directly. While the
relevant fricative feature [critical] remains faithful, input velar /x/ assumes the palatal
gesture of the front vowels /i, e/, resulting in the palatal fricative [ç]. The fundamentals of
the alignment template are still necessary, yet it is neither the labial nor tongue body
gestures that overlap, but rather the tongue tip instead.
(20)
ALIGN-TTCL(C, center, V, onset)
In a CV context, align the TTCL onset gesture of V with the center of C.
‘Consonants share tongue tip constriction location gestures with following
vowels.’
In this case, (20) requires the overlap of vocalic tongue tip gestures. As it applies to
Spanish, only /i, e/ involve active tongue tip articulations within the series of vowels,
both of which AP defines as [palatal].
Given that target /x/ is specified as [velar], IDENT(LP) is powerless to prevent
unfaithfulness since /x/ contains no labial gestures. Consequently, (21) is necessary.
(21)
IDENT(TBCL)
Input tongue body location gestures are preserved on output correspondents.
234
As stated, (21) preserves input tongue body location gestures in output correspondents.
To circumvent its satisfaction, markedness must again dominate.
(22)
ALIGN-TTCL » IDENT(TBCL)
/xema/ gema ‘gemstone’
ALIGN-TTCL
a.
[xema]
*!
b.
[ɸema]
*!
c.
[çema]
IDENT(TBCL)
*
*
The grammar from (22) selects palatal [çe], candidate (22c), as optimal since it satisfies
the dominant ALIGN-TTCL. Faithful [xe] (22a) satisfies the IDENT(TBCL) constraint
that optimal (22c) violates, but its satisfaction is as the expense of the higher ranked
ALIGN. (22b) is unfaithful to place, but its unfaithfulness is not in compliance with the
tongue tip requirement of ALIGN, rendering it the least harmonic of the three.
The three previous subsections have responded to research question 5: how to
account for the processes in an OT framework. I have argued that AP-oriented alignment
constraints motivate coarticulation to following vocalic gestures at the expense of placepreserving IDENT constraints. All other segmental alignment constraints were
subordinated, and unfaithfulness to any feature other than place was prevented by general
IDENT (i.e., manner, voice).
Furthermore, I have proposed a uniform analysis for all three processes given
their previously discussed similarities. All are motivated by gestural alignment yet differ
with respect to the place (constriction location) of the vowels that are permissible
triggers.
235
In the spirit of uniformity, the fact that the five constraints above exhibit such
similarity raises the question if they can somehow be collapsed. Since ALIGN is able to
specify all three main articulators (lips, tongue tip and tongue body constriction
locations), and two constriction locations (lips and tongue body) were required for
IDENT, one might wonder if (23) is desirable.
(23)
ALIGN-CL » IDENT(CL)
(cf. AGREE(Place) » IDENT(Place))
(23) generalizes both ALIGN(C, center, V, onset) and IDENT to the feature label CL
(constriction location). Assuming Feature Class Theory (Padgett 1995), CL is freely able
to refer to the set {LP, TTCL, TBCL} (recall that LP is the labial equivalent of
constriction location) without necessitating six (three ALIGN, three IDENT) individual
constraints.
The problem is that (23) alone cannot account for dialectal nuance. A grammar
that contains (23) compels CV assimilation in every syllable throughout the language. To
prevent such an unattested scenario while still adhering to theoretical parsimony, either
markedness or faithfulness should be generalized while the other is left specific to
account for the selective nature of triggers and targets. Since Chapter 4 concluded that the
motivating principle in all three processes is gestural overlap, I argue that a general
ALIGN-CL uniformly accounts for /f/ and /x/ allophony in Latin American Spanish, and
it is the individualization of IDENT constraints that differentiates one grammar from
another. This is in line with the argument presented in Beckman (1998) in which the
asymmetrical application of phonological processes follows from the template: specific
faithfulness » markedness » general faithfulness. (24) reflects such a scenario.
236
(24) represents a grammar that attests velar palatalization alone and contains a
hierarchy of the ‘specific faithfulness » markedness » general faithfulness’ type. In the
first tableau representing palatalization, unassimilated candidates (24a) through (24c) are
ruled out as the specific IDENT constraints are irrelevant, leaving ALIGN to determine
the winner, which only (24d) satisfies to the detriment of general IDENT(CL). Thus,
collapsing the specific markedness constraints does not seem to affect successfully
selecting a palatalized output.
To determine if the same hierarchy works elsewhere in the grammar, consider the
second tableau in which faithful [x] is expected.
(24)
IDENT(LP), IDENT(TBCL) » ALIGN-CL » IDENT(CL)
/xema/ gema ‘gemstone’
IDENT(LP)
ALIGN-CL
IDENT(CL)
a.
[xema]
*!
b.
[ɸema]
*!
*
c.
[fema]
*!
*
d.
[çema]
/xot̪ a/ jota ‘the letter <j>’
*
IDENT(LP)
ALIGN-CL
IDENT(CL)
e.
[xot̪ a]
*
f.
[ɸot̪ a]
*
*!
g.
[fot̪ a]
**!
*
h.
[çot̪ a]
**!
*
IDENT(LP)
ALIGN-CL
IDENT(CL)
/foɾmo/ formo ‘I form’
i.
[xoɾmo]
*!
*
*
j.
[ɸoɾmo]
*!
*
*
k.
[foɾmo]
l.
[çoɾmo]
**
*!
**
*
Here, the unassimilated candidate (24e) is optimal, as all other unfaithful candidates are
ruled out by virtue of either multiple ALIGN violations or an IDENT(CL) violation. In
237
the third tableau, /f/ faithfulness (24k) is also guaranteed in this grammar even in an
otherwise assimilating context, as any unfaithfulness to the labial gesture (24i, j, l)
violates IDENT(LP).
It seems to be the case that generalizing markedness and specifying faithfulness
economizes the theoretical mechanism and is still able to adequately account for
assimilation. However, three issues remain in restricting ALIGN-CL’s effect, the first of
which is the central vowel /a/ and its status as a non-trigger.
Of the five cardinal vowels in Spanish–/i, e, a, u, o/–only /i, e/ and /u, o/
participate in /x/ and /f/ allophony, respectively. As I argue above, the gestural features
[palatal] and [velar] phonologically contrast vocalic frontness and backness in AP, a
distinction exploited by the individual ALIGN constraints. Once ALIGN is generalized,
whatever featural assumption is made about central /a/ necessarily allows it to participate
in some form of assimilation, as well. Because it forms a natural class with neither /i, e/
nor /u, o/ in any of the three fricative-vowel assimilations, one of two possibilities is true.
Either /a/ phonologically contrasts with [palatal] /i, e/ and [velar] /u, o/ because it is
endowed with a third, unique constriction location feature, or it does so via presence
versus absence, i.e., [palatal] front versus [velar] back versus [Ø] central (unspecified).
In what follows, I present two principal arguments that support the latter view,
that (low) central /a/ is unspecified in Spanish. The first is articulatory. Ladefoged and
Maddieson (1996) and Martínez Celdrán and Fernández Planas (2007) both indicate that
central vowels tend to be place-neutral in that the tongue makes only a slight active
gesture (if at all) either forward or backward; rather, the jaw lowers or raises depending
238
on height. From an articulatory perspective, this seems to indicate a lack of place,
especially in AP terms since no x-axis movement (CL) is produced.
The second argument is phonological. Ladefoged and Maddieson (1996: 292) and
Bateman (2007: 273) both conclude that central vowels, when they do not participate in
phonological processes, are to be regarded as unspecified (cf. Beckman, Helgason,
McMurray and Ringen 2011 and Beckman, Jessen and Ringen 2013 for the
manifestations of specified versus unspecified features). In the particular case of Spanish,
two phenomena support such a conclusion. First is the lack of participation in the
fricative properties at hand. If /a/ is unspecified, there is no place feature to which
ALIGN-CL can refer, straightforwardly explaining the consistent faithfulness of /f/→[f]
and /x/→[x] in the context of /a/: e.g., fama ‘fame’ [fa] and jarra ‘jar’ [xa].
Second, although no Spanish dialect (as far as I am aware) exhibits front/back
vowel harmony, Valencian Catalan, a closely related Romance language, does, and a few
Castilian dialects (Eastern Andalusian, Pasiego and Tudanca; see Walker 2005,
Linebaugh 2007 and Lloret 2007 for data and discussion) attest height harmony. In every
case, central /a/ is not a trigger for harmony (see sources cited above). Thus, based on
articulatory and phonological evidence, it seems that low central /a/ is featurally inert
(with respect to CL gestures), directly accounting for its failure to condition assimilation.
Beyond /a/, there are two additional ways in which the generality of ALIGN-CL
must be restrained. First is the long-standing issue throughout this dissertation concerning
the lack of assimilation in other manners of articulation (see Chapter 2). The second is
typological in nature. If, for example, a single dialect exhibits both bilabialization and
palatalization, how is /fi/→*[çi] (the expected allophone of palatalization applied to input
239
/f/) prevented? In such a grammar, ALIGN-CL must dominate IDENT(CL) such that
neither input labiality nor dorsality is allowed to remain faithful. The consequence of this
is that [ɸ] is not the only /f/ allophone guaranteed to occur, as /fi/ is now just as
susceptible to [palatal] assimilation since the /u, o/ triggering context is no longer
encoded in the markedness constraint. Observe (25) which illustrates both incorrect
predictions given the present constraints.
In the first evaluation of (25), an input bilabial nasal in candidate (25b)
assimilates in [palatal] place to a following /i/ given that the generalized ALIGN-CL
requires CL overlap between any C and any following V. As the data show, palatalization
applies exclusively to /x/ in Spanish (yet crosslinguistically, nasal palatalization is just as
possible, as Table 1.4 shows for Kurpian and Sentani). The symbol denotes an
ungrammatical but theoretically optimal candidate given the current (incorrect) hierarchy;
indicates the attested but non-optimal candidate. Thus, in this grammar,
ungrammatical *ñisa [ɲisa] (25b) is selected at the expense of [misa], the expected output
since nasals, stridents and all other non-fricatives resist assimilation of this type (in
Spanish).
(25)
ALIGN-CL » IDENT(CL)
/misa/ misa ‘mass’ (ecclesiastical)
a.
[misa]
b.
[ɲisa]
/fila/ fila ‘line’ (queue)
ALIGN-CL
IDENT(CL)
*!
*
ALIGN-CL
IDENT(CL)
c.
[fila]
*!
d.
[ɸila]
*!
*
e.
[xila]
*!
*
f.
[çila]
*
240
In the second tableau, /fi/ is similarly susceptible to palatalization since IDENT(LP) must
be dominated to motivate bilabialization, which for the purposes of this demonstration is
also present in this hypothetical grammar. Consequently, assimilated *[çila] (25f) is
incorrectly selected while faithful [fila] (25c) is ruled out. Two additional IDENT
families must be recruited to address each of these issues. I explore both in the section
that follows.
5.3.4. Restraining the grammar
In light of the discussion above, the repertoire of IDENT constraints necessary for
the analysis must be expanded to protect from assimilation all other phonemes (in all
contexts) outside the set {/f, x/} (§5.3.4.1), yet maintain faithfulness in /f/ when adjacent
to /i, e/ and in /x/ when followed by /u, o/ in grammars (dialects) that permit a
combination of the processes (§5.3.4.2).
5.3.4.1. Manner-specific place faithfulness constraints
Recall from Chapter 1, Table 1.4 (see also Bateman 2007) the crosslinguistic
possibility of CV assimilation with any consonantal manner in the C position. That is,
while Spanish limits consonant-vowel interaction to fricatives (excluding /s/), the theory
must permit it between vowels and consonants of all types, as languages differ in this
regard.
Here I follow the basic premise of Jun (1994, 1995) and Herrick (1999)6 in
adopting manner-specific IDENT(CL) (i.e., place) constraints. In this chapter (§§5.2.2.2
and 5.2.2.3), I briefly introduced the underlying principle behind their proposals: that
6
See also Piñeros (2007), citing Padgett (1996), for IDENT-OBSTRUENT(Place). While the
motivation for this IDENT division is in line with this dissertation’s argument, obstruent
faithfulness is still insufficient, as /f, x/ must be treated separately from the other obstruents of
Spanish.
241
perceptual cues to place vary depending on manner. Better-cued segments (manners) tend
to be more likely triggers (worse targets) than the more weakly-cued segments (better
targets, worse triggers). These correlations are encoded by implicational faithfulness
constraints.
However, their analyses could not be employed at face value. In Jun’s case, the
proposal applies to CC contexts exclusively. In Herrick’s case, although applicable to CV
environments, incorrect predictions for Spanish necessitated a reanalysis. To penalize
place unfaithfulness in the non-/f, x/ segments of Spanish, an overview of the phonemic
inventory (Table 5.14 below) is necessary to see how such a constraint might be framed
theoretically.
Table 5.14. Latin American Spanish phonemic inventory with relevant manner features
Manner
Phonemes
UFT
AP
Natural class
Stops
/p b t̪ d̪ k ɡ/
[–continuant]
[closed]
[closed]
Non-stridents
/f x (h)/
[–strident]
Stridents
/s ͡tʃ/
[+strident]
Nasals
/m n ɲ/
Laterals
/l (ʎ)/
Rhotics
/ɾ r/
Vocoids
/i e a o u (j w)/
[critical]
[–strident]
[+strident]
[velum: wide]
[sonorant]
[mid]
[sonorant]
[narrow, mid, wide]
Source: The phonemic inventory is adapted from the inside cover of Piñeros (2009).
Notes: Phonemes in parentheses are dialectal (/h/, /ʎ/) or differ depending on theoretical
assumptions (/j/ and /w/, or instead non-syllabic [i̯ ] /i/ and [u̯ ] /u/).
Before continuing, a few comments are in order. Both Unified Feature Theory (UFT) and
Articulatory Phonology (AP) contain unique and exhaustive features to identify the
natural class of stops: [–continuant] and [closed], respectively.
242
Concerning the remainder of the inventory, while the AP model of gestural
overlap, its related landmark alignment constraints and its place feature inventory all
more accurately account for the coarticulation facts of the Spanish processes, its manner
features are lacking in accounting for natural classes identified by decades of
phonological research (see McCarthy 1988, Clements 1989, NC&P 1993, Clements and
Hume 1995, inter alia). With respect to the fricatives, the precise division necessary
between stridents (/s/ and the palatoalveolar affricate /t͡ ʃ/) and non-stridents is captured
only by UFT, even if binarity must be assumed in the case of non-stridents. Regarding
the sonorants, UFT contains the very feature [sonorant] that exhaustively encompasses all
non-obstruents, while AP requires at least three.
Given the crosslinguistic data that suggest that languages may refer to each
natural class as a possible target for CV assimilation while excluding all other classes, I
propose the following faithfulness constraints.
(26)
IDENT-[closed](CL), IDENT-[+strident](CL), IDENT-[–strident](CL),
IDENT-[sonorant](CL)
Generalized IDENT-manner(CL)
Corresponding input and output segments of x manner in which x{[closed],
[±strident], [sonorant]} are identical with respect to constriction location (place).
An implicational hierarchy of (26) for the CV context along the lines of Jun’s (1994,
1995) for CC (such that assimilation in one manner necessarily entails assimilation in
another) is left for future research, as such requires an extensive survey similar to
Bateman (2007) (who looked exclusively at palatalization) that describes CV
assimilations across a variety of languages and language families. Minimally, nonstrident fricatives would need to be most susceptible given that no other manner
243
assimilates in a similar fashion in Spanish. Wright (2001: 256), in fact, comments on the
phonetic basis for this asymmetry, noting that /f/ cues are more readily lost in noisemasking tasks than /s/ given the relative strength of strident frication. This is also true of
weak aperiodic cues (stop bursts or frication) relative to periodic signals (formant
structure (e.g., the sonorant series)) in general.
For present purposes, (26) responds to research question 6 concerning the Spanish
data: how can the OT account limit assimilation to /f/ and /x/? (26) accomplishes
precisely this, and is grounded in empirical and theoretical work (Jun 1994, 1995, Herrick
1999, Bateman 2007) that argues for manner-specific faithfulness constraints given the
relative strength/weakness of cues to place depending on manner and context.
Following the template specific faithfulness » markedness » general faithfulness,
it is now possible to augment the specific faithfulness inventory in order to prevent
assimilation outside the non-strident fricatives. The solution is a response to the criticism
from Chapter 2 that asked why forms such as música ‘music’ and burla ‘mockery’ do not
undergo assimilation as the two previous analyses (Lipski 1995, Mazzaro 2005)
predicted. To see how (26) fares against similar data, (27) incorporates IDENTmanner(CL) into the previous hierarchy.
As (27) demonstrates, the highly ranked IDENT-manner(CL) constraints protect
stops, stridents and sonorants from the assimilatory effects of ALIGN-CL. The palatal
stop in (27a) violates IDENT-[closed](CL), as it diverges from the bilabial place of the
input stop /b/; consequently, unassimilated (27b) is optimal, as its ALIGN-CL violation is
irrelevant.
244
ID-[closed](CL), ID-[strid](CL), ID-[son](CL) » ALIGN-CL » IDENT(CL)7
(27)
/beso/ beso ‘kiss’
a.
[ɟeso]
b.
[beso]
/sima/ cima ‘summit’
ID-[closed]
(CL)
ID-[strid]
(CL)
ID-[son]
(CL)
ALIGN-CL
*!
ID(CL)
*
*
ID-[closed]
(CL)
ID-[strid]
(CL)
ID-[son]
(CL)
ALIGN-CL
ID(CL)
c.
[ʃima]
*!
*
d.
[ɕima]
*!
*
e.
[çima]
*!
*
f.
[sima]
/misa/ misa ‘mass’
g.
*
ID-[closed]
(CL)
ID-[strid]
(CL)
ID-[son]
(CL)
ALIGN-CL
ID(CL)
[ɲisa]
*!
*
h.
[mysa]
*!
*
i.
[misa]
*
In (27c) through (27e), whatever palatalizing effect /i/ may have on /s/, the result violates
IDENT-[strident](CL), which requires place faithfulness in correspondents of input
stridents, leaving faithful (27f) optimal.
Finally, (27g) and (27h) are both ruled out, as the palatal [ɲ] under the influence
of /i/ and the front rounded vowel [y] made labial by /m/ violate IDENT-[sonorant](CL),
requiring place faithfulness of input sonorant segments. The only option is faithful (27i),
whose [mi] place discord is tolerated.
7
Alternatively, IDENT([strident]) successfully eliminates candidate (27e), as the palatal fricative
[ç] is uncontroversially [–strident]. Furthermore, IDENT([anterior]) rules out (27c) through (27e),
given that all unattested candidates are [–anterior]. ID([ant]) is also able to eliminate (27g);
however, in a grammar that exhibits /f/→[x], ID([ant]) is incapable of preventing the comparable
case /m/→[ŋ] (e.g., mono ‘monkey’ *[ŋono]), as both bilabial /m/ and velar [ŋ] are unspecified
for [anterior] (a coronal dependent). Here, the manner-specific constraint remains necessary.
245
IDENT-[–strident](CL) is not separated from general IDENT(CL) as it is exactly
this class (non-stridents) that must be left vulnerable to assimilation. It is available to the
theory for those languages that, like Spanish, make distinctions in targethood with respect
to manner yet, unlike Spanish, exclude non-stridents from participating in assimilation.
The current subsection put forth three manner-specific IDENT(Place) faithfulness
constraints as the first of two forces restraining ALIGN-CL. In the next section, I explore
a second set of faithfulness constraints necessary for the discussion in section 5.3.5 in
which I combine the three properties and discuss their typological interaction.
5.3.4.2. Constraints on coarticulatory distance
The final component of the theoretical analysis concerns dialects in which two (in
any combination) or all three properties are attested. Table 5.15 presents the possible
permutations of the three processes in Spanish.
Table 5.15. Latin American Spanish dialects of /f/ and /x/ assimilation
1 assimilation
2 assimilations
a. /f/→[x]
d. /f/→[x] /f/→[ɸ]
b. /f/→[ɸ]
e. /f/→[x] /x/→[ç]
c. /x/→[ç]
f. /f/→[ɸ] /x/→[ç]
3 possible grammars
3 possible grammars
3 assimilations
g. /f/→[x]
/f/→[ɸ]
/x/→[ç]
1 possible grammar
7 total possible grammars
A shaded cell indicates that a speaker representative of such a grammar appears in the
data from Chapter 4. They are as follows. No participant of the eight exclusively
velarized or combined velarization with palatalization (a, e). Mexican experimental
participant 35792 (female, age 38) exhibits dialect (b), pure bilabialization, with two
246
small caveats: of the 23 /f/ tokens, only one was velar, otherwise 22 were bilabial; of the
39 /x/ tokens, one was palatal, leaving 38 velar. Dialect (c)–palatalization only–is found
in every control participant (four total): Chile 15246 (female, 28); Chile 02838 (female,
40); Mexico 11111 (female, 54); Mexico 71017 (male, 31). Mexican experimental
participant 26911 (male, 35) has both /f/ velarization and bilabialization, dialect (d), in
his grammar. Dialect (f), bilabialization and palatalization, was found in Chile
experimental participant 17137 (male, 34). Lastly, dialect (g), with all three properties,
was demonstrated by Chile experimental participant 19727 (male, 40) (and marginally by
Mexican participant 35792).
Given tableau (25) in which, as the constraints stand, unattested *[çila] is selected
as the output for fila [fila] in a grammar like (f) from the table above with both
bilabialization and palatalization, it must be the case that a third series of specialized
faithfulness constraints prevent place unfaithfulness in /f/ and /x/ in the context of nontriggering vowels for their corresponding process, i.e., /i, e/ for /f/, /u, o/ for /x/.
To satisfy this final restriction on ALIGN-CV, I borrow from Dispersion Theory
(Flemming 1995) as a means to encode restraints on configurations eligible for
assimilation. While Dispersion Theory is a model of auditory contrast, I am applying the
same concepts to an articulatory perspective. As Flemming (1995: 3) argues,
phonological contrasts adhere to three main principles.
(28)
a.
b.
c.
Maximize the number of contrasts.
Maximize the distinctiveness of contrasts.
Minimize articulatory effort.
247
As (28b) relates to the auditory effectiveness of the contrasts, I will not address it in my
application of Dispersion Theory to the articulatory issue at hand. The interaction of the
goals (a) and (c), then, from (28) is schematized in Figure 5.11 (Flemming 1995: 3, his
Figure 1) assuming a hypothetical, two-dimensional space.
Figure 5.11. Interaction between maximize contrast and minimize effort
2 contrasts
Less effort
4 contrasts
More effort
As the above visualization shows, the two goals conflict: increasing the number of
contrasts comes at the expense of minimizing effort, while reducing articulatory effort (or
increasing articulatory efficiency) results in fewer contrasts. This is the core dichotomy
between faithfulness (maintain contrast) and markedness (reduce effort) in OT
terminology.
In seeking a balance between both goals, it stands to reason that languages that
opt for markedness reduction do so minimally, in the sense that they compromise by
maintaining contrast (where possible) in the least effortful way, and that assimilation is a
last resort given that it is the option most at odds with the competing goal of contrast
maintenance. This is the logic that I suggest accounts for the faithful realization of /fi, fe/
and /xu, xo/, while /fu, fo/ and /xi, xe/ are susceptible to assimilation.
The argument is represented in the hierarchy below relating effort, articulatory
distance and the phonological result in a grammar that seeks markedness reduction.
248
Table 5.16. Implicational hierarchy of articulatory efficiency
Gestures
1 gesture, 2 segments
»
2 ‘close’ gestures
»
2 ‘distant’ gestures
Effort
least effort
»
moderate effort
»
most effort
Result
assimilated output
»
faithful tolerance
/f/ [xu, xo], /x/ [çi, çe]
/f/ [fi, fe], /x/ [xu, xo]
» assimilation candidates
/f/ *[fu, fo] /x/ *[xi, xe]
To better visualize the proposal, consider Figure 5.12 below in which the relative
locations of the segments in question have been plotted in a lateral view of the vocal
tract. If one imagines a vertical division at the approximate midline of the vocal tract (cf.
[±anterior]/[±back] in SPE terms, or Ladefoged and Maddieson’s (1996: 292)
[front]/[back] distinction), the argument in Table 5.16 is that Spanish employs
assimilation between /f/ and the relatively distant /u, o/ due to markedness reduction at
the expense of contrast faithfulness.
Figure 5.12. Articulatory distance: [f], [i], [e], [o], [u], [x]
[f]
[x]
[i] [e] [o] [u]
Were /fu, fo/ to be realized faithfully, the effort necessary for two separate, non-local
gestures is greatest. Thus, markedness is favored over faithfulness in such contexts. The
249
compromise is that, while greater effort is necessary for two distinct gestures made in [fi]
and [fe] as opposed to a single gesture under assimilation, the gestures are sufficiently
close to avoid complete assimilation in order to respect faithfulness in these particular
cases. The same argument holds (in the opposite direction) for /x/. In this sense,
markedness reduction is employed only when the two gestures are sufficiently separated
to warrant faithfulness violations, otherwise two local gestures are tolerated.
Consequently, the failure of /fi, fe/ to palatalize and /xu, xo/ to bilabialize is the
desire of a markedness-reducing grammar to respect faithfulness where possible in
permitting ‘close’ gestures while drawing the line, as it were, at ‘distant’ gestures (see
Piñeros (2002) for a myriad of markedness-reducing properties in weakening dialects of
Spanish and how each (e.g., vowel nasalization, aspiration, nasal velarization) negotiates
a preference for unmarkedness while maintaining some degree of faithfulness). This
argument coincides with the results of the locus equations (indicative of gestural overlap)
from Chapter 4 that showed less coarticulation (indicative of distinct gestural targets) in
the [fi, fe] and [xu, xo] contexts.
At this point I have used the non-technical terms ‘close’ and ‘distant’ for
reference without any attempt at formalization. Without a larger corpus, it is difficult to
formalize such a proposal short of knowing the extent to which ‘closeness’ is tolerated
and/or ‘distance’ is avoided crosslinguistically. However, for present purposes I
tentatively propose the following with the above caveat and an admission that future
research is necessary. Consider the horizontal movement between each segment from
Figure 5.12 as ‘one step.’ For completeness, if low central /a/ is added, there are
maximally six steps one could make from front to back (or back to front): [f] to [i] 1, [i]
250
to [e] 2, [e] to [a] 3, [a] to [o] 4, [o] to [u] 5, [u] to [x] 6. If this is the measure by which
locality is formalized, the following IDENT constraint preserves faithfulness based on the
relative number of steps between gestures.
(29)
IDENT-MAXDIST-n(CL)
Let α and β be contiguous input segments and γ and δ their respective output
correspondents. If the places of articulation of α and β are maximally n points
apart, then the places of articulation of γ and δ are faithful to those of α and β.
Input-output faithfulness with respect to constriction location is locally respected
in consecutive segments separated by no more than n points.
(29) identifies the maximum distance n (n a whole integer that varies depending on
language) of constriction location separation that is close enough to resist the forces of
assimilation; outside this range (in the case of Spanish, n=2), ALIGN-CL determines that
the distance between the gestures is too effortful for faithfulness, and assimilation thus
results. (30) incorporates this new restriction into the previously established hierarchy.
(30)
IDENT-MAXDIST-2(CL) » ALIGN-CL » IDENT(CL)
/fila/ fila ‘line’
a.
[çila]
b.
[fila]
/xot̪ a/ jota
‘the letter <j>’
ID-MXDST2(CL)
ALIGN-CL
*!
ID(CL)
*
*
ID-MXDST2(CL)
ALIGN-CL
ID(CL)
c.
[ɸot̪ a]
*!
*
*
d.
[fot̪ a]
*!
*
*
e.
[xot̪ a]
*
In (30a), assimilated candidate *[çila] violates high ranking IDENT-MAXDIST(CL)
given that input /f/ and /i/ are only one step away, necessitating place preservation. The
251
faithful candidate (30b) [fila] is thus optimal, as its ALIGN-CL violation is minimal. In
(30c) and (30d), both candidates fatally violate IDENT-MAXDIST(CL), irrespective of
the labial output, as input /x/ and /u/ are also only one gesture away. Faithful candidate
(30e) [xot̪ a] is therefore selected. Without the high ranking IDENT-MAXDIST(CL),
determination of the optimal candidate in each evaluation falls to ALIGN-CL, which will
consistently (and incorrectly) select the assimilated output.
With the two preceding subsections, it is now possible to succinctly respond to
research question 7: what is the role of articulatory and acoustic phonetics in OT as they
pertain to the Spanish data? Work in acoustic phonetics and the relative strength of cues
to place of articulation depending on context and manner (Steriade 1993, Byrd 1994, Jun
1994, 1995, Herrick 1999, Wright 2001, 2004) underlie the motivation of mannerspecific place faithfulness constraints necessary to limit assimilation to /f, x/ while
excluding all other manners in Spanish.
Articulatory phonetics and locality combined with contributions by Flemming
(1995) motivate the constraint on configurations eligible for assimilation as they relate to
the conflicting goals in phonology of economy of articulation and contrast maintenance.
Table 5.17 summarizes the crucial rankings that account for the three processes in
Latin American Spanish. The complete ranking summary (with dashed lines for dialectal
variability to be addressed further in the next section) is provided in Figure 5.13, with the
IDENT-manner family collapsed for ease of presentation.
With every component (i.e., constraint) of the theoretical analysis in place, the
following section addresses research question 8, the dialectal implications in combining
the three Spanish processes and the crosslinguistic predictions of constraint reranking.
252
Table 5.17. Summary of constraint rankings and their effects
Ranking
Phonological effect
IDENT-MAXDIST-2(CL) » ALIGN-CL
Place faithfulness in /fi, fe/, /xu, xo/
IDENT-[closed](CL) » ALIGN-CL
Place faithfulness in stop-V contexts
IDENT-[strid](CL) » ALIGN-CL
Place faithfulness in strident-V contexts
IDENT-[son](CL) » ALIGN-CL
Place faithfulness in sonorant-V contexts
IDENT(LP) » ALIGN-CL
/f/ place preservation, /x/ palatalization only
IDENT(TBCL) » ALIGN-CL
/x/ place preservation, /f/ assimilation only
ALIGN-CL » IDENT(CL)
/f/ and /x/ assimilation
Figure 5.13. Ranking summary for Latin American Spanish fricative allophony
ID-MXDST-2(CL)
ID-manner(CL)
ALIGN-CL
IDENT(CL)
IDENT(LP)
IDENT(TBCL)
253
5.3.5. An assimilatory typology
One way in which Optimality Theory has been argued to be explanatorily superior
to its predecessors (e.g., linear and autosegmental rules) is that constraints are universal
and not language-specific; language difference is instead derived from the relative
ranking of the shared constraints within the universal set (see §5.2.1). Consequently, any
analysis for a particular property in a particular grammar (language/dialect) necessarily
results in a factorial typology–the predicted grammars resulting from all permutations of
the proposed constraints. I will first consider how the above hierarchy accounts for every
way in which Spanish grammars may combine the three properties, after which I extend
the analysis to crosslinguistic CV assimilations in general.
This discussion responds, from the theoretical perspective, to research question 8:
why do certain linguistic systems attest certain properties while other systems do not?
The OT answer to such an inquiry is found in the way in which the same set of
constraints is reranked. The auditory, articulatory and social components of the answer
are explored in section 5.4, which also provides the final component in Baker et al.’s
(2011) proposal.
Table 5.18 below provides the crucial constraint (re)rankings (from the set
identified above) necessary to capture all seven possible combinations of the three Latin
American fricative properties (see Table 5.15); i.e., each combination is a separate
Spanish grammar. (Since IDENT-MAXDIST-2(CL) and IDENT-manner(CL) are
subordinated below ALIGN-CL in no grammar of Spanish, I omit them from the table so
that the differences between dialects are highlighted.)
254
Table 5.18. Constraint rankings for /f, x/ assimilations in Latin American Spanish dialects
Properties
Crucial constraint rankings
a. /f/→[x]
ID(TBCL) » ALIGN-TBCL » IDENT(CL)
b. /f/→[ɸ]
ID(TBCL) » ALIGN-LP » IDENT(CL)
c. /x/→[ç]
ID(LP) » ALIGN-CL » IDENT(CL)
d. /f/→[x], /f/→[ɸ]
ID(TBCL) » ALIGN-CL » IDENT(CL)
e. /f/→[x], /x/→[ç]
ALIGN-TBCL, ALIGN-CL » IDENT(CL)
f. /f/→[ɸ], /x/→[ç]
ALIGN-LP, ALIGN-CL » IDENT(CL)
g. /f/→[x], /f/→[ɸ], /x/→[ç]
ALIGN-CL » IDENT(CL)
In grammars (a) through (c) in which the properties appear individually, the IDENT
constraint protecting the non-assimilating fricative (IDENT(TBCL) to preserve /x/
faithfulness in (a, b), IDENT(LP) to protect /f/ in (c)) dominates the relevant ALIGN
constraint, which in turn dominates general IDENT(CL).
Grammars (d) through (f) permit a combination of two properties. Grammar (d)
contains both /f/ allophones, requiring the dominance of IDENT(TBCL) to preserve /x/
faithfulness. Grammars (e) and (f) combine one of the two /f/ properties with /x/
palatalization, meaning that both IDENT(LP) and IDENT(TBCL) are collapsed into
IDENT(CL) since neither /f/ nor /x/ faithfulness is preserved. Whether or not the
assimilated /f/ is bilabialized or velarized is determined by the relevant ALIGN constraint
(LP and TBCL, respectively). Grammar (f) attests all three properties such that no placespecific IDENT constraint dominates ALIGN, only the IDENT-MAXDIST and IDENTmanner constraints whose dominance is shared by all.
To summarize, the seven possible grammars in Spanish (with respect to these
three properties) derive from reranking the relevant CL-specific ALIGN and IDENT
255
constraints. If IDENT-MAXDIST-n, IDENT-manner and the previously discussed
ALIGN-{CV, VC} constraints are now (re)considered, a wide variety of crosslinguistic
assimilations are explained with this small set of constraint families.
To explore the typological predictions, recall from Chapter 1, Table 1.4
(reproduced below as Table 5.19), the various manners of articulation that are vulnerable
to CV (and VC) assimilations. I argued that IDENT-manner(CL) is uniformly dominant
in Spanish, leaving only non-strident fricatives as possible assimilatory candidates.
However, expanding the constraint allows languages to differentiate among different
manners.
Table 5.19. Crosslinguistic CV assimilations redux
Manner
Stop
Fricative
Affricate
Nasal
Language
Target
Output
Context
Bemba
/t, d, k, ɡ/
[f]
__ /u/
Bulgarian
/k, ɡ/
[c, ɟ]
__ /i/
Ewe
/s, z/
[ʃ, ʒ]
__ /i/
Japanese
/h/
[ç] [ɸ]
__ /i/ __ /u/
Lezgian
[t͡ ʃ]
[t͡ ʃ, d͡ ʒ]
__ /i/
Mina
/t͡ s/
/t͡ s, d͡ z/
__ /i/ or /i/ __
Kurpian
/m/
[ɲ]
__ /i, ɛ/
Sentani
/n/
[ɲ]
__ /i, j/
To understand how such diverse properties can be reduced to a single constraint family,
consider the expanded version of IDENT-manner(CL) in (31) (cf. (26)).
(31)
ID-[closed](CL), ID-[critical](CL), ID-[±strident](CL), ID-[sonorant](CL)
256
When the manners are individualized, they may rerank freely with respect to each other.
The general formula is, therefore, to specify the non-assimilating manners in the
dominant IDENT-manner constraints, leaving the susceptible manner of articulation to
the general IDENT(CL), which is itself dominated by markedness-reducing ALIGN.
Thus, Bemba and Bulgarian require the following hierarchy, reranked from (31), to
render stops unfaithful while protecting all other manners.
(32)
IDENT-[critical](CL), IDENT-[sonorant](CL) » ALIGN » IDENT(CL)
(32) results in stop-V assimilation, as the dominant IDENT constraints preserve
faithfulness in [critical] (cf. UFT [continuant]) fricatives and affricates (as opposed to
[±strident]) and sonorant segments, leaving stop faithfulness solely in the hands of
IDENT(CL), which, unlike the other IDENT-manner constraints in this hierarchy, is
dominated by ALIGN.
Ewe, Lezgian and Mina,8 on the other hand, prioritize IDENT-[closed](CL),
IDENT-[–strident](CL) and IDENT-[sonorant](CL) above ALIGN, leaving [+strident]
segments vulnerable to assimilation. Japanese follows Spanish in that it is the [–strident]
series that is left unspecified by the dominant IDENT constraints. Lastly, Kurpian and
Sentani preserve faithfulness is all manners excluding the [sonorant] series (or more
specifically [nasal] segments if IDENT[approximant]-CL is necessary such that laterals
and rhotics remain faithful).
8
This is a slight oversimplification. If one assumes that affricates are contour segments composed
of an initial stop ([closed], [–continuant]) gesture followed by a fricative ([critical], [continuant])
gesture (see, e.g., Sagey 1986), relying on stridency is insufficient. Both IDENT[closed] and
IDENT[critical] would, therefore, need to be dominated by ALIGN in order to compel
assimilation in affricates, of which the inherent prediction is that stops and fricatives assimilate in
this language, as well. Further exploring this prediction is beyond the scope of this dissertation.
257
In the case of Mina, it is interesting to note that either a preceding or following
high vowel triggers palatalization. In this case, ALIGN-CV alone is insufficient; ALIGNVC is also necessary for the progressive context.
Not listed in the brief language survey but also logically possible are languages
with no asymmetry in assimilation (in either direction) such that all manners participate.
This grammar is the result of ALIGN-CV and/or ALIGN-VC (depending on direction)
dominating general IDENT(CL). At the other end of the spectrum are fully faithful
languages (at least concerning consonant and vowel interactions) in which IDENT(CL)
dominates ALIGN-CV and ALIGN-VC. (This is independent of CC sequences–which
may or may not assimilate in such a language–regulated by a separate ALIGN-CC
constraint (Gafos 2002), introducing into the factorial typology an entirely new set of
permutations that I do not address here.)
I omit IDENT-MAXDIST-n(CL) from the above discussion given its tentative
nature. Certainly the implicational hierarchy for the value of n and its maximal distance
of proximity for faithfulness make interesting typological predictions (ID-MAXDISTn(CL) » ID-MAXDIST-n+1(CL) » ID-MAXDIST-n+2(CL) » ID-MAXDIST-n+x(CL));
they depend, nevertheless, on a better understanding of how n is to be assessed. For these
reasons, I leave the typological implications of IDENT-MAXDIST-n(CL) for future
research.
In light of all these combinations, Figure 5.14 below provides the representative
crosslinguistic ranking summary (dashed lines indicate crosslinguistic variability) with
three possible positions for the placement of either (or both) markedness constraint(s)
(see Cho 1997 and Bateman 2007 for similar analyses).
258
Figure 5.14. Constraint reranking and factorial typology of CV/VC constraints
ALIGN-CV, ALIGN-VC
ID-[closed](CL)
ID-[critical](CL)
ID-[±strident](CL)
ALIGN-CV, ALIGN-VC
ID-[closed](CL)
ID-[critical](CL)
(no manner distinction)
(manner distinction)
ID-[±strident](CL)
ALIGN-CV, ALIGN-VC
ID-[sonorant](CL)
ID-[sonorant](CL)
(no assimilation)
Now that I have explored the ways in which languages differ concerning consonantvowel interactions and how the proposed constraints in an OT model can explain these
differences within Spanish and across other languages, the question that follows is: what
type of (social, geographic, historic) situation accounts for the same types of differences?
The following section explores the issue of dialectal/crosslinguistic variation not from an
OT perspective, but an extralinguistic one.
5.4. Initiation versus diffusion in sound patterns
The socio-historic explanation concerning the origin and propagation of a
phonological alternation and the synchronic account for the mental, systematic
representation of it are interrelated but ultimately separate questions (Mielke 2008: 191,
192) and were consequently pursued as such to this point. OT as a grammatical model
does not necessarily make claims concerning where (geographically or socially), when or
259
how a phonological property came about (and why it did so in this language/dialect but
not another), rather it seeks to represent a speaker’s knowledge at a particular point in
time. However, as Mielke (ibid.) argues, a comprehensive account of the phenomena
should address the questions of where, when and how, especially since the answer to
these issues might shed light on synchronic explanations (see, e.g., Zampaulo 2013).
In this chapter I have explored the abstract representation of the processes and
specifically, in the immediately preceding section, described how OT handles dialectal
and crosslinguistic differentiation of the properties. An alternative approach to the same
issue of differentiation seeks not an OT explanation, but a sociolinguistic and diachronic
one, which are precisely the final components expressed in Baker et al.’s (2011) threepart proposal addressing how phonological properties arise. I postponed the final piece of
the puzzle to this point in order to establish first the descriptive facts concerning dialectal
variation in the previous section. That is, once a model of a speaker’s grammar that
contains these properties has been established, what are the factors that contribute to the
diffusion of the grammar to other speakers, and why should the diffusion occur in this
way? It is at this point that the OT answer to research question 8 provided above is
supplemented with the diachronic, social and geographic perspectives.
5.4.1. Correction, hypocorrection
and the role of the listener
The first step in addressing differentiation concerns the mechanism by which a
phonological property originates between speaker and hearer in individual interactions. In
section 5.4.2, the social and geographic components complete the account.
Recall from Chapter 1 the argument that listeners are able to filter out
coarticulatory effects while parsing the acoustic signal (Ohala 1981, Beddor 2012); Ohala
260
termed these listeners corrective listeners while Beddor refers to them as conservative
since they ‘correct’ the coarticulated signal/conserve the intention of the speaker. Figure
5.15 reproduces Figure 1.1.
Figure 5.15. Three possibilities for the speaker-listener exchange redux
(a) Little to no effect
(b) Correction
(c) Hypocorrection
/fo/
/fo/
/fo/
/fo/
/fo/
/ɸo/
[fo]
[fo]
[fo] [fɸo]
[ɸfo] [ɸo]
[fo]
[fo] [fɸo]
[ɸfo] [ɸo]
[ɸo]
Context (a) is of little interest, as the speaker produced the utterance with minimal
coarticulation. Possibility (b), on the other hand, represents a situation in which a
corrective/conservative listener has filtered out the coarticulation and has successfully
perceived what the speaker intended. Exchange (c) exhibits the same amount of gestural
overlap on the part of the speaker, yet this time a hypocorrective (Ohala) or innovative
(Beddor) listener failed to undo the phonetic effect, taking at face value the speaker’s
target-deviant output. These listeners represent a (possible) source for phonologization of
a phonetic effect as they rely more on the actual perceptual cues of what is produced than
their knowledge and expectations of coarticulation along the lines of a conservative
listener. Such a scenario represents the origin of a (possible) sound change.
In terms of its spread and differentiation, the likelihood of grammaticalization of
the coarticulated sequence under scenario (c) depends on the listener, but also on the
signal itself. Consider two speakers; speaker A coarticulates to a greater extent than
261
speaker B. Under the assumption that a greater degree of coarticulation is reflected in the
corresponding acoustic signal, the same innovative listener who interacts with speaker A
will more easily and frequently misperceive the signal than s/he would during an
interaction with speaker B. I thus argue that the first source of dialectal differentiation
results from the difference in the signal itself.
As this pertains to the dialects of Latin American Spanish, there is a striking
difference in the rates of palatalization and their relationship to the degree of
coarticulation between the two geographic groups, Chile and Mexico. Consider Table
5.20 below.
Table 5.20. Rate of velar allophony in /xi, xe/ contexts by region
[x]
[ç]
Total /x/
N
%
N
%
N
%
Chile
35
15.15
196
84.85
231
100
Mexico
43
75.44
14
24.56
57
100
Chileans palatalize 85% of the time while Mexican speaker palatalization accounts for
only 25%. The faithful [x] allophone is preferred by Mexicans at a rate of 75% compared
to Chileans around 15%. As indicated by the shaded cells, Chileans prefer palatalization
while Mexicans opt for /x/ faithfulness. The point of research question 8 is to explore
why, if coarticulation is universal, it only results in grammaticalization in some
dialects/languages, and why in those dialects/languages in particular? In this case, what
accounts for the Chileans’ palatal preference?
262
When the degree of coarticulation is considered, the difference from Table 5.20
becomes clear. Table 5.21 below updates Table 4.26, the /xi, xe/ locus equations for each
region. (Recall that locus equations measure the extent of coarticulation with greater
overlap approaching 1 and lesser overlap 0.)
Table 5.21. /xi, xe/ locus equations by geolect redux
/xi/
/xe/
Mean
Chile
0.94
0.92
0.93
Mexico
0.46
0.40
0.43
As Table 5.21 shows, coarticulation between /x/ and a following front vowel is
significantly greater in Chilean than in Mexican speakers (p=0.004). Combining the
degree of coarticulation with the rate of the palatal allophone results in the positive
correlation expressed in Figure 5.16 below.
Figure 5.16. Average /xi, xe/ locus equations (slope) and rate of [ç] allophone (%)
263
As LE increases (indicating greater coarticulation), so too do rates of the unfaithful
allophone. Unsurprisingly, the less coarticulating Mexican speakers demonstrate a lower
percentage of palatal [ç]. Such results suggest that one possible explanation for dialectal
differentiation (at least for palatalization in Chilean and Mexican Spanish dialects)
concerns the degree of coarticulation. Determining why some speakers coarticulate more
than others in order to use the resulting criteria to predict who will coarticulate is a task
outside the scope of this research (if possible at all; such an endeavor would be
comparable to attempting to identify predictors of correction/conservation versus
hypocorrection/innovation in listeners). Yet acknowledging that the phonetic difference
exists and correlates with geographic variation suffices to account for the case of Latin
American Spanish palatalization in Chilean versus Mexican speakers.
In the following section, I explore a second way in which dialectal differentiation
might be explained as it relates to /f/ allophony in Chilean and Mexican Spanish.
5.4.2. Literacy and the role of social factors
When the same test for geographic difference from above is applied to /f/, the
correlation obtains, yet on a much smaller scale. The rate of /f/ assimilation by the
experimental speakers (since control participants exhibited no /f/ alternations) appears in
Table 5.22; Table 5.23 provides the locus equations for the same speakers.
Table 5.22. Rate of allophony in /fu, fo/ contexts by region (assimilators only)
[f]
[ɸ]/[x]
Total /f/
N
%
N
%
N
%
Chile
33
25.98
94
74.02
127
100
Mexico
1
4.35
22
95.65
23
100
264
Table 5.23. /fu, fo/ locus equations by geolect (assimilators only)
/fu/
/fo/
Mean
Chile
0.84
0.77
0.81
Mexico
0.96
0.92
0.94
In terms of /f/ allophonic rates, assimilating speakers in both groups prefer the unfaithful
allophones to [f]. Here, the difference in rate of assimilation between Chile and Mexico is
around 20%, while the difference in /x/ palatalization rates for [ç] was closer to 60%
between the two groups. The degree of coarticulation by the two groups is also
comparable, as the difference is not statistically significant (p=0.0788). Figure 5.17
presents the /f/ assimilation rate/locus equation correlation.
Figure 5.17. Average /fu, fo/ locus equations and rate of [ɸ]/[x] allophones (assimilators)
Although it is the case that the positive correlation between degree of coarticulation and
rate of allophony originally found in /x/ also obtains for /f/, the difference is quite subtle.
For /f/, it seems to be that both groups coarticulate and express a high degree of
265
assimilation with minimal difference between the two. This suggests that there is no
geographic difference for /f/ alternations.
There is a difference, however, between the groups in terms of experimental
versus control (not Chile versus Mexico). Table 5.24 combines the Chilean and Mexican
data concerning /f/ allophone rates, and compares them to those of the control
participants (also collapsing Chile and Mexico). The difference is clear: control speakers
exhibit no /f/ alternation at all.
Table 5.24. Rate of allophony in /fu, fo/ contexts by group
[f]
[ɸ]/[x]
Total /f/
N
%
N
%
N
%
Exp.
34
22.67
116
77.33
150
100
Control
177
100
0
0
177
100
Table 4.22 provided the complete set of /f/ locus equations by participant. For ease of
reference, Table 5.25 summarizes the relevant contexts and collapses the data into the
comparable experimental and control groups. Finally, Figure 5.18 below shows the
expected correlation.
Table 5.25. /fu, fo/ locus equations by group
/fu/
/fo/
Mean
Exp.
0.9
0.85
0.86
Control
0.7
0.7
0.7
266
Figure 5.18. Average /fu, fo/ LEs, rate of [ɸ]/[x] allophones (%) (experimental/control)
Unsurprisingly, as LE slope (indicative of coarticulation) increases, so too does the
percent of unfaithful /f/ allophones between the control and experimental groups.
However, I previously argued that, even though the control speakers attest no unfaithful
/f/ allophones themselves, their tendency to slightly coarticulate /fu, fo/ was necessary to
provide the auditory cues that led the presumed innovative listeners to mistake the
degraded [f] productions as either [ɸ] or [x]. And while the difference in degree of
coarticulation between the two groups is significant (p=0.0198), the minor difference in
slope (0.16) is disproportional to the substantial difference in allophonic rate (77.33%).
Thus, there must be some other explanation to account for the vast difference in
rate of allophony (other than the previously identified degree of coarticulation) between
these two groups of speakers who share the same respective geographic space. With this
in mind, consider Figure 5.19.
267
Figure 5.19. Years of formal education and rate of [ɸ]/[x] allophones by participant
Note: Ch exp 1 and 2, Chile experimental participants 19727 and 17137; Mx exp 1 and 2,
Mexico experimental 26911 and 35792; Ch cntr 1 and 2, Chile control 15246 and 02838;
Mx cntr 1 and 2, Mexico control 11111 and 71017.
When years of formal education are considered, a separate correlation is revealed. Figure
5.19 plots rates of /f/ allophony along the y-axis and years of formal education along the
x-axis for all eight individual production participants. What results is a negative
correlation between the two: as years of formal education increase, the percent of
unfaithful /f/ allophones decreases, with the four control participants producing no
unfaithful /f/ correspondents at all.
Considering the role of social factors–as Baker, Archangeli and Mielke (2011), in
their third component for phonologization, argue one must do–it appears to be the case
that coarticulation is necessary but insufficient in this instance. While gestural overlap
seems to be able to explain geographic differences concerning why one property
grammaticalizes where it does but not elsewhere (e.g., palatalization in Chilean versus
Mexican speakers), in the case of Latin American /f/ properties, Figure 5.19 suggests that
268
education is a factor in sociolect formation, limiting the propagation of the property
where it would otherwise spread throughout the geolect. That is, (co)articulation and
(mis)perception can only disseminate the process to the extent that other factors–here,
education–do not intervene.
This conclusion coincides with Mazzaro (2011: 164) who found the same
correlation between her high and low education groups and the frequency with which /f/
was velarized. She argues that this correlation is due to orthographic effects. She cites
work by Morais, Cary, Alegria and Bertelson (1979), who found that knowledge of
orthography has an effect on perception. Additional work has argued that literacy
increases phonemic discrimination (Wagner, Torgesen and Rashotte 1994; Byrne 1998;
Muter, Hulme, Snowling and Taylor 1998; Treiman and Bourassa 2000; Hulme, Hatcher,
Nation, Brown, Adams and Stuart 2002; Carroll, Snowling, Stevenson and Hulme 2003;
Bassetti 2006). To the extent that literacy and knowledge of orthography are corollaries
of formal education, the correlation from Figure 5.19 is supported.
The issue is further compounded by the fact that /f/ velarization is a neutralizing
process in Spanish, while /x/ palatalization is not. That is, the velar [x] output of /f/
assimilation infringes on the faithful [x] output of phonemic /x/ (see Chapter 1, §1.1,
(10)). The palatal [ç] output of /x/ palatalization, on the other hand, encounters no such
situation, and is purely allophonic. Because speakers must be able to discriminate [f] and
[x] in order to maintain the /f/~/x/ phonemic contrast–and because such is represented
orthographically (<f> versus <j>, <gi> and <ge>)–it stands to reason that those speakers
with knowledge of the orthography are less likely to adopt such a process.
269
In responding to research question 8 and accounting for the reasons for dialectal
differentiation, the data support two analyses: /x/ palatalization spreads (or not)
geographically based on the degree of coarticulation in each speech community, while /f/
allophony is divided socially as determined by years of formal education and its
associated effect of orthographic knowledge.
5.5. Conclusion
In this chapter I presented the formal analysis for Spanish /f/ and /x/ allophony
within Optimality Theory. Integrating the ideas behind Articulatory Phonology for
features and markedness–and the Production Hypothesis and Dispersion Theory for
faithfulness–I argued that a small set of well-established constraint families (ALIGN and
IDENT) can be recruited (and reranked) to account for (i) the three processes individually
and (ii) the typological implications collectively. Furthermore, I addressed the third
component in Baker, Archangeli and Mielke’s (2011) proposal accounting for the origin,
propagation and dialectal differentiation of the properties after having addressed the first
two components (motivation and variability) in Chapter 4.
The final chapter of this thesis addresses issues ancillary to the main account put
forth in the preceding pages of this chapter, frames the properties in diachronic terms
(research question 9), summarizes the dissertation’s contributions and explores avenues
for future research.
270
CHAPTER 6
CONCLUDING DIACHRONIC
AND SYNCHRONIC ISSUES
This dissertation has explored the acoustic, articulatory, phonological,
typological, geographic and social aspects of /f/ bilabialization, /f/ velarization and /x/
palatalization in Latin American Spanish by analyzing natural and elicited data from
Chilean and Mexican speakers. Specifically, nine research questions concerning the
phonetic, theoretical and sociolinguistic details of the three processes guided the
discussion. In the preceding chapters, I have responded to all but the final research
question (how can the processes be viewed from a diachronic perspective of Spanish?),
which I address in section 6.3. First, I briefly summarize the main conclusions regarding
the questions 1 through 8 below.
1. Are the fricative phonemes /f/ and /x/ synchronically involved in allophonic
alternations in Latin American Spanish? If so, what are the allophones?
Analysis of center of gravity measurements, spectral peaks and visual cues in Chapter 4
revealed the expected (based on previous literature) /f/ and /x/ allophones: labiodental [f],
bilabial [ɸ], velar [x] and palatal [ç].
2. Are the phonological processes productive (i.e., not merely lexicalized)?
The lack of effect for frequency and register (excluding the hyper-frequent ir and ser
data) suggests that the processes are indeed productive.
3. Can the processes be conceived as assimilations conditioned by the
surrounding vowel(s)? If so, are the assimilations regressive or progressive?
The quality of the preceding vowel was excluded as a triggering factor for all three
processes. This observation coincides with previous phonetic work on coarticulatory
271
preference (Lee 1997, 1999, Ghavami 2002) that found greater overlap between
consonants and subsequent vowels. Given such studies, it was unsurprising that the
quality of the following vowel conditions the processes: /u, o/ for both /f/ properties, /i, e/
for /x/ palatalization. What remains unaddressed is the fact that, in the case of /f/, while
the appearance of either unfaithful allophone is more likely before /u/ and /o/, [ɸ] and [x]
are possible (at much lower rates) before the non-back vowels /i, e, a/, as well. I address
this nuance in section 6.2.
4. Are the assimilations conditioned by stress or position in the word?
Both stress and the position of the fricative in the word were ruled out as relevant
predictors of assimilation.
5. How can a formal model of phonology–Optimality Theory–account for the
phenomena?
An ALIGN constraint that requires the imposition of a vowel’s constriction location (i.e.,
place) gestures on the preceding consonant motivates assimilation. Violated at the
expense of markedness satisfaction is IDENT(constriction location) as it pertains to the
consonant.
6. How can this theoretical analysis limit the assimilations to the subset of
fricatives {/f/, /x/} to the exclusion of all other phonemes in Spanish?
Two families of specific faithfulness constraints–IDENT-MAXDIST-n and IDENTmanner(constriction location)–that dominate ALIGN-CL were proposed to leave only the
non-strident fricatives susceptible to the effects of ALIGN.
7. What role, if any, do articulatory and acoustic phonetics play in the theoretical
model proposed?
272
Theoretically, acoustic phonetics justifies the manner-specific IDENT constraints, given
that languages may treat manner of articulation asymmetrically in terms of possible
targets for CV assimilation. As a broader account of the phenomena in question,
numerous phonetic studies have demonstrated the variability and weakness of cues to
place in (non-strident) fricatives, arguing instead for the importance of the vocalic onset
in fricative identification. This fact accounts for two aspects of the analysis in this
dissertation: directionality (CV as opposed to VC assimilation) and targethood (/f/ and /x/
are vulnerable to assimilation given the relative weakness of their cues to place).
Articulatory phonetics offers an explanation for the assimilations themselves
(gestural overlap) and for the limits on potential phonetic contexts for assimilation
(IDENT-MAXDIST).
8. What crosslinguistic typological implications does 7 entail?; i.e., why do
certain linguistic systems attest certain properties while other systems do not?
From the OT perspective, crosslinguistic properties of all CV types are accounted for by
reranking of the proposed constraints. In terms of the actuation problem in explaining
why the properties exist where they do, the degree of coarticulation and years of formal
education were argued to be two possible factors in geographic and social differentiation,
respectively.
Before exploring the historical aspect of the processes expressed in research
question 9 (§6.3), two pending (minor) elements in the theoretical model remain: the nonobligatory nature of the processes (§6.1) and the flexibility concerning the triggering
vowel in the restricted and free varieties (§6.2). I close the thesis with sections 6.4
273
(contributions of the dissertation and directions for future research) and 6.5
(general/overall conclusions).
6.1. Variability/optionality
Recall from Chapter 1 that the faithful and assimilated allophones for each
process are possible assuming a context in which a corresponding triggering vowel is
present. In other words, while particular speakers may demonstrate a preference for one
or the other, speakers may alternate freely between [f]~[ɸ]/[x] and [x]~[ç] in a single
discourse, sentence or word; assimilation is not obligatory, but rather variable/optional.
Classic OT (Prince and Smolensky 1993/2004, McCarthy and Prince 1993/2001)
was not designed to address such data in a single language. While crosslinguistic
differences arose from constraint reranking, no such mechanism could account for free
variation, e.g., [f]uerte and [x]uerte ‘strong,’ in the same grammar let alone the same
speaker; constraints were strictly ranked and their evaluation points immutable. Under
classic OT assumptions, given the hypothetical constraint inventory C1, C2 and C3, if the
ranking is C1 » C2 » C3, dominance is fixed (assuming that a ranking argument can be
made), and the winning candidate that results from such a ranking (in this language) will
consistently be the winning (attested/grammatical) output. Thus, constraints are
organized as depicted in Figure 6.1 (Boersma and Hayes 2001: 47).
Figure 6.1. Classic OT fixed constraint rankings, C1 » C2 » C3
C1
C2
C3
274
An explanatory problem arises when variable output requires the ranking C3 » C2 (x% of
the time). Boersma and Hayes (2001) propose Stochastic OT whereby the evaluation
points of constraints are not fixed. Constraints are instead envisioned as (normally
distributed) ranges of possible evaluation points that may be perturbed by noise,
increasing or decreasing the evaluation point within the range for any given evaluation. If
two constraints are close enough such that their ranges overlap, constraint reranking may
result depending on the evaluation points selected for a particular utterance.1 Consider the
evaluation in Figure 6.2.
Figure 6.2. Stochastic OT variable constraint rankings, C2 » C3
C1
C2
C3
Note that, in this grammar, C1 is highly ranked and will consistently dominate all other
constraints. However, lower in the hierarchy, domination is not as strict, and the
evaluation ranges for C2 and C3 intersect. For this specific utterance, the evaluation
points for C2 and C3 are found at their respective means (peaks), meaning that the
winning output will correspond to the ranking C2 » C3. Now consider Figure 6.3, a
separate evaluation in the same grammar.
1
Alternatively, Kiparsky (1993) proposes that each outcome (e.g., faithful or assimilated) results
from separate (fixed) rankings (‘multiple grammars’) within the mind of a single speaker. Anttila
and Cho (1998) propose ‘grammar lattices’ consisting of sets of partially ordered constraints.
Coetzee (2009) (Noisy Harmonic Grammar) incorporates lexical frequency as the weighting
mechanism for the Stochastic noise system. I do not intend to argue for or against any alternative
here. (See Cardoso 2005, Anttila 2006 and Anttila and Andrus 2006 for general literature
reviews.)
275
Figure 6.3. Stochastic OT variable constraint rankings, C3 » C2
C1
C2
C3
For this utterance, the evaluation point for C3 shifted slightly higher within its
permissible range while C2 shifted lower, inverting the ranking. As a consequence, C3
dominates C2, and the corresponding output will reflect such an order. Given a normally
distributed range of evaluation points, C3 » C2 will occur quite rarely as the point of
intersection seems to be one standard deviation away from the mean (in opposite
directions) for C2 and C3.
If we concretely apply such a mechanism to the properties at hand, the following
five figures demonstrate the necessary constraint ranges to account for variable output in
/f/ bilabialization ([f]~[ɸ]), /f/ velarization ([f]~[x]) and /x/ palatalization ([x]~[ç]).
Figure 6.4. Chile experimental /f/ bilabialization: ~55% [f], ~45% [ɸ] (see Table 4.6)
IDENT(CL) ALIGN(LP)
Figure 6.5. Mexico experimental /f/ bilabialization: ~85% [ɸ], ~15% [f] (see Table 4.6)
ALIGN(LP) IDENT(CL)
276
Figure 6.6. Experimental /f/ velarization: ~95% [f], ~5% [x] (see Table 4.6)
IDENT(CL) ALIGN(TBCL)
Figure 6.7. Chilean /x/ palatalization: ~85% [ç], ~15% [x] (see Table 5.20)
ALIGN(TTCL) IDENT(CL)
Figure 6.8. Mexican /x/ palatalization: ~75% [x], ~25% [ç] (see Table 5.20)
IDENT(CL) ALIGN(TTCL)
First recall that no control participant exhibited any /f/ allophony, indicating that the
range for the relevant constraints (ALIGN(CL) and IDENT(CL)) does not overlap in the
same vein as C1 and C2 from Figure 6.2; there is no scenario under which these
constraints may rerank in the grammar of these speakers. I thus exclude this from the
figures above. Similarly, variability is not possible with respect to manner and
coarticulatory distance, indicating that the relevant IDENT-manner(CL) and IDENTMAXDIST constraints also assume a categorically dominant (C1) position for Spanish.
277
However, the various degrees to which /f/ and /x/ exhibit allophonic variability in
each of the above groups is demonstrated by the amount of overlap in the respective
constraint ranges. Greater variability approaching 55%-45% exhibits greater overlap,
while more categorical output of the 85%-15% or 95%-5% type overlaps minimally at
the extreme ends of the curves.
Take as an example Figure 6.8, velar palatalization in Mexico. The Mexican
speakers realized /x/ faithfully approximately three-fourths of the time, meaning that the
evaluation point for IDENT(CL) (faithfulness) will outrank that of ALIGN(TTCL)
(markedness) 75% of the time, indicated by the large range in which IDENT dominates
ALIGN. However, at the relatively low end of IDENT and high end of ALIGN, the
constraints intersect, the consequence of which is possible reranking if the evaluation
points are located in this overlapped range, which occurs in approximately 25% of these
speakers’ utterances. The corresponding tableaux are provided in (1) and (2).
(1)
Faithful [x], IDENT(CL) » ALIGN(TTCL)
/xema/ gema ‘gemstone’
a.
[xema]
b.
[çema]
(2)
IDENT(CL)
ALIGN(TTCL)
*
*!
Assimilated [ç], ALIGN(TTCL) » IDENT(CL)
/xema/ gema ‘gemstone’
a.
[xema]
b.
[çema]
ALIGN(TTCL)
+n
IDENT(CL)
–n
*!
*
278
In (1), the assimilated (b) candidate is eliminated given its fatal violation of dominant
IDENT. As mentioned previously, such is the case overall in 75% of these Mexican
speakers’ utterances. However, 25% of the time, a positive noise value (+n) may shift up
the selection point at which ALIGN is evaluated, allowing it to (re)rank above
IDENT(CL) if the latter is also shifted down by a negative noise value (–n). For this
utterance, the assimilated candidate (2b) is now optimal, given its ALIGN satisfaction.
With the addition of Stochastic OT, the theory is equipped to handle
variable/optional output that its original iterations could not. The next section addresses
the longstanding issue of restricted versus free dialects in the two /f/ Spanish processes.
6.2. Context-free assimilation and phonological analogy
Recall from Chapter 1 the observation that, while /x/ palatalization is
categorically limited to the front vowels /i, e/ (*[ça], *[çu] and *[ço]), /f/ alternations are
not. On the other hand, while the probability of bilabial [ɸ] and velar [x] as outputs of /f/
(greatly) increased in the context of the back rounded vowels /u, o/, the data from
Chapter 4 showed that both were also possible (at lower rates) before the assumedly nontriggering vowels /i, e, a/ (permitting /f/→[ɸi], [ɸe], [ɸa] and [xi], [xe], [xa]).
This seems to contraindicate a vowel-conditioned assimilatory analysis if the
alternation is insensitive to the quality of the following vowel since there is neither
backness nor labiality in /i, e, a/ that could explain the velar [x] or bilabial [ɸ] outcomes.
Yet the results from Chapter 4 (higher probability of unfaithfulness before /u, o/) and the
abundance of restricting dialects (Resnick 1975 reports on 84 dialects that allow /f/
allophony; of these, 62 (approximately 73.2%) are restrictive, while 22 (26.2%) allow [x]
279
before any vowel) suggest that the /u, o/-conditioned (assimilatory) grammars represent
the default case and that the freely alternating dialects require an additional explanation.
As I argued in Chapter 1, the inverse is to assume that freely alternating dialects
without assimilation are the null hypothesis and that the dialects that permit the
alternation only before the subset /u, o/ are the exception. Since assimilation no longer
motivates the alternation under this hypothesis, the problem is that it is curious that that
particular (exceptional) subset should be the only subset allowed to exist outside the full
range as opposed to the other logically possible subsets /i, e/, /e, a/ or /i, e, a/. In other
words, there exists no dialect in Spanish that permits /f/ allophony only before the sets of
vowels /i, e/, /e, a/ or /i, e, a/; the only dialectal restrictions are /u, o/ or no restriction at
all (/i, e, a, u, o/).
This anomaly, combined with the coarticulatory data from Chapter 4 and
Resnick’s ratios, suggests that assimilatorily-motivated restricted dialects are the
diachronic origin from which freely alternating dialects arose. That is, at some point in
the development of these dialects, speakers who previously restricted the [f]~[x] and/or
[f]~[ɸ] alternation to /u, o/ due to actual coarticulation analogically extended the [x]/[ɸ]
variants to /f/ in general irrespective of the following vowel. Depending on the frequency
with which [x]/[ɸ] appear instead of [f] for a particular speaker/speech community, /f/
may come to be replaced phonemically by /x/ or /ɸ/. Mexican experimental participant
35792 (female, 38) is one such example. Of the 23 /f/ tokens that she produced, 22 were
bilabial, one was velar and none were labiodental. In her case, it thus appears that [ɸ] is,
in fact, the faithful representation of phonemic bilabial /ɸ/ (/f/ having been replaced
280
completely) and the occasional velar [x] allophone is a consequence of bilabial /ɸ/
velarization.
Such an analogical account in which the output of a phonetically motivated
process ([ɸ]/[x]) is extended to all contexts in which the target appears (/fV/) is not as
unfounded as it might seem superficially since the same rationale has been argued for the
historic development of /Cl/ clusters in Spanish.
Latin (mostly initial) CL- /kl/, PL- /pl/ and FL- /fl/ clusters have been argued to
converge on the palatal lateral [ʎ] in Old Spanish2 precisely by way of analogy (Tuttle
1975, Holt 1997, Zampaulo 2013). How and why the palatal lateral emerged in such
positions are two intriguing questions since no palatalizing yod [j] was present in these
data at this point, nor was the process restricted to front vowels at all. Consider the items
in Table 6.1.
Table 6.1. Development of Latin initial /Cl/ clusters
Latin
Old Spanish
Orthography
Gloss
CLAVE
[ʎ]ave
llave
‘key’ (noun)
PLOVERE
[ʎ]over
llover
‘to rain’
FLAMMA
[ʎ]ama
llama
‘flame’ (noun)
Notice the lack of glide and front vowels, common historical sources for palatalization.
On the other hand, note the presence of (rounded and unrounded) back vowels /o/ and /a/.
It is therefore curious that [l] palatalized in such contexts.
2
Ignoring precise phonetic details, the evolutionary path from Vulgar Latin to modern Spanish
dialects is principally: [Cl] > [Cʎ] > [ʎ] > [ʝ] > [ʒ, ʃ, j] (see Holt 1997, Penny 2002 and Zampaulo
2013 for further discussion).
281
Tuttle (1975), Holt (1997) and Zampaulo (2013) argue that there is a single source
for all three clusters: CL- /kl/. Progressive coarticulation between the dorsal [k] and the
coronal [l] slightly retracted the lateral to the palatal region, resulting in [kʎ]. Markedness
eventually reduced the complex onset to simply [ʎ] in Old Spanish, but it is the
intermediate [kʎ] step, resulting from phonetic overlap, that sets the evolution in motion.
PL- and FL- clusters are treated uniformly given their parallel evolution, yet
absent in both sequences is the dorsal gesture necessary to retract [l]. No phonetic
motivation is available in these cases, yet crosslinguistic comparisons show that the
palatal lateral existed in such contexts, given data from Upper Aragonese (Lloyd 1987).
Table 6.2. Upper Aragonese /Cl/ clusters
Latin
Upper Aragonese
Orthography
Gloss
CLAVE
[kʎ]
cllau
‘key’ (noun)
PLOVERE
[pʎ]
pllover
‘to rain’
FLAMMA
[fʎ]
fllama
‘flame’ (noun)
The initial consonant that is subsequently lost in Old Spanish remains in modern Upper
Aragonese, evidence that the /Cʎ/ stage existed for all three clusters, even in the absence
of phonetic motivation in the case of /pl/ and /fl/. Tuttle (1975), Holt (1997) and
Zampaulo (2013) argue that /l/ palatalized in the context of /pl/ and /fl/ not for phonetic
reasons, but rather under analogy after coarticulation retracted /l/ in the /kl/ contexts.
Stated evolutionarily, CL- /kl/ clusters assimilated first, after which speakers extended by
analogy the effect (outcome) to all /Cl/ contexts, even though the phonetic motivation is
absent in the expanded cases.
282
I apply this same line of reasoning to /f/ allophony; /f/ velarization and
bilabialization began in the context of /u/ and /o/ due to their dorsal and labial
components, after which (some) speakers extended the resultant [x] and [ɸ] allophones to
/f/ before any vowel. (If this might lead to eventual phonemicization of /x/ or /ɸ/ at the
expense of /f/ for (some) speakers is a prospect that I leave open here.)
In the next section, I continue the diachronic discussion and respond to research
question 9 by considering the possibility that the synchronic /f/ allophony analyzed in this
dissertation is merely the result of the same analogical argument from above as an
extension of the processes that affected Latin F (in contexts beyond FL- clusters) in its
development to modern Spanish.
6.3. The fate of Latin F
Spanish is unique in that Vulgar Latin F experiences an evolution absent in other
Romance languages. Compare the orthographic differences of the initial letter in the
following cognates.
Table 6.3. Latin F in Romance
Latin
Spanish
French
Italian
Portuguese
Gloss
FĪLIUS
hijo
fils
figlio
filho
‘son’
FERĪRE
herir
férir
ferire
ferir
‘to wound’
FACERE
hacer
faire
fare
fazer
‘to do/make’
FŪMUS
humo
fumée
fumo
fumo
‘smoke’ (n.)
FORMĪCA
hormiga
fourmi
formica
formiga
‘ant’
Source: Boyd-Bowman (1954: 62–64)
283
Every lexical item in the Spanish column is pronounced as vowel initial; i.e.,
orthographic <h> in Spanish corresponds to no phonemic or phonetic content. It is a
historic vestige that once corresponded to the glottal fricative [h] (/h/), eventually lost by
the 1500s (Pharies 2007: 87). French, Italian and Portuguese, on the other hand, are all [f]
(/f/) initial, inherited directly from Latin, straightforwardly indicated in the orthography
by <f>. Superficially, Spanish took the evolutionary path /f/ > /h/ > /Ø/, while French,
Italian and Portuguese maintained /f/ throughout. When one examines the details of F in
Spanish, however, the process is not that simple.
As overviewed briefly in Chapter 2, Penny (2002: 92) argues that F in Vulgar
Latin was actually bilabial /ɸ/ (replaced with labiodental /f/ in Rome and surrounding
areas), which underwent five stages before arriving at the modern /f/~/Ø/ distinction.
Table 6.4 below reproduces Table 2.10.
Table 6.4. Five stages of Latin F redux
Stage 1
Stage 2
Stage 3
Stage 4
Stage 5
/ɸ/→[ʍ, ɸ]
/ɸ/→[ʍ, h, ɸ]
/ɸ/→[ʍ, h, ɸ]
/h/→[f, h]
/f/ versus /Ø/
[ʍ] preceding
the glide [w]
[ʍ] preceding
the glide [w]
phonemicization
of [f]
dissimilation,
[h] preceding
syllabic [u, o]
e.g., forno
‘oven’
generalization
of [h]
preceding all
vowels
French
influence,
[f] reintroduced
preceding [w]
and [ɾ]
e.g., fuerte,
frío
[ɸ] elsewhere
[ɸ] elsewhere
[h] elsewhere
assimilation,
[ʍ] preceding
the glide [w]
e.g., fuerte
‘strong’
[ɸ] elsewhere
[h]-dropping,
[Ø] elsewhere
e.g., horno
The shaded cells will become relevant below. For now, the entire set of distinctions in
Penny’s account is necessary because F loss from Latin to Spanish was asymmetrical and
284
somewhat variable, and each stage is designed to reflect these differences. For example,
/f/ was generally conserved in /fuV/ contexts (mostly diphthongized from short, stressed
Latin O)–fuerte ‘strong,’ fui (and other past tense ir/ser forms) ‘I went/was’–and /fC/
clusters that did not otherwise palatalize (§6.2)–flor ‘flower,’ frente ‘front.’
However, a few allophones that Penny proposes at each stage (e.g., [ɸ], [ʍ]) are
conjecture since the precise phonetic details are unknown; minimally, Blake (1988)
argues that the written record supports only three main distinctions: <ff>, <f> and <h>.
Table 6.5. Orthographic evidence for F erosion
Written record
ffecha
ffuy
fiziemos
desfaga
haran
hecho
Old Spanish
[f]
[h]
[Ø]
Modern spelling
fecha
fui
Modern Spanish
[f]
harán
hecho
‘date’
‘I was/went’
‘we did/made’
hicimos
deshaga
Gloss
[Ø]
‘undo’ (3s subj)
‘do/make’ (3p fut)
‘done’ (adj)
Source: Blake (1988: 275) citing Menéndez Pidal (1919), document 106, dated 1282, La
Rioja (north central Spain).
Citing 29 historical documents originally compiled by Menéndez Pidal (1919), Blake
argues that scribes, cognizant of the then unstable nature of etymological F, resorted to a
double-f <ff> convention to represent words pronounced with labiodental [f] (or, as
Penny would argue, bilabial [ɸ]) at that time; <f> was reserved for words debuccalized to
glottal [h], and orthographic <h> represented lexical items that had already lost the initial
consonant entirely [Ø], the eventual trajectory of [h].
285
The crucial point here is that, if the synchronic /f/ allophony explored in this
dissertation is in some way related to diachronic F loss (Calvo Shadid 1996), it should be
expected that the processes share any (or perhaps a combination) of the target, trigger
and/or context components.
Indeed, both historical F erosion and modern labiodental bilabialization and
velarization target /f/. As the shaded cells from Table 6.4 indicate, Penny’s proposal also
suggests that /u/ and /o/ were the original triggering vowels after which the processes
expanded to all other contexts, as I claimed for modern /f/ allophony. I will argue against
this commonality below, but first note that the outcomes are unique. While the historical
loss of F resulted in complete deletion (after passing through intermediate glottal [h]),
current /f/ alternations result in bilabial [ɸ] or velar [x], as I demonstrated phonetically in
Chapter 4.
Concerning the shaded regions of Table 6.4, Penny argues that F loss began as
/ɸ/→[h] labial dissimilation in the environment of /u, o/. The first reason to question this
origin is that, like the Chapter 2 criticism of Lipski (1995) and Mazzaro (2005), this
dissimilation only affected /ɸu/ and /ɸo/ contexts, excluding all other manners of
articulation in labial-labial environments. Penny does not address this fact.
Second, the assumption that F erosion in /u, o/ environments preceded erosion
elsewhere (/i, e, a/) is belied by the written record analyzed in Blake (1988). The earliest
Menéndez Pidal (1919) document that Blake (1988: 274) cites is number 257, dated 1216
from Sigüenza, also in north central Spain (slightly south of La Rioja). Admittedly a
small sample from a single document, what Blake does note in the written record is
286
(early) F loss primarily in the context of non-back vowels (Table 6.6). Such evidence
seems to argue against an originally back-vowel triggered process.
Table 6.6. F erosion, document 257, 1216, Sigüenza
Written record
ffelizes
Alffonso
Old Spanish
Modern spelling
[f]
fazemos
fiziesse
felices
Alfonso
Modern Spanish
[f]
fizo
hiciese
‘happy’ (pl)
proper name
‘we do/make’
hacemos
[h]
Gloss
[Ø]
hizo
‘do/make’ (3s subj)
‘do/make’ (3s pret)
Furthermore, there is debate concerning the catalyst for F erosion regarding whether it
was an internal/phonological/phonetic process (Naro 1972, Torreblanca 1984, Penny
2002) or one resulting from an external/language contact situation specifically with
Basque, in which speakers of Old Spanish (driven north due to the Muslim conquest), via
influence from contact with /f/-less Basque speakers, replaced /f/ with /h/ (Otero 1971,
Iribarren-Argaiz 1998). If the external account is the case, such would suggest that F
erosion did not begin with /u, o/, since it would be highly curious why one-to-one /f/~/h/
replacement would start in /fu, fo/ lexical items instead of across the board. Evidence for
the language contact account is twofold.
First, returning again to Blake (1988: 272, 273), analyzing the texts globally
reveals F loss earliest in the north due to contact with Basque and /f/ maintenance for
much longer periods of time in the south prior to the Reconquest, where no contact with
Basque was possible. Compare the shared roots in Table 6.7 (documents are listed left to
right geographically (north to south) and (mostly) chronologically).
287
Assuming Blake’s argument that scribes utilized the <ff> [f]~<f> [h] distinction,
crucially note the early F loss in the north (even in the context of non-back vowels) and
the prolonged F retention in the south, suggesting a northern origin for the process.
Table 6.7. F erosion, early northern origin
Castilla del Norte
Castilla del Norte
La Rioja Baja
Valladolid
document 60
document 62
document 133
document 230
dated 1267
dated 1270
dated 1329
dated 1318
hacer
fecho, fizo
fecho
ffecho
ffacer
hijo
fijo
fijos
ffijo
Root
fasta
ffasta
Gloss: hacer ‘to do/make,’ fecho (modern hecho) ‘done/made’ (adjective), fizo (modern
hizo) ‘did/made’ (3s pret); fijo (modern hijo) ‘son’; fasta (modern hasta) ‘until.’
hasta
A second argument in favor of the Basque genesis of F erosion is the fact that many other
processes unique to Castilian Spanish have been attributed to Basque contact without
such controversy. Quoting Otero (1971: 187):
I believe that after everything [that I have discussed], it does not seem
arbitrary [emphasis mine] to attribute the revolutionary evolution of
Castilian Spanish to the influence of Basque speakers. Aside from the
historical background, the phonological coincidences are many: vowels,
difficulty with labiodentality [JR: F erosion], difficulty with the voiced
sibilants [JR: devoicing], [and] apicality of the dental fricatives [JR: apical
s][.] [my translation]3
3
Original text:
Creo que después de todo lo que precede no parecerá arbitrario atribuir la
revolucionaria evolución del español del castellano a la influencia de los
hablantes de vascuence. Aparte el fondo histórico, son demasiadas las
coincidencias fonológicas: vocales, dificultad con la labiodentalidad, dificultad
con las sibilantes sonoras, [y] apicalidad de las fricativas dentales[.]
288
The above suggests that the impetus for F erosion (present only in Spanish where, not
coincidentally, the separating factor from other Romance languages is Basque contact) is
likely external, meaning that /f/ for /h/ replacement beginning in the context of /u, o/ is
suspect. Rather, the fact that F loss occurred irrespective of the following vowel calls into
question Penny’s (now unmotivated) intermediate stages. If this is the case, historical F
loss and modern /f/ alternations only share the target /f/; both the contexts for application
and outcomes are distinct, evidence for the separate (diachronic versus synchronic)
treatment of each process.
The next section summarizes the contributions of this dissertation and explores
avenues for future research given its open questions and own shortcomings.
6.4. Theoretical and empirical contributions
and future research
This thesis has extensively explored /f/ velarization, /f/ bilabialization and /x/
palatalization in Latin American Spanish specifically as they appear in the grammars of
four Chilean and four Mexican speakers. Responding to Mielke’s (2008) appeal to
comprehensiveness, I have approached the analysis from the acoustic, articulatory,
theoretical, historical and social perspectives.
I updated the previous descriptive accounts on the topics (Perissinotto 1975,
Resnick 1975, Sanicky 1988, Lipski 1995, Calvo Shadid 1996) with detailed phonetic
analyses. The phonological proposal unified three properties that were largely
unaddressed in theoretical Spanish phonology (see, however, Lipski 1995, Mazzaro
2005). Finally, I framed the origin and dispersion of the properties in diachronic and
dialectal terms following the claims in Baker et al. (2011).
289
In terms of the OT proposal from Chapter 5, there are two predictions that offer
possibilities for future research. By limiting myself to Latin American, I excluded a
fourth (potential) fricative-vowel assimilatory process from peninsular Spanish, /x/
retraction. Piñeros (2009: 284) reports that velar /x/ retracts to uvular [χ] in some
varieties of Castilian Spanish. As is the case of the /f/ alternations, two groups of
speakers exist: those that limit retraction to the back vowels /u/ and /o/ (e.g., [χu] juguete
‘toy,’ [χu̯ i] juicio ‘judgment’ and [χo] jota ‘the letter <j>’) and those who make no such
vocalic distinction ([χe] gemelo ‘twin’).
The second prediction is that the voiced fricative/approximant allophones [, ð̪ , ɣ]
that result from post-continuant spirantization of the voiced stops /b, d̪ , ɡ/ should also be
susceptible to the type of non-strident fricative assimilation addressed in this dissertation.
In fact, such processes are well documented (see, e.g., Mazzaro 2010, 2011). In the poem
Por qué me quité del vicio, ‘Why I gave up my bad habits,’ by Manuel Bernal (cited in
Chapter 1), line 101 reads Pa’ qui aprendas a ser güeno ‘so that you learn to be good
(i.e., behave).’ The word in bold is orthographically modified to represent [ɣu̯ eno], a
dorsally assimilated bilabial [] from the standard bueno [u̯ eno] ‘good.’ Both these
properties are interesting testing grounds for the prediction made by the theoretical
analysis presented here that leaves (all) non-strident fricatives vulnerable to assimilation
in Spanish.
Future theoretical research with respect to typology is found in the proposed
IDENT-manner(CL) (implicational along the lines of Jun (1994, 1995) and Herrick
(1999) or not?) and IDENT-MAXDIST-n (crosslinguistically tenable or not?) constraints.
290
Empirically, the Mexican participant data suffered in terms of quality given the
unavoidable testing locations. In an ideal world, researchers would have access to soundattenuatued phonetic laboratories with which to acoustically and visually document
speakers of these dialects to avoid background noise. However, in my case, practical and
socioeconomic considerations required that I interview these participants at their places
of employment without the additional benefit of video equipment that was possible in the
phonetics laboratory in Chile.
6.5. Conclusions
Phonological theory at its core seeks to model speakers’ grammatical
representations of sound and sound processes at a given point in time. Increasingly this
goal has incorporated contributions by acoustic phonetics, diachronic considerations and
sociolinguistic variation. I integrated in this dissertation all these components for the
purpose of presenting a thorough, unified account of three underreported processes in two
varieties of Latin American Spanish.
291
APPENDIX A
TASK 1: SOCIOLINGUISTIC INTERVIEW
A.1. Task 1 instructions and written questionnaire
A.1.1. Spanish version
Favor de escoger un número de cinco dígitos y escribirlo en la caja arriba. Usaré
el número que escoja para identificar todas sus respuestas y así conectar cada parte del
estudio. No guardaré ninguna lista que pueda identificarlo/a a Ud. por su número. Por
esta razón, no será posible usar el código para saber sus respuestas. No se requiere su
nombre.
1. ¿Es Ud. hombre o mujer?
2. ¿De dónde es?
3. ¿En qué ciudad vive actualmente?
4. ¿Cuántos años tiene?
5. ¿Hasta qué grado escolar ha completado?
6. ¿Ha asistido a la universidad?
7. ¿Qué estudia/estudiaba?
8. ¿Tiene un trabajo? ¿En qué trabaja?
9. ¿Habla otra(s) lengua(s) además del español?
10. Si sí las habla, ¿cuál es/cuáles son la(s) otra(s) lengua(s)?
11. ¿Qué lengua considera ser su primera lengua?
12. ¿Qué lengua considera ser su lengua dominante (la que usa más
frecuentemente)?
292
13. Si Ud. se considera un/a hispanohablante nativo/a, ¿a qué edad empezó a
aprender su(s) otra(s) lengua(s)?
14. Si Ud. no se considera un/a hispanohablante nativo/a, ¿a qué edad empezó a
aprender el español?
A.1.2. English translation
Please select a five digit number and write it in the box above. I will use the
number that you select to identify your answers and connect each part of the study. I will
not save any list that will identify you based on your number. For this reason, it will not
be possible to use the code to know your answers. Your name is not required.
1. Are you male or female?
2. Where are you from?
3. In what city do you currently live?
4. How old are you?
5. What is the highest grade you have completed?
6. Have you attended college?
7. What are you studying/what did you study?
8. Do you have a job? What do you do?
9. Do you speak languages other than Spanish?
10. If yes, what are the other languages?
11. What language do you consider to be your first language?
12. What language do you consider to be your dominant language (which do you
use most frequently)?
293
13. If you consider yourself a native Spanish speaker, at what age did you begin
to learn your other language(s)?
14. If you do not consider yourself a native Spanish speaker, at what age did you
begin to learn Spanish?
A.2. Task 1 oral interview questions
A.2.1. Spanish version
1. ¿Me puede hablar de su niñez? Cuénteme dónde nació, dónde se crió,
describa estos lugares, por qué son especiales, y qué recuerdos asocia con
ellos.
2. ¿Me puede contar una anécdota cómica de su niñez?
3. ¿Me puede contar una anécdota vergonzosa de su niñez?
4. ¿Viene de una familia grande? ¿Cuántos hombres y cuántas mujeres hay
entre sus hermanos?
5. ¿Me puede contar una historia familiar? ¿Cuántas personas hay en su
familia? ¿Cómo son? ¿Dónde viven? ¿Qué hacen? ¿Con quién se lleva mejor
en su familia? ¿Qué haces o qué hacían juntos/as? ¿Qué asocia con esta
persona?
6. ¿Me puede contar acerca de sus costumbres familiares? ¿Cómo celebran
Uds. los feriados y los días familiares importantes, por ejemplo, los
cumpleaños o el santo, las bodas, etc.?
7. ¿Qué le gusta hacer para relajarse? ¿Tiene algún pasatiempo o deporte
favorito, le gusta jugar al fútbol, ir al cine, salir con amigos?
294
8. ¿Me puede hablar de su plato favorito de la comida chilena/mexicana?
Describa los ingredientes, el proceso culinario, dónde puedo conseguirlos.
9. ¿Qué aspecto del verano le gusta más? ¿Del otoño? ¿Del invierno? ¿De la
primavera?
10. ¿Me puede hablar de sus amigos/as? ¿Cuántos/as tiene? ¿Cómo son? ¿Qué
les gusta hacer juntos/as? ¿Cómo se conocieron?
A.2.2. English translation
1. Can you talk to me about your childhood? Tell me about where you were
born, where you were raised, describe these places, why they are special, and
what memories you associate with them.
2. Can you tell me a funny story from your childhood?
3. Can you tell me an embarrassing story from your childhood?
4. Do you come from a large family? How many brothers and sisters do you
have among your siblings?
5. Can you tell me a family story? How many people are in your family? What
are they like? Where do they live? What do they do? With whom do you get
along best? What do you do or did you used to do with them? What do you
associate with this person?
6. Can you tell me about your family customs? How did you celebrate important
holidays such as birthdays, weddings, etc.?
7. What do you like to do to relax? Do you have a favorite hobby or sport, such
as playing soccer, going to the theater, spending time with friends?
295
8. Can you talk to me about your favorite Chilean/Mexican dish? Describe its
ingredients, how to prepare it, where I can order it.
9. What part of the summer, winter, fall and spring do you like best?
10. Can you tell me about your friends? How many do you have? What are they
like? What do you like to do together? How did you meet?
296
APPENDIX B
TASK 2: PERCEPTION
B.1. Nativeness rating task instructions
B.1.1. Spanish version
Ud. va a escuchar unas secuencias de sonidos. Usando los siguientes números,
por favor determine si los sonidos le parecen:
1
nativos al español
2
un poco nativos
3
Ud. no está seguro/a
4
un poco extraños para el español
5
no existen en español
Esta parte del estudio dura aproximadamente 20 minutos. Hay setenta (70) sonidos. No
se puede ir hacia atrás, y es necesario dar su reacción inmediata. Ud. controla el avance
de cada página, pero los sonidos empiezan inmediatamente después del avance. Se
recomienda que Ud. suba el volumen y use audífonos.
B.1.2. English translation
You are going to hear sequences of sounds. Using the following numbers, please
determine if the two sequences of sounds seem:
1
native to Spanish
2
somewhat native
3
you are not sure
4
a little strange for Spanish
5
nonexistent in Spanish
297
This part of the study lasts approximately 20 minutes. There are seventy (70) sounds.
You cannot go backward, and it is necessary to give your immediate reaction. You
control the advancement of each page, but the sounds begin immediately after advancing.
It is recommended that you turn up the volume and use headphones.
B.2. Task 2 AX discrimination instructions
B.2.1. Spanish version
Ud. va a escuchar unas secuencias de sonidos organizadas en parejas. Usando
los siguientes números, por favor determine si las dos secuencias de sonidos de cada
pareja son:
1
iguales
2
semejantes
3
Ud. no está seguro/a
4
un poco diferentes
5
completamente diferentes
Esta parte del estudio dura aproximadamente 20 minutos. Hay cien (100) sonidos. No se
puede ir hacia atrás, y es necesario dar su reacción inmediata. Ud. controla el avance de
cada página, pero los sonidos empiezan inmediatamente después del avance. Se
recomienda que Ud. suba el volumen y use audífonos.
B.2.2. English translation
You are going to hear sequences of sounds organized in pairs. Using the
following numbers, please determine if the two sequences of sounds in each pair are:
1
identical
2
similar
298
3
you are not sure
4
a little different
5
completely different
This part of the study lasts approximately 20 minutes. There are one hundred (100)
sounds. You cannot go backward, and it is necessary to give your immediate reaction.
You control the advancement of each page, but the sounds begin immediately after
advancing. It is recommended that you turn up the volume and use headphones.
299
APPENDIX C
TASK 3: SENTENCE READING
C.1. Task 3 instructions
C.1.1. Spanish version
En esta sección del estudio, Ud. va a ver una lista de oraciones para leer. Se pide
que las lea a una velocidad y en un estilo normal/conversacional.
Dependiendo de la velocidad con la cual lea la lista, esta sección no debe durar
más de treinta (30) minutos. Durante esta sección tiene la libertad de pedir un descanso
en cualquier momento del estudio. Por favor, avise al investigador cuando necesite un
descanso, y pueda comenzar a leer de nuevo una vez que Ud. esté listo/a. Hay 110
oraciones en total. Favor de indicarle al investigador cuando esté listo/a para empezar.
C.1.2. English translation
In this section of the study, you are going to see a list of sentences to read. It is
asked that you read them at a normal/conversational rate and style.
Depending on the speed at which you read the list, this section should last no
more than thirty (30) minutes. During this section you have the freedom to take a break at
any point during the study. Please advise the researcher when you need a break, and you
may begin to read again when you are ready. There are 110 sentences in total. Please
indicate to the researcher when you are ready to begin.
C.2. Frequency analyses
The following tables present the unpaired t-test frequency analyses for the tokens
used in task 3 as per the protocol found in Marshall and van der Lely (2006) and
described in Chapter 3. Each table constitutes one condition (/f/ assimilating, /f/ non-
300
assimilating, etc.) from section 3.4. Columns are divided by ‘frequent’ versus
‘infrequent’ designation, the number of appearances (listed as ‘frequency’) in the
LEXESP corpus (Sebastián-Gallés et al. 2000) and the natural log of the ‘frequency’
value over which the p value is computed. The rows are divided according to the
phonological/phonetic variables discussed in Chapter 3: (a) initial, stressed; (b) initial,
unstressed; (c) non-initial, stressed; and (d) non-initial, unstressed positions (with the
exception of the ir/ser condition). Means, standard deviations (SD) and p values
(significance below 0.05 denoted by an asterisk ‘*’) are found in the final three rows of
each table. See section C.3 for glosses. Significance between ‘frequent’ and ‘infrequent’
tokens within each category was achieved.
Table C.1. Task 3 /fu, fuV, fo/ assimilating frequency analysis
Frequent
Frequency
Natural log
Infrequent
Frequency
Natural log
fuga
131
4.88
fuco
2
0.69
fuente
664
6.50
fuelle
8
2.08
fondo
1191
7.08
fobia
47
3.85
futuro
905
6.81
funesto
24
3.18
fortuna
299
5.70
fogoso
26
3.25
profundo
699
6.55
difuso
62
4.13
afuera
168
5.12
refuerzo
24
3.18
uniforme
228
5.43
aforo
247
1.95
difusión
148
5.00
efusivo
7
1.95
metáfora
105
4.65
aforismo
19
2.94
Mean
453.80
5.77
46.60
2.72
SD
384.04
0.89
72.76
1.04
a.
b.
c.
d.
p
<0.001*
301
Table C.2. Task 3 /fi, fe, fa/ non-assimilating frequency analysis
Frequent
Frequency
Natural log
Infrequent
Frequency
Natural log
fino
327
5.79
físico
4
1.39
fecha
421
6.04
ferro
2
0.69
falta
1264
7.14
fario
3
1.10
final
1668
7.42
filete
17
2.83
fenómeno
501
6.22
festón
5
1.61
familiar
609
6.41
farsante
5
1.61
beneficio
298
5.70
grafito
2
0.69
efecto
1027
6.93
prefecto
17
2.83
artefacto
61
4.11
calafate
2
0.69
dificultad
538
6.29
sáfico
2
0.69
diferente
982
6.89
plumífero
2
0.69
tarifa
73
4.29
pífano
2
0.69
Mean
647.42
6.10
5.25
1.29
SD
491.99
1.03
5.61
0.81
a.
b.
c.
d.
<0.001*
p
Table C.3. Task 3 /uf, of/ progressive assimilation frequency analysis
Frequent
Frequency
Natural log
Infrequent
Frequency
Natural log
bufanda
29
3.37
bufete
11
2.40
profano
30
3.40
ofidio
11
2.40
suficiente
681
6.52
acúfeno
2
0.69
oficial
774
6.65
estrofa
15
2.71
Mean
378.50
4.99
9.75
2.05
SD
404.78
1.85
5.50
0.92
a.
b.
p
0.0294*
302
Table C.4. Task 3 /xi, xiV, xe/ assimilating frequency analysis
Frequent
Frequency
Natural log
Infrequent
Frequency
Natural log
giro
174
5.16
gili
2
0.69
gente
1855
7.52
geno
2
0.69
gigantesco
162
5.09
jineta
2
0.69
general
1892
7.55
geógrafo
2
0.69
registro
136
4.91
ilegible
12
2.49
religión
340
5.83
elogioso
7
1.95
ejemplo
619
6.43
manijero
2
0.69
lógico
545
6.30
frígido
13
2.56
imagen
1310
7.18
hereje
34
3.53
Mean
781.44
6.22
8.44
1.55
SD
716.91
1.04
10.58
1.10
a.
b.
c.
d.
<0.001*
p
Table C.5. Task 3 /xu, xo, xa/ non-assimilating frequency analysis
Frequent
Frequency
Natural log
Infrequent
Frequency
Natural log
justo
535
6.28
judo
2
0.69
joven
1126
7.03
jota
10
2.30
jarro
94
4.54
jade
2
0.69
juventud
402
6.00
jurel
2
0.69
jornada
320
5.77
joroba
10
2.30
jamás
744
6.61
jayán
2
0.69
ajuste
62
7.12
bajura
2
0.69
enojoso
14
2.64
piojoso
6
1.79
tajante
69
4.23
majada
2
0.69
lujurioso
24
3.18
sojuzgado
2
0.69
trabajo
2229
7.71
navajo
2
0.69
hoja
412
6.02
sonaja
2
0.69
Mean
502.58
5.59
3.67
1.05
SD
639.39
1.60
3.17
0.66
a.
b.
c.
d.
P
<0.001*
303
Table C.6. Task 3 /ix, ex/ progressive assimilation frequency analysis
Frequent
Frequency
Natural log
Infrequent
Frequency
Natural log
quijote
51
3.93
rijoso
6
1.79
prejuicio
100
4.61
viejuco
2
0.69
hijo
2215
7.70
prefijo
6
1.79
lejos
437
6.08
hollejo
2
0.69
Mean
700.75
5.58
4
1.24
SD
1023.98
1.67
2.31
0.64
a.
b.
0.0029*
p
C.3. Task 3 Spanish sentences (non-randomized)
with corresponding translation
Please note that targeted words appear in bold; words that are not targeted in a
specific sentence yet that contain phonological contexts relevant for other conditions
under investigation appear with a single underscore.
/f/ assimilating
1.
El plan de los prisioneros resultó en una fuga masiva de la cárcel.
The prisoner’s plan resulted in a mass jailbreak.
2.
Había mucho fuco en la costa durante nuestras últimas vacaciones.
There was a lot of alga on the coast during our last vacation.
3.
Es común ver una fuente en la plaza central de muchas ciudades grandes
del mundo.
It is very common to see a fountain in the central plaza of many of the
world’s large cities.
4.
Tener mucho fuelle es una gran ventaja para los atletas profesionales.
Having a lot of stamina is a big advantage for professional athletes.
5.
Según los jefes no hay fondos suficientes para un aumento de sueldo este
año.
According to the bosses there are not enough funds for raises this year.
6.
A la vecina que le tiene fobia a los animales no le gusta ir de caminata.
The neighbor who has a fear of animals does not like to go hiking.
304
7.
Seguramente la futura generación humana va a tener que enfrentar mucha
contaminación.
Certainly future generations are going to have to confront a lot of
contamination.
8.
El viaje fue muy funesto después de que perdimos los pasaportes.
The trip was a disaster after we lost our passports.
9.
Por haber ganado el partido el equipo tuvo la fortuna de jugar en el
campeonato.
For having won the game, the team had the fortune of playing in the
championship.
10.
Mi hermano tiene un amigo fogoso que siempre me está gritando.
My brother has a fiery friend who is always shouting at me.
11.
Nadar en los ríos profundos puede ser peligroso si uno no tiene cuidado.
Swimming in deep rivers can be dangerous if one is not careful.
12.
Leer autores que escriben en una manera difusa es un reto.
Reading authors who write in a vague manner is a challenge.
13.
Sería mejor que los niños jugaran afuera para evitar accidentes en la casa.
It would be better if the children were to play outside to avoid accidents in
the house.
14.
El ejército pidió refuerzos después de una batalla intensa.
The army asked for reinforcements after an intense battle.
15.
Las bailarinas llevaron uniformes ornamentados para la competencia.
The ballerinas wore decorated uniforms for the competition.
16.
La banda popular necesita un estadio con un aforo de cien mil plazas.
The popular band needs a stadium with a (seating) capacity of 100,000
seats.
17.
Después del artículo periodístico la difusión del evento aumentó mucho.
After the newspaper article, news of the event spread considerably.
18.
Prefiero pasar tiempo con gente efusiva que con gente reservada.
I prefer to spend time with effusive people rather than reserved people.
19.
Es arduo para los estudiantes entender poemas con muchas metáforas.
It is arduous for students to understand poems with many metaphors.
305
20.
Para que un aforismo sea efectivo, necesita hablar de temas universales.
For an aphorism to be effective, it needs to talk about universal themes.
/f/ ir and ser
21.
La niña fue a la película temprano para conseguir los asientos perfectos.
The girl went to the movies early to get the perfect seats.
22.
La madre quería que su hijo fuera al doctor ya que su dolor había
persistido.
The mother wanted her son to go to the doctor because his pain had
persisted.
23.
Los niños fueron al parque todos los fines de semana ese verano.
The children went to the park every weekend that summer.
24.
Cuando estaba en Perú fui a Lima, pero no tuve tiempo para ir a Cuzco.
When I was in Peru I went to Lima, but I did not have time to go to Cusco.
25.
Les recomendé a mis padres que fueran al mar para celebrar su
aniversario.
I recommended that my parents go to the beach to celebrate their
anniversary.
26.
Mis hermanos y yo fuimos a un concierto el fin de semana pasado.
My brother and I went to a concert last weekend.
27.
Cuando me llamó tu esposa le dije que fuiste al mercado.
When your wife called me I told her that you went to the store.
28.
Te pedí que fueras a la tienda para comprarme comida.
I asked you to go to the store to buy me food.
/x/ assimilating
29.
Para llegar al destino tienes que tomar cuatro giros a la izquierda.
To get to the destination you have to take four left turns.
30.
Esta gila hizo algo tan bobo anoche que no lo vas a creer.
This silly person did something so foolish last night that you are not going
to believe it.
31.
Mis amigos no se divirtieron porque había demasiada gente en la playa.
My friends did not have fun because there were too many people on the
beach.
306
32.
Los miembros de una monarquía que vienen de geno real nunca tienen que
trabajar.
The members of a monarchy who come from royal lineage never have to
work.
33.
Las ballenas son gigantescas comparadas con el krill que comen.
Whales are gigantic compared to the krill that they eat.
34.
Montar a la jineta es una manera específica de montar a caballo.
The jumping position is a specific way of riding a horse.
35.
Por lo general mis vecinos son respetuosos pero a veces hacen mucho
ruido.
Generally my neighbors are respectful but sometimes they make a lot of
noise.
36.
Vinieron cincuenta geógrafos a la conferencia para discutir sus
investigaciones.
Fifty geographers came to the conference to discuss their research.
37.
Los datos que buscas deben estar en algún registro gubernamental.
The facts that you are looking for should be in a governmental registry.
38.
En ciertas profesiones la escritura ilegible puede causar problemas graves.
In certain professions, illegible handwriting can cause serious problems.
39.
Este semestre Dana toma clases de filosofía, religión, y literatura.
This semester Dana is taking philosophy, religion and literature classes.
40.
Todo el elogioso trabajo que ha hecho su ayudante explica su ascenso.
All the praiseworthy work that her/his assistant has done explains her/his
promotion.
41.
Antes de cada proyecto el jefe siempre demuestra un buen ejemplo de lo que
espera.
Before each project the boss always shows a good example of what he
expects.
42.
Todos los obreros renunciaron a causa de un manijero muy exigente.
All the workers quit because of a very demanding foreman.
43.
Los padres discutieron la manera más lógica de cómo dividir las
responsabilidades.
The parents discussed the most logical way to divide the responsibilities.
307
44.
Las temperaturas frígidas de los polos causan problemas para los
investigadores allí.
The frigid polar temperatures cause problems for the researchers there.
45.
A los niños les gusta encontrar imágenes de animales en las nubes.
Children like to find images of animals in the clouds.
46.
Por haber actuado como herejes, los jefes los despidieron.
For having behaved like heretics, the bosses fired them.
/f/ progressive assimilation
47.
La madre de Sebastián le tejió una bufanda roja, su color favorito.
Sebastian’s mother wove him a red scarf, his favorite color.
48.
Paula fue al bufete del abogado para discutir su próxima demanda.
Paula went to the lawyer’s office to discuss her next lawsuit.
49.
El experto y el profano sorprendentemente tuvieron una buena discusión.
The expert and the novice surprisingly had a good discussion.
50.
Dentro de la clasificación de ofidio se encuentran las serpientes.
Within the ophidian classification, snakes are found.
51.
La madre le dijo al niño que no tienen dinero suficiente para un juguete
hoy.
The mother told the child that they do not have enough money for a toy
today.
52.
Si uno sufre los síntomas del acúfeno es necesario ir a un especialista.
If one suffers from symptoms of tinnitus it is necessary to go to a specialist.
53.
La Real Academia Española intenta tener la palabra oficial de la lengua.
The Royal Spanish Academy tries to have the final say about the language.
54.
Los instructores de literatura esperan estrofas organizadas de sus
estudiantes.
Literature instructors expect organized stanzas from their students.
/x/ progressive assimilation
55.
Don Quijote es un libro que todos los estudiantes en España tienen que leer.
Don Quijote is a book that every student in Spain must read.
56.
Los chicos rijosos se gritaron por horas después de la pelea.
The quarrelsome boys shouted at each other for hours after the fight.
308
57.
No es justo juzgar a alguien a base de prejuicios o estereotipos.
It is not fair to judge someone based on prejudices or stereotypes.
58.
Tras los años el vejuco ya no se parece al hombre guapo que era.
After many years the old man no longer resembles the handsome man he
once was.
59.
A la madre y a su hijo les encanta caminar por el parque los sábados.
The mother and son love to walk through the park on Saturdays.
60.
La lingüista es experta en morfología, especialmente de los prefijos de
sustantivos.
The linguist is an expert on morphology, especially noun prefixes.
61.
Cuando Braulio va de vacaciones, viaja muy lejos para escapar del estrés.
When Braulio goes on vacation, he travels far to escape the stress.
62.
No se debe comer el hollejo de algunas frutas, la naranja, por ejemplo.
You should not eat the skin of some fruits, the orange, for example.
/f/ non-assimilating
63.
A la novia no le gusta el pelo muy fino que tiene.
The bride/girlfriend does not like the really fine hair that she has.
64.
Es mejor tener clase en un aula física en vez de electrónica.
It is better to have class in a physical classroom than in an electronic one.
65.
El comité todavía no sabe la fecha para la próxima reunión.
The committee still does not know the date for the next meeting.
66.
La galera no puede parar porque este ferro está roto.
The galley cannot stop because this anchor is broken.
67.
La profesora se ha dado cuenta de la falta de esfuerzo por parte de los
alumnos.
The professor has realized the lack of effort by the students.
68.
Según la niñita, le trae mucho fario su muñeca.
According to the little girl, her doll brings her a lot of (good) luck.
69.
A finales de diciembre la familia García se va a mudar.
At the end of December the García family is going to move.
309
70.
Arturo compró veinte filetes porque esa noche iba a tener una fiesta.
Arturo bought twenty fillets because that night he was going to have a party.
71.
Era fenómeno el proyecto del grupo que recibió la nota más alta de la
clase.
The project from the group that received the highest grade in the class was
phenomenal.
72.
El edificio turístico incluye puertas adornadas de festones complejos.
The touristic building includes doors adorned with complex festoons.
73.
Su cara me parece familiar, pero todavía no lo reconozco.
His face seems familiar to me, but I still do not recognize him
74.
La vendedora farsante me estafó la semana pasada.
The phony seller ripped me off last week.
75.
Aumentaron mucho los beneficios de la empresa después del nuevo
producto.
The company’s profits increased a lot after the new product.
76.
La compañía requiere mucho grafito porque los lápices son su producto
principal.
The company requires a lot of graphite because pencils are its principal
product.
77.
El nacimiento de su primer hijo tenía un gran efecto en su vida.
The birth of her/his first son had a large effect on her/his life.
78.
El nuevo prefecto está a cargo de todos los asuntos financieros.
The new prefect is in charge of all the financial matters.
79.
La paleontóloga descubrió un artefacto que data del siglo dieciséis.
The paleontologist discovered an artifact that dates to the 16th century.
80.
Después de que el calafate termine arreglando el barco, podemos salir.
After the caulker finishes fixing the ship, we can leave.
81.
El estudiante tiene mucha dificultad con las matemáticas.
The student has a lot of difficulty with math.
82.
Contiene muchos versos sáficos las obras de Unamuno.
The works of Unamuno contain many Sapphic verses.
83.
Juana busca un estilo personal diferente después de años de lo mismo.
Juana is looking for a different personal style after years of the same (style).
310
84.
Hay un pájaro bien plumífero que vive en el bosque cerca de mi casa.
There is a very feathery bird that lives in the woods near my house.
85.
Imponen una tarifa de cierta cantidad cuando pasas por la aduana.
They impose a tax of some amount when you pass through customs.
86.
Mi amiga María empezó a tocar el pífano cuando tenía diez años.
My friend María began to play the fife when she was ten years old.
/x/ non-assimilating
87.
El niño se quejó de que no era justo su castigo.
The boy complained that his punishment was not just.
88.
Todos los viernes practico judo en vez de ir al gimnasio.
Every Friday I do judo instead of going to the gym.
89.
Cuando era joven Ramón les solía causar muchos problemas a sus padres.
When he was young, Ramon tended to cause a lot of problems for his
parents.
90.
Durante el examen de ortografía Víctor no sabía si la palabra contenía jota
o ge.
During the writing exam Victor did not know if the word contained ‘j’ or ‘g.’
91.
Manuela tiene una jarra favorita que le gusta usar sólo para reuniones
elegantes.
Manuela has a favorite pitcher that she likes to use only for elegant
reunions.
92.
Los mineros creían que era jade la piedra que encontraron.
The miners thought that the rock that they found was jade.
93.
Mis abuelos recordaron la juventud por horas durante su aniversario.
My grandparents reminisced about their youth for hours during their
anniversary.
94.
Vimos un banco de treinta jureles durante nuestro último crucero.
We saw a school of thirty scads during our last cruise.
95.
Julia se siente cansadísima después de su jornada de diez horas.
Julia feels really tired after her ten hour workday.
96.
El camello dromedario tiene una joroba mientras el bactriano posee dos.
The dromedary camel has one hump while the Bactrian has two.
311
97.
Te juro que nunca jamás vuelvo a la casa de ese hombre.
I swear to you that I am never again returning to that man’s home.
98.
Todos los niños respetaron la jayana de la escuela por su estatura.
All the children respect the strong girl of the school because of her stature.
99.
La directora hizo un ajuste del programa porque uno de los músicos
canceló.
The director made an adjustment to the program because one of the
musicians cancelled.
100. La bajura del niño desafortunadamente le hace susceptible al acoso
escolar.
The boy’s shortness unfortunately makes his susceptible to bullying.
101. Los otros niños no quieren jugar con el niño enojoso porque los molesta.
The other children do not want to play with the annoying boy because he
bothers them.
102. Las hijas querían cuidar al perro piojoso pero los padres les dijeron que no.
The girls wanted to care for the flea-ridden dog but their parents told them no.
103. Le dieron un «no» tajante a la propuesta porque cuesta demasiado.
They gave her/him a forceful “no” because the proposal costs too much.
104. Los pastores descubrieron que el ganado escapó de la majada anoche.
The shepherds discovered that the cattle escaped from the pen last night.
105. Le dio una mirada lujuriosa que lo cautivó a él.
She gave him a lustful look that captivated him.
106. Los esclavos sojuzgados se rebelaron contra el amo.
The subjugated slaves rebelled against the master.
107. Los maestros insisten en que los estudiantes hagan trabajo individual.
The instructors insist that the students do individual work.
108. Las niñas se mojaron y ensuciaron en el navajo después de la lluvia.
The girls got wet and dirty in the puddle after the rain.
109. Todos los alumnos sacaron una hoja de papel para la prueba.
All the students took out a sheet of paper for the exam.
110. Al bebé le encantaba jugar con su sonaja por horas sin fin.
The baby loved to play with her/his rattle for hours without end.
312
APPENDIX D
TASK 4: PICTURE DESCRIPTION
D.1. Task 4 instructions
D.1.1. Spanish version
En esta sección del estudio, Ud. va a ver una serie de imágenes. Solamente va a
tener quince (15) segundos por imagen para:
1
dar el nombre de la imagen
2
dar cualquier otro sinónimo/otra palabra que pueda describir la imagen
3
describir los atributos físicos y/o emocionales de la imagen
4
describir qué está haciendo la gente
5
explicar quién(es) suele(n) usarla (si se aplica)
6
explicar para qué sirve o para qué se usa (si se aplica)
7
explicar dónde se encuentra o dónde se usa (si se aplica)
8
cualquier otro aspecto que le parezca relevante o importante si hay tiempo
Después de los quince segundos, el programa va directamente a la próxima imagen. Por
eso, se necesita su primera reacción.
En esta sección, tiene la libertad de pedir un descanso en cualquier momento del
estudio. El investigador puede pausar el programa, y comenzar de nuevo cuando Ud.
esté listo/a. Hay 50 imágenes en total. Favor de indicarle al investigador cuando esté
listo/a para empezar.
D.1.2. English translation
In this section of the study, you will see a series of images. You will only have
fifteen (15) seconds per image to:
313
1
provide the name of the image
2
provide any synonym/other word to describe the image
3
describe the image’s physical and/or emotional attributes
4
describe what the people are doing
5
explain who tends to use it (if applicable)
6
explain for what it is used (if applicable)
7
explain where it is found or used (if applicable)
8
any other aspect that might seem relevant to you if there is time
After the fifteen seconds, the program advances directly to the next image. Because of
this, your immediate reaction is necessary.
In this section you have the freedom to take a break at any point during the study.
The researcher can pause the program and begin again when you are ready. There are 50
images in total. Please indicate to the researcher when you are ready to begin.
D.2. Task 4 images (non-randomized) with target word
and corresponding translation
In the following pages, the 51 total images (50 unique target words per country)
are provided in the following order (note that participants viewed all images in color and
in randomized order):
/f/ assimilating
/x/ assimilating
/f/ progressive assimilation
/x/ progressive assimilation
/f/ non-assimilating
/x/ non-assimilating
314
/f/ assimilating
fútbol ‘soccer’
perfume ‘perfume’
fumar ‘to smoke’
confundir ‘to confuse’
fuego ‘fire’
foto ‘photo’
fotocopiar ‘to photocopy’
alfombra ‘rug’
315
/x/ assimilating
teléfono ‘telephone’
giba ‘hump’
jirafa ‘giraffe’
mejilla ‘cheek’
página ‘page’
región (Chile) ‘region’
región (Mexico) ‘region’
gema ‘gem’
316
gemelos ‘twins’
mujer ‘woman’
/f/ progressive assimilation
imagen ‘image’
bufanda ‘scarf’
estufa ‘stove’
sofá ‘couch’
/x/ progressive assimilation
profesor ‘professor’
frijoles ‘beans’
317
hijo ‘son’
oreja ‘ear’
/f/ non-assimilating
conejo ‘rabbit’
fila ‘line’ (queue)
figura ‘figure’
edificio ‘building’
científico ‘scientist’
feria ‘fair, market’
318
feliz ‘happy’
enfermo ‘sick, ill’
cafetería ‘cafeteria’
falda ‘skirt’
familia ‘family’
elefante ‘elephant’
/x/ non-assimilating
gafas ‘glasses’
jugo ‘juice’
319
juguete ‘toy’
juez ‘judge’
brújula ‘compass’
joya ‘jewel’
joroba ‘hump’
callejón ‘alley’
ojo ‘eye’
jaula ‘cage’
320
jabón ‘soap’
pijamas ‘pajamas’
pájaro ‘bird’
321
REFERENCES
Álvarez-Santullano Busch, M. Pilar. 1992. Variedad interna y deterioro del dialecto
huilliche [Internal variety and the deterioration of the Huilliche dialect]. Revista
de lingüística teórica y aplicada [Journal of Theoretical and Applied Linguistics]
30, 61–74.
Anttila, Arto. 2006. Variation and opacity. Natural Language and Linguistic Theory 24
(4), 893–944.
Anttila, Arto and Curtis Andrus. 2006. T-orders. Manuscript, Stanford University,
Stanford, California. Rutgers Optimality Archive 873.
Anttila, Arto and Young-mee Yu Cho. 1998. Variation and change in Optimality Theory.
Lingua 104, 31–56.
Backley, Phillip. 2011. An Introduction to Element Theory. Edinburgh, United Kingdom:
Edinburgh University Press.
Baker, Adam, Diana Archangeli and Jeff Mielke. 2011. Variability in American English
s-retraction suggests a solution to the actuation problem. Language Variation and
Change 23, 347–374.
Bamgbos̩ e, Ayo̩ . 1966. A Grammar of Yoruba. New York, New York: Cambridge
University Press.
Bassetti, Benedetta. 2006. Orthographic input and phonological representations in
learners of Chinese as a foreign language. Written Language and Literacy 9 (1),
95–114.
Bateman, Nicoleta. 2007. A Crosslinguistic Investigation of Palatalization. Ph.D.
dissertation, University of California, San Diego, San Diego, California.
Bayley, Robert. 2002. The quantitative paradigm. In J. K. Chambers, P. Trudgill and N.
Schilling-Estes (eds.), The Handbook of Language Variation and Change, 117–
141. Malden, Massachusetts: Blackwell.
Beckman, Jill. 1998. Positional Faithfulness: An Optimality Theoretic Treatment of
Phonological Asymmetries. Ph.D. dissertation, University of Massachusetts,
Amherst, Amherst, Massachusetts.
Beckman, Jill. 1999. Labial opacity as labial attraction. Department of Linguistics
colloquium, University of Iowa, Iowa City, Iowa, 18 November.
322
Beckman, Jill. 2000. Partial copying and emergent unmarkedness in Igbo reduplication.
In K. Megerdoomian and L. A. Bar-el (eds.), WCCFL 20: Proceedings of the 20th
West Coast Conference on Formal Linguistics, 68–81. Somerville, Massachusetts:
Cascadilla Press.
Beckman, Jill, Michael Jessen and Catherine Ringen. 2013. Empirical evidence for
laryngeal features: aspirating vs. true voice languages. Journal of Linguistics 49
(2), 259–284.
Beckman, Jill, Pétur Helgason, Bob McMurray and Catherine Ringen. 2011. Rate effects
on Swedish VOT: evidence for phonological overspecification. Journal of
Phonetics 39 (1), 39–49.
Beddor, Patrice Speeter. 2012. A coarticulatory path to sound change. Language 85 (4),
785–821.
Bernal, Manuel. “Por qué me quité del vicio” [“Why I gave up my bad habits”].
Blake, Robert. 1988. Ffaro, Faro, or Haro?: F doubling as a source of linguistic
information for the Early Middle Ages. Romance Philology 41 (3), 267–289.
Blevins, Juliette. 2004. Evolutionary Phonology: The Emergence of Sound Patterns. New
York, New York: Cambridge University Press.
Boersma, Paul and Bruce Hayes. 2001. Empirical tests of the Gradual Learning
Algorithm. Linguistic Inquiry 32, 45–86.
Boersma, Paul and David Weenink. 2013. Praat: doing phonetics by computer [computer
program]. Version 5.3.41, retrieved 23 February 2013 <http://www.praat.org/>.
Boyd-Bowman, Peter. 1954. From Latin to Romance in Sound Charts. Washington, D.C.:
Georgetown University Press.
Bradley, Travis. 2014. Optimality Theory and Spanish phonology. Language and
Linguistics Compass 8 (2), 65–88.
Bradley, Travis. forthcoming. Labialization and palatalization in Judeo-Spanish
phonology. Selected Proceedings of the 42nd Linguistic Symposium on Romance
Languages. Philadelphia, Pennsylvania: John Benjamins.
Brody, Michal. 2004. The Fixed Word, the Moving Tongue: Variation in Written Yucatec
Maya and the Meandering Evolution Toward Unified Norms. Ph.D. dissertation,
University of Texas, Austin, Texas.
Browman, Catherine P. and Louis Goldstein. 1986. Towards an articulatory phonology.
Phonology Yearbook 3, 219–252.
323
Browman, Catherine P. and Louis Goldstein. 1989. Articulatory gestures as phonological
units. Phonology 6, 201–251.
Browman, Catherine P. and Louis Goldstein. 1990. Tiers in Articulatory Phonology, with
some implications for casual speech. In J. Kingston and M. Beckman (eds.),
Papers in Laboratory Phonology I: Between the Grammar and the Physics of
Speech, 341–397. Cambridge, United Kingdom: Cambridge University Press.
Browman, Catherine P. and Louis Goldstein. 1991. Gestural structures: distinctiveness,
phonological processes, and historical change. In I.G. Mattingly and M. StuddertKennedy (eds.), Modularity and the Motor Theory of Speech Perception, 313–
338. Hillsdale, New Jersey: Lawrence Erlbaum Associates.
Browman, Catherine P. and Louis Goldstein. 1992. Articulatory Phonology: an overview.
Phonetica 49, 155–180.
Browman, Catherine P. and Louis Goldstein. 1995. Gestural syllable position effects in
American English. In F. Bell-Berti and L. Raphael (eds.), Producing Speech:
Contemporary Issues for Katherine Safford Harris, 19–33. New York, New York:
American Institute of Physics.
Bybee, Joan. 2010. Language, Usage and Cognition. New York, New York: Cambridge
University Press.
Byrd, Dani. 1994. Articulatory Timing in English Consonant Sequences. Ph.D.
dissertation, University of California, Los Angeles, Los Angeles, California.
Byrd, Dani. 1996. Influences on articulatory timing in consonant sequences. Journal of
Phonetics 24, 209–244.
Byrne, Brian J. 1998. The Foundation of Literacy: The Child’s Acquisition of the
Alphabetic Principle. Hove, East Sussex, England: Psychology Press.
Calvo Shadid, Annette. 1996. Variación del fonema /f/ en dos sociolectos costarricenses
[/f/ phonemic variation in two Costa Rican sociolects]. Filología y Lingüística
[Philology and Linguistics] 22 (1), 119–127.
Campbell, Lyle. 1997. American Indian Languages: The Historical Linguistics of Native
America. New York, New York: Oxford University Press.
Canfield, D. Lincoln. 1981. Spanish Pronunciation in the Americas. Chicago, Illinois:
University of Chicago Press.
324
Cardoso, Walcir. 2005. An integrated approach to variation in OT: evidence from
Brazilian Portuguese and Picard. In T. Geerts, I. van Ginneken, and H. Jacobs
(eds.), Romance Languages and Linguistic Theory 2003: Selected Papers from
‘Going Romance’ 2003, Nijmegen, 20-22 November, 1–14. Philadelphia,
Pennsylvania: John Benjamins.
Carroll, Julia M., Margaret J. Snowling, Jim Stevenson and Charles Hulme. 2003. The
development of phonological awareness in preschool children. Developmental
Psychology 39 (5), 913–923.
Cerrón-Palomino, Rodolfo. 1987. Lingüística quechua [Quechuan linguistics]. Cuzco,
Peru: Centro de Estudios Rurales Andinos Bartolomé de Las Casas [Bartolomé de
Las Casas Center for Rural Andean Studies].
Cho, Hyeon-Kwan. 1997. A Phonological Analysis of Consonantal Assimilations and
Consonant-Vowel Interactions. Ph.D. dissertation, University of Minnesota,
Minneapolis, Minnesota.
Cho, Taehong and Patricia Keating. 2009. Effects of initial position versus prominence in
English. Journal of Phonetics 37, 466–485.
Chomsky, Noam and Morris Halle. 1968. The Sound Pattern of English. New York, New
York: Harper and Row.
Clark, Mary. 1990. The Tonal System of Igbo. Providence, Rhode Island: Foris
Publications.
Clements, G. Nicholas. 1985. The geometry of phonological features. Phonology
Yearbook 2, 225–252.
Clements, G. Nicholas. 1989. A unified set of features for consonants and vowels.
Manuscript, Cornell University, Ithaca, New York.
Clements, G. Nicholas. 1990. The role of the sonority cycle in core syllabification. In J.
Kingston and M. Beckman (eds.), Papers in Laboratory Phonology I, 283–333.
Cambridge, Massachusetts: Cambridge University Press.
Clements, G. Nicholas and Elizabeth V. Hume. 1995. The internal organization of speech
sounds. In J. Goldsmith (ed.), The Handbook of Phonological Theory, 245–306.
Oxford, United Kingdom: Blackwell.
Coetzee, Andries W. 2009. An integrated grammatical/non-grammatical model of
phonological variation. In Y.-S. Kang, J.-Y. Yoon, H. Yo, S.-W. Tang, Y.-S.
Kang, Y. Jang, C. Kim, K.-A. Kim and H.-K. Kang (eds.), Current Issues in
Linguistic Interfaces, volume 2, 267–294. Seoul, South Korea:
Hankookmunhwasa.
325
Coetzee, Andries W. and Shigeto Kawahara. 2013. Frequency biases in phonological
variation. Natural Language and Linguistic Theory 31 (1), 47–89.
Coleman, John. 1992. The phonetic interpretation of headed phonological structures
containing overlapping constituents. Phonology 9, 1–44.
Coleman, John. 2001. The phonetics and phonology of Tashlhiyt Berber syllabic
consonants. Transactions of the Philological Society 99 (1), 29-64.
Cowan, Hendrik Karel Jan. 1965. Grammar of the Sentani Language. The Hague,
Netherlands: Martinus Nijhoff.
Davidson, Lisa. 2003. Articulatory and perceptual influences on the production of nonnative clusters. In M. J. Solé, D. Recasens and J. Romero (ed.), Proceedings of the
15th International Congress of Phonetic Sciences, 3–9. Barcelona, Spain.
Du Feu, Veronica. 1996. Rapanui. New York, New York: Routledge.
Deng, Li and Douglas O’Shaughnessy. 2003. Speech Processing: A Dynamic and
Optimization-Oriented Approach. New York, New York: Marcel Dekker, Inc.
Dolphyne, Florence Abena. 1996. A Comprehensive Course in Twi (Asante) for the NonTwi Learner. Accra, Ghana: Ghana University Press.
Duthie, Andrew. 1996. Introducing Ewe Linguistic Patterns: A Textbook of Phonology,
Grammar, and Semantics. Accra, Ghana: Ghana University Press.
Elugbe, Ben Ohiomamhe. 1989. Comparative Edoid: Phonology and Lexicon. Port
Harcourt, Nigeria: University of Port Harcourt Press.
Feijóo, Sergio and Santiago Fernández. 2003. Auditory identification and acoustic
representation of the voiceless fricatives and affricates. Revista Electrónica de la
Red Temática en Tecnologías del Habla [Online Journal of the Thematic Network
on Speech Technologies] April 2003, 1–12.
Fernández, Santiago, Sergio Feijóo, R. Balsa and N. Barros. 1998. Acoustic cues for the
auditory identification of the Spanish fricative /f/. In R. H. Mannell and J. RobertRibes (eds.), Proceedings of the International Conference on Spoken Language
Processing, volume 5, 2151–2154. Sydney, Australia: Australian Speech Science
and Technology Association, Inc.
Flemming, Edward. 1995. Auditory Representations in Phonology. Ph.D. dissertation,
University of California, Los Angeles, Los Angeles, California.
326
Fougeron, Cécile and Patricia A. Keating. 1997. Articulatory strengthening at edges of
prosodic domains. Journal of the Acoustical Society of America 101 (6), 3728–
3740.
Foulkes, Paul, James M. Scobbie and Dominic Watt. 2010. Sociophonetics. In W. J.
Hardcastle, J. Laver and F. E. Gibbon (eds.), The Handbook of Phonetic Sciences,
2nd ed., 703–754. Malden, Massachusetts: Blackwell.
Fowler, Carol A. 1980. Coarticulation and theories of extrinsic timing. Journal of
Phonetics 8, 113–133.
Fowler, Carol A. 1983. Converging sources of evidence on spoken and perceived
rhythms of speech: cyclic production of vowels in monosyllabic stress feet.
Journal of Experimental Psychology: General 112 (3), 386–412.
Frajzyngier, Zygmunt, Eric Johnston and Adrian Edwards. 2005. A Grammar of Mina.
Berlin, Germany: Mouton de Gruyter.
Fujimura, Osamu. 1979. An analysis of English syllables as cores and affixes. Zeitshrift
für Phonetik, Sprachwissenschaft und Kommunicationsforschung 32, 471–476.
Fujimura, Osamu. 1988. Some remarks on consonant clusters. Annual Bulletin: Research
Institute of Logopedics and Phoniatrics 22, 59–64.
Fujimura, Osamu. 1990. Demisyllables as sets of features: comments on Clements’
paper. In J. Kingston and M. Beckman (eds.), Papers in Laboratory Phonology I:
Between the Grammar and the Physics of Speech, 334–340. Cambridge,
Massachusetts: Cambridge University Press.
Gafos, Adamantios. 1999. The Articulatory Basis of Locality in Phonology. New York,
New York: Garland.
Gafos, Adamantios. 2002. A Grammar of Gestural Coordination. Natural Language and
Linguistic Theory 20 (2), 269–337.
García-Bellido, Paloma. 1999. Local movement and the syntax of Spanish sounds. In R.
Folli and R. Middleton (eds.), Oxford University Working Papers in Linguistics
Philology and Phonetics 4.
Garrett, Andrew and Keith Johnson. 2013. Phonetic bias in sound change. In A. C. L. Yu
(ed.), Origins of Sound Change: Approaches to Phonologization, 51–97. New
York, New York: Oxford University Press.
Gerdel, Florence. 1974 [1955]. Tzeltal (Maya) Phonemes. Summer Institute of Linguistics
Mexico Workpapers, volume 1.
327
Ghavami, Golnaz Modarresi. 2002. The Effects of Syllable Boundary, Stop Consonant
Closure Duration, and VOT on VCV Coarticulation. Ph.D. dissertation,
University of Texas at Austin, Austin, Texas.
Göksel, Aslı. 2005. Turkish: A Comprehensive Grammar. New York, New York:
Routledge.
Goldsmith, John. 1976. Autosegmental Phonology. Ph.D. Dissertation, Massachusetts
Institute of Technology, Cambridge, Massachusetts.
Gordon, Matthew, Paul Barthmaier and Kathy Sands. 2002. A cross-linguistic acoustic
study of voiceless fricatives. Journal of the International Phonetic Association 32,
141–174.
Guthrie, Malcolm. 1967–1971. Comparative Bantu (4 volumes). Gregg International
Publishers.
Hall, Nancy. 2006. Cross-linguistic patterns of vowel intrusion. Phonology 23, 387–429.
Halle, Morris. 1992. Phonological features. In W. Bright (ed.), International
Encyclopedia of Linguistics, 207–212. Oxford, United Kingdom: Oxford
University Press.
Hardman, Martha James. 2001. Aymara. Munich, Germany: Lincom Europa.
Harris, John. 1994. English Sound Structure. Cambridge, Massachusetts: Blackwell.
Harris, John and Geoff Lindsey. 1995. The elements of phonological representation. In J.
Durand and F. Katamba (eds.), Frontiers of Phonology: Atoms, Structures,
Derivations, 34–79. Harlow, Essex, United Kingdom: Longman.
Harris, Katherine Safford. 1958. Cues for the discrimination of American English
fricatives in spoken syllables. Language and Speech 1 (1), 1–7.
Haspelmath, Martin. 1993. A Grammar of Lezgian. Berlin, Germany: Mouton de Gruyter
Herrick, Dylan. 1999. Catalan cluster simplification and nasal place assimilation.
Linguistics Research Center: Phonology at Santa Cruz 6, 25–37.
Hillenbrand, James M., Michael J. Clark and Terrance M. Nearey. 2001. Effects of
consonantal environment on vowel formant patterns. Journal of the Acoustical
Society of America 109 (2), 748–763.
Hirayama, Manami. 2005. Place asymmetry and markedness of labials in Japanese.
Toronto Working Papers in Linguistics 24, 125–168.
328
Holt, D. Eric. 1997. The Role of the Listener in the Historical Phonology of Spanish and
Portuguese: An Optimality-Theoretic Account. Ph.D. dissertation, Georgetown
University, Washington, D.C.
Hualde, José Ignacio. 1997. Spanish /i/ and related sounds: an exercise in phonemic
analysis. Studies in the Linguistics Sciences 27 (2), 61–79.
Hualde, José Ignacio. 2004. Quasi-phonemic contrasts in Spanish. In V. Chand, A.
Kelleher, A. Rodríguez and B. Schmeiser (eds.), WCCFL 23: Proceedings of the
23rd West Coast Conference on Formal Linguistics, 374–398. Somerville,
Massachusetts: Cascadilla Press.
Hualde, José Ignacio. 2005. The Sounds of Spanish. New York, New York: Cambridge
University Press.
Hualde, José Ignacio, Miquel Simonet and Francisco Torreira. 2008. Postlexical
contraction of non-high vowels in Spanish. Lingua 118, 1906–1925.
Hudak, Thomas John. 1990. Thai. In B. Comrie (ed.), The World’s Major Languages,
757–775. New York, New York: Oxford University Press.
Hulme, Charles, Peter J. Hatcher, Kate Nation, Angela Brown, John Adams and George
Stuart. 2002. Phoneme awareness is a better predictor of early reading skill than
onset-rime awareness. Journal of Experimental Child Psychology 82 (1), 2–28.
Hume, Elizabeth. 1992. Vowels, Coronal Consonants and their Interaction in Non-linear
Phonology. Ph.D. dissertation, Cornell University, Ithaca, New York.
Huskey, Liberty Celeste. 2010. Velar palatalization: a cross-dialectic study of Chilean,
Mexican, and Castilian Spanish. Manuscript, University of Arizona, Tucson,
Arizona. <http://arizona.openrepository.com/arizona/bitstream/10150/146905/1/
azu_etd_mr20100196_sip1_m.pdf>.
Ikekeonwu, Clara. 1999. Igbo. In Handbook of the International Phonetic Association: A
Guide to the Use of the International Phonetic Alphabet, 108–110. New York,
New York: Cambridge University Press.
Instituto Nacional de Estadística y Geografía de México [National Institute of Statistics
and Geography of Mexico]. 2005. <http://www.inegi.org.mx/>.
Iribarren-Argaiz, Mary Carmen. 1998. Influencia del vascuence en los cambios
fonológicos del castellano: un caso de lenguas en contacto [Basque influence on
the phonological changes of Castilian: a case of language contact]. Romance
Quarterly 45 (1), 3–34.
329
Itô, Junko and Armin Mester. 1986. The phonology of voicing in Japanese: theoretical
consequences for morphological accessibility. Linguistic Inquiry 17, 45–73.
Itô, Junko and Armin Mester. 1993. Licensed segments and safe paths. Canadian Journal
of Linguistics/Revue canadienne de linguistique 38, 197–213.
Itô, Junko and Armin Mester. 1999. Realignment. In R. Kager, H. van der Hulst and W.
Zonneveld (eds.), The Prosody Morphology Interface, 188–217. Cambridge,
United Kingdom: Cambridge University Press.
Jakobson, Roman, C. Gunnar M. Fant and Morris Halle. 1952. Preliminaries to Speech
Analysis: The Distinctive Features and Their Correlates. Cambridge,
Massachusetts: MIT Press.
Jassem, Wiktor. 1968. Acoustical description of voiceless fricatives in terms of spectral
parameters. In W. Jassem (ed.), Speech Analysis and Synthesis, volume 1, 189–
206. Warsaw, Poland: Państwowe Wydawnictwo Naukowe.
Johnson, Keith. 2012. Acoustic and Auditory Phonetics, 3rd ed. Malden, Massachusetts:
Wiley-Blackwell.
Jongman, Allard, Ratree Wayland and Serena Wong. 2000. Acoustic characteristics of
English fricatives. Journal of the Acoustical Society of America 108 (3, 1), 1252–
1263.
Jun, Jongho. 1994. A constraint-based analysis of place assimilation typology. Studies in
the Linguistic Sciences 24 (1/2), 263–277.
Jun, Jongho. 1995. Perceptual and Articulatory Factors in Place Assimilation: An
Optimality Theoretic Approach. Ph.D. dissertation, University of California, Los
Angeles, Los Angeles, California.
Ka, Omar. 1994. Wolof Phonology and Morphology. Lanham, Maryland: University
Press of America.
Keating, Patricia, Taehong Cho, Cécile Fougeron and Chai-Shune Hsu. 1997. Domaininitial articulatory strengthening in four languages. UCLA Working Papers in
Phonetics 97, 139–151.
Kim, Nam-Kil. 1990. Korean. In B. Comrie (ed.), The World’s Major Languages, 881–
898. New York, New York: Oxford University Press.
Kiparsky, Paul. 1993. An OT perspective on phonological variation. Paper presented at
Rutgers University Optimality Workshop, Rutgers University, and NWAVE 1994,
Stanford University. <http://www.stanford.edu/~kiparsky/Papers/nwave94.pdf>.
330
Kochetov, Alexei. 2002. Production, Perception, and Emergent Phonotactic Patterns.
New York, New York: Routledge.
Kohler, K. J. 1990. Segmental reduction in connected speech in German: phonological
facts and phonetic explanations. In W. J. Hardcastle and A. Marchal (eds.),
Speech Production and Speech Modeling, 69–92. Dordrecht, The Netherlands:
Kluwer.
Kornfilt, Jaklin. 1990. Turkish and the Turkic languages. In B. Comrie (ed.), The World’s
Major Languages, 619–644. New York, New York: Oxford University Press.
Kozhevnikov, Valeriĭ Aleksandrovich and Liudmila Andreevna Chistovich. 1965.
Speech: Articulation and Perception. Joint Publications Research Service 30, 543.
Washington, D.C.: United States Department of Commerce.
Kropp Dakubu , Mary Esther. 2002. Gã Phonology. Legon, Ghana: Institute of African
Studies, University of Ghana.
Ladefoged, Peter. 2001. Vowels and Consonants: An Introduction to the Sounds of
Language. Malden, Massachusetts: Blackwell.
Ladefoged, Peter. 2006. A Course in Phonetics, 5th ed. Boston, Massachusetts: Thomson
Wadsworth.
Ladefoged, Peter. 1989. Representing phonetic structure. UCLA Working Papers in
Linguistics 73, 1–79. University of California, Los Angeles.
Ladefoged, Peter and Ian Maddieson. 1996. The Sounds of the World’s Languages.
Cambridge, Massachusetts: Blackwell.
Lamb, William. 2001. Scottish Gaelic. Munich, Germany: Lincom Europa.
Launey, Michel. 2011. An Introduction to Classical Nahuatl. New York, New York:
Cambridge University Press.
Leben, William. 1973. Suprasegmental Phonology. Ph.D. dissertation, Massachusetts
Institute of Technology, Cambridge, Massachusetts.
Lee, Hyun Bok. 1993. Korean. Journal of the International Phonetic Association 23 (1),
28–31.
Lee, Joo-Kyeong. 1997. A phonetic examination of C/V place assimilation in OT. In J.
Austin and A. Lawson (eds.), Proceedings of the Fourteenth Eastern States
Conference on Linguistics 1997, 115–126. Cornell University, Ithaca, New York:
CLC Publications.
331
Lee, Joo-Kyeong. 1999. A Phonetic Examination of C-to-V Place Assimilation. Ph.D.
dissertation, University of Illinois at Urbana-Champaign, Urbana, Illinois.
Lefebvre, Claire and Anne-Marie Brousseau. 2002. A Grammar of Fongbe. New York,
New York: Mouton de Gruyter.
Lewis, M. Paul (ed.). 2009. Languages of Chile. Ethnologue: Languages of the World,
16th ed. Dallas, Texas: SIL International.
Lindblom, Björn. 1963. On vowel reduction. Rep. No. 29, The Royal Institute of
Technology, Speech Transmission Laboratory, Stockholm, Sweden.
Linebaugh, Gary Dean. 2007. Phonetic Grounding and Phonology: Vowel Backness
Harmony and Vowel Height Harmony. Ph.D. dissertation, University of Illinois at
Urbana-Champaign, Urbana, Illinois.
Lipski, John. 1994. Latin American Spanish. London, England: Longman.
Lipski, John. 1995. [round] and [labial] in Spanish and the “free-form” syllable.
Linguistics 33, 283–304.
Lipski, John. 2005. A History of Afro-Hispanic Language. New York, New York:
Cambridge University Press.
Lloret, Maria-Rosa. 2007. On the nature of vowel harmony: spreading with a purpose. In
A. Bisetto and F. E. Barbieri (eds.), Proceedings of the XXXIII Incontro di
Grammatica Generativa (Bologna, March 1-3, 2007), 15–35. Bologna, Italy:
Università di Bologna.
Lloyd, Paul M. 1987. From Latin to Spanish: Historical Phonology and Morphology of
the Spanish Language, volume 1. Philadelphia, Pennsylvania: American
Philosophical Society, volume 173.
Lombardi, Linda. 1999. Positional faithfulness and voicing assimilation in Optimality
Theory. Natural Language and Linguistic Theory 17, 267–302.
Macaulay, Monica. 1996. A Grammar of Chalcatongo Mixtec. University of California
Publications in Linguistics, volume 127. Los Angeles, California: University of
California Press.
Manrique, Ana María Borzone de and María Ignacia Massone. 1981. Acoustic analysis
and perception of Spanish fricative consonants. Journal of the Acoustical Society
of America 69 (4), 1145–1153.
Marshall, Chloe and Heather K. J. van der Lely. 2006. A challenge to current models of
past tense inflection: the impact of phonotactics. Cognition 100 (2), 302–320.
332
Martínez Celdrán, Eugenio and Ana M. Fernández Planas. 2007. Manual de fonética
española: articulaciones y sonidos del español [Manual of Spanish Phonetics:
Articulations and Sounds of Spanish]. Barcelona, Spain: Editorial Ariel.
Mazzaro, Natalia. 2005. Aspiration and velarization of /f/ in Argentine Spanish. Toronto
Working Papers in Linguistics 25, 58–67.
Mazzaro, Natalia. 2010. Changing perceptions: the sociophonetic motivations of the
labial velar alternation in Spanish. In M. Ortega-Llebaria (ed.), Selected
Proceedings of the 4th Conference on Laboratory Approaches to Spanish
Phonology, 128–145. Somerville, Massachusetts: Cascadilla Proceedings Project.
Mazzaro, Natalia. 2011. Experimental Approaches to Sound Variation: A Sociophonetic
Study of Labial and Velar Fricatives and Approximants in Argentine Spanish.
Ph.D. dissertation, University of Toronto, Toronto, Canada.
McCarthy, John J. 1986. OCP effects: gemination and antigemination. Linguistic Inquiry
17, 207–263.
McCarthy, John. 1988. Feature geometry and dependency: a review. Phonetica 43, 84–
108.
McCarthy, John and Alan Prince. 1993. Generalized alignment. Manuscript, University
of Massachusetts, Amherst and Rutgers University.
McCarthy, John and Alan Prince. 1993/2001. Prosodic morphology: constraint
interaction and satisfaction. Manuscript, University of Massachusetts, Amherst,
Amherst, Massachusetts, and Rutgers University, New Brunswick, New Jersey.
Technical Report No. 3, Rutgers University Center for Cognitive Science. Rutgers
Optimality Archive 482.
McCarthy, John and Alan Prince. 1995. Faithfulness and reduplicative identity. In J.
Beckman, L. Walsh Dickey and S. Urbanczyk (eds.), University of Massachusetts
Occasional Papers 18, 249–384. Amherst, Massachusetts: GLSA.
McGuire, Grant and Molly Babel. 2012. A cross-modal account for synchronic and
diachronic patterns of /f/ and /θ/ in English. Laboratory Phonology 3 (2), 251–
272.
McMurray, Bob and Allard Jongman. 2011. What information is necessary for speech
categorization? Harnessing variability in the speech signal by integrating cues
computed relative to expectations. Psychological Review 118 (2), 219–246.
Menéndez Pidal, Ramón. 1919. Documentos lingüísticos de España: El Reino de Castilla
[Linguistic Documents: The Kingdom of Castile]. Madrid, Spain: Centro de
Estudios Históricos [Center of Historical Studies].
333
Merrill, Elizabeth D. 2008. Tilquiapan Zapotec. Journal of the International Phonetic
Association 38, 107–114.
Mielke, Jeff. 2008. The Emergence of Distinctive Features. New York, New York:
Oxford University Press.
Mielke, Jeff. 2012. A phonetically-based metric of sound similarity. Lingua 122, 145–
163.
Morais, José, Luz Cary, Jésus Alegria and Paul Bertelson. 1979. Does awareness of
speech as a sequence of phone arise spontaneously? Cognition 7 (4), 323–331.
Morén, Bruce. 2003. The Parallel Structures Model of feature geometry. Working Papers
of the Cornell Phonetics Lab 15, 194–270.
Morén, Bruce. 2006. Consonant-vowel interactions in Serbian: features, representations
and constraint interactions. Lingua 116 (8), 1198–1244.
Moreno Fraginals, Manuel. 1984. Cultural contributions and deculturation. In M. Moreno
Fraginals (ed.), Africa in Latin America: Essays on History, Culture, and
Socialization, 5–22. L. Blum (trans.). New York, New York: Holmes and Meier
Publishers.
Munro, Pamela. 2005. Zapotec grammar without tears (except perhaps for the
grammarian). In R. Beam de Azcona and M. Paster (eds.), Conference on
Otomanguean and Oaxacan Languages, March 19-21, 2004, University of
California at Berkeley, 87–106.
Muter, Valerie, Charles Hulme, Margaret Snowling and Sara Taylor. 1998.
Segmentation, not rhyming, predicts early progress in learning to read. Journal of
Experimental Child Psychology 71 (1), 3–27.
Naro, Anthony J. 1972. On ‘f > h’ in Castilian and Western Romance. Zeitschrift für
Romanische Philologie 88 (4/6), 435–447.
Nartey, Jonas. 1982. On fricative phones and phonemes: measuring the phonetic
difference within and between languages. UCLA Working Papers in Phonetics 55.
Nawrocki, Grzegorz. 2008. Laryngeal articulations of /x/ in Southern Polish. ZAS Papers
in Linguistics 49, 145–165.
Nespor, Marina and Irene Vogel. 1986. Prosodic Phonology. Dordrecht, The
Netherlands: Foris Publications.
334
Ní Chiosáin, Marie and Jaye Padgett. 1993. Inherent VPlace. Manuscript, LRC-93-09,
Linguistics Research Center, Department of Linguistics, University of California,
Santa Cruz.
Nirgianaki, Elina. 2011. Formant correlates in Greek fricative-vowel sequences.
Proceedings of the 17th International Congress of Phonetic Sciences (ICPhS
XVII), 1498–1501.
Nirgianaki, Elina, Anthi Chaida and Marios Fourakis. 2010. Acoustic structure of
fricative consonants in Greek. In Proceedings of the 3rd ISCA Workshop, 125–
128.
Nittrouer, Susan and Joanna H. Lowenstien. 2009. Does harmonicity explain children’s
cue weighting of fricative-vowel syllables? Journal of the Acoustical Society of
America 125 (3), 1679–1692.
Norman, Geoff. 2010. Likert scales, levels of measurement and the “laws” of statistics.
Advances in Health Sciences Education: Theory and Practice 15 (5), 625–632.
North, Joanne and Jäna Shields. 1977. Silacayoapan Mixtec phonology. In W. R.
Merrifield (ed.), Studies in Otomanguean Phonology, 21–35. Dallas, Texas: The
Summer Institute of Linguistics and The University of Texas at Arlington.
Núñez Cedeño, Rafael A. and Alfonso Morales-Front. 1999. Fonología generativa
contemporánea de la lengua española [Contemporary Generative Phonology of
the Spanish Language]. Washington, D.C.: Georgetown University Press.
Ohala, John J. 1981. The listener as a source of sound change. In C. S. Masek, R. A.
Hendrick and M. F. Miller (eds.), Papers from the Parasession on Language and
Behavior, 178–203. Chicago, Illinois: Chicago Linguistics Society.
Ohala, John J. 1989. Sound change is drawn from a pool of synchronic variation. In L.
Breivik and E. Jahr (eds.), Language Change: Contribution to the Study of its
Causes, 173–198. Berlin, Germany: Mouton de Gruyter.
Ohala, John J. 1993. The phonetics of sound change. In C. Jones (ed.), Historical
Linguistics: Problems and Perspective, 237–278. London, United Kingdom:
Longman.
Ohala, John J. 2010. The relation between phonetics and phonology. In W. J. Hardcastle,
J. Laver and F. E. Gibbon (eds.), The Handbook of Phonetic Sciences, 2nd ed.,
653–677. Malden, Massachusetts: Blackwell.
Ohala, John J. and Manjari Ohala. 1993. The phonetics of nasal phonology: theorems and
data. In M. Huffman and R. Krakow (eds.), Phonetics and Phonology, 225–249.
San Diego, California: Academic Press.
335
Öhman, Sven E. G. 1966. Coarticulation in VCV utterance: spectrographic
measurements. Journal of the Acoustical Society of America 39 (1), 151-168.
Otero, Carlos. 1971. Evolución y revolución en romance [Evolution and Revolution in
Romance]. Barcelona, Spain: Seix Barral.
Padgett, Jaye. 1995. Feature classes. In J. Beckman, L. Walsh Dickey and S. Urbanczyk
(eds.), University of Massachusetts Occasional Papers in Linguistics 18: Papers
in Optimality Theory, 385–419. Amherst, Massachusetts: GLSA.
Padgett, Jaye. 1996. Partial class behavior and nasal place assimilation. In K. Suzuki and
D. Elzinga (eds.), Proceedings of the 1995 Southwestern Workshop on Optimality
Theory, Tucson, Arizona: University of Arizona. Rutgers Optimality Archive 113.
Padgett, Jaye and Marzena Żygis. 2007. A perceptual study of Polish fricatives, and its
relation to historical sound change. Manuscript, University of California, Santa
Cruz.
Pater, Joe. 1999. Austronesian nasal substitution and other NC effects. In R. Kager, H.
van der Hulst and W. Zonneveld (eds.), The Prosody-Morphology Interface, 310–
343. Cambridge, United Kingdom: Cambridge University Press.
Penny, Ralph. 2002. A History of the Spanish Language. Cambridge, United Kingdom:
Cambridge University Press.
Perissinotto, Giorgio. 1975. Fonología del español hablado en la ciudad de México
[Phonology of the Spanish spoken in Mexico City]. Guanajuato, Mexico: El
colegio de Mexico.
Perkell, Joseph S. 1969. Physiology of Speech Production: Results and Implications of a
Quantitative Cineradiographic Study. Cambridge, Massachusetts: MIT Press.
Petrova, Olga, Rosemary Plapp, Catherine Ringen and Szilárd Szentgyörgyi. 2006. Voice
and aspiration: evidence from Russian, Hungarian, German, Swedish, and
Turkish. The Linguistic Review 23, 1–35.
Pharies, David A. 2007. A Brief History of the Spanish Language. Chicago, Illinois:
University of Chicago Press.
Piñeros, Carlos-Eduardo. 2002. On the interplay between consonant alignment and
feature faithfulness in a Caribbean Spanish dialect. Manuscript, Department of
Spanish and Portuguese, University of Iowa.
Piñeros, Carlos-Eduardo. 2007. The phonology of nasal consonants in five Spanish
dialects. In F. Martínez-Gil and S. Colina (eds.), Optimality Theoretic Studies in
Spanish Phonology, 146–171. Philadelphia, Pennsylvania: John Benjamins.
336
Piñeros, Carlos-Eduardo. 2009. Estructura de los sonidos del español [Structure of the
Sounds of Spanish]. Upper Saddle River, New Jersey: Pearson Prentice Hall.
Prince, Alan and Paul Smolensky. 1993/2004. Optimality Theory: constraint interaction
in generative grammar. Manuscript, Rutgers Center for Cognitive Science,
Rutgers University, New Brunswick, New Jersey, and Computer Science
Department, University of Colorado, Boulder, Boulder, Colorado. Technical
Report No. 2, Rutgers University Center for Cognitive Science. Rutgers
Optimality Archive 537.
Reighard, John. 1972. Labiality and velarity in consonants and vowels. In P. Peranteau, J.
N. Levi and G. C. Phares (eds.), Papers from the 8th Annual Regional Meeting of
the Chicago Linguistic Society, 533–543. Chicago, Illinois: Chicago Linguistic
Society.
Resnick, Melvyn C. 1975. Phonological Variants and Dialect Identification in Latin
American Spanish. The Hague, Netherlands: Mouton de Gruyter.
Rice, Keren and Peter Avery. 1991. Laterality and coronality. In C. Paradis and J.-F.
Prunet (eds.), Phonetics and Phonology: The Special Status of Coronals, 101–
124. San Diego, California: Academic Press.
Rodríguez Champi, Albino. 2006. Quechua de Cusco [Cuzco Quechua]. In S. A. Marlett
(ed.), Ilustraciones fonéticas de lenguas ameríndias [Phonetic illustrations of the
Amerindian languages]. Lima, Peru: The Summer Institute of Linguistics
International and Universidad Ricardo Palma.
Ronquest, Rebecca. 2012. An Acoustic Analysis of Heritage Spanish Vowels. Ph.D.
dissertation, Indiana University, Bloomington, Indiana.
Rosner, Burton S. and John B. Pickering. 1994. Vowel Perception and Production.
Oxford, United Kingdom: Oxford University Press.
Rout, Leslie B., Jr. 1976. The African Experience in Spanish America: 1502 to the
Present Day. New York, New York: Cambridge University Press.
Rubach, Jerzy. 1994. Affricates as strident stops in Polish. Linguistic Inquiry 25 (1), 119–
143.
Rubach, Jerzy. 2008. Prevocalic faithfulness. Phonology 25 (3), 433–468.
Rubach, Jerzy. 2014. Soft labial conspiracy in Kurpian. Journal of Linguistics 50, 185–
230.
337
Sadwosky, Scott and Gastón Salamanca. 2011. El inventario fonético del español de
Chile: principios orientadores, inventario provisorio de consonantes y sistema de
representación (AFI-CL) [The phonetic inventory of chilean spanish: guiding
principles, provisional inventory of consonants and system of representation
(IPA-CL)]. Onomázein 24 (2), 61–84.
Sagey, Elizabeth. 1986. The timing of contour segments. MIT Working Papers in
Linguistics 8, Department of Linguistics Massachusetts Institute of Technology,
208–220.
Sagey, Elizabeth. 1990. The Representation of Features in Non-linear Phonology: The
Articulator Node Hierarchy. New York, New York: Garland.
Sanicky, Cristina A. 1988. El comportamiento de /f/ en el habla misionera [The behavior
of /f/ in the speech of Misiones]. Bulletin of Hispanic Studies 65 (3), 273–278.
Scatton, Ernest A. 1984. A Reference Grammar of Modern Bulgarian. Columbus, Ohio:
Slavica Publishers.
Schwegler, Armin. 2006. Bantu elements in Palenque (Colombia): anthropological,
archaeological, and linguistic evidence. In J. B. Haviser and K. C. MacDonald
(eds.), African Regenesis: Confronting Social Issues in the Diaspora, 204–222.
London, England: University College London Press.
Sebastián-Gallés, Núria, M. Antònia Martí, Manuel Carreiras and Fernando Cuetos.
2000. LEXESP: léxico informatizado del español [LEXESP: Computerized
Lexicon of Spanish]. Barcelona, Spain: Universitat de Barcelona.
Selkirk, Elisabeth. 1986. On derived domains in sentence phonology. Phonology
Yearbook 3, 371–405.
Silbert, Noah and Kenneth de Jong. 2008. Focus, prosodic context, and phonological
feature specification: patterns of variation in fricative production. Journal of the
Acoustical Society of America 123 (5), 2769–2779.
Smyth, David. 2002. Thai: An Essential Grammar. New York, New York: Routledge.
Soli, Sigfrid D. 1981. Second formants in fricatives: acoustic consequences of fricativevowel coarticulation. Journal of the Acoustical Society of America 70, 976–984.
Steriade, Donca. 1993. Positional neutralization and the expression of contrast.
Manuscript, University of California, Los Angeles, Los Angeles, California.
Stevens, Kenneth. 1985. Evidence for the role of acoustic boundaries in the perception of
speech sounds. In V. A. Fromkin (ed.), Phonetic Linguistics: Essays in Honor of
Peter Ladefoged, 243–255. New York, New York: Academic Press.
338
Stevens, Kenneth N. 1989. On the quantal nature of speech. Journal of Phonetics 17, 3–
46.
Stevens, Kenneth N. and Arthur S. House. 1963. Perturbation of vowel articulations by
consonantal context: an acoustical study. Journal of Speech and Hearing
Research 6, 111–128.
Strange, Winifred, Robert R. Verbrugge, Donald P. Shankweiler and Thomas R. Edman.
1976. Consonant environment specifies vowel identity. Journal of the Acoustical
Society of America 60, 213–224.
Sussman, Harvey M., Helen A. McCaffrey and Sandra A. Matthews. 1991. An
investigation of locus equations as a source of relational invariance for stop place
categorization. Journal of the Acoustical Society of America 90, 1309–1325.
Sussman, Harvey M. and Jadine Shore. 1996. Locus equations as phonetic descriptors of
consonantal place of articulation. Perception and Psychophysics 58, 936–946.
Tateishi, Koichi. 1990. A morphological analysis of Mende consonant mutation. In J. P.
Hutchison and V. Manfredi (eds.), Current Approaches to African Linguistics,
volume 7, 99–110. Providence, Rhode Island: Foris Publications.
Thomas, Erik R. 2013. Sociophonetics. In J. K. Chambers, P. Trudgill and N. SchillingEstes (eds.), The Handbook of Language Variation and Change, 2nd ed., 108–
127. Malden, Massachusetts: John Wiley and Sons, Inc.
Thornton, John K. 1992. Africa and Africans in the Making of the Atlantic World, 14001800. New York, New York: Cambridge University Press.
Tingsabadh, M. R. Kalaya and Arthur S. Abramson. 1993. Thai. Journal of the
International Phonetic Association 23 (1), 25–28.
Tiran cadáver muerto ya fallecido de un dijunto [Dead body of a previously deceased
victim thrown]. 9 October 2012. La Extra Grupo Diario de Morelia [The Extra
Daily Group of Morelia]. Accessed 15 April 2013. <http://laextra.mx/tirancadaver-muerto-ya-fallecido-de-un-dijunto/>.
Torreblanca, Máximo. 1984. La “f” prerromana y la vasca en su relación con el español
antiguo [The pre-Roman and Basque “f” and their relation with Old Spanish].
Romance Philology 38 (3), 273–281.
Torrence, William Harold. 2005. On the Distribution of Complementizers in Wolof. Ph.D.
dissertation, University of California, Los Angeles, Los Angeles, California.
Treiman, Rebecca and Derrick C. Bourassa. 2000. The development of spelling skill.
Topics in Language Disorders 20, 1–18.
339
Tuttle, Edward F. 1975. The development of PL, BL, and FL in Italo-Romance:
distinctive features and geolinguistic patterns. Revue de linguistique romane 39
(155/156), 400–431.
Utman, Jennifer Aydelott and Sheila E. Blumstein. 1994. The influence of language on
the acoustic properties of phonetic features: a study of the feature [strident] in
Ewe and English. Phonetica 51, 221–238.
Wagner, Richard K., Joseph K. Torgesen and Carol A. Rashotte. 1994. Development of
reading-related phonological processing abilities: new evidence of bidirectional
causality from a latent variable longitudinal study. Developmental Psychology 30,
73–87.
Walker, Rachel. 2005. Weak triggers in vowel harmony. Natural Language and
Linguistic Theory 23, 917–989.
Weinreich, Uriel, William Labov and Marvin Herzog. 1968. Empirical foundations for a
theory of language change. In W. Lehmann and Y. Malkiel (eds.), Directions for
Historical Linguistics, 95–195. Austin, Texas: University of Texas Press.
Welmers, William. 1973. African Language Structures. Berkeley, California: University
of California Press.
Wright, Richard. 2001. Perceptual cues in contrast maintenance. In E. Hume and K.
Johnson (eds.), The Role of Speech Perception in Phonology, 251–277. New
York, New York: Academic Press.
Wright, Richard. 2004. A review of perceptual cues and robustness. In D. Steriade, R.
Kirchner and B. Hayes (eds.), Phonetically Based Phonology, 34–57. Cambridge,
United Kingdom: Cambridge University Press.
Yeon, Jaehoon. 2011. Korean: A Comprehensive Grammar. New York, New York:
Routledge.
Youssef, Islam. 2013. Place Assimilation in Arabic: Contrasts, Features, and
Constraints. Ph.D. dissertation, University of Tromsø, Tromsø, Norway.
Zampaulo, André. 2013. When Synchrony Meets Diachrony: (Alveolo)Palatal sound
Patterns in Spanish and Other Romance Languages. Ph.D. dissertation, Ohio
State University, Columbus, Ohio.
Zeng, Fan-Gang and Christopher W. Turner. 1990. Recognition of voiceless fricatives by
normal and hearing-impaired subjects. Journal of Speech and Hearing Research
33, 440–449.
340
Zubritskaya, Ekaterina L. 1995. The Categorical and Variable Phonology of Russian.
Ph.D. dissertation, University of Pennsylvania, Philadelphia, Pennsylvania.
Zúñiga, Fernando. 2000. Mapudungun. Munich, Germany: Lincom Europa.
© Copyright 2026 Paperzz