A New Perspective on Vowel Variation Across the 19Th and 20Th Centuries in Columbus, OH Dissertation Presented in Partial Fulfil

Home , Canadian Shift, Chain shift, Fronting (sound change), Philadelphia English

A New Perspective on Vowel Variation across the 19th and 20th Centuries in

Columbus, OH

Dissertation

Presented in Partial Fulfillment for the Degree of Doctor of Philosophy in the

Graduate School of The Ohio State University

David Durian, M.A., B.A.

Graduate Program in Linguistics

The Ohio State University

2012

Dissertation Committee:

Donald Winford, Advisor

Cynthia G. Clopper

Brian D. Joseph

David Durian

2012

Abstract

The research in this dissertation focuses on the documentation and analysis of vowel variation and sound change in Columbus, OH. Columbus is a Midland speech community located in the Midwestern United States. Specifically, this work documents and analyzes vowel variation among 62 native-English speaking European American informants, who are divided into 4 generation cohort groups (speakers born 1896-1913;

1924-1938, 1945-1968, and 1976-1991), 2 social class groups (middle and working), and

2 sex groups (male and female), so that patterns of vowel variation and sound change can be measured quantitatively across a number of speakers' vowel systems which are representative of the Columbus vowel system throughout the course of the 20th Century.

Some vowel system data, obtained from 19th Century Central Ohioans, is also analyzed for use in establishing an "initial state" for the Columbus vowel system, from which patterns of vowel variation and sound change observed in the data diverge over time.

Although vowel variation and sound change across the entire vowel system is documented and analyzed, several specific patterns of vowel variation are given extensive focus for analysis, description, and discussion. These patterns include the parallel fronting of the back diphthongs /uw/ and /ow/; the conditioned tensing and raising of /ae/, also known as split short-a system raising; and the Third Dialect Shift, a pattern of covariant vowel movement in which the low vowels /a/ and /ae/ are involved in

ii a backing chain shift, and the front vowels /E/ and /I/ are involved in a backing parallel shift. Each of these patterns of vowel variation are quantitatively analyzed using linear mixed effects regression (lmer) analyses. The data analyzed are vowel formant data normalized using Lobanov's (1971) z-score technique.

The influence of the social factors sex, social class, and age on the social conditioning of vowel variation is analyzed using the lmers, as are the linguistic factors of following and preceding consonant. Additional factors, such as speakers' social evaluations of vowel variations, speakers' language attitudes towards social and regional dialects spoken in Ohio and in the US more generally, and speakers' attitudes towards

Columbus as a community, are also investigated. In addition, the possible influence of demographic changes, economic changes, and urban growth patterns in Columbus during the 20th Century on the patterns of vowel variation exhibited by speakers is explored.

Our study is notable in being the first to document the occurrence of the split short-a system in Columbus vowel systems. It is also notable as being the first statistical analysis of vowel variation trends in a US Midland community across 4 generations of middle and working class European American speakers. In addition, it is the first to feature a systematic overall pattern analysis of the full vowel systems of 19th Century- born Central Ohioans impressionistically transcribed for the Linguistic Atlas of the North

Central States [LANCS] in 1933 (and collected as McDavid & Payne, 1976-1978).

iii Dedication

This work is dedicated to two scholars who were significant influences on me while I conducted my research for, and completed my writing of, this dissertation: Dr.

Henry Hunker (1924-2009), Professor of Geography at The Ohio State University; and

Dr. Robert Stockwell (1925-2012), Professor of Linguistics at the University of

California, Los Angeles.

iv Acknowledgements

There are many people I wish to acknowledge as I complete the writing of this dissertation. As many readers may know, during the course of the time that I did the fieldwork and writing of this dissertation, I went through quite a life journey. As a result, the period of writing and analysis was longer than it often is, which has contributed to the list of acknowledgements becoming even longer than it would have been otherwise.

First and foremost, I would like to thank my informants. Without them, this dissertation would not have been possible, because I would have nothing to study.

Thanks to all of you who opened you homes to me, or took the time to meet with me at

Oxley Hall, to talk about talk in Columbus.

Next, I would like to thank my committee. I would like to thank my advisor, Don

Winford, for all of his help and guidance during the 9 and 1/2 years I was a graduate student at Ohio State. The advice you gave me throughout my time here as a researcher and as an academic has proved invaluable to me. I would also like to thank Brian Joseph for being both a committee member and a friend throughout my time in graduate student, as well. The advice and input you gave on the historical and phonological aspects of my work are very much appreciated. In addition, I would like to thank Cynthia Clopper for serving on my committee, as well. The advice you provided on vowel analysis, normalization, and linear multiple effects regression proved to be invaluable. Finally, I

v would also like to thank my undergraduate mentors Don Hardy and Lisa Ann Cairns, whose guidance and friendship led me to going to grad school, and thus, ultimately influenced the work presented here by getting me starting down this path in the first place.

Beyond my committee, I would also like to thank a variety of linguists who provided me with additional advice and insights, not just during the dissertation writing process, but also more generally throughout my graduate school career. First and foremost, I wish to thank Erik Thomas for the help and advice he gave me regarding vowel analysis during the earlier years of my graduate career. I likely would not have been as excited to study the Columbus vowel system had it not been for Erik and his earlier work on Columbus vowels. In addition, I would like to thank William Labov,

Dennis Preston, Charles Boberg, and Matt Gordon for comments after conference presentations on material that I ultimately wrote about here in Chapters 3, 4, 5, 6, and 7. I would also like to thank Walt Wolfram, Peter Trudgill, Betty Birner, Scott Schwenter,

John Lawler, Bartek Plichta, Malcah Yaeger-Dror, Ben Munson, Rob Hagiwara, Joe

Salmons, Craige Roberts, Chris Brew, Julie McGory, Rich Janda, Aaron Dinkin, Daniel

Erza Johnson, Josef Fruehwald, Michael Friesner, Kara Becker, Hilary Prichard, Tyler

Kendall, David Bowie, Benjamin Tobert, Christine Mallinson, Doug Bigham, Kate Shaw

Points, Jessica White Susaita, Jennifer Nycz, Meredith Tamminga, Laurel MacKenzie,

Sonya Fix, Cara Shousterman, Brian Jose, Molly Babel, Anita Szakay, and Maeve

vi Eberhardt, for their interest, support, and good conversations over the years that I worked on this.

Another group of colleagues I wish to thank is my student interns--the folks who helped me complete large chunks of the analytical work. Without their help, vowel coding and formant analysis of the instrumental data would never have gotten done in any kind of timely fashion. Specifically, I thank Erica Crawford, Jenn Schumacher,

Missy Reynard, Josh Roush, Val Lindak, Andrew Smith, and Emily Dorrian.

Beyond these folks, I would also like to thank a number of my Ohio State colleagues for friendship, intellectual camaraderie, and general excellent times, throughout my time in graduate school. In particular, I wish to thank Grant McGuire,

Na’im Tyson, Angelo Costanzo, Mike Armstrong, Stacey Bailey, Adriane Boyd, Bridget

Smith, Kathleen Hall, Julia Papke, Salena Sampson, Christin Wilson, Mike Phelan, Katie

Carmichael, Abby Walker, Rachel Burdin, Vedrana Mihalicek, Chris Worth, Marivic

Lesho, Andy Plummer, Brice Russ, Meghan Armstrong, Robin Dodsworth, Ila Nagar, and Anouschka Bergman. In addition, I would like to thank my parents, as well as my two brothers, for their support while I dissertated.

Finally, one last set of folks I would like to thank is the organizations and individuals who contributed data to me for use in this dissertation. First, I wish to thank the Ohio Historical Society for allowing me to use recordings of 7 informants, born 1898-

1913, who were tape recorded for various projects conducted by the OHS in the 1970s

vii and 1980s. Second, I would like to thank Joan Hall, on behalf of The Dictionary of

American Regional English [DARE] for allowing me to use the tape-recorded interview for DAREOH098, made in 1970. Third, I wish to thank Erik Thomas, of North Carolina

State University, for allowing me to use his tape recorded interview with ColumbusMK, the Columbusite who also appears plotted in Thomas (2001), for analysis in my work, as well. This recording was made in 1984. Fourth, I wish to thank Bill Kretzchmar, on behalf of the Linguistic Atlas of the North Central States [LANCS], for letting me use

Raven McDavid's recorded interview with LANCSOH28C, made in 1957. At the time we first analyzed this audio for formant data to use in this dissertation (2008), only a roughly

34-minute fragment of usable audio from the interview could be found. So this is all we used. As of 2011, however, a much larger portion is now available. We hope to use this larger portion in an updated version of our analysis at a later date. Fifth, and finally, I wish to thank the Department of Psychology at the Ohio State University for allowing me to use several informants, all born circa 1938-1980, who were recorded for the Buckeye

Speech Corpus project, in 2000-2001.

viii Vita

1999 ...... B.A., Manga Cum Laude, English, Northern Illinois University 2008 ...... M.A., Linguistics, The Ohio State University 2005-2010 ...... Graduate Teaching Assistant, The Ohio State University 2011...... Adjunct Professor, Joliet Junior College 2011-2012 ...... Adjunct Professor, Waubonsee Community College 2011-2012 ...... Adjunct Professor, College of DuPage

Publications

Durian, David. 2011. Review: Mesthrie, Rajend, Joan Swann, Ana Deumert & William Leap. 2009. Introducing sociolinguistics (2nd ed.). Language in Society, 40.3:373-377.

Durian, David, Robin Dodsworth, and Jennifer Schumacher. 2010. Convergence in blue collar Columbus, OH African American and White vowel systems? In Malcah Yeager-Dror and Erik R. Thomas (Eds.). African American English speakers and their participation in local sound changes: A comparative study. Publication of the American Dialect Society, 94. Durham: Duke University Press. p. 161-190.

Durian, David, Julia Porter Papke, and Salena Sampson. 2009. Exploring social, regional, and ethnic variation in the undergraduate classroom. American Speech 84.2:227-238.

Durian, David. 2008. The vocalization of /l/ in urban blue collar Columbus, OH African American Vernacular English: A quantitative sociophonetic analysis. The Ohio State University Working Papers in Linguistics, 58:30-51.

Durian, David. 2007b. Getting [S]tronger every day? Urbanization and the sociogeographic diffusion of (str) in Columbus, OH. NWAV 35 Conference Proceedings. University of Pennsylvania Working Papers in Linguistics 13.2:65-79.

ix Durian, David. 2007a. File 10.3: Regional variation (Content expansion and revision). In Anouschka Bergmann, Kathleen Hall and Sharon Ross (Eds.), Language files (10th ed.). Columbus, OH: The Ohio State University Press. p. 418-426.

Durian, David. 2006. NWAV at 35: A look at the history, directions, and development of NWAV(E), 1972-2006—Conversations with Jack Chambers, Ralph Fasold, William Labov, Dennis Preston, John Rickford, Gillian Sankoff, Roger Shuy, Peter Trudgill, and Walt Wolfram. NWAV 35 Conference Program. Columbus, OH: Zip Publishing.

Durian, David. 2002. Review: Corpus-based text analysis from a qualitative perspective: A closer look at NVivo. Style, 36.4: 738-742.

Hardy, Donald E., and David Durian. 2000. Grammar and seeing in Flannery O'Connor's fiction: The stylistics of syntactic compliments. Style, 34.1:92-116.

x Table of Contents

Abstract ……………………………………………………………………………. ii

Dedication …………………………………………………………………………. v

Acknowledgements ……………………………………………………………….. vi

Vita ………………………………………………………………………………… x

Publications ………………………………………………………………………... x

List of Tables ……………………………………………………………….……... xviii

List of Figures …………………………………………………………………… xxii

List of Maps ……………………………………………………………………… xxviii

Chapter 1: Introduction: Revisiting the "Mysterious Midland" (via

Columbus) ………………………………………………………………… 1

1.1 Introduction ………………………………………………………... 1

1.2 Motivations for the Present Study …………………………. ……... 8

1.3 A Road Map to This Dissertation …………………...... 18

1.4 Some Notes on Terminology and Vowel System Analysis ……… 22

1.5. Notation Conventions Used for Vowel Classes ……...... 28

Chapter 2: Columbus--A Profile of the Speech Community ………………...... 31

2.1 Introduction …………………………………………...... ….……... 31

xi 2.2 Settlement History and Dialect Formation Influences ……………. 34

2.3 Demographic Change and Urban Growth as Social Change Factors

in Columbus ………..……………………………....………...... 35

2.4 Economic Growth as a Social Change Factor in Columbus ……… 43

2.5 The Possible Linguistic Influence of Demographic Change,

Urban Growth, and Economic Change in Columbus …….……….. 46

2.6 Speaker Social Evaluations of, and Attitudes towards, Vowel

Variation in Columbus ……………………………………………. 56

2.7 Measuring the Social, Cultural, and Linguistic Influence of

Demographic Change, Urban Growth, and Economic Change

on Vowel Variation in Columbus………………..………………… 57

Chapter 3: A New Perspective on Vowel Variation in Columbus, 1850-2000 …… 59

3.1 Introduction ………………………………………………. ……… 59

3.2 Sample Population Characteristics of, and Recruitment

Strategies for, the Present Study …..……………………………… 65

3.3 The Assignment of Social Class to Informants…………………… 68

3.4 Methods of Acoustic Analysis of Vowel Formants for the

Present Study …………………………………………………….. 71

3.5 Speaker Vowel System Normalization for the Present Study ……. 73

xii 3.6 Earlier Studies of Columbus Speech Used for Comparative

Analysis …………………………………………………………… 75

3.7 Patterns of 19th Century Vowel Variation ………………...... 81

3.7.1 Central Ohio Vowel Systems among Speakers

Born 1846-1854 as Analyzed in Previous Studies … ……... 84

3.7.2 Central Ohio Vowel Systems among Speakers

Born 1846- 1854 as Based on Our Reanalysis of

LANCS Field Records …………………………….……... 85

3.8 Patterns of 20th Century Vowel Variation ………………...... 91

3.8.1 Recessive Features ………………………………….…….. 93

3.8.2 Patterns Showing Continuation/Incrementation

From 19th Century Vowel Systems ………………………. 99

3.8.3 Innovations in 20th Century Vowel Systems……………… 109

3.8.4 New Innovations Found in Our Data Not

Discussed in Previous Studies …………………………….. 115

3.8.5 Comparison of Our Data with the JFSP Data …………….. 116

3.9 General Trends in the Data across Social Groups and Covariant

Vowel Classes …………………………………………….. 120

Chapter 4: Phonetic Analogy and the Parallel Fronting of the Back Diphthongs … 126

4.1 Introduction ……………………………………………...... 126

xiii 4.2 Previous Studies of Back Diphthong Fronting in US English …..... 130

4.3 Previous Discussions of Parallel Shifting ………………………… 136

4.4 Linear Mixed Effects Regression (lmer) Analyses of Nuclear

F2 of SHOES, BOOT, and BOAT in Our Data …………………... 145

4.4.1 F2 of SHOES …………………………………………….. 148

4.4.2 F2 of BOOT ………………………………………………. 151

4.4.3 F2 of BOAT ………………………………………………. 154

4.4.4 Summary of lmer Analyses ………………………………. 157

4.5 Phonetic Analogy and the Fronting of the Nuclei of SHOES, BOOT,

and BOAT ………………………………………………………… 158

Chapter 5: The Inception and Development of Third Dialect Shift in Columbus … 173

5.1 Introduction ……………………………………………………….. 173

5.2 Previous Studies of the Third Dialect Shift in North American

English …………………………………………………………….. 175

5.3 Linear Mixed Effects Regression (lmer) Analyses of Nuclear F1

and F2 of the Third Dialect Shift Vowel Series (BAT, BET,

BIT, BOT (and BOUGHT)) ………………………………………. 181

5.3.1 F1 of BAT (+BAG)………………………………...... 183

5.3.2 F2 of BAT (+BAG) ……………………………………….. 187

5.3.3 F1 of BET …………………………………………………. 190

xiv 5.3.4 F2 of BET …………………………………………………. 191

5.3.5 F1 of BIT ………………………………………………….. 193

5.3.6 F2 of BIT ………………………………………………….. 197

5.3.7 F1 of BOT…………………………………………………. 199

5.3.8 F2 of BOT ………………………………………………… 201

5.3.9 F1 of BOUGHT …………………………………..,……… 203

5.3.10 F2 of BOUGHT …………………………..………………. 206

5.4 The Merger of BOT and BOUGHT ………………………………. 208

5.5 The Inception and Development of the Third Dialect Shift in

Columbus …………………………………………………………. 218

Chapter 6: The Rise and Fall of Short-a in Columbus ……………………...…….. 233

6.1 Introduction ……………………………………………………….. 234

6.2 Previous Studies of Split Short-a Systems in US English ……….. 238

6.3 Our Analysis of Split Short-a Systems in Columbus …………….. 256

6.4 Revisiting the History and Development of Split Short-a Systems

in US English ……………………………………………………... 277

6.5 An Alternative Perspective on the History and Development of Split

Short-a Systems in US English …………………………………… 290

6.6 The Possible Formation and Development of the Columbus Split

Short-a System ……………………………………………………. 293

xv 6.7 Some Final Thoughts on the Rise and Fall of Split Short-a in

US English ………………………………………………………… 296

Chapter 7: “We Speak ‘Normal Midwestern English’…Most of the Time”: On the

Social Motivation of Vowel Variation and Sound Change in Columbus … 298

7.1 Introduction ……………………………………………………….. 298

7.2 Sex, Social Class, and Generational Cohort Membership ...... 301

7.2.1 Blue Collar G2, G3, and G4 Vowel Variation Patterns …... 303

7.2.2 White Collar G2, G3, and G4 Vowel Variation Patterns .... 304

7.3 The Influence of Demographic and Economic Changes on Vowel

Variation…………………………………………………………... 305

7.4 The Influence of Social Evaluation and Language Attitudes on Vowel

Variation ………………………………………………………….. 314

7.5 The Influence of Attitudes Regarding Columbus as "Midwestern

Place" and Vowel Variation ...... 326

Chapter 8: Conclusion ………………………………………………………….... 330

8.1 Introduction ……………………………………………………….. 330

8.2 Summation and Closure……………………………………..…….. 331

Bibliography ……………………………………………………………… 339

Appendix A: Central Ohio Dialect Survey Interview Protocol …………………… 356

Appendix B: Normalized Individual Speaker Vowel Plots ……………………….. 362

xvi List of Tables

Table 3.1: Demographic Characteristics of the 62 Speakers Analyzed in

the Present Study ………………………………………………………… 68

Table 3.2: Previous (Mostly) Published Columbus Vowel Variation Studies of

Informants Born 1846-1908 …………………………. ………………….. 78

Table 3.3: Previous (Mostly) Published Columbus Vowel Variation Studies of

Informants Born 1920-1987 …………………………. …………………... 81

Table 3.4: Variation in Central Ohio Vowel Systems Noted in Reynard's 1933

LANCS Field Notes ……………………………………………………… 89

Table 3.5: Summary of 19th Century Central Ohio Vowel Variation Patterns ….. 91

Table 3.6: Vowel Classes Undergoing Variation among Columbus Speakers

Born 1896-1987 as Analyzed in Previous Studies …………………...…… 96

Table 3.7: Vowel Classes Undergoing Variation among White Collar Columbus

Speakers Born 1896-1991 in Our 62 Speaker Corpus …………………… 97

Table 3.8: Vowel Classes Undergoing Variation among Blue Collar Columbus

Speakers Born 1896-1991 in Our 62 Speaker Corpus …………………… 98

Table 4.1: Coefficients in the Linear Mixed Effects Model Predicting the F2 of

SHOES from the Social Variables (Sex, Generation, Social Class), Their

Interactions, and Following Phonological Environment ………………… 149

xvii Table 4.2: Coefficients in the Linear Mixed Effects Model Predicting the F2 of

BOOT from the Social Variables (Sex, Generation, Social Class), Their

Interactions, and Following Phonological Environment …………………. 153

Table 4.3: Coefficients in the Linear Mixed Effects Model Predicting the F2 of

BOAT from the Social Variables (Sex, Generation, Social Class), Their

Interactions, and Following Phonological Environment …………………. 155

Table 4.4: Coefficients in the Linear Mixed Effects Model Predicting the F2 of

/uw/ (SHOES/BOOT) from the Social Variables (Sex, Generation, Social Class),

Their Interactions, and Following Phonological Environment …………… 165

Table 5.1: Coefficients in the Linear Mixed Effects Model Predicting the F1 of

BAT from the Social Variables (Sex, Generation, Social Class), Their

Interactions, and Following Phonological Environment …………………. 186

Table 5.2: Coefficients in the Linear Mixed Effects Model Predicting the F2 of

BAT from the Social Variables (Sex, Generation, Social Class), Their

Interactions, and Following Phonological Environment …………………. 188

Table 5.3: Coefficients in the Linear Mixed Effects Model Predicting the F1 of

BET from the Social Variables (Sex, Generation, Social Class), Their

Interactions, and Following Phonological Environment …………………. 191

xviii Table 5.4: Coefficients in the Linear Mixed Effects Model Predicting the F2 of

BET from the Social Variables (Sex, Generation, Social Class), Their

Interactions, and Following Phonological Environment …………………. 192

Table 5.5: Coefficients in the Linear Mixed Effects Model Predicting the F1 of

BIT from the Social Variables (Sex, Generation, Social Class), Their

Interactions, and Following Phonological Environment …………………. 195

Table 5.6: Coefficients in the Linear Mixed Effects Model Predicting the F2 of

BIT from the Social Variables (Sex, Generation, Social Class), Their

Interactions, and Following Phonological Environment …………………. 198

Table 5.7: Coefficients in the Linear Mixed Effects Model Predicting the F1 of

BOT from the Social Variables (Sex, Generation, Social Class), Their

Interactions, and Following Phonological Environment …………………. 200

Table 5.8: Coefficients in the Linear Mixed Effects Model Predicting the F2 of

BOT from the Social Variables (Sex, Generation, Social Class), Their

Interactions, and Following Phonological Environment …………………. 202

Table 5.9: Coefficients in the Linear Mixed Effects Model Predicting the F1 of

BOUGHT from the Social Variables (Sex, Generation, Social Class), Their

Interactions, and Following Phonological Environment …………………. 205

xix Table 5.10: Coefficients in the Linear Mixed Effects Model Predicting the F2 of

BOUGHT from the Social Variables (Sex, Generation, Social Class), Their

Interactions, and Following Phonological Environment …………………. 208

Table 5.11: Analysis of the Extent of Low Back Merger in Our Data …………… 215

Table 5.12: Correlation of BOT and BAT/BAG Backing Across All Speakers in

Our 62 Speaker Corpus …………………………………………………… 221

Table 5.13: Correlation of BET and BAT/BAG Backing Across All Speakers in

Our 62 Speaker Corpus …………………………………………………… 227

Table 6.1: Raw IPA Transcriptions from Reynard's LANCS Field Notes for

Short-A Words …………………………………………………………… 258

Table 6.2: Columbus Short-A Systems by Type Across All 62 Speakers ……….. 273

Table 6.3: The 4 Stages of Short-a System Development and Growth in US

English ……………………………………………………………………. 288

xx List of Figures

Figure 2.1: Annexation in Columbus …………………………………………….. 38

Figure 2.2: Population Growth in Central City Columbus (1815-2010) …………. 44

Figure 4.1: Following Segments Conditioning /uw/-Fronting …………………… 167

Figure 4.2: Following Segments Conditioning BOAT-Fronting …………………. 167

Figure 4.3: Progression of SHOES, BOOT, and BOAT Fronting in Blue Collar

Women's Speech ………………………………………………………….. 169

Figure 4.4: Progression of SHOES, BOOT, and BOAT Fronting in White Collar

Men's Speech ……………………………………………………………… 170

Figure 5.1: Examples of Distinct and Fully Merged BOT and BOUGHT Classes

among Speakers …………………………………………………………... 213

Figure 5.2: Examples of Partially Merged BOT and BOUGHT Classes among

Speakers …………………………………………………………………... 214

Figure 5.3: Correlation of BOT Backing and BAT/BAG Backing among All 62

Speakers in Our Data Set …………………………………………………. 222

Figure 5.4: Stages of the Third Dialect Shift in White Collar Men's Vowel

Systems …………………………………………………………………… 225

Figure 5.5: Possible Chain Shift of BOT, BAT, and BET in G2 White Collar

Women's Vowel Systems ………………………………………………… 226

xxi Figure 5.6: Correlation of BET Backing and BAT/BAG Backing among All 62

Speakers in Our Data Set …………………………………………………. 228

Figure 5.7: The Third Dialect Shift, Depicted as The Canadian Shift in Clarke,

et al. (1995 )……………………………………………………………….. 229

Figure 5.8: The Third Dialect Shift, Depicted as The Canadian Shift in Boberg

(2005) …………………………………………………………………….. 230

Figure 5.9: The Third Dialect Shift in Columbus, with Low Vowel Chain Shift

and Front Vowel Parallel Shift Noted …………………………………… 231

Figure 6.1: "Classic" New York City Split System ……………………………… 241

Figure 6.2: "Classic" Philadelphia Split System …………………………………. 241

Figure 6.3: Split Short-a Systems among Speakers Born c. 1810-1865 …………. 247

Figure 6.4: Split Short-a Systems among Speakers Born c. 1850-1912 …………. 248

Figure 6.5: Split Short-a Systems among Speakers Born c. 1890-1971 …………. 249

Figure 6.6: A Vowel System Showing an Allophonic Split Short-a System …….. 263

Figure 6.7: A Vowel System Showing a Continuous Short-a System …………… 264

Figure 6.8: The Columbus Split Short-a System …………………………………. 268

Figure 6.9: A Vowel System Showing a Nasal Short-a System ………………….. 271

Figure 6.10: A Vowel System Showing a Nasal Short-a System ………………… 272

Figure 7.1: Population Growth in Central City Columbus (1815-2010) …………. 307

Figure 7.2: Annexation in Columbus (1834-2008) ………………………………. 311

xxii Figure B.1: Vowel System of White Collar Male LANCS028C, Born 1896 …….. 363

Figure B.2: Vowel System of White Collar Male Tobias, Born 1898 ……………. 364

Figure B.3: Vowel System of White Collar Male William, Born 1905 ………….. 365

Figure B.4: Vowel System of White Collar Male Don, Born 1924 ……………… 366

Figure B.5: Vowel System of White Collar Male Tom C, Born 1928 ……………. 367

Figure B.6: Vowel System of White Collar Male Tom W, Born 1937 …………... 368

Figure B.7: Vowel System of White Collar Male Ben, Born 1952 ………………. 369

Figure B.8: Vowel System of White Collar Male David, Born 1954 ……………. 370

Figure B.9: Vowel System of White Collar Male Steve, Born 1955 ……………. 371

Figure B.10: Vowel System of White Collar Male Peter, Born 1956 ……………. 372

Figure B.11: Vowel System of White Collar Male Ed, Born 1967 ……………… 373

Figure B.12: Vowel System of White Collar Male Abe, Born 1985 ……………. 374

Figure B.13: Vowel System of White Collar Male Bryan, Born 1987 …………… 375

Figure B.14: Vowel System of White Collar Male Matt M, Born 1987 ………….. 376

Figure B.15: Vowel System of White Collar Male Meesha, Born 1988 ………… 377

Figure B.16: Vowel System of White Collar Male Justin L, Born 1990 ………… 378

Figure B.17: Vowel System of Blue Collar Male Arthur, Born 1898 …………… 379

Figure B.18: Vowel System of Blue Collar Male ColumbusMK, Born 1908…….. 380

Figure B.19: Vowel System of Blue Collar Male Frank, Born 1913 …………….. 381

Figure B.20: Vowel System of Blue Collar Male Joe, Born 1926 ………………...382

xxiii Figure B.21: Vowel System of Blue Collar Male Tony, Born 1930 …………….. 383

Figure B.22: Vowel System of Blue Collar Male Jim S, Born 1932 …………….. 384

Figure B.23: Vowel System of Blue Collar Male Bill, Born 1949 ………………. 385

Figure B.24: Vowel System of Blue Collar Male Ted, Born 1951 ………………. 386

Figure B.25: Vowel System of Blue Collar Male Terry, Born 1959 …………….. 387

Figure B.26: Vowel System of Blue Collar Male Michael, Born 1962 ………….. 388

Figure B.27: Vowel System of Blue Collar Male Jim H, Born 1967…………… 389

Figure B.28: Vowel System of Blue Collar Male Jason, Born 1980 …………… 390

Figure B.29: Vowel System of Blue Collar Male Justin R, Born 1982 …………. 391

Figure B.30: Vowel System of Blue Collar Male Vincent, Born 1984 ………… 392

Figure B.31: Vowel System of Blue Collar Male Travis, Born 1985 …………… 393

Figure B.32: Vowel System of Blue Collar Male Louie, Born 1986 …………… 394

Figure B.33: Vowel System of White Collar Female Theadora, Born 1903 …….. 395

Figure B.34: Vowel System of White Collar Female DAREOH28, Born 1906 …. 396

Figure B.35: Vowel System of White Collar Female Joyce, Born 1924 …………. 397

Figure B.36: Vowel System of White Collar Female Nancy C, Born 1927 ……… 398

Figure B.37: Vowel System of White Collar Female Barbara, Born 1938 ………. 399

Figure B.38: Vowel System of White Collar Female Toni, Born 1948 …………... 400

Figure B.39: Vowel System of White Collar Female Marsha, Born 1950 ………. 401

Figure B.40: Vowel System of White Collar Female Sheri, Born 1950 …………. 402

xxiv Figure B.41: Vowel System of White Collar Female Linda, Born 1956 ………… 403

Figure B.42: Vowel System of White Collar Female Margaret, Born 1962 ……… 404

Figure B.43: Vowel System of White Collar Female Janna, Born 1978 …………. 405

Figure B.44: Vowel System of White Collar Female Molly, Born 1978 ………… 406

Figure B.45: Vowel System of White Collar Female Mandy, Born 1981 ……….. 407

Figure B.46: Vowel System of White Collar Female Jayne, Born 1984 …………. 408

Figure B.47: Vowel System of White Collar Female Liz, Born 1988 ……………. 409

Figure B.48: Vowel System of Blue Collar Female Louise, Born 1908 …………. 410

Figure B.49: Vowel System of Blue Collar Female Ann, Born 1912………………411

Figure B.50: Vowel System of Blue Collar Female Marie, Born 1925…………… 412

Figure B.51: Vowel System of Blue Collar Female Mary, Born 1928 …………… 413

Figure B.52: Vowel System of Blue Collar Female Helen, Born 1930 …………… 414

Figure B.53: Vowel System of Blue Collar Female Linda G, Born 1945 ………… 415

Figure B.54: Vowel System of Blue Collar Female Susan, Born 1950 …………… 416

Figure B.55: Vowel System of Blue Collar Female Molly, Born 1952 ……………417

Figure B.56: Vowel System of Blue Collar Female Paula, Born 1958 …………… 418

Figure B.57: Vowel System of Blue Collar Female Michelle, Born 1961 ………... 419

Figure B.58: Vowel System of Blue Collar Female Rachael, Born 1976 ………... 420

Figure B.59: Vowel System of Blue Collar Female Angela, Born 1977 ………..... 421

Figure B.60: Vowel System of Blue Collar Female Julia, Born 1982 ………...... 422

xxv Figure B.61: Vowel System of Blue Collar Female Kelly, Born 1988 ………...... 423

Figure B.62: Vowel System of Blue Collar Female Kaitlyn, Born 1991 ………..... 424

xxvi List of Maps

Map 2.1: Columbus and Surrounding Suburbs in 1950 ………………………….. 39

Map 2.2: Columbus and Surrounding Suburbs in 2000 ………………………….. 40

Map 2.3: The Columbus Metropolitan Statistical Area (CMSA) in 2000 ……….. 42

xxvii Chapter 1: Introduction--Revisiting the “Mysterious Midland” (via Columbus)

1.1 Introduction

Since 1930 (Kurath, 1930), the speech of the Central Ohio city known as

Columbus, OH has received a fair amount of attention in studies of vowel variation. In that initial study, Hans Kurath made some fairly inventive use of a device known as the

Telechord, a recording device which allowed the user to capture an acoustic signal onto a wax paper cylinder. The same machine also allowed playback of the signal, a fact that

Kurath made much use of as he used the machine to play back the signal repeatedly, allowing him to make an extremely fine-grained phonetic transcription of the speech of a young female Ohio State University college student reading the familiar (at least to dialect researchers) tale "Arthur the Rat."

In sum, Kurath estimated, the playback facility of the machine allowed him to listen to each segment produced by the 20 year old subject "15 to 20 times in all" (93) so that he could check and recheck his work for accuracy, until the wax cylinder was no longer listenable. Although he himself did not perform a systematic analysis of the dialect features noted in his transcript, in the article, Kurath notes several features of vocalic variation in the woman's speech that would become quite well known in the years following the publication of his study. These included mild centralization of /aw/ and

/ow/, tensing and raising of /ae/ before nasal, and some reduction of the glide of /ai/.

1 Beyond serving as the first study of Columbus vowel variation, it is, as far as we know, one of the first detailed studies of vowel variation in the US Midland more generally. In addition, it would also serve as a useful pilot study in the establishment of the phonetic transcription methodology that Kurath would implement in the Linguistic

Atlas of the United States and Canada [LAUSC] project that he would begin and oversee shortly after the publication of this article. However, when implemented for large-scale use in the Atlas project, the technique utilized by field workers would not include the use of the Telechord, likely because it was not a fieldwork friendly device.

As the years rolled on, several more studies on vowel variation in Columbus would be published, although until 1993, with the publication of Thomas ([1989]/1993), the data on which these additional studies were based on either just 3 speakers who were interviewed by Cassil Reynard, in 1933, and Raven McDavid, in 1956, for the Linguistic

Atlas of the North Central States (a subproject of Kurath's LAUSC) (Marckwardt, 1940;

1942; Nobbelin, 1980), or one speaker audio-recorded for the Dictionary of American

Regional English [DARE] in 1970 (Frazer, 1978). Thus, even though nearly 40 years separated the Nobbelin and Frazer studies from the two Marckwardt studies, and nearly

50 years separated Nobbelin (1980) from Kurath (1930), all of the information that the field was working from during this period for Columbus was based primarily on 2 speakers born in the middle of the 19th Century, and 2 speakers born around the beginning of the 20th Century (1896 and 1906, respectively).

2 With the publication of Thomas's ([1989]/1993) study, the story began to change, as Thomas presented data from Columbusites born in the late 1960s. This meant that other researchers, when commenting on patterns in Columbus, were able to draw on much more recent data on which to make their observations than the earlier studies. In addition, Thomas provided data not only on the speech patterns of European Americans, like the earlier studies had, but also African American Columbusites, as well.

Following this study, three more would be conducted in the 1990s and early-to- mid-2000s which would build on Thomas's study by providing additional data from subjects born later than the turn of the 20th Century, and also by conducting instrumental analyses of the vowel systems investigated. The first two of these studies—Keiser, et al.

(1997) and Dodsworth (2005)–were conducted by Ohio State University researchers, who would follow in the footsteps of Kurath as innovators, presenting instrumental analyses of vocalic data during a period when impressionistic studies were still fairly commonly found. The analyses they presented were only for select vowel classes, however—/ae/ in the case of Keiser, et al. (1997) (drawing on data from 2 speakers born during the 1950s and 1960s), and only /ow/ and /ae/ in the case of Dodsworth (drawing on 21 speakers from 3 general age groups, born during the period 1920-1987). Given this fact, it would not be until the third study of this era—Thomas (2001) —that the first study to present an instrumental analysis of the entire vowel system of a Columbus informant would be presented. This came in the form of Thomas's presentation of the complete systems of

3 three informants, two of whom also appeared in his ([1989]/1993) study and were born in

1970, and one of whom was born in 1908, in vowel plot form.

Following the completion of Dodsworth (2005), studies of Columbus began to focus more extensively on the analysis of multiple generations of speakers and also the analysis of whole vowel systems. This series includes three projects that spawned multiple publications. The first was Labov, et al.'s (2006) Atlas of North American

English, which resulted in two Columbus-related publications—Labov, et al. (2006) itself, and Ash (2006). In Ash (2006), data on several specific vowel variables is presented, but not the entire vowel system. 16 subjects, born during two different time periods, 1920-1930, and 1960-1980, were analyzed. In Labov, et al. (2006), data across the entire vowel system for 6 of the 16 speakers-—the largest group of speakers up to that time who had their entire systems analyzed—was presented. The bulk of the data were obtained from female speakers born in the 1960s, although a male born in 1920 is also included in the analysis to represent "earlier speech patterns."

The second project, which began in 2004, was our own ongoing pilot study for this dissertation—The Century of Language Change in Columbus (CLCC) project. This study featured 40 subjects at its largest, with male and female speakers of two class groups (working and middle), and born during 4 time periods, analyzed. As with Labov, et al.., the pilot featured whole vowel system analyses (for 22 of the speakers in Durian

(2008), and for 40 speakers, in Durian (2009, 2011), all of whom also appear here in our

4 dissertation analysis), although only certain vowel classes were given extensive focus in the series of conference presentations that resulted from the pilot phase of this work

(Durian & Smith, 2005; Durian, 2007, 2008b, 2009, and 2011). The age groups for the speakers included those born in: 1896-1908; 1924-1938; 1945-1967; and 1976-1990. The vowel classes and phenomena selected for analysis in these studies included /uw/- and

/ow/-fronting (Durian, & Smith, 2005), the near-merger of /a/ and /ɔ/ (Durian, et al.,

2008b, 2009), a series of vowels we first argued in Durian (2008) show Columbus participating in the Canadian Shift—/a/, /ae/, /E/, and /I/—with each showing backing

(see also Durian, 2009), and /aw/-fronting (Durian, 2011). Selected subgroups of these speakers were also featured in comparative work on similarities and differences between

European American and African American working class and middle class vowel systems. These analyses were presented in Durian, et al. (2007) and Durian, et al. (2010).

In addition, speakers from the pilot were also used for analyses of /s/-retraction and /l/- vocalization in Columbus speech presented in Durian (2007) and Dodsworth, et al.

(2006). Additional theoretical discussions relating to our data, including portions of the first presentation of the material on phonetic analogy discussed in Chapter 4, the

Canadian Shift as the Third Dialect Shift presented in Chapter 5, and the inception and development of the Third Dialect Shift, also discussed in Chapter 5, were presented initially in Durian & Joseph (2011), Durian & Gordon (2011), and Durian (2012).

A third project which began after our work was underway is that of Jacewicz

5 Fox, and Salmons. It includes a focus on Central Ohio (including Columbus) vowel systems, as well, although taking a different angle from our own work or that of other researchers discussed so far. In their work, they have approached the study of sound change across 3 generations of speakers in 3 different dialect areas—North Carolina,

Southeastern Wisconsin, and Central Ohio. However, unlike previous studies, they focus on somewhat different aspects of vowel variation. This includes the analysis of spectral change shown throughout the course of a vowel, vowel duration, and trajectory length of the vowel. The main studies of relevant interest here published by the time of writing from this project are Jacewicz, et al. (2011) and Jacewicz, et al. (2011b), in particular, the sections dealing with Central Ohio. Their analysis includes 78 speakers, divided into speakers born between 1919 and 1940, 1958 and 1973, and 1994-1998. Sex is also studied as a social factor, although others, such as social class or race, like we used in our pilot work, are not used. The type of speech analyzed is read citation form speech.

Thus far, among their studies dealing with cross-general generational vowel change in Central Ohio, although the whole vowel system has received some attention in their publications, only the Canadian Shift vowel series of /a/, /ae/, /E/, and /I/ has received detailed attention as "changes of significance." Both Jacewicz, et al. (2011 a,b) confirm the findings of our own earlier (2008) and (2009) papers, albeit with a larger group of speakers per age group than our studies. However, they only find the low vowel component (/ae/-backing and /a/-backing) to be significant, with little evidence of /E/ or

6 /I/ backing. We find /ae/- and /E/-backing to be significant among many speakers. with

/I/-backing and /a/-backing being notable although not statistically significant. They do find significant /E/-lowering and /I/-lowering, but it is due to their undergoing monophthongization among speakers. Although we did not report this trend in our studies, Thomas ([1989]/1993) has reported this trend previously. Jacewicz (2011b) suggests it may be simply a stress-related difference due to their use of read speech, however. In addition, the explanation Jacewicz et al (2011a) provide regarding the

Canadian Shift (which they call the North American Shift) having a parallel shift component in their data, is also similar to one we posited earlier for the Canadian Shift in

Durian (2009), although they do not credit us with having done so1. We present this analysis in expanded form based on our expanded study population in Chapter 5.

This is not to say the Jacewicz, et al. studies do not provide any additional results our pilot study or Thomas's work has not already covered, however. They cover other aspects of the patterns of variation we did not. For instance, the focus on changes in spectral characteristics of vowels is quite interesting, and the study provides us with changes in the dynamics of vowels we have not seen discussed in previous studies. As well, the cross-dialectal comparisons on these characteristics, as well as formant value changes across the three generations of informants from North Carolina and Wisconsin as well as Central Ohio is also interesting reading. In addition, the information on duration

1 Why this omission occurs is unclear to us since they discuss other findings from Durian (2009) in another part of the article. 7 and trajectory length they provide is interesting to learn about, although its role in helping us determine linguistic motivations for vowel variation and sound change in real world conversational speech beyond what earlier discsusions of these characteristics have told us (e.g., Thomas, 2001, 2003, 2011; Labov, et al, 1972) requires further exploration.

Generally, what this this side-by-side comparison also suggests is that, our work, both before, and especially in the expanded version we will present in the remaining 8 chapters of this dissertation, and Jacewicz, et al.'s work, as discussed above, is best seen as being complementary to one another. This will be made clearer in section 1.2, where we detail how the expanded version of our study differs not only from theirs, but also our pilot work and other previous studies of Columbus. As we discuss, it does so by covering much ground not addressed in any previous study of Columbus vowel variation.

1.2 Motivations for the Present Study

The general story provided by the study of vowel variation from the time of

Kurath's initial (1930) study to the analysis, 81 years later, of our pilot study and the

Jacewicz, et al. study, discussed in section 1.1, is one of interesting observations made by authoritative researchers and innovations in the science of the study, analysis, and observation of significant patterns of vowel variation using Columbus data. However, like much of the rest of the Midland dialect region—the major speech area to which

Columbus belongs, a placement which has been made frequently in a variety of dialect

8 studies over time, both on phonetic/phonological grounds (e.g., Kurath, 1930; Allen,

1977; Pederson, 1977; McDavid & Payne, 1976-1978; Thomas, 1989[1993]; Thomas,

2001; Clopper, et al.., 2005; Labov, et al.., 2006; Ash, 2006) and on lexical grounds (e.g.,

Davis, 1948; Marckwardt, 1957; Allen, 1977; McDavid & Payne, 1976-1978; Carver,

1987; Frazer, 1993; Ash, 2006) —it has also continued to remain understudied in several key ways, leaving the field with a somewhat marred perspective on the patterns of vowel variation to actually be found in the community.

Specifically, Columbus has yet to be the subject of a multi-generational quantitative statistical instrumental analysis of vowel variation patterns across the entire vowel system that takes into account social factors beyond age and sex, addresses the significant linguistic factors of preceding and following segmental environment, and deals with the social, as well as the linguistic, motivations of vowel variation and sound change found in the community. That is, Columbus has not been the subject of the kind of study that has been shown in other geographic areas, in particular, cities like New

York City (Becker, 2010; Becker & Wong, 2012; 2009); Philadelphia (Labov, 2001);

Montreal (2005); Chicago (McCarthy, 2011), and Raleigh (Dodsworth & Kohn, 2012), to provide a robust analysis of the patterns of cross-generational sound change in progress available in a community because of the focus on these additional factors and motivations. This criticism includes not only the earlier studies discussed in section 1.1, but the Jacewicz, et al. study and our own previous work (except for the focus on class).

9 One primary social factor that has been missing in most previous studies of

Columbus is social class. Other than our own previous work (Durian, 2007, 2008b, 2009,

2010; Durian, et al., 2007; Durian, et al., 2010) and the work of Keiser, et al. (1997), social class has usually not been an explicit focus of Columbus linguistic studies. The studies of Labov, Boberg, and Dodsworth & Kohn also included a social class component, with an explicit comparison made between linguistic patterns of two or more social class groups, in an attempt to better hone in on the leaders of linguistic change in the communities they have analyzed, who are often found among working class speakers.

Many studies, not just those conducted previously in Columbus, do not add this additional dimension of analysis, although doing so would likely allow for a more detailed understanding of the patterns of language variation and change to be acquired, as studies such as Labov (2001), Boberg (2005), Dodsworth &Kohn (2012), and our own previous work in Columbus (e.g., Durian, 2007, 2008b, 2009, 2010; Durian, et al., 2007;

Durian, et al., 2010) show.

Meanwhile, the lack of focus on the social motivations, as well as the linguistic motivations, of vowel variation and sound change has left us with at times less satisfactory explanations for why some patterns of vowel variation persist and become incrementational, while others recede and fade with time. As 50 years of sociolinguistic research on language variation and change has demonstrated, a full explanation of patterns of vowel variation and sound change are usually not able to be satisfactorily

10 completed by only looking at sex and age as basic social factors, or by only looking at linguistic factors like vowel movement trends or changes in the spectral characteristics of vowels, even when they are looked at in conjunction with one another. Speakers' perceptions, social evaluations, and language attitudes must also be assessed in conjunction with an analysis of the patterns of vowel variation and sound change involved to obtain the most detailed understanding of why and how sound changes and vowel movement patterns happen the way they do (e.g., Labov, [1966]/2006, 2001;

Trudgill, 1986; Preston, 1996; Eckert, 2000)

And when those linguistic patterns are investigated, the same body of 50 years of sociolinguistic research has also demonstrated that many vowel variation patterns which can be systematically observed as unfolding in real or apparent time are motivated not just by issues such as vowel system symmetry, changes in spectral characteristics of vowels, or the maintenance of margins of security for vowels, but also the powerful conditioning influence of following and preceding consonants and consonant groups

(e.g., Labov, et al., 1972; Labov, 1981, 1989, 1994, 2001, 2010; Labov, et al., 2006). This work builds on earlier studies going back to the beginning of the Neogrammarian study of sound change (e.g., Hockett, 1965; Bloomfield, 1933; Paul, 1888/1920; Ostoff &

Brugmann, 1878; see also Schuchardt, 1885 for an early alternative perspective). No study of Columbus or Central Ohio yet conducted has presented a statistical analysis of conditioning environments impacting the patterns of vowel variation shown by speakers

11 across the entire vowel system for more than a handful of full speaker systems at one time (Labov, et al., 2006). As our dissertation will demonstrate, this power cannot always be determined simply by making simple word class or word subclass divisions for vowel classes, or only looking at the spectral, durational, or trajectory length characteristics of vowels.2 In fact, phonetic conditioning involving following segments plays a key role in the incrementation of a number of the vowel shifts and mergers currently underway in

Columbus, as well as the historical occurrence of the split short-a system. Phonetic conditioning also plays a key role in the view of phonetic analogy (Schuchardt, 1885;

Vennemann, 1972; Durian & Joseph, 2011) that we put forth in Chapter 4.

The same is also true of studies of US Midland cities beyond Columbus. Thus far, the closest the Midland has come to receiving this kind of robust quantitative analysis is two analyses conducted nearly 30 years apart: Lusk (1976) and Fogle (2008). Lusk

(1976) is a multigenerational study, conducted in Kansas City, KS. Having been conducted in the mid-1970s, the study makes use of only a simple correlational analysis of the patterns of vowel variation (e.g., Labov, [1966]/2006), with impressionistic judgments, rather than instrumental measurements, of the patterns of vowel variation and language change found in the data. That being said, the study was still rather impressive for several reasons. First, it was, until Thomas's (1996) study of vowel variation in

Johnstown, OH, the only multi-generational study of vowel variation in the US Midland

2 Most likely, we suspect what we really need to observe all of these additional factors in concert with the analysis of segmental conditioning. This is work we will take up in a future study. 12 that had been conducted during the modern sociolinguistic era. Second, it analyzed data from both working- and middle-class speakers, an approach not often utilized in studies of Midland speech, either before or since. Third, it was rather forward looking—Lusk in fact identified the inception of the Canadian Shift in Kansas City among her youngest informants. However, at the time, she did not realize this was the case, and instead, like

Keiser, et al.. (1997) would do in Columbus nearly 20 years later (albeit by only investigating /ae/ variation rather than the entire vowel system, like Lusk did), identified the vowel system as being potentially a version of the Northern Cities Shift, except that it differed in showing backing of BOT rather than fronting, as is normally expected. In her defense, the Canadian Shift would not be found in another community for another 10 years (Luthin, 1987, in California) and not identified specifically as a chain shift of significance until almost 20 years later (as the Canadian Shift, in Ontario, by Clarke, et al., 1995), so the fact that she did not identify it as a new and distinct vowel shift, in the absence of information from other regions to provide additional confirmation of its being something new and different, can perhaps be overlooked as simply a case of a young researcher new to the field making a conservative call to avoid being seen as a making an error of judgment in case she was mistaken.

Fogle (2008), in contrast, presents the results of a multi-generational study in

Indianapolis, IN, a prototype of modern-day US Midland speech, at least as it has been defined by Clopper, et al. (2005) and Labov, et al. (2006). However, the study only

13 presents results for a subset of the vowel system, focusing almost exclusively on the back vowels (SHOES/BOOT, BOAT, and BOUT, along with BOT and BOUGHT), with the one exception being that it also (briefly) discussed BIDE/BITE variation, as well. The lack of analysis of BAT, BET, BIT, BAIT, and BEET appears to have resulted from their not being noted as showing "significant" variation in Labov, et al. (2006), since, as Fogle notes, her study only focused on confirming Labov, et al.'s (2006) notion that

Indianapolis is a prototype of US Midland speech.

Perhaps because of the lack of systematic quantitative statistical instrumental analyses of the whole vowel system in US Midland communities (and larger, very recent, speaker-rich studies such as Durian (2008, 2009) and Jacewicz, et al. (2011a, b)), until quite recently, the Midland has tended to be depicted in discussions of vocalic dialect trends as either being "lacking in the occurrence of major chain shifts," as stated by

Labov, et al. (2006), "generally stable" (at least compared to the Northern Cites or areas of the South showing the Southern Shift), as stated by Labov (1991), or even somewhat bland, an impression left in the minds of many readers after reading early descriptions of

"General American" vowel variation in "the Middle West," such as those presented by

Palmer (1926), Krapp (1925), or Kenyon (1924).3 Much less frequently, the Midland has been referred to as "mysterious" by researchers, given the lack of such a study. 4

3 Even with recent studies like our pilot work, our conversations with sociolinguists at recent conferences like NWAV and ADS suggests these stereotypes continue to live on. 4 This description was reportedly provided to Keiser and his associates by Labov at NWAV when they presented an early version of Keiser, et al. (1997) in the mid-1990s. 14 Given the lack of systematic instrumental analysis of the entire vowel system in most previous studies of Columbus specifically, and the US Midland more generally, and also the lack of a quantitative analytical multi-generational study with similar characteristics to the studies of cities like New York City (Labov, [1966]/2006; Becker,

2010; Becker & Wong, 2012; 2009); Philadelphia (Labov, 2001); Montreal (2005);

Chicago (McCarthy, 2011), and Raleigh (Dodsworth & Kohn, 2012) we have discussed above, especially the class-oriented studies of urban cities such as Labov ([1966]/2006,

2001), Becker (2010), Boberg (2005), and Dodsworth & Kohn (2012), we have decided to undertake such an analysis here in our work on Columbus. Specifically, we conduct a quantitative instrumental analysis of 4 generations of speakers living in the greater

Columbus area. The speakers included in our analysis are subjects who have been sampled to represent the larger Columbus population during the time period we have chosen for analysis. This time period is the entire 20th Century, and thus, the speakers we have chosen to analyze were born throughout most of this time period. To do this, we have selected speakers who were born between 1896 and 1991, a period that reflects the speech norms of the time period we wish to analyze, since in sociolinguistic studies, we usually assume that a speaker's speech is representative of the time period in which that speaker was raised and spent their formative years in a given speech community

(Chambers & Trudgill, 1998; Milroy & Gordon, 2003).

To provide a historical context into which to place these results, we have also

15 chosen to supplement our analysis of 20th Century patterns of vowel variation with a

(re)analysis of impressionistic data obtained from 2 Central Ohio males born in 1846 and

1854 and interviewed by Linguistic Atlas of the North Central States fieldworker Cassil

Reynard. One of these men was born in Reynoldsburg, while the other was born in

Worthington. Each man is born in a city near Columubs in which modern-day residents who were also interviewed for our study were born. Each city is also close enough to

Columbus that we assume, based on available evidence, that the speech norms there in the mid-19th Century were likely quite similar to Columbus. This additional data is used to provide an "initial state" for the Columbus vowel system—a literal "beginning map" of the phonological space, in 2 dimensions, from which the later generations of speakers we have chosen to analyze the vowel systems of vary and change away from. As we also found after completing our analysis of the data, it provided us with a deeper understanding of the triggering event for the back diphthong parallel shift, the history of historical split short-a realization in the community, and the triggering event for the

Canadian Shift in Columbus.

Via our analysis, we focus on several concerns. First, just as in our pilot study, we have included four generations of speakers in our analysis. This includes 3 generations of speakers we interviewed ourselves, as well as 1 additional generation of speakers born during the late 1800s and early 1900s, most (if not all) of whom were deceased by the time we conducted our field work in 2004-2005 and 2008-2009. As will be discussed in

16 more detail in Chapter 3, the 4 generational groups we have assembled for our study specifically include speakers born 1896-1913 (Generation 1), 1924-1938 (Generation 2),

1945-1968 (Generation 3), and 1976-1991 (Generation 4). Including speakers from all 4 of these generational groups allowed us to make our data comparable not only to older studies, such as Labov ([1966]/2006) or Labov (2001), where the oldest speakers were born in 1890 (New York City) and 1888 (Philadelphia), or Thomas (2006) and McCarthy

(2011), which make use of informants born as early as 1880, but also more recent studies, such as Boberg & Strassel (2000), Labov, et al. (2006), Dinkin (2009), or Becker (2010), where the older speakers tended to be born circa 1920 or later. In that way, our data can be compared with the largest range of previous studies possible.

Second, we completed an instrumental analysis of variation and change in major vowel classes of North American English for all 62 of the speakers we audio recorded.

This was done for two reasons. One, we did so to obtain a full and detailed picture of all vowel variation patterns to be found in the data. This includes mergers, splits, and individual vowel movements, as well as chain shifts and multiple class vowel shifts. Two, we did so to facilitate our use of a robust normalization procedure, so that statistical and other numerical comparisons between all speakers based on their vowel formant measurements could be made. Third, we present a statistical analysis of the patterns of vowel variation and sound change in our data, that takes into account sex, age, and social class, as well as preceding and following segment, so as to be able to more concisely

17 discern the influence of linguistic and social factors on the realization of the vowels by our informants. This is done using the relative recently developed approach involving linear mixed effects regression analysis. As will be explained in Chapter 4, this method is employed because of the advantages it allows the analyst over more traditional analyses, such as ANOVA or VARBRUL analysis.

Throughout the following pages of this dissertation, we present the results of our multigenerational, quantitative statistical analysis of vowel variation and sound change across middle and working class groups in the 20th Century in Columbus. Where applicable and relevant, we will also present the results of our (re)analysis of 19th

Century vowel variation patterns in Reynard's IPA field notes. In doing so, we hope to provide a series of results that finally treats the Midland, not as an area of "nonsignificant normality," as it was often depicted in early studies of the Midland, where it was characterized as an area where simple, "General American" was spoken (Krapp, 1924;

Kenyon, 1925), the area of the US defined by "a lack of major vowel shifts," as per

Labov, et al. (2006), or an area of "mystery," but as an area of interesting and vibrant vowel variation, including a major chain shift—The Third Dialect Shift—which our results most certainly indicate it is.

1.3 A Road Map to This Dissertation

The content of this work is divided into a total of 8 chapters. Chapter 1 serves as

18 the introduction to the work, and it presents an overview of the general scope of the dissertation. It also includes a road map of the remaining chapters of the work, and a brief introduction to the formal notation used to discuss vowel classes (and their underlying word classes) throughout this dissertation.

Chapter 2 provides a profile of the community in which we have conducted our research. Included is a discussion of relevant geographic, demographic, and economic factors which we believe have had an impact on the realization of vowel variables, and on the progression of vocalic sound changes and vowel shifts, in the community during the course of the 19th and 20th Centuries. In addition, a brief discussion of the settlement history and the influence settlement patterns may have had on dialect formation during the early 19th Century in Columbus is provided. Some discussion of informant selection for our study, as well as our methods of sampling speakers from the community and socially stratifying our informants, is also presented.

Chapter 3 presents an overview of the major patterns of vowel variation we have found in the community via our instrumental analysis of speaker vowel systems. The analysis focuses on the whole vowel system, and it includes all major vowel classes of

North American English. The analysis includes the patterns we have found not only in our instrumental data, collected from speakers born c. 1896-1991 in the community, but also those we have found in the data of the two men, born 1846 and 1854, for whom we have only impressionistic data, transcribed by Reynard in 1933. In addition, we compare

19 the patterns we found in our data sets with the findings of previous studies of Columbus vowel variation, so that our work can be better understood in relationship to the work that came before it. This is done via a detailed meta-reanalysis of the results of the previous studies, where we tie the results together chronologically in a way that has not actually been done in any earlier study.

Chapters 4 5, and 6 present detailed analyses of major patterns of significance initially discussed in Chapter 3, several of which have not been discussed in any previous study of the Columbus vowel system. In contrast to the presentation of the material in

Chapter 3, where a simple analysis of speaker and speaker group mean vowel class normalized formant values is presented, the analyses in Chapters 4, 5, and 6 present more detailed statistical analyses of the data, with chapters 4 and 5 presenting the results of linear mixed effects regressions for the vowel classes discussed. Chapter 4 deals with the back diphthong subsystem. Chapter 5 deals with the vowels comprising the Third Dialect

Shift, looking specifically at its inception and development in Columbus during the course of the 20th Century. These are two topics (its development and its inception) that have not been discussed in a previous study of Columbus, nor in any previous discussion of any US Midland speech community. Chapter 6 presents a detailed analysis of the

Columbus short-a system, focusing on its change from a split short-a system to a continuous and then finally a nasal system among most speakers born after 1976. This is another variation pattern not discussed in earlier analyses of the Columbus vowel system.

20 In addition, Chapters 4, 5, and 6 deal with important theoretical contributions the data are able to make to our understanding of vowel variation and vocalic sound change, when and where relevant. Chapter 4 discusses parallel shifting and the role phonetic analogy (Schuchardt, 1885; Vennemann, 1972) appears to play in the generalization process that accompanies the occurrence of parallel shifts and their incrementation in speaker vowel systems, not just in Columbus, but also in varieties of US English more generally. Chapter 5 discusses the Third Dialect Shift in light of different claims by researchers in recent years concerning its status as either a chain shift (Clarke, et al.,

1995; Labov, et al.., 2006) or a combination chain and parallel shift (Boberg, 2005), to determine which type of scenario best fits its occurrence specifically in Columbus.

Chapter 6 revisits the results of previous studies of the occurrence of split short-a systems in US English and reexamines them in light of our own findings regarding the occurrence of split short-a in our own data as well as in the results of several historical studies of split-a systems in US English which appear to have been "forgotten" in later studies of split a systems until quite recently. Via this analysis, we present a novel explanation for the occurrence of this system in areas outside of New York City and Philadelphia that presents an alternative to that proposed previously by Labov (2007) and Dinkin (2009).

In Chapter 7, we turn to presenting a more detailed discussion of the impact of social factors and socio-geographic forces, such as urbanization, annexation, and community-wide demographic and economic change on the patterns of vowel variation

21 and sound change exhibited by the Columbus vowel system during the course of the 19th and 20th Centuries. Included in this analysis is a discussion of the impact that community-internal and -external speaker migration patterns may have had on the transmission and diffusion of patterns of vowel variation such as back diphthong fronting and the recession of the split short-a system. In addition, some consideration of the influence of speaker conceptions of Columbus as a "place" may have had on the diffusion of the Third Dialect Shift across social class groups in Columbus is also provided.

Finally, in Chapter 8, a summary of the major findings of the dissertation is presented. This includes some consideration of what the findings add to our deeper understanding of patterns of vowel variation and sound change in English. This consideration includes not only Columbus English, but also English in the US Midland and North America more generally.

1.4. Some Notes on Terminology and Vowel System Analysis

Throughout this dissertation, several terms and concepts will be returned to quite often. Given that this is the case, we wish to define some of these terms of concepts here, in Chapter 1, so that they are presented to the reader at the earliest possible juncture. Like much work dialectology conducted since the early 20th Century, our work here is focused extensively on the documentation of sound changes impacting the Columbus vowel system, as they unfold in both real and apparent time. Real-time changes are those which

22 can be observed as having occurred literally in actual calendar time, while apparent-time changes are those that we assume have happened given differences in the vowel realization patterns of older speakers versus younger speakers as they were observed in one population of speakers at one point in calendar time.

Although we touch on this fact only briefly in the main body of the dissertation, all patterns of vowel change in our data that differentiate the oldest generation of speakers in our study (speakers born 1896-1913) from the next oldest generation

(speakers born 1924-1938) are not only apparent time changes, but they are real time changes as well. We say this because not only are the changes being observed at one point in time—that is, we have set up an analysis of the oldest speakers with younger speakers who were all recorded at just one point in history—but, given the difference in times of recording between the oldest and next oldest groups, that difference has also allowed us to make a real-time comparison of differences for these groups. The bulk of the recordings for the oldest speakers were made between 1970 and 1983, and most speakers during that time period were recorded when they were in their late 60s, 70s, or

80s. When we recorded the next oldest group, all were in their 70s or 80s. Thus, when we compare the vowel systems of 70- and 80- year olds recorded roughly 30-40 years ago with the vowel systems of 70- and 80- year olds recorded today, we can observe real time change because a real time gap exists in the data collection period that is large enough to to allow one to study actual change between the speaker groups. Given that many of the

23 trajectories can be observed in apparent time among the younger groups of speakers, we assume throughout this dissertation that these apparent trends show the continuation of the real time trends.

Another point we should raise here is how we use the terms "sound change" and

"vowel variation" throughout this work. Sound change is used to describe what Joseph

(2008) and Anderson, Dawson, and Joseph (2011) would call "sound change proper."

That is, we actually do mean that if we call the patterns sound changes, the changes we have observed qualify as changes in the pronunciation of vowel class from an earlier first state to a later second state. Following the widely agreed upon modern definitional viewpoint of Neogrammarian sound change (Hock, 1991), a sound change proper in our data means that the change began in a small, highly restricted set of environments, usually conditioned phonetically by a following consonantal segment or set of segments.

The change then spreads to an increasingly larger set of circumstances, becoming less restricted as it spreads. Eventually, all candidates for the change become affected.

It should be noted, however, that although we discuss sound change proper in the instances of sound change we discuss throughout this dissertation, most are what Labov would call "sound change in progress" (Labov, 1994). That is, many of the changes are observed at some point along the way from a beginning state to some later second state, with the assumption being that they haven't fully arrive yet at the later second state. This assumption is typically borne out by the data, with a younger generation often showing

24 the incrementation and further development of a trend shown in the speech of an older generation that came before those younger speakers. This is, of course, not always the case--some changes can proceed in one direction for a while, before changing course and heading another direction. For instance, in Columbus during the late 19th Century, /ae/ appeared to be on a raising and fronting path. Later, among speakers born progressively later in the 20th Century, however, this path shows complete reversal, and in fact, a new course of change is now occurring, with /ae/ showing a primarily back trajectory among most young speakers. This doesn't mean the pattern of variation is not change in progress in both cases—it is. It just happens to be that the type of change path taken has changed.

This is because, as our data will show, sound change does not occur on a predestined, unidirectional trajectory. It may look like that at times, especially if we base our conclusions on completed sound changes—those which are no longer active, such as, for instance, the series of changes typically grouped as the English Vowel Shift. But this is not always the case. The ability for sound change to modify its direction in progress instead highlights the fact that the change we are observing is in progress, with progress meaning in part, its being open to possible fluctuation in directionality. Some changes sometimes change direction, others do not.

This fact also further highlights the need to draw a distinction between sound changes in progress and completed sound changes, which to us, are ultimately both forms of "sound change proper." To highlight the fact that fluctuation can be a part of sound

25 change in progress and attempt to better get at the "in progress" versus "completion" sound change distinction, we discuss many of the patterns here as "vowel variation" in the body of the text, rather than using the "sound change in progress". This allows us to reserve the use of the term "sound change" for discussion of the process of sound change more generally, regardless of status (completed or in progress.), and the use of the term

"completed sound change" to mean a completed sound change.

In addition, to better define trajectories of change for the sound changes in progress we are studying, we will adopt a set of terms commonly used in descriptive dialectology for describing dialect features undergoing change--recessive, incremental/continuous, and innovative. Recessive changes in progress are those that initially appear as an innovation in the community among speakers born at an earlier point in time than later groups of speakers, but then show decreasing use, or cease to continue appearing at all, in vowel systems as an innovation among speakers born at a later point in time. In contrast, innovative changes in progress are those that appear among speakers belonging to a particular speaker age cohort but did not appear as an innovative pattern in use among the speakers belonging to the group who immediately precede that cohort. These patterns show continued use in the speaker cohort who immediately follows the innovating cohort, and this continuation is usually accompanied by increasing use of the feature among each younger cohort of speakers (incrementation).

A continuous/incremental change in progress is one that appears in all speaker vowel

26 systems across age groups in our study. Like the innovation, it also shows increasing use among increasingly younger speakers. The use of these terms, as well as cogent examples of their occurrence in real world speech situations, will be provided in Chapter 3.

One final concept we also wish to comment on briefly here at the outset is also that, in terms of the diagnosis of sound changes as recessive, innovative, or incremental/continuous, we ultimately make use of the two concepts for vowel analysis which ultimately descend from structural dialectology. First, vowel systems are best analyzed from an overall pattern perspective to allow us to best observe the changes involved in them (e.g., Trager & Smith, 1951; Stockwell, 1964, 1978; Labov, 1994,

2010; Labov, et al., 2006). To do so, an "initial state" for the vowel system is assumed, and then patterns of divergence away from this initial state are analyzed, documented, and discussed by the analyst looking to investigate change trends in vowel realization.

Second, in order to do this, we need to analyze vowel classes based on their historical occurrence in English, so that long term changes to them can be observed and ascertained. This involves the use of historical vowel class documentation based on word classes. For our analysis throughout this dissertation, we follow Labov (1994) in the use of word classes based on the post-English Vowel Shift late-Middle English vowel system. Adopting this approach, our perspective on the quantification of vowel variation is thus squarely aligned with the approach adopted by Labov (e.g, Labov, [1966]/2006;

Labov, 1994, 2001, 2010; Labov, et al, 2006), and thus, most present-day variationists,

27 who in turn have adopted the same perspective. In addition, our approach to the observation of vowel shifts is influenced to a large extent by Martinet (e.g., 1952, 1955) and Labov (e.g., 1994, 2010; Labov, et al, 1972), although to a lesser extent, we have also been influenced by the work of Luick (1903), Jespersen (1908), Lass (e.g., 1976),

Stockwell (e.g., 1972, 1978, 2002), and Donegan (e.g., 1978).

1.5 Notational Conventions Used for Vowel Classes

Note however, that our approach to referring to vowel classes is somewhat different from Labov's. Rather than relying on Trager & Smith's (1957) notation, we instead adopt the use of key words to symbolize vowel classes and the word classes underlying them. This makes our approach more like that of Wells (1982 In the discussion of vocalic variation involving the major vowel classes of modern day North

American English throughout this dissertation, we have adopted the following notational conventions. First, all major vowel classes will be discussed using, in most cases, B_T key word notation. There are several reasons for the use of this convention. First, it is useful for indicating contrasts between conditioning environments which have consistently been shown to have “across the board” conditioning effects on North

American English vowels. For instance, in almost all dialect areas within North America,

/ae/ has been found to undergo nucleus raising when a nasal follows the vowel. In addition, in many dialects of North American English, the nucleus of /ae/ may undergo

28 raising, non-raising, or even lowering and retracting when /ae/ is followed by non-nasal segments, even when /ae/ is undergoing nucleus raising when a nasal segment occurs immediately after the vowel. Furthermore, in some dialects as well, /ae/ when followed by velar segments (/k/, /g/, and /ŋ/) may also show different kinds of covariant raising or lowering behaviors to /ae/ before front nasals and /ae/ before non-nasals and non-velars.

In addition, following liquids (/l/ and /r/) also often condition the realization of a preceding /ae/. Hence, the influence of these segments also needs to be noted.

Therefore, to allow meaningful contrasts to be referenced in our discussion, we have adopted the following convention for dealing with /ae/, whereby pre-nasal /ae/ is referred to as BAN, pre-velar /ae/ is referred to as BAG, and non-pre-nasal/non-pre- velar/non-pre-liquid /ae/ is referred to as BAT. For other vowel classes where following velar segments have not been found to significantly condition raising, lowering, fronting, or backing of a given vowel class, we combine following velars with other non-nasal and non-liquid following segments. Thus, /ɛ/ before nasals is BEN, /ɛ/ before /r/ is BEAR,

/ɛ/ before /l/ is BELL, and /ɛ/ before any other segments is simply BET.

Although a full listing of all possible vowel class groupings could certainly be given following the kind of conventions we have just described, the reality is, most studies of vocalic variation in Columbus conducted over the years have not exhaustively studied all of the possible conditioning segment + vowel class combinations that exist in modern day North American English. Therefore, we have chosen to list only those that

29 are relevant to this dissertation.

/u:/- SHOES (SH= preceding coronals), BOOT (all other preceding segment types)

/ʊ/-PUT

/oʊ/ - BOAT, FORCE

/aʊ/- BOUT

/ɔ/- BOUGHT, BALL, DAWN, NORTH

/ɑ/- BOT, DOLL, DON

/aI/- BITE, BIDE (here, the distinction is whether a voiceless or voiced segment follows)

/ae/- BAT, BAN, BAG (BAD is used in cases where /ae/-raising is specifically discussed before

voiced stops; BAN is used for the front nasals /n/ and /m/; BAG is used for /k/, /g/, and /ŋ/)

/ɛ/- BET, BEN

/e/- BAIT

/I/- BIT, BIN

/i/- BEET

The use of key words should not be taken to imply that the findings discussed only refer to tokens of, say, b_t words. Within any B_T group, for instance, the T simply means “voiced stops, voiceless stops, voiced fricatives, voiceless fricatives, and velars are included” (with the exception of the BAT class, of course). Tokens included in _T groups can therefore include all sorts of actual lexemes uttered by a speaker, as long as the word uttered contains a following segment belonging a qualifying segment group.

Also, unless otherwise noted in the text, any nasal subgrouping (i.e., DON, DAWN, or

BAN, BIN, and BEN) clusters only the front nasals /n/ and /m/ together. 30 Chapter 2: Columbus--A Profile of the Speech Community5

2.1 Introduction

As we discussed in Chapter 1, one of the principal foci of our study is to look at the development of the Columbus vowel system during the course of the 19th and 20th

Centuries. During the course of that time period, Columbus has grown significantly as a population. In 1840, roughly 6 years before the oldest speaker for whom we have data, was born, the population was only 6,048. By 1850, the population nearly tripled, to

17,882. 4 years before the second oldest speaker was born. By 1990, one year before the final speaker was born, the population was 632, 910, while in 2010, shortly before the time of writing, the population was 797, 434. This means the population has grown by nearly 132 times over the course of the time-depth of this study.

This set of facts about Columbus's size during the time period covered by the data discussed throughout this dissertation highlights several important points. First, it highlights the significance of population growth as it has impacted the size and shape of the community in the years immediately following the original settlement of the community, in 1812. Second, it highlights the changes to the size and scale of the community over the course of time that the informants from whom we have obtained data--either instrumental or impressionistic--have actually lived, worked, and engaged in

5 The work in this chapter owes a large debt of gratitude to the work of Geographers Henry Hunker and Robert Beauregard for influencing our approach to the urban geography issues presented. 31 the community and community-life. Third, it also highlights the general stability of

Columbus as a community over a two-century life span. Unlike many other larger urban communities located throughout the United States, Columbus stands out as a member of a diverse minority group of cities that have managed to sustain continued growth throughout the entire time the city has existed (Lentz, 2003; Rusk, 2003). Since 1815, when the first census was taken in the community, until 2010, no decennial census has shown a loss in population estimates, and in fact, every census has shown at least some population growth (Lentz, 2003; Hunker, 2001).

As we will see in Chapter 7, we believe this community attribute, along with

Columbus's continued spatial growth and economic stability, have actually contributed to the formation of several of the patterns of vowel variation and language change that we will ultimately describe, discuss, and analyze in the chapters that will follow in this dissertation. Before we present this discussion, however, we first must provide the reader with ample background material to understand our reasoning for thinking that

Columbus's growth throughout this time period has had the linguistic impact we believe it has. We then must present the reader with ample analysis of the linguistic data, so that the patterns of variation and change to which we refer are sufficiently understood. The detailed discussion of the linguistic patterns we refer to here will be presented in the next

4 chapters, first in overview format, and then with detailed discussion provided for three of the major patterns of vowel variation that our analysis of the Columbus vowel system

32 has provided.

Before this discussion is provided, however, we turn here in Chapter 2 to first providing the reader with a discussion of the important settlement, demographic, economic, and urban growth trends that have occurred in Columbus during the last 200 years, so as to provide the reader with sufficient knowledge of these patterns for reference in later chapters of this dissertation. In section 2.2, we first discuss the settlement history of Columbus and how the early settlers set the cultural pace of

Columbus in the following years. We also briefly discuss the impact that the linguistic backgrounds of the early settlers likely had on Columbus speech during the early part of the 19th Century, as a way of providing a background for our discussion of linguistic patterns of change and vowel variation which will follow in Chapters 3, 4, 5, and 6. We follow this discussion in sections 2.3 and 2.4 with a look at how Columbus has grown and changed since settlement times, focusing on various indicators of growth and change in the community, including geographic, demographic, and economic indicators. This includes a discussion of the impact geographic, demographic, and economic growth as a change factor have had on the community. In doing so, we will provide useful background information for the analysis of the influence of social, economic, and geographic change factors on language use among Columbusites that we present in

Chapter 7.

33 2.2 Settlement History and Dialect Formation Influences

When first settled in 1812, Columbus as a city was literally created from the very beginning to be the site of the state capital for Ohio. Because of this, Columbus was guaranteed more or less from day one to be a fast-growing settlement community on what was, at the time, the Western edge of the country. Although detailed census records of

Ohio were not kept until 1850 (Wilhelm, 1982), other sources provide details which allow historians to be able to say with a fair amount of certainty where the principle groups who settled Central Ohio during the earliest years of Columbus’s history originally hailed from. These groups include settlers of Mid-Atlantic (Pennsylvania, New

Jersey, Maryland, and to a much lesser extent New York), Northeastern (Connecticut,

Massachusetts), and Upper Southern (West Virginia, Virginia, Kentucky) origins. Each of these groups were heavily involved with the settlement of Central Ohio beginning in the late 1790s, and all continued to have a significant impact on settlement in the area through the late 1800s (Chaddock, 1908; Brown, 1967; Wilhelm, 1976, 1982). Although the relative ratio of settlement groups in Central Ohio to one another was not known until the first US census of Ohio in 1850, at that time, Mid-Atlantic settlers were the most predominant settlement group, followed by Northeasterners, and then Upper Southerners

(Wilhelm, 1982).

In terms of the impact these groups had on dialect formation in Columbus, studies discussing regional dialects as they likely existed in the early 1800s suggest Mid-

34 Atlantic-influenced features included the centralization of the nuclei of the back upgliding vowels BOOT and BOAT; close or merged realization of BOT and BOUGHT before /t/; r-fullness, and possibly the raising of the nucleus of /ae/ (Thomas, 2004;

Kurath & McDavid, 1961; Kurath, 1928, 1930), whereas Upper Southern-influenced features likely include centralization of the nucleus of BOUT, mild raising of PUT and

BIT before /ʃ/, and deletion of the glide of BITE (Thomas, 2004; Frazer, 1978). In addition, New England-influenced features likely included merger of the NORTH and

FORCE classes (Thomas, 2004; Kurath & McDavid, 1961; Kurath, 1928). The development of these features over time in the area, particularly as they developed in the speech of Central Ohioans in the mid-1800s, will be discussed in Chapter 3, via our analysis of IPA impressionistic transcription field records from the Linguistic Atlas of the

North Central States [LANCS] for two Central Ohio men born in 1846 and 1854, respectively, and transcribed by fieldworker Cassil Reynard in 1933.

2.3 Demographic Change and Urban Growth as Social Change Factors in Columbus

Two social change factors of significance to our dissertation that have had a significant impact on the community since its founding in 1812 have been urban growth and demographic population change. Throughout the past 200 years, urban growth and demographic population change have been a continuous process in Columbus, with increases in the residential population and the spatial size of Columbus as an urban area

35 occurring almost every decade throughout the city’s history (Stevens, 1996).

During the first 50 years, the population of Columbus showed a slow, but steady growth pattern, as the area evolved from one of first settlement to the state capital of

Ohio (Cole, 2000). Following the Civil War period and continuing through the early

1900s, the present-day “central city” was formed, and the city found itself transformed from being a fairly small town to a true urban entity (Speer, 1972). Yet, even with this first period of expansive growth, Columbus in the early 20th century never became industrialized like other large cities in Ohio, such as Cleveland and Cincinnati. This lack of extensive industrialization explains why it never suffered from the same issues accompanying rapid industrialization that affected these other cities during this time period, such as large-scale labor union strikes or pronounced civil unrest due to racial tensions. Instead, it continued to be a haven of light industry and a growing goods- and services-based industry, focused heavily on regional and intraregional production demands rather than national markets (Hunker, 2001). Economic growth as a change factor is discussed in the next section.

During the second half of the 20th century, two powerful changes occurred in the community that led to a second period of expansive growth and a dramatic transformation of the Columbus landscape. The first was urban growth via geographic expansion, which resulted from Columbus’s adoption of a policy of annexation of surrounding areas—a concerted effort to expand the interests of business and residential

36 development, as well as political influence, within the area—beginning in 1952 (Hunker,

1996; Nobel & Peacefull, 1996). The second was the extensive internal out-migration of

European Americans from the central city area to surrounding suburbs beginning in the early 1950s and continuing at a fairly continuous pace through the 1980s (Jacobs, 1994;

Foster, 1997). This movement occurred partly due to new residential construction drawing residents away from the central city as land was annexed and developed by

Columbus and partly due to the building of the I-70, I-71, and I-270 expressways, which facilitated travel to and from the central city and more peripheral areas.

Throughout the Columbus area, annexation and outmigration from the central city were accompanied by extensive and pronounced urbanization and suburbanization. The net impact of urbanization and suburbanization during this period on Columbus is most clearly seen in the expansion of roadways, the building of shopping malls, and the growth of business in the transitional areas separating the urban core from the suburbs. As a result of the building of the I-70, I-71, and I-270 expressways during the late-1940s to mid-1970s, residents moving within the region were given more choices as to where they could settle and raise their families (Adolphus, 1982; Burgess, 1994). As well, due to the expansion of business along I-270, communication networks, and the interactions within these networks among speakers residing in areas that were once considered “more remote,” have increased, and areas that were once sparsely populated have been extensively built up (Hunker, 2001).

37 As a result of annexation, by 2000 (the last decade for which we have data),

Columbus had expanded from roughly 42 square miles in 1950 to 220 square miles

(Rusk, 2003) The impact of this growth is displayed by decade in the graph in Figure 2.1, as well as in Maps 2.1 and 2.2, which display the size of Columbus in 1950 and again in

Figure 2.1: Annexation in Columbus, from 1834 to 2007 (Source: Columbus Dispatch, 2008) 38

Map 2.1: Columbus and Surrounding Suburbs in 1950 (Source: Columbus Planning and Development

Division, n.d.)

Map 2.2: Columbus and Surrounding Suburbs in 2000 (Source: Columbus Planning and Development

Division, n.d.)

40 2000. Compared to other geographically Midwestern cities, Columbus’s growth has been significant, representing a 434% change over this time period. In contrast, Cleveland and

Detroit grew by 3% and 0%, respectively, between 1950 and 2000 (Rusk, 2003:18).

Furthermore, during this same time period, the population expanded to include nearly 1.6 million residents within the Columbus Consolidated Metropolitan Statistical Area

(CMSA) (U.S.Census, 2000). This number includes residents living within the central city, which includes the original 1950s-era 42 square mile Columbus area which continues to function as the “principal center of influence” for the larger Columbus metropolitan area, as well as annexed areas within Franklin County that fall within 5-10 miles of the I-270 beltway; Columbus’s suburbs (Grove City, Hilliard, Dublin, New

Albany, Westerville, Reynoldsburg, Canal Winchester, Obetz, Grandview Heights,

Marble Cliff, Upper Arlington, Worthington, Bexley, and Whitehall) and 5 additional counties that contribute a significant commuter population to businesses located within the central city area: Delaware, Licking, Madison, Pickaway, and Fairfield (Hunker,

2001). The CMSA is included in Map 2.3.

Within the CMSA, the central city and its surrounding suburban areas have been the most strongly affected by annexation, urbanization and suburbanization. As defined using census data calculations, the central city, which is demarcated as “Columbus proper” in Maps 1.2 and 2.2 (and excludes Columbus’s suburban population), had a population of roughly 797,434 in 2010 (U.S. Census, 2010). Figure 2.2 displays

41 population growth in the central city area from the first census of the population taken in

Columbus, in 1815, to the latest census, in 2010.

Given that the central city and the suburbs that fall within 10-15 miles of I-270 are the area most residents and non-residents think of when they refer to Columbus as an urban entity, this entire area will henceforth be referred to as Columbus in the remaining discussion in this dissertation. In this context, Columbus is essentially used as a synonym

Map 2.3: The Columbus Consolidated Metropolitan Statistical Area (CSMA) (Source: Hunker, 2001, p. 30) 42 for what we might also call the greater metropolitan Columbus area. Within the context of this definition, the 42 square mile area of the 1950s-era Columbus is defined as the

“urban core” or "central city," using the portion of the Columbus area demarcated as

“Columbus Proper” in Map 1. In contrast, the portion of Columbus that became apart of

“Columbus proper” via annexation throughout the period 1950-2000, as detailed in Map

2, are defined both as the central city, as well as the “non-suburban periphery,” depending on the context in which this area is referred to. If pre-annexation, then it is the

"non-suburban periphery," whereas if it is post-annexation, than it would be a part of the

"central city." In addition, suburban areas of the greater metropolitan Columbus area are defined using the area designated for each in Map 2, and are referred to as "the suburbs" or "suburban periphery" depending on the context of the discussion.

2.4 Economic Growth as a Social Change Factor in Columbus

A third social change factor that has had a significant impact on Columbus that is relevant to our dissertation has been economic growth. Since the founding of the city in

1812, economic growth has been one of several important change factors that have impacted Columbus. Since that time, the gradual evolution of the Columbus economy has been dictated heavily by the status of the city as a center of state government, by its location geographically at the center of the state, and, after 1870, its status as home to a major state university, The Ohio State University (Lentz, 2001). Its status as a state

43 government center provided a stable economic base for middle class workers who serve as the workers for the state bureaucracy. Its geographic location has provided useful to

Columbus being situated as a regional trade and commerce center, a status that was first enhanced in 1833, when the National Road first passed through the city, and then later in the 1870s, when railroads first began to pass through the city (Lentz, 2003). Its status as home to a significant state university has also allowed Columbus an additional channel for the provision of a stable economic for the middle class (Wright, 1958; Hunker, 2001).

During the early 1900s, Columbus also began to expand as an economic center for

Blue Collar light industry and manufacturing jobs. Industries that began to develop and prosper in Columbus between 1900 and 1940 included companies such as Jeffrey

Manufacturing, which made mining machinery, Timkin Roller Bearings (maker of Roller

Figure 2.2: Population Growth in Central City Columbus (1815-2000) (Source: U.S. Census Bureau, n.d.) (Note: *1815 is the date of the first population census, per Studer, 1873)

44 Bearings), Jaeger Machinery (makers of portable cement mixers), Buckeye Steel Casings

(makers of railroad castings), Ralston Steel Car Company (manufacturers of railroad freight cars), Seagraves (makers of fire engines), and Peruna Drugs (makers of patent medications), among others (Wright, 1958; Hunker, 1958, 2001).

During the Second World War, manufacturing continued to expand in Columbus, with a boon for factory jobs being provided by the building of the Curtis-Wright plant in

1941. Following the war, Curtis-Wright continued to operate in Columbus, and after they war, they began to recruit inexpensive labor by recruiting workers from Appalachia who were looking for work. Other companies that also opened during this time that further encouraged Blue Collar laborers to migrate from Appalachia included the opening of the

Lustron Corporation (maker of porcelain-construction homes), General Motors,

Westinghouse, and Western Electric. Jobs at these factories continued to provide a labor base for Blue Collar workers that lasted well into the 1970s (Hunker, 2001, Lentz, 2003).

However, during the late 1970s, 1980s, and early 1990s, many of the factory jobs that were available in Columbus began to move, either to overseas locations or to other parts of the country, as a part of the so called "Sun-belt" migration of jobs. During this time period, manufacturing jobs decreased, as Columbus switched from a light industry center to a customer-service-based industry center (Lentz, 2003; Hunker, 2001). As a result, today, Blue Collar jobs now tend to be customer-service-oriented rather than manufacturing- or factory-oriented, or they revolve around other types of service-oriented

45 work, like grocery store clerk jobs, maintenance staff work, or handyman-type professions. Meanwhile, middle class jobs have continued to focus on state government positions, university-oriented work, and, during the second half of the 20th Century, grew to include banking and insurance-industry work (Hunker, 2001; Lentz, 2003).

2.5 The Possible Linguistic Influence of Demographic Change, Urban Growth, and Economic Change in Columbus

Turning to studies of language variation and change in the community, three investigations of consonantal variation (Durian, 2004, 2007; Dodsworth, et al.., 2006) reveal the influence that urban growth and demographic change may be having on linguistic and cultural behaviors of Columbus residents. Specifically, these studies reveal the impact that speakers' patterns of social and geographic mobility have had on the transmission and diffusion of innovative patterns of realization of retracted and palatalized variants of /s/ in /_tr/ clusters and the vocalization of /l/ across the Columbus metropolitan area during the second half of the 20th century. More importantly, these studies have revealed an impact on the mindset of Columbusites that is of most relevance for our present purposes in this dissertation.

In our (2007) study, we drew on two data sources, collected in distinct phases, to analyze variable realization of innovative variants of /s/ in /_tr/ clusters. The first source was a rapid anonymous survey (RAS) of 120 European American Columbus clerks, modeled after Labov's use of the technique to study the social distribution of (r) in New

46 York City department stores. The second source, collected slightly later, was a collection of digital tape recorded hour-long interviews with 32 middle class European American

Columbusites that we interviewed in 2004-2005.6 As no previous research was available on the social distribution of (str), either in Columbus or in any other US city, before we began the study, the RAS was used because it allowed us to collect a large body of information concerning its distribution by social factors such as age, sex, and social class quickly and inexpensively, and determine a base line for its social distribution in the community. We then collected and analyzed the sociolinguistic interview data to obtain a more detailed view of the impact of sex and age on, and the impact of linguistic factors influencing innovative /s/ use in, the speech of middle class Columbusites.

For the RAS, we collected data from 120 store clerks who worked at three mall locations in Columbus (Polaris, Easton, and City Center) and who were further equally subdivided by age (three groups: 15-30; 35-50; 55-70, with 40 informants per group), sex

(two groups: male and female, with 60 informants per group), and social class (three groups: upper middle class, lower middle class, working class). A total of 240 tokens were elicited7: 2 per clerk, 1 a more emphatic and 1 a less emphatic realization. Results of a GoldVARB analysis of the 240 token corpus revealed social class, age, sex, and speech

6 A good number of these subjects are the same speakers whose vowel systems we have chosen to analyze in this dissertation. 7 In Columbus, /s/ in /_tr/ clusters has three variably gradient realizations. There is the standard variant [s]; a retracted variant, [ʂ], typified by an /s/ that shows retroflexion without pronounced rounding; and a rounded and palatalized variant [ʃ], which is the more prototypical vernacular realization. In my initial coding of tokens for the RAS, I coded tokens simply as [s] or [ʃ], whereas for the sociolinguistic interview 47 environment to be significant factors conditioning innovative [ʃ] realization among store clerks (p<.01). Of these, the most relevant to the present discussion were social class and age. The youngest clerks (15-30 year olds) showed the strongest lead in [ʃ] realization overall (38%), regardless of speech environment, when compared to older clerks. This lead was less significant for the next oldest group (35-50), and was strongest over the oldest group (55-70), who realized [ʃ] only 8% of the time. For social class, the strongest

[ʃ] realizers were located within the working class (39%), followed closely by the lower middle class (26%), and then the upper middle class (8%).

The sociolinguistic interview data sample was comprised of 32 middle class speakers who were further subdivided by age, sex, and speaker's residential mobility history. Residential mobility history included two sub-factors: a) location in which a speaker was raised, and b) location in which a speaker currently lives. Mobility history was included given the pervasiveness of informants of all ages discussing urban growth and the impact of Columbus's post World War II annexation policy as a significant change factor in the community in our sociolinguistic interviews with them, as well as the pervasiveness of most speakers over the age of 40 also discussing the impact they felt the migration of European Americans from the central city area to suburban areas within

Columbus had on the community over the course of their lives. portion of the study, I included a distinction between [s], [ʂ], and [ʃ]. For the purposes of the, GoldVARB and cross-tabulation analyses presented in this discussion, however, the [ʂ] and [ʃ] distinction was collapsed to a single category—innovative [ʃ]. 48 Dividing the sample this way yielded a population of 16 males and 16 females, further subdivided into two age groupings: 38-69 and 19-32. Furthermore, this division yielded a second subdivision of 16 speakers who were raised in central city Columbus, and 16 speakers who were raised in the suburbs. The use of the age and location-raised subdivisions reflect two sociologically relevant facts about the speakers. First, the age groups reflect a division among speakers based on those who grew up in the community as Columbus's post-World War II annexation policy was beginning to be implemented versus those who grew up as it was well underway. Second, in terms of the patterns of migration shown by speakers for present-day location, they tend to reflect the general patterns of out-migration from the central city area to the suburbs among the larger

Columbus population during the period 1950-2000.

In a GoldVARB analysis, age, location in which the speaker was raised, and word environment (whether the /_tr/ cluster was word initial or word medial) were significant factors conditioning innovative [ʃ] realization (p<.01). Whereas location in which a speaker currently lives, phonetic segment immediately preceding, and immediately following, the /_tr/ cluster were not significant. Further recoding of the data, and additional cross-tabulation analyses to determine significant interaction effects between social factors revealed a significant effect for innovative [ʃ] realization by age-by location in which a speaker was raised (p<.002), but no significant effect for innovative

[ʃ] realization by-age-by-present-location. In addition, the interaction of innovative [ʃ]

49 realization by-age-by-speaker-by residential-mobility pattern was found to be inconclusive, given the inconsistent distribution of the speakers across categories, which did not allow an adequate statistical comparison to be made. The significant effect for innovative [ʃ] realization by age-by location in which a speaker was raised means there was a greater increase in [ʃ] realization among younger speakers born/raised in the suburbs than among older speakers.

In the (2007) analysis, we argued the results suggest two possible interpretations.

The first is fairly simple—the location in which the speaker now lives is not significant because it merely reflects the shift of a significant number of informants included in the

"Columbus column" based on where a speaker was raised to the "Suburb column" based on where the speaker now lives. The leveling resulting from such a shift is simply an artifact of the movement of the speakers from one column to another. A second interpretation is that, although the first interpretation is quite sensible from the perspective of "bean counting," it overlooks the social significance of what the numbers in the columns actually represent. That is, the movement of a substantial number of speakers who robustly realize more innovative /s/ variants (in particular, palatal [ʃ]) born in the central city area to areas of the Columbus metropolitan area to suburban areas, where speakers, historically, used less of the innovative variants before this significant movement of speakers occurred. As a result, the leveling of differences between central city and suburban residents found in the present day results versus the location raised

50 results may actually be reflective of larger patterns of change occurring throughout

Columbus during this time period, given that, as mentioned earlier, the distribution of the informants based on geographic location in the sample provides a fairly accurate representation of patterns of migration for the general population of residents who were raised in, and stayed in, the area throughout their lives.

That such a movement may be having such linguistic consequences for the community, was further suggested by additional facts that emerged when individual characteristics, attitudes, and linguistic behaviors of a number of the speakers who

“changed columns” were considered. Viewed as archetypes of Columbus community speakers, Durian argued that an understanding of these individuals' life histories provides a key to understanding how the transmission and diffusion of innovative /s/ variants may have occurred during the second half of the 20th century in Columbus. The migrating speakers showed some of the strongest patterns of innovative /s/ realization among their cohorts, specifically the palatal [ʃ] variant, and Durian posited that they likely do so as a result of being raised in more working-class areas. Having resettled later in their lives to the suburbs, they appeared to be at the forefront in advancing the spread of the innovative variants of /s/ within the larger community, and were ultimately found to show strong potential as transmission agents for innovative /s/ variants in Columbus. Of these speakers, the largest cluster is a group of several older women who moved to the suburbs after they completed college and have since raised families there.

51 Generally speaking, the background characteristics of this group of older female speakers was found to be comparable with speakers found to be at the forefront of linguistic innovation in other studies of urban communities in the US, such as New York,

New Orleans, Albany, Cincinnati (Labov, 2007), and Philadelphia (Labov, 2001). In

Labov’s study of Philadelphia, he found that women of similar socio-economic backgrounds who, in their adult lives, “crossed” class groups, tended to be what he refers to as “linguistic leaders of change” (cf, 323-411). Specifically, he found these speakers to be women who belonged to multiple social groups, were upper working class speakers who placed a strong emphasis on upward mobility, and who tended to generally have stronger attitudes of non-conformity to societal norms than other speakers. 8 Furthermore, he found that the leaders of change in Philadelphia are women who are socially active in the community, via daily contact with multiple groups in multiple locations throughout the Philadelphia metropolitan area and as individuals having contact with other individuals at multiple levels of the socio-economic hierarchy. Given their social mobility, these “leaders” served as transmission agents for the diffusion of linguistic innovation, such as the fronting of the nucleus of BOAT, BOOT, and SHOES, as well as the fronting of BAN, in the 5 neighborhoods he studied within the greater Philadelphia metropolitan area.

8 See also Eckert’s discussion of the role of “in betweens” in the transmission of innovations between “jocks” and “burnouts” in her study of Belton High in Detroit (2000:213-228). Attempting to tie his results to Eckert’s, Labov argues that the leaders of change in Philadelphia were likely much like Eckert’s non- conformist “in-betweens” at a younger age, hence the parity in findings regarding the role of “agents of change” in their studies (2001:512-518). 52 In Columbus, as was the case with Labov’s (2001) Philadelphia women, the social background of the female Columbusites who use innovative realizations of [ʃ] places them in a strong position to be linguistically influential. They grew up in areas in which working-class speech would have been pervasive. If the patterns revealed by the RAS are accurate, it is indeed possible that older speakers in these areas were using more innovative /s/ realizations when the women were growing up. Thus, during the language acquisition process, Columbus "leaders" would likely have been exposed to these innovative speakers, who served as their speech models.

After college, Columbus "leaders of change" also shared the experience of having moved geographically to more affluent areas in the suburbs, while at the same time, they moved upwards on the ladder of social mobility. This marks their behavior as "leaders of change" as similar to Labov’s (2001) leaders in that they are socially mobile, while it marks them in contrast to Labov’s leaders in that they are also geographically mobile. In present-day Columbus, the "leaders of change" are now securely situated in the middle class, both socially and geographically, placing them in an ideal position to be the transmitters of innovative linguistic variables such as innovative /s/ variants. As well- situated transmitters, they are in a position not only to influence other older speakers with whom they come into contact on a daily basis, but also younger speakers with whom they interact in their professional lives.

To conserve space, we have chosen not to focus in detail on the results of other

53 studies of linguistic features of Columbus speech which may show urban growth and population redistribution having a similar influence on the transmission and diffusion of innovative realization to the results for innovative /s/. However, it should be noted that

Durian (2004) and Dodsworth, et al.. (2006) found similar results concerning the influence of residential and social mobility on the vocalization of /l/. That is, location- raised was a statistically significant factor influencing variation, with Columbus-born speakers showing a significant lead over suburban-born speakers, but present location was non-significant, as a result of similar leveling of the difference between speaker groups for vocalized realizations of /l/. Although not discussed originally in Dodsworth, et al.. (2006), we used the same speakers in that study as I used in Durian (2007), and a reinvestigation of patterns of variation among those speakers I made recently reveals the same group who stood out as "leaders" in Durian (2007) also stand out as "leaders" in their use of vocalized /l/ variants.

In the Durian (2007) study, we also found evidence suggesting urban growth may be influencing speakers’ attitudes about culture and community in the larger area—that is, Columbus as a "place"—and that these attitudes may have an influence on speakers’ realizations of innovative /s/. Tentative evidence for this conclusion was presented in comments made by some speakers born and raised in Columbus proper during my sociolinguistic interviews with them. The majority of speakers in this group expressed pride about growing up in the “urban” part of Columbus. Unlike the speakers who were

54 born and raised in the suburbs, these speakers seemed more optimistic about the expansion of Metropolitan Columbus into surrounding areas, and they expressed a qualitatively higher level of satisfaction in the local arts, restaurant, and cultural activities

“scene” than suburban-born speakers. They also saw Columbus’s growth as “progress,” as for them, urban growth over the last 35 years has had the positive effect of making

Columbus become more like other urban cities (e.g., Chicago, New York, Philadelphia), and the majority were vocally dismissive of the popular stereotype of Columbus as a

“cowtown.”

The contrast in the more positive ideologies of these speakers regarding

Columbus versus the less positive ideologies of suburban-raised speakers suggests that urban-raised-speakers may be more receptive to using innovative /s/ variants than their suburban-raised counterparts, perhaps as a kind of marker of "urban affiliation."

Generally, speakers showing a higher level of “urban affiliation” (as indicated by their more positive attitudes towards Columbus and Columbus culture) showed a higher use of innovative variants than speakers who showed lower levels of affiliation, regardless of the location in which they presently live. Given these results, we decided to probe informants' perceptions of Columbus as a "place" to determine if their attitudes regarding

Columbus have any influence on their increased or decreased use of innovative variants of the vocalic variables we plan to study. These results are discussed in Chapter 7.

55 2.6 Speaker Social Evaluations of, and Attitudes Towards, Vowel Variation in Columbus

Coupled with our analysis of speaker perceptions of "place," in Chapter 7, we present an analysis of the influence of informants' social evaluations of the use of innovative variants in the community, to determine what role, if any, attitudes concerning language use in Columbus had on the increased or decreased use of more innovative variants by both middle and working class informants. In a variety of previous sociolinguistic studies, informants' social evaluation of the speech of others, as well as their own vowel pronunciations, have been found to be a significant factor impacting how the use of innovative variants of vowel variables diffuse throughout a community, as well as the social motivation for speakers' use or non-use of innovative linguistic variants

(e.g., Labov, [1966]/2006, 2001; Trudgill, 1986; Preston, 1996; Eckert, 2000). For instance, previous studies have found that a linguistic variable infused with notions of prestige or correctness may be used more frequently by middle-class informants (Labov,

[1966]/2006, 2001; Trudgill, 1986), while variables linked with social group affiliation may be used more by members of the social group for which the variable is seen as being a marker of in-group status (Eckert, 2000). In contrast, variables stigmatized as "non- standard," "incorrect," or "hickish" may be used less by middle-class speakers or by those who are not members of the social group being stigmatized (Labov, [1966]/2006, 2001;

Preston, 1996).

56 In previous perceptual dialectology studies of Ohio, findings suggest that some

Columbusites' social evaluations may be playing a role in some speakers' decreasing use of fronted realizations of BOUT, a feature marked as "hick" or "hillbilly" (Benson, 2005), and possibly the decreasing use of raised realizations of BAT and BAG, features marked as "Clevelandish" (Campbell-Kibler, et al., 2008), over time. Given these findings, we expect that we may find social evaluation influencing the use of vocalic variation in

Columbus. The findings of our analysis of speakers' social evaluations of linguistic variables will also be presented in Chapter 7.

2.7 Measuring the Social, Cultural, and Linguistic Influence of Demographic Change, Urban Growth, and Economic Change on Vowel Variation in Columbus

To gauge the impact of the social change forces we discussed in sections 2.3 and

2.4 on our speakers, and especially on their patterns of vowel variation and sound change in our study, and to also take into account the relevant findings from the previous linguistic studies of Columbus discussed in sections 2.5 and 2.6, we have chosen to focus on the influences of these various social, geographic, and economic forces in our data in several ways. The first is to select informants from both Blue Collar and White Collar backgrounds, so that we can gauge the possible influence of economic changes on patterns of vowel variation and language change in our data. We were also able to gauge the influence of differences in speaker social class backgrounds on the patterns of vowel variation shown by informants using this approach. We have also chosen to ask our

57 informants of both social class backgrounds direct questions pertaining to their perception of the influence of the geographic and demographic change forces discussed in Sections 2.3, 2.4, 2.5, and 2.6 on their lives and on the possible use of speech patterns by Columbusites our sociolinguistic surveys with informants.

The influence of informants' social class backgrounds will be discussed directly in each of the following chapters of our dissertation from this point forward: Chapter 3, 4, 5,

6, and 7 as a key social factor impacting the patterns of vowel variation and sound change we have observed. Meanwhile, the possible influence of economic changes to the community on informants' vowel realization patterns will be explored later, in Chapter 7.

At the same time, also in Chapter 7, we will explore the possible influence of speakers' perceptions of the influence of the geographic and demographic change forces discussed in Sections 2.3, 2.4, 2.5, and 2.6 on their lives and on their possible use of speech patterns.

58 Chapter 3: A New Perspective on the Columbus Vowel System, 1850-20009

3.1 Introduction

As discussed in Chapter 1, although a number of studies of patterns of vowel variation and change in the Columbus vowel system have been conducted since 1930

(Kurath, 1930), no previous study yet conducted has presented a detailed, generation-by- generation account of these patterns across the entire vowel system of multiple speakers born throughout the 20th Century. As well, no previous study has yet to compare, side- by-side, patterns of variation and change across the entire vowel system among European

American speakers of both Blue Collar and White Collar backgrounds born throughout this time period. In addition, no previous study has yet to compare, side-by-side, patterns among speakers born throughout this time period and belonging to these social class groups using instrumental analysis of normalized vowel formant data as an analysis technique.

In the next four chapters, such comparisons of the vowel systems of speakers born in the Columbus area during the years 1896-1991, a time period that will allow analysis of vowel variation and sound change in the community throughout the course of the 20th

Century, will be presented. Via the course of this discussion, the systems of 62 speakers—31 of Blue Collar and 31 of White Collar background—will be analyzed in

9 This chapter owes a debt of gratitude to Don Winford for encouraging us to think creatively about exploring the dialectal history of Columbus and to Erik Thomas for helping us to think practically about how to approach it. 59 significant detail. In this chapter, Chapter 3, we begin our analysis by first presenting an overview of all of the major and signficant patterns of variation which occur in the

Columbus vowel system among our speakers. This will be followed, in Chapters 4, 5, and

6, by a discussion of several of the major covariant vowel shift patterns that are to be found in Columbus speech across the generational groups we have chosen for analysis. In

Chapter 4, we will focus on the back diphthong parallel shift, a pattern of vowel shift in which the diphthongs SHOES, BOOT, and BOAT are each undergoing a process of nuclear fronting among speakers of all ages and social backgrounds in Columbus. As speakers get younger, the vowels are getting fronter, as we will also see discussed here in

Chapter 3. In Chapter 5, we will focus on the Third Dialect Shift, a chain shift which is often found in the vowel systems of middle class speakers born after 1945 in Columbus.

This shift involves the backing of the nucleus of BIT, BET, and BAT, linked to the backing of BOT, and possibly also the merger of BOT with BOUGHT, at least when the shift has been defined previously for Canada, as in, for instance, Labov, et al.. (2006) and

Clarke, et al.. (1995). In Chapter 6, we will focus on the complex series of changes that the Columbus short-a system has undergone, beginning in the vowel systems of speakers born in the mid-19th Century and continuing through today, as seen in the vowel systems of speakers born at least as late as 1991.

For the overview of vowel variation to be presented here in Chapter 3, material for our analysis and discussion will be drawn directly from the analysis of the 62

60 speakers we have recorded for analysis for this dissertation. As well, the results of earlier studies of speakers born around the same time period as our speakers will be discussed, so that the patterns found in our data can be compared with those of previous studies. To allow this comparison to be most successfully accomplished, speakers in all studies will be divided into similar speaker age groupings to these we have used in our study and compared. These age groups include speakers born in c. 1896-1913 (Generation 1), c.

1924-1937 (Generation 2), c. 1945-1967 (Generation 3), and c. 1976-1991 (Generation

4). These speakers represent 20th Century speech patterns since all were born either during, or just before the start of, the 20th Century, and all would have learned the bulk of their adult speaking habits during childhood years spent in Columbus during the 20th

Century.

To obtain a snapshot of an "initial state" of the vowel system that will allow trajectories of change to be determined for each of the vowel classes observed in our study among the 20th Century speakers, we will also present a detailed analysis of the full systems of two speakers born in the Central Ohio area in 1846 and 1852, respectively, who were interviewed for the Linguistic Atlas of the North Central States

[LANCS] (McDavid & Payne, 1976-1978) by fieldworker Cassil Reynard in 1933. These men's vowel systems allow us to get a sense of the patterns of vowel variation that likely typified 19th Century Columbus area speech around the middle of the 19th Century. (For sake of transparency, the two men born in 1846 and 1854 are not assigned a generation

61 group number like the 20th Century speakers are, although, technically, they belong to the second group of native-born speakers to live in the Columbus area following the settlement of Columbus in 1812.) To allow a chronological comparison and discussion of vowel variation and vocalic sound change in Columbus, the discussion of 19th Century patterns will be presented first, followed by discussion of the 20th Century patterns.

In the chapter that follows, we begin our analysis of major vowel variation and change patterns in Columbus by first presenting an overview of all of the significant patterns of variation to be found in the Columbus vowel system. This aspect of our analysis will incorporate both a descriptive analysis of our own 62-speaker data set and a discussion of previous studies of Columbus speech conducted by other researchers since

1930, of vowel variation among speakers born from 1846 through 1991. By descriptive here, we mean an analysis that incorporates normalized data and some reliance on comparisons of speaker patterns using either group or individual speaker mean normalized formant values drawn from our own data set. It will not include more extensive quantitative statistical analysis involving methods such as linear mixed effects regression, multiple regression, or logistic regression (such as one might find in an analysis relying on GoldVarb or VARBRUL). Instead, we will reserve the use of quantitative statistical analysis for later chapters, where we will explore several of the major vowel shift patterns described in this chapter in much greater detail.

We have chosen to use the qualitative descriptive approach in this chapter for

62 several reasons. First, most other studies of Columbus that have been conducted to date have been descriptive rather than quantitative in nature. Thus, comparisons across data sets are best made using a descriptive approach. Second, many studies of Columbus speech conducted previously are impressionistic rather than instrumental. Therefore, again, a descriptive comparison between data sets needs to be used here, since there is no logical way to compare impressionistic results with instrumental results quantitatively.

Third, even with studies that have been conducted using instrumental and/or quantitative methods of analysis, most analysts have used somewhat different methods of normalizing their formant data, analyzing the patterns of vowel variation exhibited, and analyzing the type of variation exhibited by a given vowel class. Hence, we find it best to utilize a descriptive approach to compare results even across these types of studies to ensure that accurate comparisons across data sets are made. In this case, a more general assessment will be more accurate, since doing so allows us some leeway in accounting for differences among studies. Fourth and finally, using a descriptive approach for this overview chapter allows us to cast our results within both the descriptive and quantitative traditions of studies of language variation and change, thus allowing us to demonstrate more fully how our results compare and contrast with the largest body of extant research possible.

In each section of this chapter, previous research (if available) on the variation shown by the vowel class or vowel classes under discussion will be discussed first. This

63 will then be followed by an analysis and discussion of the variation shown by the same vowel class or classes in our data set. For the analysis of data sets included in this chapter, mean formant values for all vowel classes measured at the group level are discussed, with generational cohort used as the basic unit of comparison. This means, for instance, that if a comparison of Blue Collar men's speech with White Collar men's speech is presented, and variation across all 4 generational groups is discussed, then the comparison contrasts Blue Collar men of generation 1 with White Collar men of generation 1, Blue Collar men of generation 2 with White Collar men of generation 2, and so on. If variation among men within a class group is discussed, then the comparison would be made between Blue Collar men of generation 1 and Blue Collar men of generation 2, and so on. The same approach also typifies comparisons of speakers by sex within or across class groups, as well. Thus, generation 1 Blue Collar men are compared with generation 1 Blue Collar women, and so on. For our data set, the mean formant values used for analysis and comparison are z-score normalized formant values.

Before we move on to presenting the overview and summary of the patterns of variation and change impacting the Columbus vowel system, we first need to discuss the specifics of our own data set, including how our speakers were selected and who the speakers are. In addition, we also need to briefly introduce the previous studies of vowel variation in Columbus and Central Ohio more generally that are used in our comparative analysis of vowel variation patterns presented in the remainder of this chapter.

64 3.2 Sample Characteristics of, and Recruitment Strategies for, the Present Study

For the majority of the analysis of our own data set, which was obtained for the purposes of this dissertation, the data to be analyzed are drawn from a 62-speaker corpus that we created from audio recordings drawn from several different sources. The author made the bulk of these recordings, either in 2004-2005, during a pilot study of Columbus vowel variation patterns, or in 2008-2009, during the main phase of fieldwork for this dissertation. A total of 44 of the 62 speakers analyzed in this study were drawn from these two sources.

In addition, recordings of 18 speakers were drawn from a variety of other projects, both linguistic and non-linguistic in nature, that were conducted in Columbus between 1970 and 2001. This includes 10 speakers recorded for earlier linguistic projects in Columbus: 8 speakers recorded by Ohio State University researchers for the Buckeye

Speech Corpus project in 2000 and 2001, 1 speaker recorded for the Dictionary of

American Regional English [DARE] project in 1970, and 1 speaker recorded by Raven

McDavid for the Linguistic Atlas of the North Central States [LANCS] project in 1956.

Beyond this, our corpus includes 1 speaker recorded by North Carolina State University professor Erik Thomas for a life history narrative project he conducted while a high school student living near Columbus in 1984, and 7 speakers recorded during several community history projects by researchers at the Ohio Historical Society during the years

1970-1983. The Ohio Historical Society tapes come from three projects: Boryczka's

65 "History of Ohio Labor Unions" project (see Boryczka and Cary, 1982 for more details);

Raphael's "History of the Jewish Community of Columbus" project (see Raphael, 1979 for more details); and a project on "Public Welfare in Ohio," conducted by various researchers, in 197010.

With the exception of the audio recordings made for the LANCS and DARE dialect studies, all audio recordings used for analysis in this project share several attributes in common, even though they come from a number of diverse sources. First, all were taken from one-on-one conversational narrative-oriented interviews, where the subject being interviewed discussed their life history in additional to a variety of topics that tended to yield significant portions of conversational speech from the subject.

Second, all contain at least 30 minutes of speech elicited specifically from the informant.

In fact, in most cases, as much as 60-90 minutes of speech was available for analysis.

Third, all included at least some discussion of changes to the Columbus community that have occurred over time that the subjects have perceived as being socially and historically significant. This final common strand is important for our analysis purposes here, as these discussions have provided us with a large corpus of discourse material that feeds the conversational analysis of informants' perceptions of Columbus and significant changes to the community during the 19th and 20th Centuries that we present in Chapter

8 of this dissertation.

10 As far as we could determine, no work from this final project was ever published. It may have resulted in an unpublished Ohio State University dissertation, but we were unable to determine this definitively when we were doing our research. 66 For the speakers we interviewed and recorded ourselves, speakers were recruited for the project in one of several ways. The first was through flyers that we distributed in undergraduate linguistics classes and to subjects as they completed interviews with us.

The second was via our linguistics Web site, which was housed at the time on the Ohio

State University Linguistics web server. The third was through the Ohio State University

Linguistics Department's "Talks and Experiments" subject pool. The fourth was through word of mouth recruitment, either by OSU faculty, graduate students, or undergraduates.

In these cases, the OSU person would discuss the project with a friend, and the friend would tell the person, or be asked by the person if, they wanted to participate. Most older subjects and some younger subjects were contacted via word of mouth or flyer recruitment, while many younger subjects were recruited using the "Talks and

Experiments" pool. Much of the time, subjects participated in the research without compensation. It was offered, but subjects would turn down the offer. In the cases where it was accepted, $10.00 was paid to subjects for their time. Speakers were interviewed for our interviews using the interview schedule provided in Appendix A.

The 62 speakers whose data are analyzed for the main portions of our analysis for this dissertation represent 4 generations of Columbusites born between 1896 and 1991.

All speakers were born, raised, and continue to live in Columbus, and all are native-

English speakers. As mentioned in 3.1, the age cohort divisions for these 4 groups are: a)

10 speakers born circa 1896-1913; b) 12 speakers born circa 1924-1937; c) 20 speakers

67 born circa 1945-1967; and d) 20 speakers born circa 1976-1991. The speakers are also nearly evenly divided by sex, with 32 men and 30 women being included in the study for analysis. In addition, the 62 speakers represent informants who were evenly selected from both White Collar and the Blue Collar speakers--with 31 speakers belonging to each group. The sample population characteristics for our study are presented in table 3.1.

3.3 The Assignment of Social Class Categories to Informants

As discussed in Chapter 1, we choose to use socioeconomic status as a social factor in our study, given the usefulness of this factor for helping researchers analyze the social and linguistic motivations of vowel variation in prior research in urban communties located through the United States (e.g., Labov, [1966]/2006, 2001; Shuy, et al, 1967; Wolfram, 1969; Lusk, 1976; Johnstone, et al., 2002; Boberg, 2005; Dodsworth

& Kohn, 2012). In many previous studies (e.g., Shuy, et al.., 1967; Trudgill, 1974;

Labov, 2001), socio-economic status has tended to be assigned to speakers using some

Generational Social Class Groups Group Blue Collar White Collar 1 (Born c. 1896- Men Women Men Women 1913) 3 2 3 2 2 (Born c. 1924- Men Women Men Women 1938 3 3 3 3 3 (Born c. 1945- Men Women Men Women 1968 5 5 5 5 4 (Born c. 1976- Men Women Men Women 1991 5 5 5 5 Table 3.1: Demographic Characteristics of the 62 Speakers Analyzed in the Present Study

68 kind of variable weighting scale and composite score scheme. Thus, occupation might be assigned a variable weight of 1-5, with lower paying occupations assigned a lower number, and higher occupations being assigned a higher number, and education level being determined by assigning a lower number to the completion of only high school and a higher number being assigned to the completion of college. Following the assignment of a weight to each criteria, the weighted scores are then combined and a composite score assigned to the informant to signify their class membership.

In our research, we did not find a strong need to assign weights to our criteria or an overall composite score for social class membership. We did plan to assign weights to the 5 criteria we originally used to select informants to fit the class categories of our study that attempted to capture both the speaker's current socio-economic status, as well as their "class background." These criteria included: a) education level; b) occupation; c) mean household income of the location in which the speaker was raised; d) mean household income of the location in which the speaker currently lives; e) occupation of the speaker’s mother and father. However, once we got into actually doing our research, we found most informants' socioeconomic status was fairly easy to determine using only my 5 criteria without weights, since all of our informants fell naturally into three broad groups:

a) the speaker's occupation clearly matches that of their parents, with both parents and the speaker's occupations being identifiable as either middle or working class, and the socio-economic status of the neighborhoods in which they grew up and now live is also well matched to occupation level. 69 b) the speaker's occupation is clearly identifiable as being different from their parents’, with the parents' occupations being clearly identifiable as working class, and the speaker's occupation being clearly identifiable as middle class. In addition, the socio-economic status of the neighborhood a speaker was raised in has clearly matched that of their parents' occupation level, while the neighbor in which the speaker now lives matches their own occupation level.

c) the speaker's occupation is clearly identifiable as being different from their parents’, with the parents' occupation levels being clearly identifiable as middle class, and the speaker's occupation being clearly identifiable as working class. The socio-economic status of the neighborhood a speaker was raised in has clearly matched that of their parents' occupation level, while the neighbor in which the speaker now lives matches their own occupation level.

The only attribute that did not always "match up" among speakers we interviewed is education level (higher education levels do not always equate with higher paying jobs, or vice versa). However, given that 4 out of 5 of the criterion have allowed for accurate classification of subjects, we chose ultimately to ignore any education level mismatches.

For the purposes of this dissertation, only speakers belonging to group a) were used for analysis. In the analysis itself, working class speakers are called Blue Collar, while middle class speakers are called White Collar. These terms are used since these are the terms speakers most often self-identified by when issues of social class came up in interviews that we conducted with them. The interview protocol we used to interview informants is included at Appendix A.

70 3.4 Methods of Acoustic Analysis of Vowel Formants for the Present Study

Recorded data used for instrumental analysis of vowel formants came to us initially in a variety of formats, given the series of different sources we used for material.

The Ohio Historical Society tapes, as well as those made for LANCS and DARE, were initially recorded on reel-to-reel tape, but were then transferred to digital WAV files at

44.1 kHz. These files were provided to us either by the OHS, DARE, or Erik Thomas, since Thomas made the digital transfer of the LANCS reel himself. In addition, Thomas provided us with a 44.1 kHz WAV file of his 1984 interview recording, which initially existed as a cassette tape recording. Files from the Buckeye Speech Corpus were provided to us via 44.1 kHz WAV files stored on computer servers at the Ohio State University

Department of Linguistics. For our own recordings, the 2004-2005 recordings were initially made using a Sony DAT recorder. The files were then transferred to 44.1 kHz

WAV files for analysis by us shortly after they were made. The 2008-2009 recordings were made using a Zoom H2 digital recorder. Here, the recordings were made directly to

44.1 kHz WAV files.

The acoustic analysis of all WAV files was conducted in the computer program

Praat (Boresma and Weenick, 2006), using a variable window of 10-14 LPC coefficients depending on the quality of the token. Vowels were first marked off in PRAAT by either the author or one of seven student interns (Erica Leigh Crawford, Jennifer Schumacher,

Melissa Reynard, Joshua Roush, Valerie Lyndak, Andrew Smith, or Emily Dorrian) who

71 worked on our project during 2005-2010. Boundaries were then checked by the author to ensure consistency of marking. Vowels were then measured by the author, aided by a custom-made formant extraction script in PRAAT, with adjustments made by hand as necessary. For the purposes of our analysis here, F1 and F2 were the only formants measured, and they were done so in Hz. 7 to 10 tokens of every vowel class we chose to study were measured. Measurements for both F1 and F2 were obtained from 13 distinct points in the duration of the vowel token (0%, 10%, 20%, 25%, 30%, 40%, 50%, 60%,

70%, 75%, 80%, 90%, and 100%), although for the purposes of the analysis we present throughout the dissertation, only 3 of these points are used. These points are 25% and

75%, which we use to represent the nucleus target and glide target for diphthongs, and

50%, which we use to represent the nucleus target of monophthongs. In addition, duration information for all vowels was also obtained, although this information is not used in the analyses we present in this dissertation. The more simplified approach to analysis we choose to use here (no duration and 1-2 measurement points depending on vowel type) in order to facilitate faster analyses of the vowels for the purposes of the dissertation. All measurement points were double checked by the author after measurements were taken, as even having just 3 points to check was quite time-consuming. 11.12

In addition, No more than 3 repetions of an lexical item were used, as per the best

11 Measurements were taken and double checked only by the author to ensure consistency of measurement across vowel tokens. 12 In a future study we plan to look at the spectral change information (which we can obtain from the 13 measurement points) and also the vowel duration information, to determine the impact that changes in spectral dynamics and vowel duration across generational groups may have on the vowel data.. 72 pratices for vowel analysis suggested by DiPaolo, et al. (2011) and Thomas (2001). As well, measurements were taken of tokens where no /r/ or /l/ proceded the vowel, and where no /r/, /l/, or nasals followed the vowel. The exceptions to this rule are the BAN class, where word final nasal tokens were used, the SCHOOL class, where word final /l/ tokens were used, and the BOARD class, where word final /r/ tokens were used. This was done to avoid vowel coloring complications which can occur with preceeding nasals, as well as preceding or following /r/ and /l/ (See DiPaolo, et al., 2011 for more details).

3.5 Speaker Vowel System Normalization for the Present Study

Throughout this dissertation, when instrumental data analysis is drawn upon, the formant measurements have been normalized. Normalized formant measurements are used so that speakers of different ages and different sexes can be compared. When natural formant data is produced by a speaker, differences in the formants they produce are influenced to some extent by biological differences between speakers, such as vocal tract length, height, and weight. Thus, to eliminate the influence of these differences on the formant data, we normalize the data using a procedure that allows us to attempt to eliminate this influence.

In our analysis, we use Lobanov's (1971) z-score method. However, because

Lobanov's method was originally made to deal with only monophthongal data, we have needed to implement a customized version of this procedure, to deal with diphthongal

73 data, since we have extensively analyzed those in our study. In addition, we wanted to better account for skewing effects that can be introduced into the data because of the extensive back vowel fronting shown by our G3 and especially G4 speakers.

To account for possible skewing effects we used corner vowels, including

SCHOOL for /uw/, to help establish a reliable outline of the general parameters of the vowel system--particularly the backest and highest points, as well as the lowest and frontest, as suggested by Watt & Fabricius (2002)--and also account for skewing causes by extensive SHOES- and BOOT-fronting. We then also using mid-front and mid-back points, so as to offset skewing caused by excessive BOAT-fronting.

The vowel classes we ultimately used for normalization are then, are the following (referred to as VG=Vowel Group in the normalization calculations below):

BEET BIT BAIT BET BAT (A composite average of BAT, BAG, and BAN) BOUT BOT BOUGHT BUT BOARD PUT SCHOOL

We calculated the Lobanov normalization as follows. F1 and F2 measurements for monophthongs were taken at 50% of the vowel's duration.. Measurements for

74 diphthongs were taken at two points. A point representing the nucleus was taken at 25% of the vowel's duration, while a point representing the glide was taken at 75% of the vowel's duration:

F1 Lobanov = F1 - μVG

δVG

F2 Lobanov = F2 - μVG

δVG

where μVG is the average formant frequency across the "Vowel Group" vowels listed above (calculated as the mean across individual vowel class means), and δVG is the standard deviation for the "Vowel Group" average (again across vowel class means).

Calculations were made using individual vowel system measurements, means, and standard deviations. VG measurements were taken at 50% of the vowel's duration.

3.6 Earlier Studies of Columbus Speech Used for Comparative Analysis

Across the extant studies of vowel system variation among European American

Central Ohioans born between 1846 and 1910 (Kurath, 1930; Markwardt, 1940, 1942;

Frazer, 1978; Nobbelin, 1980; Thomas, 2001, 2004) data pertaining to either the partial or entire vowel systems of six European American speakers have been made available.

75 Three of these speakers are men interviewed for LANCS (one from Columbus, born

1896; one from Worthington, born 1854; and one from Reynoldsburg, born 1846), one is a woman born in 1906 in Columbus and recorded for DARE, one is a woman born circa

1905-1910 in Columbus recorded by Hans Kurath for analysis in Kurath (1930), and one is a man born circa 1908 interviewed by Erik Thomas in 1984 and analyzed by Thomas

(2001). In these studies, 3 of the speakers (the 1896 born man interviewed for LANCS, the 1906 born woman interviewed for DARE, and the 1908 born man interviewed by

Thomas) were tape recorded during their interviews and each has received at least some instrumental or impressionistic analysis in previous studies. The other three speakers have only impressionistic transcripts in IPA notation available for analysis, made by the original interviewers. Of these speakers, all save the women analyzed by Kurath (1930) are the exact same speakers we have used for analysis in our own vowel analysis work for this dissertation.

It should be noted at the outset that no explicit reference to the social class backgrounds of the speakers analyzed was made in any of these earlier studies, although the 1896 born LANCS speaker, the speaker in Kurath (1930), and the 1906 born DARE speaker have all been categorized in the various studies as speakers of “cultivated” speech. In addition, personal communication with Thomas reveals the 1908 born speaker analyzed in Thomas (2001) is from a Blue Collar background. Therefore, in the discussion included in this chapter, this latter speaker’s patterns will be addressed

76 separately from the previously mentioned speakers’ patterns. In addition, the two LANCS men born in 1846 and 1852 will be discussed separately from the other speakers described above, since there is at least a 44 year gap in time between when they were born and the next closest speaker in age to them among the other speakers. A summary of the information provided in section 3.7 is included in table 3.2.

Turning to speakers born post-1920, the most intensive previous studies of patterns of full system vowel variation among European Americans in the greater

Columbus area are Labov, et al.. (2006), Thomas ([1989]/1993), Thomas (2001), Durian

(2008, 2009), Durian, et al.. (2010), Durian et al.. (2007), and Jacewicz, et al. (2011a,b).

Additional studies which supplement the findings of these studies for variation involving individual vowel classes include Hartman-Keiser, et al.. (1997) for the BAT and BAN classes; Dodsworth (2005) for the BAT and BOAT classes in Worthington, OH; Durian and Smith (2005) for the BOAT and BOUT classes; Durian (2011) for the BOUT class; and Ash (2006) for the BOT and BOUGHT classes. In contrast to the studies of vowel variation conducted among speakers born 1846-1910, with the exception of Labov, et al.. on in each of these studies. However, personal communication with Thomas has revealed

(2006), Thomas ([1989]/1993; 2001), and Ash (2006), social class is explicitly focused that the speakers in Thomas ([1989]/1993; 2001) are of Blue Collar backgrounds, so this addtional information is incorporated in our discussion even though it is not discussed explicitly in the original study. Given the usefulness of this information to the objectives

77 Study Type of Analysis Social Class Number of Informant Birth Group Informants Year Ranges Marckwardt (1940, Impressionistic Not specified 3 (All interviewed Male speakers 1942) (BOT and for LANCS) born 1846, 1852, BOUGHT) and 1896 Nobbelin (1980) Impressionistic Not specified 3 (All interviewed Male speakers (BOT and for LANCS) born 1846, 1852, BOUGHT) and 1896 Kurath (1930) Impressionistic Not specified 1 (Female speaker) 1 Female born c. (Whole Vowel ("Cultivated 1905-1910 System) Speech") Frazer (1978) Impressionistic Not specified 1 from Columbus 1 Female speaker (BOUT, ("Cultivated (28 from Ohio born 1906 SHOES/BOOT, Speech") more generally; All BOUGHT, BIT, from DARE) BET, BITE/BIDE) Thomas (2004) Instrumental Not specified 2 (1 speaker each 1 Male born 1896 (Some) ("Cultivated from LANCS and (LANCS) Speech") DARE) 1 Female born 1906 (DARE) Thomas (2001) Instrumental Not Specified 1 (Male speaker) 1 Male born c. (Whole Vowel (Blue Collar-- 1908 System) Thomas, P.C.)

Table 3.2: Previous (Mostly) Published Columbus Vowel Variation Studies of Speakers Born c. 1846-1908

of my dissertation, I have chosen to include comments relevant to class-based variation when possible and where applicable. However, given the lack of representation for Blue

Collar speakers born between 1908 and 1966 in previous studies, I do not comment on

Blue Collar patterns of vowel variation in previous studies for speakers born in

Generations 2 or 3. In addition, since Durian (2008b, 2009, 2011) all draw on the same speaker population, as well as Jacewicz (2011a,b), each of these sets is dealt with as simply one study, rather than individual studies. They will be referred to Century of

Language Change in Columbus (CLCC) Pilot for the Durian studies, and the Jacewicz,

Fox, and Salmons Project (JFSP) for Jacewicz, et al. (2011a,b). Speaker population 78 counts, social class background characteristics (when known), and other relevant characteristics of each of the studies mentioned above, are included in table 3.3.

Throughout this discussion, we will refer only to the results of the full 62 speaker analysis we completed for the dissertation, rather than the results of our earlier CLCC

Pilot results in the textual descriptions of results compared below, since our dissertation results supersede the pilot results. Meanwhile, given the difference in the nature of the data collected for the Jacewicz et al.. project versus others, where full vowel results for multiple generations of speakers are available for comparison, but the speech analyzed is read citation form rather than natural conversation (all other studies in this section are either conversational or mixed conversation and citation form speech), we will present a side-by-side comparison of results for our G2, G3, and G4 speakers with their 3 generations of speakers in a separate section than all of the other studies we compare and contrast our results with. This comparison is presented in section 3.5.5.

From this point on, the reader should note that, unless otherwise specifically stated, all patterns of linguistic variation discussed refer to patterns found among speakers living within the greater Columbus metropolitan area. (That is, either Columbus

“proper” or nearby surrounding suburbs, such as, for example, Worthington, Dublin,

Hilliard, Upper Arlington, Grove City, Bexley, etc.). In addition, to conserve space, we only discuss vowel classes found to be undergoing significant patterns of variation in the discussion below. Thus, if a vowel class is not discussed, the reader should assume that

79 Study Type of Analysis Social Class Number of Informant Birth Group Informants Year Ranges Thomas Impressionistic Blue collar 32 (16 European Speakers born ([1989]/1993) (Full Vowel Americans/16 1966-1971/ System) African Americans) Thomas (2001) Hybrid Blue collar 2 (both with full General focus on (Impressionistic + system plots) speakers born Instrumental) 1908-1971 (Full Vowel System) Labov, Ash & Instrumental (Full Not specified 1 with full vowel 1 Male, born Boberg (2006) Vowel System) plot; 6 in map- 1920; 5 Females, based results; 16 born 1930-1973 included in Ash (2006) follow up study Durian, Dodsworth Instrumental (Full Blue Collar 4 European Speakers born & Schumacher Vowel System) Americans/4 post-1940 in 2 (2010) African Americans generation groups in general analysis (c. 1950-1960 and (with full system c. 1975-1980) plots) 15 European Americans in BOAT-specific analysis Durian, Instrumental (Full White Collar 4 European Speakers born Schumacher & Vowel System) Americans/4 post-1940 in 2 Reynard (2007, African Americans generation groups Unpublished (all with full (c. 1950-1960 and Squibb) system plots) c. 1978-1985) Century of Instrumental (Full Blue and White 22 European Speakers born in 4 Language Change Vowel System) Collar Americans (11 generation groups in Columbus Blue Collar/11 (c. 1896-1908; c. (CLCC) Pilot White Collar) 1924-1938; c. 1945-1967; c. (Durian 2008b, 1976-1990) 2009, 2011) Jacewicz (2011, Instrumental (Full White Collar 78 European Speakers born in 3 a,b) (JFSP) Vowel System) Americans generation groups (c. 1919-1940; c. 1958-1973; c. 1994-1998)

Table 3.3: Previous (Mostly) Published Columbus Vowel Variation Studies of Speakes Born c. 1920-1987

80 Hartman-Keiser, et Instrumental (Only White Collar and 2 speakers (1 of White Collar al.. (1997) BAT and BAN Blue collar each background) speaker born circa studied) 1950; Blue collar speaker born circa 1970 Dodsworth (2005, Instrumental (Only White Collar 21 speakers Speakers born Unpublished BOAT, BOT, BAT, circa 1945-1987 Dissertation) and BAN) (10 Males, 11 Females) Durian & Smith Instrumental (Only White Collar 32 speakers Speakers born (2005) BOAT and BOUT) 1945-1986 Ash (2006) Instrumental (Only Not specified 16 speakers Speakers born BOT and circa 1920-1980 (6 BOUGHT related Males; 10 classes and Females) BIN/BEN)

Table 3.3 (Cont.): Previous Published Columbus Vowel Variation Studies of Speakes Born c. 1920-1987

vowel class has not been found to show a significant pattern of variation among speakers of the given birth year cohort, class group, or sex group under discussion.

Finally, although several recent studies have also focused on ethnicity via the study of European and African American vowel systems (Thomas, [1989]/1993; Durian, et al.., 2007; Durian, et al., 2010), because our present study is focused exclusively on vowel variation patterns among European Americans, we have chosen to exclude findings pertaining to patterns among African American Columbusites in the discussion below. (For patterns of vowel variation in African American English, the reader is directed to Thomas ([1989]/1993), Durian, et al. (2010), and Durian, et al. (2007).)

3.7 Patterns of 19th Century Vowel Variation

The first set of vowel systems we discuss here are those of the members of the 81 oldest generational cohort from whom we have been able to obtain data: two men who were born in the Central Ohio area in 1846 and 1852. As mentioned earlier in this chapter, these vowel systems were analyzed in our study as a way of determining an

“initial state” for the Columbus vowel system. By initial state here, we mean a beginning or base line state for each of the primary vowels of interest in our study from which we can determine patterns of either stability or change in the vowels of each generational group born after the initial group.

According to McDavid & Payne (1976-1978), the two men were born in

Reynoldsburg and Worthington, two cities located quite near to Columbus. The

Reynoldsburg-born man was born 1846, and his father's family came from Pennsylvania, while his mother's family came from Vermont. He was a landlord of fames and houses in his working life and also was a Civil War veteran. He was 87 at the time of Reynard's interview and is listed in McDavid & Payne (1976-1978) as LANCS028A. The

Worthington-born man's father's family came from Virginia, while his mother's family origins are not discussed. He was born in 1854, was 79 at the time of interview, and was a farmer. He is listed in McDavid & Payne (1976-1978) as LANCS28B13.

As was also mentioned earlier, data for these men are entirely impressionistic,

13 Note that in Davis (1948), the speakers are listed as 33a and 33b. For reasons that are not clear to us, although LANCS28a should be 33a and LANCS28b should be 33b, the information for the two men is reversed in Davis's listing. That is, the Worthington born man in McDavid & Payne (1976-1978) is listed as being from Reynoldsburg, while the Reynoldsburg born man is listed as being from Worthington. This may also be the case in Markwardt (1940, 1942), since Davis appears to use Markwardt's notes for his categorization, his being a student of Markwardt, as well as Hans Kurath, at the time. We are unable to verify this, however, since Markwardt's articles do not provide informant listing information. 82 having been transcribed phonetically by Cassil Reynard in 1933 during interviews for

LANCS. As such, the data are not fully comparable with the instrumental data we have analyzed from the speakers born between 1896 and 1991 in later sections of this dissertation. However, a broad comparison between these patterns can actually be made once we take into account how the narrow transcriptions made for at least some words by

Reynard likely match up with the modern instrumental measurements we will be drawing on for our analysis and discussion of 20th Century patterns of vowel realization.

At this point, we should also mention some relevant caveats about our use of this data set here, both in our own work, but also that of other analysts, such as Nobbelin,

Marckwardt, or Thomas, who have also used this source data to varying degrees in previous studies. First, it should be pointed out that it represents the phonetic judgments of only one field worker. Although Reynard was trained to use the proper IPA notation and notational conventions of narrow transcription that were widely in use during the

Linguistic Atlas of the United States and Canada [LAUSC] project (Kurath, et al..,

1939:Chapter 4), he may have been subject to converting some of the vowel variation he heard to the norms of his own vowel system. In particular, he was noted by Kurath to be less adept at transcribing low back vowels than other researchers (Kurath, et al..,

1939:51-52). A more common criticism leveled not only at his work, but in fact, many researchers of the era is that he may also have been less aware of vowel variation trends that were either new or else less commonly known about at the time than those which

83 were more widely known about (e.g., Chambers & Trudgill, 1998; Thomas, 2006; Labov,

[1966]/2006, Johnson, 1998). Although this is certainly possible, information we have had access to for Ohio more generally suggests to us that, although Reynard may not have been as detailed as other fieldworkers, such as McDavid or Guy Lowman, he does seem to generally capture most of the dialect trends of others who transcribed speakers born around the same period in other parts of Central and Southern Ohio (McDavid &

Payne, 1976-1978; Marckwardt, 1940, 1942; Nobbelin, 1980).

Regardless of these criticisms, we have chosen to use this data because it is all that is currently available. As Labov (1994) and Janda & Joseph (2003) both note, work in historical reconstruction (such as we do here) often involves the researcher having to make the best use he or she can from imperfect data. In is in that spirit, and with an awareness of the problems and caveats noted above, that we present, first, the results of previous studies using Reynard's notes, and then our historical reanalysis of the material.

3.7.1 Central Ohio Vowel Systems among Speakers Born 1846-1854, as Analyzed in Previous Studies

For the two Central Ohio men interviewed for LANCS, born 1846 in

Reynoldsburg and 1854 in Worthington, the results of four analyses (Marckwardt, 1940,

1942; Nobbelin, 1980; Thomas, 2004) of vocalic variation patterns noted in the raw IPA field notes reveals them showing the following features:

• Mild raising of the nucleus of BIT and PUT before /ʃ/ in words such as "dish" and "fish" or "push" and "bushels"

84 • Reduction of the offglide of BIDE in words such as "five" • Near merger of BOT and BOUGHT before /t/ • Partial merger of the NORTH and FORCE classes • Mild raising and lengthening of /ae/ before voiced stops (BAD) and nasals (BAN)

For the patterns noted above, the only strong difference noted between the speakers is that the 1846 born man shows a stronger tendency toward close realizations of

BOT and BOUGHT before /t/ than the 1854 born man. One additional difference, noted by Thomas (2004), is that the 1854 born man also shows some glide reduction for BITE, whereas the 1846 born man does not. Both men show the same tendency toward glide reduction for BIDE, however.

Beyond these patterns of vowel variation, it has been previously unknown if other variation in vowel class realization might be found among these men. This is because a number of vowel classes for which data were obtained by Reynard have not been analyzed, at least as far as we are aware, in any previous published or unpublished study.

These classes include SHOES, BOOT, BUT, BAIT, BET, BIT, and BEET. An analysis of these classes is presented in the next subsection.

3.7.2 Central Ohio Vowel Systems among Speakers Born 1846-1854, as Based on Our Reanalysis of LANCS Field Records

Given the lack of analysis of SHOES, BOOT, BUT, BAIT, BET, BIT, and BEET in these previous studies, we decided to go back and revisit the raw data ourselves, to

85 determine if any patterns of variation are to be found among these classes in this early speaker data. To do so, we obtained copies of Reynard's actual field notes via the LANCS microfilm series made available via McDavid & Payne (1976-1978). Specifically, we obtained copies of all of the available field records for Reynard's interviews. The copies were made from the LANCS microfilm series, available as McDavid & Payne (1976-

1978), as hardcopy print outs of the microfilms. These copies were then scanned as a series of PDF files and saved to a series of computer hard drives, so that the notes would not be lost or damaged. In addition, other materials from the McDavid & Payne (1976-

1978) films, such as text discussions of speaker's family histories and maps of informants locations within Ohio, were also saved to PDF at this time and stored on the hard drives.

We then went through and systematically attempted to locate all transcribed realizations of all tokens of each of the major vowel classes for both of the Central Ohio men. We then recorded the narrow IPA symbols used by Reynard on a separate tally sheet, and then tallied up the tokens corresponding to each symbol used for a given vowel, so that the full range of variant pronunciations recorded by Reynard for the informant could be observed, analyzed, and counted.

The results of our own analysis are provided in table 3.4. In this table, we begin at the back of the vowel system, working our way down from the highest vowels to the lowest, we then move forward, beginning with the low front vowels and working our way back up to the highest vowels. In the cases where we chose to separate vowel subclasses,

86 LANCS 28A (b. 1846) LANCS 28B (b. 1854) BOOT (1) ʊ (1: hoops) ʊ (1: hoops) SHOES (9) ɪ>u (2: due_x, dues) juu (1: music) juu (1: music) uu (6: two_x, suit, uu (4: new_s, suit, student, tube, dues, student, tube) due_x) ʊu (1: two_x) u (2: coop, new_s) ʊ (1: coop) PUT (3) ɤ (1: butcher) uV (1: push) ʊ∧ (1: bushels) ʊ (2: butcher, bushels) ʊ (1: push) BUT (7) ʌ (4: soot, puss, musk, ʊV (1: musk) mush) ʌ (6: soot, puss, shut, ʌ >∧ (1: bug, toward ʊ) mush, husks, bug) ʌV (1: husks) ɛ (1: shut) BOAT (10) o (2: z_over, hotel) o (2: z_over, hotel) oʊ (8: ago_x, sofa, posts, oʊ (7: ago_x, sofa, posts, coat, post, coast, x_oats, coat, post, coast, x_oats, toad) toad) o<ʊ (1: toad) FORCE (4) oV (4: boar, mourning, oV (4: boar, mourning, four, hoarse) four, hoarse) BOUGHT (14/13) ɔ (2: hog, wash) ɔ (4: hog, sausage, ɔV (2: dog, cough) cough, hogs) ɒ⋅∧ (1: t_off) ɔ⋅ (1: dog) ɒ⋅ (1: fog) ɔə (1: fog) V ɒ (7: t_office, sausage, ɔ (4: wash, t_off, foggy, hogs, foggy, daughter, taught) taught, caught) ɒ⋅(1: daughter) ɒ< (1:water) ɒ (2: t_office, water)

DAWN (1/2) ɒ (1: long) ɔ (1: long) ɔV(ɔVə) (1: launch) BALL NONE NONE NORTH (4) ɔ (4: x_orphan, morning, ɔ (4: x_orphan, morning, forty, horse) forty, horse) BOT (5/6) ɔə (1: god [curse]) ɔV (1: hospital) ɒ⋅ (1: god [name]) ɒə (1: pot) ɒə (1: watch) ɒ<⋅ (1: god [name]) ɒ (1: hospital) ɒ< (1: god [curse]) ɑ (1: f_oxen) ɑ (2: f_oxcn, watch) Table 3.4: Central Ohio Vowel Variation Noted in Reynard's 1933 LANCS Field Notes 87 LANCS 28A (b. 1846) LANCS 28B (b. 1854) (1) ɔV(ɒ) ɔV(ɒ) (1) ɒ ɒ< (1) a A BITE (2/3) ɐVɪ (2:twice, wife) ɐVɪ (1:white) aɛ (2:twice, wife)

BIDE (2) ɐVɪ (1:died) ɐVɪ (1:died) aɛ (1:five) aɛ (1:five) BOUT (7) aʊ (3:house, house, aʊ (4:house, house, n_out) n_out, how_f)) ao (4: mouth, cow, ao (3: mouth, cow, θ_out θ_out, how_f) ) BAT (3) æ⋅(1:mad) æ (3: mad, Saturday, æ (2: Saturday, catty) catty) BATH (9/9) æ⋅ (2: mass, calf) æə (1:mass) æ (7: half, basket, past, æ⋅ (2: calf, class) pasture, ashes, class, æ (6: half, basket, past, casket ) pasture, ashes, casket) BAG (2/3) æ (2: sack, actress) æ (3: bag, sack, actress) BAN (5/6) æə̃ (1:dance) æ⋅̃ (2: aunt, ma'am) æ⋅̃ (1:ram) æ⋅(1:ram) æ ̃ (1:aunt) æ̃ (1: dance) æ (2:handy, hammer) æ (2:handy, hammer) BET (7) ɛ (7: seven, second, ɛ (7: seven, second, chest, sweat, catch, deaf, chest, sweat, catch, deaf, nephew) nephew) BEN (5/5) ɛ (5:them, ten, fence, ɛ (5:them, ten, fence, fence, hen) fence, hen) BAIT (6) eɪ (3: ways, vase, case) eɪ (2:vase, eggs) eVɪ (2:eight, paper) eVɪ (2:ways, case) eV(1: eggs) eV ɪ< (1:eight) e (1:paper) BIT (6) i V(1: dishes) i V (1: fish) ɪ⋅ (1: six) ɪ (4: fifth, dishes, whip, ɪ (4: fifth, fish, whip, wish) wish) <ɪ (1: six) BIN (1) ɪ (1:been) ɪ (1:been) BEET (3/3/2) i (2: cheese, sweet) i (2: cheese, sweet) i ⋅ (1:cheese) < i ⋅ (1: cheese) Table 3.4 (Cont): Central Ohio Vowel Variation Noted in Reynard's 1933 LANCS Field Notes

88 all members of a given subclass are grouped together, based on the historical word class to which they belong for modern English, as presented in sources such as Labov, et al.

(2006), Wells (1982), or Kurath (1964). This is the case even with vowel classes showing tendencies toward merger, such as NORTH and FORCE. (Thus, NORTH is grouped with

BOUGHT and FORCE is grouped with BOAT.)

Our analysis reveals that the speakers do show some additional variation beyond what earlier studies have discussed, as well as some similar patterns to what has been discussed. Like the earlier studies, we find mild raising of the nucleus of BIT and PUT before /ʃ/ in words such as "dish" and "fish" or "push" and "bushels", reduction of the offglide of BIDE in words such as "five," and near merger of BOT and BOUGHT before

/t/. This should not be surprising, since we, just like the previous authors, are drawing from the same data source for our analyses. Turning to the added data for additional vowel classes we have looked at, we also find the following additional variation patterns:

• Evidence of some lowering of the nucleus of BAIT • Some lowering of the glide of BOUT (but no fronting of the nucleus) • Variable lengthening of the glide (diphthongization) of BOAT • Some lowering of the nucleus of BUT and PUT (in environments other than before /ʃ/) • The beginnings of mild centralization/fronting of the nucleus of SHOES

In addition, we also find mild raising of /ae/ not only before nasals and /d/, as in previous studies, but also before voiceless fricatives, as suggested by Reynard's indication of lengthening before /f/ in "calf" and /s/ in "class" and "mass" in tokens spoken by both

89 men. Meanwhile, based on our analysis of the other non-previously discussed vowel classes, we find that BIT, BIN, BET, and BEN do not show notable patterns of variation,

Vowel 19th Century Males Class (b. 1846 and 1854) SHOES Mild Centralization BOOT

BOAT Mild Diphthongization

BOUT

PUT Mild raising before /ʃ/

BUT BOUGHT Close realization before /t/ BOT DAWN DON BALL DOLL BITE Mild glide reduction BIDE Mild glide reduction

BAT Tensing/Raising before voiceless fricatives and /d/ BAN Tensing/Raising BET BIT Mild raising before /ʃ/ BEN BIN BAIT BEET

Table 3.5: Summary of 19th Century Central Ohio Vowel Variation Patterns (Empty boxes indicate vowel classes for which no significant patterns of vowel variation are noted)

90 other than the previously mentioned raising of BIT before /ʃ/ found among both speakers.

Finally, we also do find some evidence of some variation in length for BEET, but given that only one token of "cheese" for each speaker shows this, it's not clear whether this represents a significant pattern of variation or simply a change in stress emphasis placed on the realization of a repeat instance of a token during the course of an interview. The close proximity of these tokens to the initially recorded instance of "cheese" for both speakers in both cases suggests the latter may actually be the case.

A summary of all patterns of variation discussed in this section is included in

Table 3.5. This information is used to define the "initial state" of the Columbus vowel system as we are defining it for the purposes of our analysis of 20th Century vowel variation and sound change trends throughout the remainder of this dissertation.

3.8 Patterns of 20th Century Vowel Variation

In this section, we will discuss patterns of vowel variation and sound change among speakers in all 4 of the generational groups we have chosen to focus on extensively in this study. To remind the reader, these generational groups include speakers born c. 1896-1913 (Generation 1, referred to as G1 from this point onward), c.

1924-1937 (Generation 2 or G2), c. 1945-1967 (Generation 3 or G3), and c. 1976-1991

(Generation 4 or G4). As with the previous section, we present the results of previous studies of Columbus speech that have focused on speakers who are members of each of

91 these age cohorts. In this section, the results of any relevant previous studies are presented first, followed by a discussion of the results of our analysis of our dissertation data set. In addition, results are often broken down further by which social class group and which sex group makes use of which patterns, since both of these speaker characteristics typically influences the type of variation shown by a vowel class.

It is also important to note that vowel plots for all 62 of our speakers that were born during the time periods noted above and analyzed for this section are available as

Appendix B of this dissertation. The reader may find it useful to refer to these plots for a more detailed understanding of the patterns discussed. Often, the visual details provided by the plot can make the pattern of variation described much more transparent.

Also like the previous section, we present results as broken down by vowel class.

In addition, to make use of the "initial state" information we were able to develop as a result of the discussion presented in section 3.7, we will also make use of three additional grouping distinctions to group and discuss the results presented in this section. These distinctions are a three-category system for describing innovative and non-innovative patterns of sound change in the data. These categories include recessive, innovative, and continuous and incremental patterns of change in the data. Recessive patterns are those that initially appear as an innovation in the community among speakers born at an earlier point in time than later groups of speakers, but then show decreasing use, or cease to continue appearing at all, in vowel systems as an innovation among speakers born at a

92 later point in time. In the context of this discussion, this term is used to refer to patterns of vowel variation that show recession, or decreasing use over time, in 20th Century vowel systems relative to 19th Century vowel systems. Conversely, innovative patterns are those that appear among speakers belonging to a particular speaker age cohort but did not appear as an innovative pattern in use among the speakers belonging to the group who immediately precede that cohort. These patterns show continued use in the speaker cohort who immediately follows the innovating cohort, and this continuation is usually accompanied by increasing use of the feature among each younger cohort of speakers

(incrementation). A continuous and incremental feature is one that appears in all speaker vowel systems across age groups in our study.14 Like the innovation, it also shows increasing use among increasingly younger speaker groups. Unlike other innovations discussed in the bulk of this chapter, however, its use can also be traced back to variation originally shown in 19th Century vowel systems, an attribute not shared by innovations as formally defined for our descriptive purposes here.

3.8.1 Recessive Features

Turning to patterns of vowel variation and change among informants born between 1896 and 1991 and found in previous studies of Columbus vowel variation, shown in summary form in table 3.6, previous studies (Kurath, 1930; Frazer, 1978;

14 . For our purposes, continuous features are always incremental, since a non-incremental continuous feature would be classified instead as a recessive feature based on the definitions we use here. 93 Vowel Generation 1 Generation 2 Generation 3 Generation 4 Class (b. 1896-1910) (b. 1925-1945) (b. 1945-1965) (b. 1965-1987) SHOES Mild Centralization Strong Centralization Fronting Fronting BOOT Mild Centralization Centralization/ Fronting Some fronting BOAT Mild Centralization Centralization Centralization/ Fronting Some fronting BOUT Mild Centralization Centralization Fronting Some retraction of the nucleus PUT Mild raising before /ʃ/ Some fronting Some fronting

BUT Some fronting Some fronting BOUGHT Close realization Close before /t/ Close or merged Close or merged before before /t/ before /t/ and /k/ /t/ and /k/ BOT DAWN Close realization Close or merged Close or merged DON BALL Close or merged Close or merged DOLL BITE Mild glide reduction Mild glide reduction BIDE Mild glide reduction Mild glide reduction Glide reduction mostly Glide reduction mostly found among some found among some Blue Collar speakers Blue Collar speakers BAT Raising/lengthening Retraction BAN Raising/lengthening Mild raising Mild raising Mild raising BET BIT Mild raising before /ʃ/ BEN Close or merged Close or merged Close or merged BIN BAIT BEET Table 3.6: Vowel Classes Undergoing Variation among Columbus Speakers Born 1896-1987 as Analyzed in Previous Studies (Note: Shaded cells indicate no data are currently available; Blank cells indicate no significant variation was found involving that class)

Thomas, 2001, 2004; Labov, et al.. 2006; Durian, et al.., to appear; Durian, et al.., 2007) indicate that several of the features noted in the speech of the mid-19th century born

Central Ohio speakers were recessive in the vowel systems of speakers born after 1896.

94 These include the raising of PUT and BIT before /ʃ/, and glide reduction of BITE and

BIDE. Only one speaker, a man born in 1896, shows PUT raising, and no speakers show

BIT-raising, while glide reduction of BITE is consistently found only among speakers born before World War II, and is more robust among men (Thomas, 2001, 2004; Labov, et al.. 2006; Durian, et al.. 2007). For BIDE, speakers born before World War II generally show stronger tendencies toward glide reduction than speakers born after World War II, with men again showing stronger tendencies towars reduction than women. However, the results of Thomas ([1989]/1993) and Durian, et al. (to appear) suggest glide reduction may only be recessive in White Collar speech, as a number of the post-1966 born Blue

Collar speakers analyzed in their studies continue to show strong reduction tendencies

In our speaker data for the present study, summarized in tables 3.7 (White Collar speakers) and 3.8 (Blue Collar speakers), we find confirmation for all of the above trends described in previous studies for speakers of similar age in our data, as well as two additional trends. These additional trends include raising of BET and BAT before /ʃ/, in addition to raising of BIT and PUT. In sum, the recessive trends we find in our 62 speaker data set are:

• Raising of PUT, BIT, BET, and BAT before /ʃ/ among G1 and G2 speakers • Some glide reduction of BITE among G1 and G2 speakers • Some glide reduction of BIDE among G1 and G2 white collar speakers

For PUT and BIT before /ʃ/, we find several Blue Collar speakers born before 95 1945 who show raising tendencies in words such as "push," "bush," and "fish," but no evidence of the raising of either vowel before /ʃ/ among speakers born after 1945. In particular, we find these raised realizations used by one White Collar man born in 1896

(the LANCS speaker, who is the same speaker reported to show the raised realization in previous studies), two Blue Collar men born in 1898 and 1913, and one Blue Collar woman born in 1930. These same speakers also produced raised realizations of BET and

BAT before /ʃ/ in words like "mesh" and "cash," a trend not discussed specifically for

Columbus in previous studies.15 Among speakers born after 1945, several Blue Collar men and women born in G3 (c 1945-1967) indicated overt awareness of raised pronunciations of BIT and PUT, as well as BET and BAT, as spoken by both older speakers and others in their own age group, but none claimed to produce them, and none were observed by us, in their own speech.

For BITE and BIDE, we find generally the same trends as previous studies (e.g.,

Thomas, [1989]/1993; Thomas, 2001, 2004; Labov, et al.. 2006; Durian, et al.. 2007).

That is, for BITE, we find glide reduction of BITE to be consistently found only among speakers born before World War II, with men showing a stronger tendency towards reduction than women. The new observation in our data is that we find this trend among both Blue Collar and White Collar speakers. For BIDE, as in previous studies, we find speakers born before World War II generally showing stronger tendencies toward glide

15 In addition, at least two more men--one white collar and one blue collar--born in G2 (c. 1924-1937), also produced raised realizations of BAT and BET before /ʃ/ during our recorded interviews with them, although they did not produce raised realizations of BIT or PUT. 96 Patterns of 20th Century Vowel Variation among White Collar Speakers in Our 62 Speaker Corpus

Vowel Generation 1 Generation 2 Generation 3 Generation 4 Class (b. 1896-1913) (b. 1924-1937) (b. 1945-1967) (b. 1976-1991) SHOES Mild Centralization Strong Centralization Fronting Fronting BOOT Mild Centralization Centralization/ Fronting Some fronting BOAT Mild Centralization Centralization Centralization/ Fronting Some fronting BOUT Mild Centralization Centralization Fronting Some retraction of the nucleus PUT Mild raising before /ʃ/ Some fronting Fronting Fronting

BUT Some fronting Some fronting Fronting Fronting BOUGHT Close realization Close before /t/ Close or merged Close or merged before before /t/ before /t/ and /k/ /t/ and /k/ BOT DAWN Close realization Close or merged Close or merged DON BALL Close or merged Close or merged DOLL BITE Mild glide reduction Mild glide retraction BIDE Mild glide reduction Mild glide retraction

BAT Raising/lengthening Raising/lengthening Lowering and Retraction among Women; Retraction Lowering among Men BAN Raising/lengthening Raising Raising Raising BET Backing among Backing among Men Backing among Men Women BIT Mild raising before /ʃ/ Mild Backing among Men BEN Close or merged Close or merged Close or merged BIN BAIT BEET

Table 3.7: Vowel Classes Undergoing Variation among White Collar Columbus Speakers Bon 1896-1991 in Our 62 Speaker Corpus (Note: Shaded cells indicate no data are currently available; Blank cells indicate no significant variation was found involving that class)

97 Patterns of 20th Century Vowel Variation among Blue Collar Speakers in Our 62 Speaker Corpus

BOAT Mild Centralization Centralization/Some Fronting Fronting Fronting BOUT Mild Centralization Centralization/Some Fronting Some retraction of the Fronting nucleus among Women/Fronting among Men PUT Mild raising before /ʃ/ Some fronting and Fronting and some Fronting Raising rasing BUT Some fronting Some fronting Fronting Fronting BOUGHT Close realization Close before /t/ Close or merged Close or merged before before /t/ before /t/ and /k/ /t/ and /k/ BOT DAWN Close realization Close or merged Close or merged DON BALL Close or merged Close or merged DOLL BITE Mild glide reduction Mild glide retraction BIDE Mild glide reduction Mild glide retraction Mild glide retraction Mild glide retraction among Men among Men

BAT Raising/lengthening Raising/lengthening Lowering and Retraction among Women; Retraction Lowering among Men BAN Raising/lengthening Raising Raising Raising BET Mild raising before /ʃ/ Mild raising before /ʃ/ Backing among Men BIT Mild raising before /ʃ/ Mild backing BEN Close or merged Close or merged Close or merged BIN BAIT BEET

Table 3.8: Vowel Classes Undergoing Variation among Blue Collar Columbus Speakers Born 1896-1991 in Our 62 Speaker Corpus (Note: Shaded cells indicate no data are currently available; Blank cells indicate no significant variation was found involving that class)

98 reduction than speakers born after World War II, with men, especially Blue Collar men, showing stronger tendencies toward reduction than women. Among speakers born after

World War II, we continue to find some glide reduction among Blue Collar male speakers, although it is found to a lesser extent than pre-World War II born speakers.

White collar G3 and G4 speakers do not show glide reduction, however.

3.8.2 Patterns Showing Continuation/Incrementation from 19th Century Vowel Systems

Among the remaining features found in 19th century vowel systems—partial merger of the NORTH and FORCE classes, the fronting of the nucleus of SHOES, the raising of /ae/ before front nasals, voiced stops, and front voiceless fricatives, and the near merger of BOT and BOUGHT before /t/—all save /ae/-raising continue to be found in 20th Century vowel systems across all 4 generations of speakers who have been analyzed either in our own data or in previous studies. Across studies, the findings generally indicate increased frequency in degree of innovative variant usage among successively younger generational cohorts. /ae/-raising, in contrast is found in increasing amounts among G1 and G2 speakers versus the 19th Century speakers, but then not among G3 and G4 speakers, except for before front nasals. For this reason, it is counted as a continuous rather than recessive, since nasal raising, even among G3 and G4 speakers, is a continuation of the 19th Century pattern, even if raising in other environments is not continued all the way across the 4 groups. All of these trends save

99 BOT and BOUGHT near merger have generally tended to be found in most studies of

Columbus speech that have focused on these vowel classes, with findings generally be similar across studies. However, the findings for the exact behavior of BOT and

BOUGHT have been the subject of considerable debate in previous studies, as our discussion below will demonstrate.

For SHOES, all studies which have been conducted on nuclear F2 variation involving this vowel class have tended to find the same results. That is, women lead over men in SHOES-fronting across generation groups in Columbus (Thomas, [1989]/1993,

2001, 2004; Labov, et al.., 2006; Durian, et al.., 2007; Durian, et al.., 2010). In addition, a consideration of the results of Thomas ([1989]/1993), Labov, et al. (2006), Durian, et al..

(2007), and Durian, et al.. (2010) suggests Blue Collar speakers typically front the nucleus of SHOES more than White Collar speakers within any given age group.

For the merger of NORTH and FORCE into /oɹ/ (or BOARD, as we have it classified in our vowel plots in Appendix B), previous studies (Thomas ([1989]/1993,

2001, 2004; Durian, et al. (2007), Durian, et al. (2010)) report that this merger appears to have been completed during the early part of the 20th Century in Columbus. As mentioned in the previous section, the merger was already clearly underway even among the oldest observed speakers in Columbus in LANCS, albeit only partially at the time.

Among the speakers analyzed impressionistically by Thomas in his studies of Columbus vowel variation, he finds evidence of near completion of the merger among speakers born

100 in 1896 and 1908 in (2001, 2004), and complete merger of the classes among speakers born after 1966 in Thomas ([1989]/1993, 2001). Both of the studies by Durian and his associates also found evidence of the completion of the merger among Blue Collar and

White Collar Columbusites born after 1945.

Turning to the raising of /ae/, raised realizations of BAN (/ae/ before nasals) are found to a great extent among speakers born before World War II, regardless of sex, and have been found to be near categorical among all speakers, regardless of sex or class background, born following World War II (Kurath, 1930; Frazer, 1978; Thomas,

[1989]/1993, 2001, 2004; Hartman-Keiser, et al.., 1997; Dodsworth, 2005; Labov, et al..

2006; Durian, et al.., to appear; Durian, et al.., 2007). For non-pre-nasal /ae/-raising, studies of speakers born before 1925 (Kurath, 1930; Frazer, 1978; Thomas, 2004, 2006) have found that both male and female speakers realize /ae/ with a mildly raised nucleus before voiced stops (BAD) and typically before velars (BAG), as well, a pattern showing partial continuation from the vowel systems of the 19th Century born Central Ohioans discussed earlier. Here, we say the continuation is only partial, since pre-front voiceless fricative raising was not found among these speakers, although it is found in 19th

Century vowel systems, as our reanalysis of 19th Century LANCS data in the previous section, demonstrated.

Among speakers born after 1925, however, somewhat different tendencies have been found, although the direction of these tendencies is presently the topic of some

101 debate, with different studies finding different results. In their corpus of speakers born in

1925-1973, Labov, et al.. (2006) find that most of their speakers do not show raised (or lowered) realizations of /ae/, except before front nasals, with the exception of several women born after 1960 who realize /ae/ with a higher nucleus before /d/, as in "bad" or

"mad." In contrast to their findings, Dodsworth (2005) found no raising of non-pre-nasal

/ae/ for post-1925 born Columbusites, even before /d/, but the study did find women born after 1950 in the Columbus suburb of Worthington showing a tendency to lower and retract /ae/ in non-pre-nasal and non-pre-velar environments. Durian, et al.. (2007), on the other hand, found tendencies toward higher realizations of /ae/ before /d/ among some women born after 1950, and a tendency toward retraction, but not lowering, for men born after 1970, as well as retraction, with lowering, among women born after 1970. For post-

1966 born Blue Collar speakers, speakers in both Thomas's ([1989]/1993) and Durian, et al.'s (2010) studies show no strong patterns of raising for BAT. Clearly, given the mixed results of all of these previous studies, non-pre-nasal /ae/ realization requires further investigation. We provide such an investigation below.

For BOT and BOUGHT, studies of Columbusites born before 1925 show that

BOT and BOUGHT show close/partially merged rather than fully merged realization before /t/, with no strong differences among men and women, a pattern which shows continuation from the pattern found among the previously discussed 19th Century Central

Ohio men (Kurath, 1930; Frazer, 1978; Thomas, 2001). Among speakers born after

102 1925, as with BAT, different studies have reported different findings regarding the frequency of merger among speakers. Campbell-Kibler, et al.. (2008) found BOT and

BOUGHT, as well as DON and DAWN and DOLL and BALL, to be a fully merged set of classes. Labov, et al.. (2006), Ash (2006), and Durian, et al.. (2007) found the same phonetic environments to be most likely to condition phonetic variation, but most speakers in all 3 studies show stronger tendencies towards close/partially merged production rather than fully merged production. Although this is the case, however,

Labov, et al.. (2006) and Ash (2006) claim full merger is on the increase, with Columbus being one of many Midland cities studies in ANAE labeled as belonging to a zone of

"merger in progress." 16

Among Blue Collar Columbusites born after 1966, Thomas’s ([1989]/1993) impressionistic analysis found BOT and BOUGHT to be merged among most speakers.

However, as comments included with his analysis reveal, the merger is limited only to tokens occurring in the following phonetically conditioned environments in his data: pre- nasally, before voiceless stops, and before /l/. In contrast, Durian, et al..'s (2010) instrumental analysis found speakers of similar age to show only close rather than merged realizations before /t/ and /k/. Variation involving the realization of DOLL and

BALL, as well as DON and DAWN, were not investigated in their study, however, so no further results regarding pre-nasal and pre-/l/ merger are presently available.

16 In fact, the map on page 40 of Ash (2006) suggests the vast majority of Columbus ANAE informants in the full 16 speaker ANAE data set are close/partially merged rather than fully merged. 103 In our data set, we find confirmation for only one of the continuation trends noted above--the merger of NORTH and FORCE. However, we find somewhat different results for the continuation trends involving SHOES, BOT and BOUGHT, and /ae/ (BAN, BAT, and BAG). For SHOES, we find fronting as an overall trend for the vowel class in our data just as in earlier studies, but the leaders of change are somewhat different. For BOT and BOUGHT, the data are much less conclusive concerning full merger of the classes among our speakers. For /ae/ (BAN, BAT, and BAG), we find a more complex form of raising behavior among older speakers than previous studies of /ae/ in Columbus. Each of these trends will now be discussed in more detail.

For NORTH and FORCE, as just mentioned, we find much the same results as have been found in previous studies. That is, the two classes are merged in most speaker vowel systems among speakers in our study, with the only exceptions being speakers born in G1 and one G2 Blue Collar male. For the G1 speakers, as well as the Blue Collar

G2 man, NORTH and FORCE show partial merger. (Note: Given the closeness of the classes even in our G1 speaker data, and the completeness of the merger in the speech of our G2, G3, and G4 speakers, we have chosen to include only the merged category

BOARD, which includes tokens drawn from both the NORTH and the FORCE classes, in our plots.)

For SHOES-fronting in our data, we find somewhat different results than previous studies. In our data, Blue Collar speakers lead over White Collar speakers across our 4

104 generational groups, although this lead has become leveled out among G3 and G4 speakers. Across G1, G2, and G3, women show a lead over men regardless of social class background, but in G4, men of both groups have begun to take the lead over women, a new finding that has not been reported previously. Overall, in our data, we believe these trends suggest that the fronting of SHOES may have reached its maximum limit in the community among younger speakers, especially women. That is, it may not be able to be fronted any further without imposing on the acoustic space occupied by BEET without damaging the margin of security separating the two vowels in the vowel systems of younger women.

Turning next to BOT and BOUGHT, we again find somewhat different results in our data from those found in earlier studies.17 In our data, just as in earlier studies, we find that BOT and BOUGHT show close rather than merged realization only before /t/ for

G1 speakers, but only among some speakers. Several others show a clear distinction between the two classes, with a goodly amount of differentiation of BOT and BOUGHT, even before /t/. Among those speakers who do show closer realization before /t/, there are no strong differences among men and women, nor are there strong social class differences.

Among speakers born after 1925, we find stronger tendencies among speakers to show merger of BOT and BOUGHT, although more often than not, speakers in our data

17 Note: We do not have enough data from our speakers to comment on variation involving DOLL and BALL or DON and DAWN, so we will not comment on their occurrence in our data here. 105 show only close/partially merged realization of BOT and BOUGHT before /t/ and /k/, and not full merger of all tokens in the classes. In G3, we find 45% of speakers showing full merger (9 out of 20 speakers), with 55% showing either partial merger (10 out of 20 speakers) or distinct realization (1 of 20). In G4, 35% show full merger (7 out of 20 speakers), while 65% show only partial merger (13 out of 20 speakers). This makes our results more like those discussed for Labov, et al.. (2006), Ash (2006), and Durian, et al.

(2010) than Thomas ([1989]/1993) or Campbell-Kibler, et al.. (2008), although we find a higher incidence of full merger than these studies do. As with G1 speakers, there are not strong differences between male and female speakers among G2, G3, or G4 speakers, although Blue Collar speakers do show a somewhat stronger tendency towards merger in

G4 and especially G3.

Although there was an increase in our data in the incidence of full merger among speakers between G1 and G2, and a significant increase from G2 to G3, we also do not find evidence of any marked increase from G3 to G4. In fact, in G4, speakers showing partial merger are a larger group than speakers showing full merger, as noted above, and as also noted, in fact, less speakers in G4 in our data show full merger than speakers in

G3. Thus, this also makes our results different from Labov, et al.. (2006) and Ash (2006), who claim full merger is on the increase based on their data, as discussed earlier. The reasons as to why we believe this is the case in our data are more complex than we have room to discuss in this overview, however, and so we will hold off on discussing them

106 until later in the dissertation in Chapters 5.

Finally, for /ae/, we also find somewhat different results than previous studies, particularly among speakers born before 1945. In our data, we find raising not only before voiced stops and front nasals, as in the previous studies of early 20th Century

Columbus speech discussed earlier, but also before voiceless front fricatives, as in 19th

Century vowel systems, making for a more complete pattern of continuation and incrementation of prior patterns of variation in our data than has been found in previous studies. We do not, however, find many instances of raising before velar consonants among our speakers, however, as some of the previous studies have.

Among most of our G1 speakers (7 of 10), we find raising before voiced stops, voiceless front fricatives, and front nasals. This trend is also found among our G2 speakers, although to a somewhat lesser extent (7 of 12 speakers). In both generational groups, both men and women, as well as Blue Collar and White Collar speakers, show this type of raising trend. However, in G2, more White Collar speakers raise than Blue

Collar speakers, with Blue Collar women showing less raising than all other sex/social class cohorts. Among G3 speakers, we find lowering and backing beginning to dominate as the primary realization trend among speakers in place of raising, although a good number of speakers still do raise /ae/ before voiced stops and front nasals, but not before other consonants. These trends typify both Blue Collar and White Collar speech, although

Blue Collar speakers lower and back more robustly than White Collar speakers. In G4,

107 we find all speakers raising /ae/ before nasals, and some raising /ae/ before voiced stops, but less so than speakers in G3. We also find women (regardless of social class) lowering more, and White Collar men retracting more, than G3 speakers of similar sex and social class backgrounds. Blue Collar G4 men, meanwhile, differ from other speakers by showing a fronting, rather than backing, trend.

The we find raising among G1 and G2 in the particular environments in which they raise is quite interesting, as it suggests that Columbus speakers of this vintage are actually making use of a split short-a like series of realizations for /ae/, akin to the realization of short-a as a split system found historically in east coast US cities such as

New York City, NY; Baltimore, MD; and New Haven, CT., among speakers of similar age. The use of this similar type of split-a system in Ohio has been found among speakers of similar age in several recent studies (Boberg & Strassel, 1995, 2000; Thomas, 2006) in

Cincinatti and nearby central OH cities such as Westfield Center, Johnstown, Belleville,

Dover, and Clarington, but not in previous studies of Columbus, as shown earlier. At the same time, the robustness of the lowering and backing of /ae/ shown by our G3 and G4 speakers is also quite interesting versus findings from earlier studies, because it suggests

(along with the data on BET and BIT backing discussed in the next subsection) that younger speakers in Columbus are engaging in the Canadian Shift (referred to as the

Third Dialect Shift in Chapter 5).

We believe both of these findings are interesting enough, and also unique enough

108 (not having been discussed in previous studies) to require further exploration than we have room to include in this overview chapter. Thus, we will return to explore both of these trends in much greater detail later, in Chapter 5 of this dissertation. For now, we instead wish to move to a discussion of innovations found in 20th Century vowel systems, both in previous studies and our own data set.

3.8.3 Innovations in 20th Century Vowel Systems

Turning to the other patterns of variation in table 3.6 found to be pervasive in the speech of Columbusites born throughout the 20th century, the remaining patterns represent innovations not found in the 19th Century vowel systems. These include the centralization and fronting of the back upgliding vowel subsystem, which involves the vowel classes BOOT and BOAT; the centralization and fronting of BOUT; the fronting of PUT and BUT; close or merged realization of BIN and BEN, and the monophthongization of BEET. Unlike the findings of studies of variation involving BAT,

BOT, and BOUGHT discussed in 3.8.2, previous studies of the above features (e.g.,

Kurath, 1930; Frazer, 1978; Thomas, [1989]/1993, 2001, 2004; Labov, et al.., 2006;

Dodsworth, 2005; Durian, et al.., 2010; Durian, et al.., 2007) show much stronger agreement regarding the patterns of variation involved. For all of these features, younger speakers have been found to use more innovative realizations than older speakers. In addition, within each age cohort, women have been found to use more innovative

109 realizations than men, with the exception of close or merged realization of BIN and BEN, where men have been found to be more innovative.

For BOOT and BOAT, innovation involves the realization of the nuclei of each of these upgliding back vowels either with a centralized nucleus, as among pre-World War

II born speakers, or a fronted nucleus, as among post-World War II born speakers

(Thomas, [1989]/1993, 2001, 2004; Labov, et al.., 2006; Durian, et al.., 2007; Durian, et al.., 2010; Dodsworth, 2005). For BOUT, innovation also involves the realization of the nucleus of BOUT with either a centralized nucleus, as among pre-World War II born speakers, or a fronted nucleus, as among some post-World War II speakers (Kurath,

1930; Frazer, 1978; Thomas ([1989]/1993, 2001). Other post-World War II speakers, particularly White Collar speakers born after 1970, have been found to show some reversal of fronting trends however, showing retraction (backing) of the nucleus instead

(Durian, et al., 2007). In addition, BOUT has also tended to show further evidence of glide shortening over time, such that younger speakers often show shorter glides than older speakers (Thomas, [1989]/1993, 2001; Durian, et al., 2007; Durian, et al.., 2010). In contrast to this, it should also be noted that BOAT has tended to become more diphthongal over time, although this increase appears to have slowed down among speakers born after World War II, as it is often not commented on by analysts in studies of post World War II speakers (Thomas, 2001; Dodsworth, 2005; Durian, et al.., 2007).

For PUT and BUT, innovation involves fronted realizations. PUT and BUT

110 fronting have been found only among post-1950 born speakers, with speakers born after

1970 showing stronger tendencies than speakers born before 1970 (Labov, et al.., 2006;

Durian, et al.. 2007; Durian, et al.., 2010). In addition, among Blue Collar speakers, innovative realization of BUT includes raising, a tendency which was found by Durian, et al. (to appear) and Durian, et al. (2007) to differentiate Blue Collar from White Collar speech for this vowel class among post-1970 born speakers. For BIN and BEN, innovative realization involves the increased tendency among speakers to realize BIN and

BEN as either close or merged overlapping vowel classes. In Blue Collar speech,

Thomas's ([1989]/1993) and Durian, et al.'s (2010) analyses indicate BIN and BEN to be fully merged in Columbus, among post 1970 born speakers, while Labov, et al.. (2006) and Ash (2006) have found BIN and BEN to typically be close rather than merged among speakers (class not specified) born after 1925. Finally, for BEET monopthongization,

Thomas ([1989]/1993, 2001) found evidence of BEET monopthongization among speakers born after 1966 versus the 1908 born speaker analyzed in Thomas (2001).

Durian, et al. (2007) and Durian, et al. (2010) confirm this finding among speakers born after 1945, but do so without direct explicit comparison with speakers born earlier.

In our data, we find similar trends to typify the fronting of BOOT and BOAT to those found in previous studies, but only among speakers in G1, G2 and G3; for G4, we actually find different and somewhat surprising findings compared to those reported in previous studies. For BOOT, we find similar results to SHOES as discussed in the

111 previous subsection, with Blue Collar speakers showing a general lead on fronting versus

White Collar speakers across all 4 generational groups, and women leading over men in all generations save the youngest (G4). Unlike SHOES, however, the Blue Collar lead still remains among G4 speakers, rather than becoming leveled out.

For BOAT, we again find similar results to earlier studies of Columbus vowel variation, with Blue Collar speakers generally fronting the nucleus of BOAT more extensively than White Collar speakers. However, like with SHOES, we find in our data that this lead holds in only the first 3 generation groups. In G4, we again find the differences between Blue Collar and White Collar speakers becoming leveled out. Also like SHOES, we again find that although women, regardless of class group, were leading over men across the older 3 generational groups, among G4 speakers, we find men in both class groups have pulled ahead of women, so that now men generally show the lead over women regardless of class group, in G4. As with SHOES, this result differs from previous studies, where women, even among G4, have been found to lead.

Overall, for BOAT and BOOT, as well as the previously discussed SHOES, our results indicate that Blue Collar men in G4 are ultimately the linguistic leaders of change for all three vowel classes. This is a finding that contrasts with earlier studies, where Blue

Collar women in G4 were seen to be the leaders for the three classes. This difference in results in our data versus earlier studies will be discussed in further detail in Chapter 4.

For BOUT, we find evidence of an overall fronting trend across G1, G2, and G3

112 speakers, regardless of social class background or sex, as found in previous studies, and just like the other back vowel classes SHOES, BOOT, and BOAT. However, in G4, our results differ from previous studies, in that we find more robust retracting (backing) trends for the nuclear F2 of BOUT than in previous studies. In our data, we find the

White Collar men, and both Blue Collar as well as White Collar women, showing backing trends in our data. The only group that does not show an overall backing trend in

G4 is Blue Collar men, who, as a group, continue to show fairly robust fronting tendencies in our data.

For PUT and BUT, we find somewhat different results in our data than previous studies. We find evidence of PUT and BUT fronting as early as G1, rather than beginning with G3, as suggested by previous results. For PUT, we find fronting to be led across generational groups by Blue Collar speakers over White Collar speakers, and women leading over men within each class group. For BUT, we also find fronting to be led across class groups by Blue Collar speakers, and again, women lead over men within both White Collar and Blue Collar speaker groups, regardless of speaker age. For both

BUT and PUT, we find the female Blue Collar lead to hold in G4, a finding consistent with previous studies of BUT and PUT fronting in Columbus. Like Durian, et al.. (2010), we find evidence of some raising of BUT among some Blue Collar men which has not been reported in other studies. We find the raising to be mild, only in the speech of men in G2 and G3, and only in the speech of a few our speakers, of these age groups.

113 For BIN and BEN, as we did not focus explicitly on these variables in our analysis of vowel variation patterns in Columbus, we are unable to compare realization trends with previous studies. Tokens we did note in passing do suggest merger of the classes is found among G4 Blue Collar men--the males Thomas ([1989]/1993) notes to have complete merger in his data. However, our observations are not systematic enough to comment on the extent to which full or only partial merger can be said to typify the realizations in our data set among these speakers.

Finally, turning to BEET, although it is true that we do find an overall trend towards shorter realizations of BEET in our data across generational groups, as in earlier studies, such that younger speakers typically realize a shorter BEET than older speakers, we do not find this trend to hold up for every speaker in the data set. For instance, among

White Collar women, BEET has always been realized as a fairly short, monopththong.

Thus, we do not find evidence of robust shortening among this group, since no evidence of pronounced diphthongization is found to begin with. Among other speaker groups, less consistency among speakers of different age groups and social class backgrounds is found. For men, both Blue Collar and White Collar speakers generally do realize shorter

BEETs when younger men are compared with older men, but in every generational group in our study beyond G1, there is always at least one speaker who realizes a longer BEET than other speakers his own age, and also speakers older than he is. The same pattern is also true for Blue Collar women in our data set. Given the variance in our data for BEET,

114 we suggest a detailed study of BEET may be warranted in a future study of Columbus vowel variation.

3.8.4 New Innovations Found in Our Data Not Discussed in Previous Studies

Beyond the innovations in 20th Century vowel systems we discussed in section

3.8.3, there are several other patterns of innovation we find among 20th Century

Columbus speakers that have not been discussed in any previous studies of Columbus speech that we are aware of18. These patterns of innovation involve the short front vowel classes BET and BIT. In particular, the innovative variation we find that has not been discussed in previous studies includes variation in F2 of these vowel classes--specifically backing of the nucleus of both BET and BIT among some speakers.

For BET, we initially find nucleus backing among White Collar women in G2, which then continues in the speech of G3 and G4 women. In addition, we find BET- backing in the speech of both G2 and G3 White Collar and Blue Collar men's speech, as well as G2 and G3 Blue Collar Women's speech. G4 Blue Collar women show a reversal of the trend, however, now starting to show BET-front. In contrast, G4 White Collar and

Blue Collar men continue to show an increase in frequency of backed realizations in G4,

18 Here, we mean specifically the work of others. At the time of writing (2012), these innovations have been discussed in several of the author's own conference presentations, as well as one publication by the author and others (Durian, et al.. (2010)). However, since all of these materials have been written by the author, and almost all draw from subsets of the dissertation data set we discuss and analyze here, we have chosen to mark this material as "not discussed previously" in these presentations and publications, and instead, present that information as "new" throughout the remainder of this dissertation. 115 increasing both the amount of backing and also the frequency of backed variant use. G4 men show the strongest increase of all, which suggests these men may be the leaders of linguistic change for a new BET-backing trend in Columbus.

For BIT, we find some evidence of mild backing tendencies in the vowel systems of Blue Collar G3 men and women, but like with BET backing among G3 Blue Collar men, this tendency is not carried on by speakers in G4. In addition, we find a backing tendency among G4 White Collar men. Since this group is the youngest speaker group in the study, we are unable to determine at present whether this will be a continuing tendency or not, however. As will be discussed in Chapter 5, this tendency is not statistically significant at present, so it may also simply be "noise" in the data that is neither linguistically or socially significant. However, as we also will discuss more in

Chapter 5, it may represent the start of a third phase of front vowel parallel shifting

(specifically, parallel backing), that, along with BET-backing and BAT-backing among

G4 White Collar men, signifies the full presence of the Third Dialect Shift as a pattern of notable vowel shift in Columbus.

3.8.5 Comparison of Our Data with the JFSP Data

As mentioned earlier, Jacewicz (2011 a,b) feature Columbus data in their tri-state, tri-generation dialect study. Thus far, the study focuses on data obtained from speakers living in Ohio, Wisconsin, and North Carolina. 3 generations of speakers are analyzed: an

116 older group born 1919-1940, a middle group born 1958-1973, and a young group born

1994-1998. According to their description of speaker occupations and educational levels of their speakers, their speakers are most like our white-collar speakers. (Like Labov, et al., 2006, they did not stratify their data by social class, only by sex and age). Regarding the data presented in the most relevant studies to our work here, the material used for analysis has been primarily read word list citation form speech. This marks the data in contrast to our own, where the data is primarily spoken conversational speech. As we can see from the age distribution of the JFSP (Jacewizc, Fox, and Salmons Project) data set, their groups match quite closely the age groups we use for our study. Given this fact, we will refer to their speakers by the same G distinctions we use for our speakers, except for their speakers we will use g2, g3, g4, rather than G2, G3, and G4, which will continue to be used for our data. Note that there is no g1 in JFSP, because they did not record or include speakers of this vintage. In addition, we compare only patterns of white-collar speech in our data with their data, given the class background of their speakers.

In Jacewicz (2011a), differences in vowel realization are made by analyzing differences in the duration of the vowels at 5 distinct points (20%, 35%, 50%, 65%,

80%). This contrasts with our data, where diphthongs were measured at 25% and 75%, and monophthongs were measured at 50%. Given the differences, we will compare the

JFSP data at 20% with our data at 25% for diphthong nuclei and 50% in both data sets for monophthong nuclei. According to Jacewicz (2011a), the only statistically significant

117 patterns of vowel variation in their data are backing of /a/ and backing of /ae/. This was determined using ANOVA. However, the normalized comparative diagrams of speakers presented in Figure 2 (page 55) suggests that significant patterns of variation are also found for SHOES/BOOT-fronting (they appear to use a mixed SHOES/BOOT class),

PUT-fronting, some BOUGHT lowering, BET-lowering, and BIT-lowering. Other vowel classes are reported as not undergoing significant variation.

SHOES/BOOT-fronting looks much like it does in our data for G2 and G3, with g3 being fronter in their data than g2 and women being slightly fronter than men. In their data, g4 is also fronter than g3, although men and women appear to be about equally as front as one another, which is comparable to G4 white collar speakers in our data. For

PUT-fronting, fronting of the nucleus is found to increase in their data, as speakers get younger, just as it does in our data. Women appear to lead slightly over men, which is similar to our data for G2 and G3, but not similar for G4, since in G4, we find men taking the lead over women.

For BOUGHT, speakers in g3 are lower compared to g2, but this trend seems to stop in g4, and there is no strong difference between speakers for gender. This is the reverse of what we find in our data, with G4 showing raising compared to G3, who in turn show some raising compared to G2. In our data, women show more raising than men. For BOT-backing and raising, they find a significant increasing in backing and raising among speakers, with an increase in backing and raising shown in g3 and g4

118 compared to g2, with no strong sex difference. This is similar to what we find in our data.

Given the lowering also shown by BOUGHT in the JFSP data, however, it would appear as though more cases of full-merger are likely to be found in the JFSP than we find in our data. Specifics on the merger are not reported, however, so it is difficult to know this for certain.

Turning to BAT-lowering and backing, JFSP report similar backing and lowering trends to what we find, with g2 speakers using much higher and more diphthongal realizations, just as our G2 speakers do, and then g3 and g4 speakers showing lower, backer, and more monophthongal realizations, again, just as in our data. Also in the JFSP set, men tend to lag behind women in showing lowering and backing trends, another similarity with our data set. Finally, turning to BET- and BIT-lowering, JFSP report a lowering and monophthongizing trend that no previous study of white collar speech in

Columbus has reported, although Thomas ([1989]/1993) does report a similar trend for older blue collar speech versus his 1966-1970 born speakers in that study. However,

Thomas does not indicate how much older the "older" speakers he refers to here are. It should be noted however, that although Jacewicz (2011a) makes some hay of this finding, suggesting it is a previously unreported language change trend, Jacewicz (2011b) reports it may not be a change trend so much as speech-emphasis related. When more emphatic realizations made in the citation speech form speech was compared with speech collected in other tasks that was less emphatic, the more emphatic speech was found to be

119 more diphthongal. The description given for the less emphatic speech, instances, however, suggests a more monophthongal realization among both older and younger speakers. As we have chosen not to deal with BET and BIT as diphthongs in our study, we do not, at present, have information we can add either way on this finding. We plan to address this question in a follow up study.

As we can see, for almost all trends where JFSP find notable results, our study generally reports the same trends. However, a key difference we will begin to observe between our data and theirs in coming chapters is in the number and type of vowel variation trends that we find to be statistically significant in our data versus theirs, with our study finding more significant trends. In our data, we find statistically significant trends for SHOES/BOOT-fronting, BOAT-fronting, BOUT-fronting, BOUGHT-raising,

BAT-lowering and backing among younger speakers, BET-backing, and BIT-backing.

Although we do not focus on in our dissertation, we also find significant PUT-fronting and BUT-fronting trends. In addition, we find a linguistically significant pattern of BOT- backing and raising. The difference in our findings is likely due to the fact that JFSP use primarily read speech, whereas we use primarily conversational speech.

3.9 General Trends in the Data Across Social Groups and Covariant Vowel Classes

Taking the results of the analysis of vowel variation across 19th and 20th Century speakers presented in sections 3.6-3.8, and summarized in tables 3.7 and 3.8, we have

120 now been able to obtain a general picture of patterns of variation and change involving the Columbus vowel system during the period 1850-2000. As mentioned earlier, vowel plots of speakers showing all of these patterns can be found in Appedix B. As we can now see, generally speaking, the most conservative speaker group overall for G1 and G2 is White Collar men, followed by White Collar women, then Blue Collar men, and finally, Blue Collar women, who are the most innovative speaker group overall. This can be seen most clearly across most vowel classes discussed in sections 3.7 and 3.8, where

Blue Collar women are more often the leaders of change, and White Collar men are often the most conservative speakers for a given change. Vowel classes that are especially salient for showing social class differentiation along these lines include SHOES, BOOT,

BOAT, PUT, BUT, BOUT, and BAT.

In G3 and G4, however, these patterns begin to change. For SHOES-, BOOT-, and BOAT-fronting, Blue Collar men move to the lead over Blue Collar women, especially in G4, so that by G4, Blue Collar men are now in the lead as linguistic leaders of change for these vowel classes. Meanwhile, for BAT-backing, as well as BET-backing and BIT-backing, which represent a recent change in the systems of G3 speakers, we find

White Collar men begin to move into the lead as the linguistic leaders of change for these vowel classes in G3, and especially in G4, so that in G4, they are clearly established as the leaders for this entire set of vocalic sound changes. It should be noted however that

Blue Collar women still do remain the leaders of change in G3 and G4, in particular for

121 BUT- and PUT-fronting, as well as BAT-lowering. But the fact that they are only clearly the leaders for a couple of vowel classes is a quite surprising finding, given the results of many previous studies of vowel variation in urban communities in the US (e.g., Labov,

1966, 1994, 2001; Eckert, 2000; Johnstone, et al., 2002; Dodsworth, 2005; Fogle, 2008;

Becker, 2010), where women have usually been found to be the leaders of linguistic change for most vocalic variables that have been studied previously. The lack of a female lead among younger speakers in our study is a more complex issue than we have room to discuss in just one subsection in this chapter, and so it is one we will return to in more detail in Chapter 7 of this dissertation.

Other general patterns of note that emerge from the combined consideration of sections 3.4-3.6 are the covariant patterns of vowel variation that are found in the data once one begins to look at the variation patterns as they co-occur in speaker vowel systems across speakers in various generational cohort groups. Four especially salient covariant patterns of vowel variation emerge: 1) the frontward shift of the back diphthongs SHOES, BOOT, and BOAT; 2) the conditioned rasing of BAT among older speakers, particularly those in G1 and G2; 3) the backward shift of BAT among G3 and

G4 speakers and the backward shift of BOT, among G1, G2, G3, and G4 speakers, a pattern which suggest that these two vowels appear to be involved in a possible chain shift; and 4) the backward shift of the front short vowels BET and BIT. All 4 of these patterns are notable because these patterns of variation have been noted to occur in other

122 dialect areas outside the US Midland, with 1), 3), and 4) following the same covariant configurations as detailed here. In those areas, they have been studied as covariant vowel shifts. The frontward shifts of SHOES, BOOT, and BOAT have typically been grouped together as a parallel shift, and this shift has been found to occur not only in the US

Midland, but the US West and US South, as well, in the recent Atlas of North American

English (Labov, et al.., 2006). Meanwhile, the backward shift of BAT and BOT, along with the backward parallel shift of BET and BIT and the merger of BOT and BOUGHT, have all been grouped together as the Canadian Shift in the ANAE. In the ANAE, however, the Canadian Shift was noted to occur only in Canada, hence it's name, and not in the Midland, as we find here. In fact, as far as we are aware, ours is the first study of

US Midland vowel variation to find, describe, and report extensively on, the occurrence of the Canadian Shift in a US Midland city.19

As sections 3.7 and 3.8 also make clearer, these patterns of vowel shift are actually quite a bit more complex in our data than the discussion so far has acknowledged. This is so because we have attempted to conserve space in this rather long chapter, choosing to focus only on an overview of the vowel variation patterns found in our data, rather than an extensive and detailed survey. In particular, the Canadian Shift warrants additional research, since a) our pilot study, and now the full study here, is the

19 Here we mean chronologically in terms of when we reported on the phenomena in the Midland initially. Our first report dates from January 2009, via an oral presentation of early findings from this dissertation at the Annual Meeting of the Linguistics Society of America (Durian, 2009). Since that time, we are of at least four publications that have used the term to describe the BAT/BOT/BOUGHT low vowel chain shift component: Durian, et al. (2010), Bigham (2010), and Jacewizc, et al (2011a,b). 123 only study of Columbus speech so far to report significant findings for BET-backing as well as BAT-and BOT-backing, b) the only study of Columbus to investigate the Shift in both blue and white collar speech; c) our data also allow us to study not only its occurrence in our data, but, as well will demonstrate in Chapter 6, also its inception and full development over the course of the entire 20th Century in Columbus. As well, we also believe the field's understanding of the back vowel parallel shift would be enhanced by our taking a more detailed look at it in this dissertation as well.

A third pattern of vowel variation that warrants more detailed attention in a later portion of this dissertation mentioned above is the complex change of the Columbus short-a system from the split-like short-a of the 19th Century data set to the nasal short-a system of the G4 speakers of the 20th Century data set. There are several reasons for this.

First, the split short-a pattern found among the 19th Century- and early 20th Century- born speakers has not been found specifically in Columbus in previous studies (although it has been found in other areas in Ohio in recent studies among speakers of similar age to our late 19th Century- and early 20th Century-born speakers). Thus, our exploring it more detail here will provide the field with details on a new "sighting" of this pattern of realization in historical data. Second, data on the occurrence of the split-a pattern among speakers born in the mid-19th Century in the Midland more generally has not been found or explored in previous studies. Therefore, discussing our findings will prove beneficial to expanding our understanding of the occurrence of this system in US English more

124 generally. Third, our understanding of the occurrence of the Canadian Shift in Columbus will be further enhanced by a more detailed exploration of the Columbus short-a system.

As we argue in Chapter 7, the backing of BOT may be tied not only to the inception of the Canadian Shift in Columbus, but also the collapse of the historical short-a system.

Given the potential of more detailed analyses of all three of these patterns of vowel shift to enhance the knowledge of the field on their occurrence not only in

Columbus, but also in the US Midland, we will now turn to provided these more detailed analyses in the next 3 chapters of this dissertation. In Chapter 4, we focus on the parallel fronting of the back diphthongs BOAT, BOOT, and SHOES. In Chapter 5, we focus on the parallel backing of the low vowels BAT and BOT, and the parallel backing of the front vowels BET and BIT. In Chapter 6, we discuss the historical conditioned tensing and raising of BAT, known as a split short-a system.

125 Chapter 4: Phonetic Analogy and the Parallel Fronting of the Back Diphthongs20

4.1 Introduction

As we saw in Chapter 3, several significant interrelated patterns of vowel variation developed during the course of the 20th Century in Columbus, none of which were found to any great extent in the vowel systems of the speakers analyzed who were born during the mid-19th Century in the Central Ohio area. In the next three chapters, we will explore in greater detail the three most significant patterns of vowel shift that emerged in the general analysis and discussion presented in the previous chapter:

Here in chapter 4, we begin this series of more detailed discussions by presenting a detailed analysis of the covariant vowel shifting process that the vowel classes SHOES,

BOOT, and BOAT are undergoing in Columbus, known as the back diphthong parallel shift. We present a detailed analysis of this parallel shift in this chapter for two reasons.

First, although this process of shift has been noted to occur in a number of varieties of US

English, most accounts have not attempted to account for the linguistic motivation of the shift; rather, they have tended instead to document the shift descriptively. Second, our data, and the technique of quantitative analysis we have chosen to employ, allow us to

20 Aspects of the content of this chapter were first presented in conference presentation format in Durian (2008b) and Durian & Joseph (2011). In addition, this chapter owes a debt of gratitude to Brian Joseph and Robert Stockwell for encouraging us to think beyond the confines of later 20th Century sound change theory orthodoxy and to Cynthia Clopper for helping us implement the lmers which are vital in our approach to the study of the vowel variation trends in this chapter as well as the next. We also wish to acknowledge the influence of George Clinton and Parliament on our view of cloning. 126 explore in more detail the linguistic details of this process of shift. In doing so, we argue that although parallel shifts are often grouped with chain shifts as being one of two types of vowel shift, parallel shifts are a different kind of shift. This difference is in part a factor of the way parallel shifts operate.

First, they involve the movement of vowels that are parallel to one another in phonetic space rather than occurring in series, as chain shifts usually do. Second, they appear to be motivated by the cross-categorical generalization of vowel shifting behaviors. When one vowel class begins to move in phonetic space, the second vowel class appears to mimic fairly closely the movement behaviors of the first class, although lagging behind the first in the demonstration of that behavior in real as well as apparent time. This movement is phonetically conditioned, with the conditioning beginning in a narrowly defined set of phonetic environments. This set of environments initially conditions movement in one vowel class before the behavior begins to expand. As we will argue in the final section of this chapter, we believe this expansion ultimately involves change in realization via phonetic analogy (as defined by Schuchardt, 1885 and

Vennemann, 1972).

This contrasts with chain shifts, which involve the coordinated movement of two

(or more) vowels, such that movement of one vowel class induces the movement of a second vowel class. Chain shifts are of two types: minimal chain shifts and extended chain shifts. A minimal chain shift is defined as "a change in the position of two

127 phonemes in which one moves away from an original position that is then occupied by the other” (Labov 1994:118). This movement is typically depicted in terms of the movement of A ->B->, with A being one vowel class and B the second vowel class.

When A moves first, the chain shift is called a push chain. When B moves first, it is called a pull chain (Durian & Gordon, 2011). In the push chain, the first vowel literally pushes or nudges the second vowel out of its position in phonetic space. In a pull chain, the first vowel moves, and the space that opens up becomes occupied by the second vowel, which also begins to move after the first vowel moves. An extended vowel shift involves the movement of several vowel classes, along the same lines as the minimal chain shift .

Unlike our earlier discussion of this vowel shift in Chapter 3, here in Chapter 4, we will use the fine grained analysis of phonetic conditioning factors employing the method of quantitative statistical analysis known as linear mixed effects regression modeling (henceforth referred to here as lmer analysis) (see Baayen, et al., 2008; Jaeger,

2008; and Johnson, 2009 for more detail) to explore the causes and propagation of the back diphthong parallel shift in Columbus. This method of analysis is useful for several reasons. First, it allows all tokens of all vowel classes to be included in the statistical analysis of variation trends, rather than just the means across tokens, as with methods such as ANOVA. Second, since it includes all individual tokens, it allows continuous data to be analyzed, an advantage over methods such as VARBRUL analysis, since

128 VARBRUL requires the data to be chunked into discrete categories before an analysis can be conducted. Third, it allows both fixed effects and random effects to be taken into account in the analysis. This means that the patterns of variation shown both by individual speakers, as well as the larger social groups they belong to, are included in the analysis. Fourth, it allows a baseline social group to be established for each analysis of variation conducted. This makes for a more realistic account of the patterns of variation and change than a simple analysis of speaker and group means, as we employed in

Chapter 3 (and is common in descriptive analyses of instrumental data). This is so because usually one group will clearly be the most conservative group, while others will be more innovative than this conversative group. When multiple social categories are used, such as we employ here, one group will also be most innovative. Using this approach, then, we can determine which is most conservative, which is most innovative, and set the proper baseline to the most conservative group accordingly. As in the portion of our analysis in Chapter 3 focused on the 62 speakers we have selected for extensive analysis, these categories include speaker generational cohort membership, social class background, and sex.

Before we present the lmer analysis results, however, in section 4.2, we first present a discussion of previous studies of back diphthong fronting in US English, to provide context for our analysis in the later half of the chapter. Then, in section 4.3, we present a general discussion of how parallel shifts have come to be defined as per our

129 usage of the concept here, as well as how previous studies have claimed they work from a theoretical perspective. We will then move in section 4.4 to discussing the results of our lmer analyses of SHOES, BOOT and BOAT. In sections 4.5 and 4.6, we then discuss why we think our results show that parallel shifts are best described as a process involving phonetic analogy, rather than rule simplification or rule extension/generalization, as in earlier discussions, such as those of King (1969), Labov, et al.. (1972), Labov (1991, 1994, 2010), and Labov, et al.. (2006).

4.2 Previous Studies of Back Diphthong Fronting in US English

The study of back diphthong fronting has a long history in US English.

SHOES/BOOT-fronting, in particular, has long been noted to be a feature of Southern US

English, with reports of the early signs of SHOES/BOOT fronting being reported in the literature among speakers born as early as the 1840s by Thomas (2001). In areas outside of the South, early signs of SHOES/BOOT-fronting have been reported among somewhat younger speakers, such as speakers born in the 1860s and later in New York State,

Brooklyn, NY, and Portland, Maine, as well as among speakers born c. 1868 and later in parts of Pennsylvania outside of Pittsburgh and Philadelphia (Kurath & McDavid, 1961).

Labov (1994, 2001) provides evidence of the beginnings of SHOES/BOOT fronting among Philadelphians born c. 1890, while speakers analyzed by Thomas (2001) in areas of the US Midland, such as Central and Southern Ohio, who were born as early as the

130 1880s, have also been found to show early signs of fronting. In other studies, speakers born after the turn of the 20th Century have been found to show SHOES/BOOT-fronting in a variety of other US areas, including Central Illinois (Habick, 1980), California

(Luthin, 1987), New England (Thomas, 2001), and Canada (Clarke, et al., 1995). More recently Labov, et al. (2006) have found signs of /uw/-fronting beginning to appear among speakers born during the second half of the 20th Century and later in many parts of the US, including much of the US West, the US Midland, and the US South. This trend towards fronting clearly shows incrementation from earlier time periods, as also indicated by the results of earlier studies, such as those just mentioned, as well as in areas where it had been previously thought that SHOES/BOOT-fronting would quite likely not develop.

This includes cities in the Inland North and the Northeast, such as Chicago, Toledo,

Detroit, Rochester, Syracuse, and even New York City.

BOAT-fronting also has a long history of being studied in US English, and it is often found in many of the same areas as SHOES/BOOT-fronting. This is because fronting of BOAT very often, although not always, accompanies SHOES/BOOT- fronting, not only in dialects of English, but many Germanic languages (Martinet, 1952;

Labov, et al., 1972; Labov, 1994). When BOAT fronting does not accompany

SHOES/BOOT-fronting, it is because only SHOES/BOOT-fronting occurs; BOAT- fronting does not appear to occur on its own. In all observed cases we are aware of, when

BOAT-fronting accompanies SHOES/BOOT-fronting, it always develops after the

131 inception of SHOES/BOOT-fronting and lags somewhat behind SHOES/BOOT as the process of fronting proceeds in speaker vowel systems (Labov, 1994).

According to Thomas (2001), based on his analysis of raw field records from the

Linguistic Atlas of the Middle and South Atlantic States [LAMSAS] and Linguistic Atlas of the North Central States [LANCS], BOAT-fronting in US English was originally confined to three geographic centers, among speakers born during the second half of the

1800s. These areas were also early centers of SHOES/BOOT-fronting. These areas include: a) Philadelphia and adjacent parts of Southeastern Pennsylvania, Southern New

Jersey, and Delaware; b) Western Pennsylvania, Northern West Virginia, and parts of

Central and Southern Ohio; and c) Eastern North Carolina.

As noted by Thomas, Central Ohio was an early center of SHOES/BOOT and

BOAT-fronting, and Columbus as a Central Ohio city was no exception. As discussed in

Chapter 3, most studies of Columbus vowel systems that have been conducted previously have focused on SHOES/BOOT and BOAT fronting, and all of those that have focused on the vowel systems of speakers born after 1896 and before 1924 have found at least some amount of both SHOES/BOOT- and BOAT-fronting (Kurath, 1930; Frazer, 1978;

Thomas, 2001, 2004). Among the vowel systems of speakers born after 1924,

SHOES/BOOT- and BOAT-fronting shows significant increases in the vowel systems of increasingly younger speakers (Thomas, [1989]/1993; Dodsworth, 2005; Durian &Smith,

2005; Durian, et al.., 2007; Durian, et al., 2010; Labov, et al.., 2006).

132 Being coupled together as a covariant parallel shift as they quite often are,

SHOES/BOOT-fronting and BOAT-fronting have been argued to share a common triggering event in several previous studies (Labov, et al., 1972; Labov, 1991; Labov,

1994; Thomas, 2001; Labov, et al., 2006). In particular, given that SHOES/BOOT- fronting always precedes BOAT-fronting, the event would be the trigger specifically for

SHOES/BOOT-fronting. As with so many aspects of the study of vowel shifts in US

English, the triggering event for SHOES/BOOT-fronting would ultimately be determined by Labov and his associates.

In several different studies of the parallel fronting of BOAT and SHOES/BOOT in US English, Labov and his associates posited several different possible triggering events that induced the fronting of these vowel classes in English. In Labov, et al. (1972),

Labov took his first stab at attempting to posit a triggering event. There, he posited that the back diphthong shift in US Southern English results from the raising of /ɔ/, which forces /uw/ to front as a result of a push shift. Although this explanation seems plausible for the US South, it turned out that it did not work as an explanation for fronting in other

US English areas, such as Philadelphia, the US West, or the US Midland.

Thus, a number of years later, Labov (1994) posited a second possible triggering event. This time, he suggested that back diphthong fronting in these other areas occurs as a "response" to the vocalization of /l/, a process found in all of these non-Southern dialect areas, as a way of maximizing the distinction between /uw/ and /ow/ before /l/, where

133 vocalization occurs, and other /uw/ and /ow/ following environments. This explanation seemed quite plausible, given that in each area where data were available for analysis,

/uw/ and /ow/ were often shown to centralize or front when they occurred before consonantal segments other than /l/, but usually not before /l/. In fact, by the time Labov posited this hypothesis, younger speakers in these areas commonly showed what appeared to be a phonemic split for /uw/ and /ow/ before /l/ and /uw/ and /ow/ before other environments (Labov, 1994: 332). This explanation also seemed logical given that available data in these areas also suggest /l/-vocalization is likely an older process in these areas than back diphthong fronting (Thomas, [1989]/1993; 2001; Labov, 1994).

However, even with its plausibility, this explanation was still problematic since it essentially treated the Southern Shift version of back vowel fronting more like a chain shift, while it treated the other forms of US back vowel fronting as a parallel shift. This was problematic because although this was once thought to be the case for US English

(as in Labov, et al., 1972, see section 4.3 for more details), once data from across the different regions of the US were compared directly side by side, via the data and analysis provided by the Atlas of North American English (Labov, et al.., 2006), it became apparent that back vowel fronting in the Southern Shift and in other areas of the US (and

Canada) was actually quite similar, enough so to suggest they probably all stem from a similar triggering even, rather than two (or more), as had been posited earlier on.

This led Ash (1996), and later Labov (2010), to posit one triggering event for

134 SHOES/BOOT-fronting in the US and Canada. Using the ANAE data to support their analysis, the singular triggering event Ash (1996) and Labov (2010) posit for the fronting of SHOES/BOOT in these areas is the merger of the historical /iw/ and /uw/ classes over time in much of the US and Canada. As Ash (1996) discusses, historically, the /iw/ and

/uw/ classes stood in opposition to one another, albeit in a distribution that was quite uneven, as an outcome of the English Vowel Shift. To "correct" for this unevenness, speakers of US English began to front the nucleus of the SHOES subclass of the /uw/ vowel class, as its members began to more closely resemble those of the /iw/ class, once the members of the /iw/ class had begun to shift backwards and also began to lose their glide behavior. This "disruption" in the equilibrium of the /iw/ and /uw/ classes also caused members of the /uw/ class to begin to shift forward and "meet up" with /iw/. As

Labov (2010) adds, the leading members of the /uw/ class that began to shift were tokens of /uw/ that occurred in words where coronal consonants preceded /uw/. Once this occurred, the fronting of /uw/ began--at different times, in different places, hence the different "beginning times" for /uw/ fronting in different parts of the US, as revealed by the studies mentioned previously of different dialect areas, and by the ANAE, for cities located in the North and Northeast where /uw/-fronting was not found previously.

In Columbus, this triggering event, as proposed by Ash (1996) and Labov (2010), does seem quite plausible. Our data definitely show signs of /iw/ and /uw/ overlap (for tokens of historical /iw/ and post-coronal /uw/) among speakers who are showing the first

135 stage of /uw/-fronting--SHOES fronting--in our data. Our data also show the expected progression of /uw/ fronting first, followed by /ow/ fronting. As discussed previously in

Chapter 3, we find the earliest signs of SHOES-fronting among the 1846 and 1854 born men in the LANCS data we reanalyzed, while we find the earliest signs of BOAT- fronting among our born c. 1896-1913 speakers.

However, given the progress shown for BOAT in the speech of the more advanced speakers in this cohort--Blue Collar women--we suspect this is not the first cohort to have actually shown BOAT-fronting in Columbus. Based on the relative difference between G1 and G2 middle class men in our data--the most conservative speakers in our data for BOAT, and the group that appears to be most likely to have shown the actual inception of BOAT-fronting in their vowel systems. We hypothesize that BOAT-fronting may have begun among working class men in the previous generation (speakers born roughly 1870-1890 or so), while fronting among working class women may have also begun during that generation, if not possibly even 2 generations earlier. (That is, among speakers born roughly 1850-1870 or so.) This is obviously something that will need to be investigated further in a later study, if available data can be found to make better informed observations.

4.3 Previous Discussions of Parallel Shifting

Having now discussed the relevant previous literature on the parallel fronting of

136 the back diphthongs in previous studies of US English, we will now move to a discussion of the previous literature discussing the theoretical study of parallel shifts as they pertain to our data on Columbus vowel variation. As we noted in Chapter 1 and the introduction to this chapter, one of the contributions we hope our data will make is to provide a deeper understanding of the processes of language variation and change that we find occurring in our data. Given the time depth of the data set we have available for analysis and also the lack of extant robust quantitative studies of Midland whole vowel system variation among speakers of similar age and social class backgrounds to those whom we have assembled in our study, we believe our analysis can provide this understanding.

Furthermore, analysis of the parallel fronting of the back diphthongs allows us to explore in some detail how parallel shifts operate as a process of language change and variation.

However, we need to first discuss our usage of the term parallel shift to describe this parallel fronting process, as some readers may view the term as controversial. To others, it may be unfamiliar, even to readers who have some experience with analyzing or reading about vowel shifts in previous discussions. This is so for two reasons. First, the term has only been used sparingly in much of the previous literature (e.g., Labov, et al..

1972; Labov, 1991, 1994, 2010; Gordon, 2001; Boberg, 2005), enough so that it has not necessarily gained wide-use currency in the field. Thus, although parallel shifts, in particular the covariant shift of BOAT and SHOES/BOOT, have been described in many studies in the past 30-40 years, the term parallel shift, or the related term "parallel

137 movement," is not always used to describe the shift. Second, when the term has been used, it has not always been clear what sort of vowel shift is being described by it. Is it some type of chain shift or some other type of vowel movement (e.g., Labov, et al.,

1972)? This ambiguity is related in large part to early descriptions of chain shifts, where shifts such as the covariant shift of BOAT and SHOES/BOOT were often described purely as chain shifts, without any distinction being made between this type of vowel shift and the lockstep-style of vowel shift normally described by the term chain shift.

This includes, for instance, the English Vowel Shift, the Northern Cities Shift of US

English, or the back vowel chain shift of the Hauteville dialect of French (cf, Jespersen,

1909; Luick, 1903; Martinet, 1952, 1955; Lass, 1976; Labov, et al.., 1972).

In early, pre-Labovian studies of chain shifting, parallel shifts such as the covariant frontward movement of BOAT and SHOES/BOOT were either not discussed, because the studies focused primarily on vowel shifts such as the English Vowel Shift where the concept of parallel shift is unnecessary to use (e.g., Jespersen, 1908; Luick,

1903)21, or they were described as a traditional chain shift (Martinet, 1952, 1955). One of the primary examples linguists often use to exemplify Martinet's thinking on the motivation of chain shifting is in fact his discussion of overcrowding of the back of the vowel space. To relieve this "phonetic pressure," languages will often front the back

21 Although several trajectories of shift have been proposed for the English Vowel Shift, all explanations of the Shift we are aware of, save one (those of Stockwell & Minkova, 1997, and related papers), describe it purely in terms of being some kind of chain shift. The alternate explanation is one in which no chain shift at all is posited to explain the changes. Rather, it has been argued to be a series of mergers and individual vowel shifts, which Stockwell & Minkova call "vowel drifts". 138 vowels--quite often SHOES/BOOT and BOAT specifically--as Martinet (1952, 1955) argues, the two sources most commonly cited for his perspectives on chain shifting.22

When Labov began studying vowel shifts in detail (Labov, et al., 1972), he was the first linguist we are aware of to introduce a second vowel shifting term and type-- parallel shifting--under the term "parallel movement." However, his early use of the term was somewhat confusing, since he seemed to tag back vowel fronting as both a parallel movement and a chain shift in the same study. As discussed by Labov, et al.. (1972), fronting of SHOES/BOOT and BOAT was defined as a chain shift in three of the four vowel systems of present-day English that were analyzed in the study--the Southern Shift of US English (at least initially), the London Chain shift, and the vowel shift noted to occur in the systems of the Norwich speakers analyzed in the study. In the fourth vowel system, Philadelphia, however, Labov and his associates analyzed SHOES/BOOT and

BOAT fronting using the specific term "parallel movement." Specifically, they note in

Philadelphia English that:

"The paths of /ow/ and /uw/ do not make up a chain shift; they are parallel movements responding to some common cause. It is not at all obvious that a chain shift is involved in this situation, since there seems to be no back upgliding vowel which moves up behind /uw/ to assume their positions and which might have been held back by high back /uw/" (124-125).

Although this is the case, they do find fronting of these classes to be a chain shift in the London Chain Shift (here, /ɔ/ raises upward in the back of the vowel system,

22 See also Haurdicourt & Julliard (1949) for another set of linguists who make use of this argument to explain why the back vowels front. 139 causing /uw/ and /ow/ to front), the Southern Shift (where they initially argued /ɔ/ raising also leads to fronting of /uw/ and /ow/, although Labov later reclassified the Southern

Shift as a parallel shift, with movement akin to Philadelphia as described above), and

Norwich (/uw/ fronts as /ow/ raises up into the space of /uw/, then /ow/ gets dragged by

/uw/ in a chain-link progression). In addition, many cases of back vowel movement resulting from chain shifts have been reported historically in languages such as Swedish and East Norwegian (Haugen, 1970), Greek and the Portuguese of San Miguel

(Haudicourt & Julliand, 1949), and French (Haudicourt & Julliand, 1949, Martinet,

1952). Because this is so, Labov, et al. (1972) posit back vowel fronting as a principle of chain shifting. Specifically, they posit "Principle III: In chain shifts, back vowels move to the front" (106).

The positing of such a principle suggests that all back vowel fronting must therefore be a form of chain shift. Yet, Labov and his associates stand by their statement that the Philadelphia version of back vowel fronting (and it turned out also the Southern

Shift version) is not a chain shift in their detailed analyses of the occurrence of back vowel fronting in the various locales they studied (133-145) after making their initial statement concerning parallel movement that we cited above (from 124-125). At the time, the "common cause" for /uw/-fronting and /ow/-fronting continued to remain undetermined, but it would later be determined (quite a bit later, in fact, by Ash (1996),

Labov, et al. (2006), and Labov (2010)) to be the merger of /iw/ and /uw/, as we

140 discussed earlier in section 4.2.

Although Labov, et al. (1972) clearly set up a contrastive distinction where parallel movement and chain shifts are defined as distinct forms of vowel shifting, the fact that Labov and his associates established chain shifting principle III in the same study, and in fact, in the same chapter of their original study as they also discuss parallel movement for the first time, seems to have caused some definitional confusion in the years since Labov, et al. (1972) was first published. In addition, this confusion seems to have been furthered by the fact that they did not establish what the "common cause" for the parallel movement of /uw/ and /ow/ might be at the time. This confusion can be seen in the research that followed the publication of Labov, et al. (1972) for a number of years afterwards, where parallel shifts were often treated by researchers simply as a chain shift, following the "lead" of Labov, et al. (1972) and their chain shift principle III.

In Principles of Linguistic Change, Volume One (1994), Labov addressed this situation in his discussion of Principle III in a section entitled "Fronting of the Mid

Vowels as a Parallel Motion." He writes:

"The first description of Pattern 3 in [Labov, et al. (1972)] included the fronting of the mid-back nucleus (/o:/ or /ow/), since it was commonly associated with the fronting of /uw/. Yet further reflection indicates that the fronting of /ow/ has no connection with either chain shifting or the principles of chain shifting. It is never linked directly with the raising of a back vowel that would otherwise merge with the mid vowel, but instead represents a generalization of the fronting of the high back vowel" (208).

In doing so, he more succinctly states how back vowel parallel movement differs from

141 back vowel chain shifting. That is, back vowel chain shifting may occur as a result of

"collision avoidance" of /uw/ and a mid back vowel that shifts through raising (such as in the London Chain Shift example discussed above), while back vowel parallel shifting occurs when one vowel class begins to behave like a second vowel class, generalizing the movement behavior of that second class to the first vowel class. However, much like the earlier discussion of the differences between back vowel fronting and back vowel chain shifting in Labov, et al. (1972), Labov (1994) is a bit thin on details further distinguishing the two types of shifts from one another beyond statements like the one provided above and a few others scattered throughout that volume, as well as in his (2001) volume.

A second aspect of Labov's definition and conception of parallel shifts that is further explored and developed more extensively in Labov (1994), however, is a discussion of what of "generalization" in the context of parallel movement means. In an early portion of the volume, Labov states:

"The study of chain shifts cannot be isolated from the study of parallel movements, which have been widely discussed in the search for principles of linguistic change. Parallel shifts, or shifts that restore parallelism, formed much of the evidence for the principle that linguistic change can be explained as a process of simplification (King 1969)" (30-31).

For Labov, generalization means "rule generalization" (or "rule extension"), with rule being defined as a generative rule, as per their usage by generative philologists (and

Labov himself, albeit from the perspective of variable generative rules) in the late 1960s and 1970s. From this perspective, part of why parallel shifts often develop in vowel

142 systems is because they represent a basic form of regular neogrammarian sound change

(a point emphasized by Labov in his detailed analysis of SHOES/BOOT and BOAT fronting in the speech of Carol Meyer (a Philadelphia speaker whose speech from one entire day, in a variety of settings, was originally analyzed by Hindle initially for his

(1980) dissertation) in Labov, 1994:460-465, and via the use of data collected from across the US via ANAE, first discussed in Labov, 2001:475-497, and later, in significantly shortened and revised form, in Labov, et al., 2006:152-157). That is, they demonstrate a process of regular phonetic conditioning of a sound change. The change begins in a very limited core of environments showing robust phonetic conditioning, and then it expands to eventually include all possible phonetic environments, such that, eventually, all tokens of all words for a given vowel class show the change.

Rather interestingly, although Labov discusses parallel shifts as a form of rule simplification, and although he extensively presents rules for various chain shifts that he discusses in the Labov (1994) volume, the reader is not presented with explicit rules demonstrating the parallel shift rule simplification process. We assume it would likely be written as a variable rule which allows for an expanding array of phonetic environments to be specified as the generalization of the rule expands in the mental grammar of speakers, encompassing more and more environments, until a very simplified, expanded, and generalized rule would finally be written, to encapsulate the end state of this change process. The closest Labov comes to providing a set rule for the process that we are

143 aware of is in a much earlier, and rather obscure squib that appeared in Italian as Labov

(1976), where Labov presents the following rule for the fronting of both SHOES/BOOT and BOAT in New York City English, where the fronting of these vowels is quite similar to the fronting that occurs in Philadelphia English (43):

[-low] [-back]/ ___ [-cons] [-back]

In our view, this rule is too basic to describe back diphthong fronting during its initial stages, nor as the rule develops over time in the community grammar during the incrementation of back vowel fronting.

In more recent discussions of parallel movement and chain shifts, Labov, et al..

(2006) and Labov (2010) do not pursue a rule-based approach to analyzing or discussing parallel movement. Thus, we have no additional material to comment on regarding a rule- based approach to the analysis of parallel shifts and the role of generalization in their development, incrementation, and propagation. This appears to be the case not because

Labov has changed his perspective from that which he put forth earlier, but rather, he has moved away from casting the discussion within a linguistic framework that, by the mid-

2000s, had become quite dated.

Instead, he presents more material relating to the development of parallel shifts from a general, non-generative analytical perspective. In doing so, he demonstrates that

144 BOAT fronting does seem to be a generalization of SHOES/BOOT-fronting to a second vowel class in a framework-neutral way, and he also demonstrates that both vowel classes seem to share similar phonetic conditioning constraints in terms of the influence of, in particular, preceding consonantal segments on fronting. In the Labov, et al. (2006) and Labov (2010) analysis, preceding coronals as a group show a strong influence on conditioning fronting, with other preceding consonants also showing some influence, but much less strongly. In addition, Labov finds that a following /l/ has a very strong influence on inhibiting fronting of both SHOES/BOOT and BOAT. Furthermore,

SHOES/BOOT and BOAT show a clear chronology in the data, with SHOES/BOOT- fronting clearly preceding BOAT fronting in many cases. In the other cases of

SHOES/BOOT-fronting, no BOAT-fronting occurs, but there are no cases where BOAT- fronting precedes SHOES/BOOT-fronting or where BOAT-fronting occurs without

SHOES/BOOT-fronting (and SHOES/BOOT-fronting first). These trends further suggest a chronology, where SHOES/BOOT fronts first, and BOAT (if it fronts) fronts second.

A more detailed discussion of these recent results will be included in section 4.5, when we compare our results with them. However, before we compare our results, we must first introduce and discuss them. We now turn to doing so in section 4.4.

4.4 Linear Mixed Effects Regression (lmer) Analysis of Nuclear F2 of SHOES, BOOT, and BOAT in Our Data

To begin the detailed analysis of the fronting of the nuclei of the back diphthong

145 subsystem in our own Columbus data, we turn first to a lmer analysis of the nuclear F2 of the three vowel subclasses tagged in our data set that are involved in this particular vowel shift: SHOES, BOOT, and BOAT. In particular, we will investigate the impact that the linguistic (phonetic) factors of preceding and following segment, as well as the social factors we have chosen to stratify our data by (age, social class, and sex) have had on the fronting of the nuclei of all 3 vowel subclasses among speakers born during the course of the 20th Century. To do so, we will investigate the values of specific coefficients that allow us to predict the F2 of each subclass in a lmer model from the chosen social and linguistic factors.

In the discussion that follows, unless otherwise mentioned, it should be noted that all analyses assume the vowel systems of older middle class men provide a baseline from which all other vowel systems discussed diverge. Specifically, the vowel systems of middle class men in the G1 group are used as the baseline system in all analyses. As shown in Chapter 3, this group has the most conservative vowel system for most of the vowel variation trends that developed in Columbus speech during the course of the 20th

Century, and so it therefore seems logical to use their systems as a general baseline.

As discussed in Chapter 3, in Columbus, since at least the beginning of the 20th

Century, speakers in the area show a robust tendency toward realizing /uw/ after coronals significantly differently than how they realize /uw/ after non-coronal segments, with /uw/ after coronals being realized significantly fronter on average than /uw/ following other

146 segments. This tendency mirrors a more general tendency among most US English speakers since about the same time period, as demonstrated in data analyzed in dialect atlas studies since the 1930s (Kurath, et al., 1939; Thomas, 2001; Labov, et al., 2006).

Given this fact, in our analyses for chapter 4, we have chosen to maintain the subdivision of /uw/ into post-coronal and post-non-coronal subclasses, just as we did earlier in chapter 3. Following similar conventions, we again refer to these subgroups as SHOES and BOOT in the analyses featured below.

When one conducts a lmer analysis, one can include any number of linguistic factors one wishes to include, much like one would do in a more traditional statistical analysis of sociolinguistic data, such as in a GoldVarb or ANOVA analysis. Lmer also allows for random factors to be taken into account, which gives lmer analysis a clear advantage over GoldVarb or ANOVA analyses. For our analyses, two random factors were included: speaker and token. Both factors were included unless log likelihood tests showed only one should be included. If one of the factors was excluded, this fact will be referenced in the text.

In our analyses, the linguistic factors to be included are preceding and following phonetic segment. In addition, one can also include any number of social factors suspected to influence variation in the pronunciation of vowels among speakers. For the present analysis, we have chosen to include all of the basic social factors discussed in

Chapter 3 that we suspect are having an influence on vowel variation across time in

147 Columbus. These factors include sex (quantified as a binary male/female division), social class (quantified as a split into middle and working class groups, per the exact same parameters discussed previously in Chapter 3), and age. As per our discussion in Chapter

3, age is again quantified using generational cohort groupings. These groups are composed as follows: Generation 1 (G1) includes speakers born 1896-1913; Generation 2

(G2) includes speakers born 1924-1938; Generation 3 (G3) includes speakers born 1945-

1968; Generation 4 (G4) includes speakers born 1976-1991.

4.4.1 F2 of SHOES

Turning first to the lmer analysis of F2 of SHOES, our model reveals a number of coefficients that are significant predictors of F2 realizations among speakers. These significant coefficients are listed in table 4.1 in bold. Positive coefficients indicate a fronter /uw/ production, whereas negative coefficients indicate a backer /uw/ production.

The model used for this analysis includes speaker as a random effect, but not token because it did not contribute significantly to model fit.

Turning first to significant social factors, both social class (listed as Blue), and age (listed as Generation 2, 3, and 4) are significant as independent factors. In addition, interactions of sex*age (listed as Woman*G4) and sex*social class (listed as G3*Blue and G4*Blue) are also all significant statistically. Considered together, these results suggest that, as our initial analysis in Chapter 3 also suggested, fronting of the nucleus of

148 Coefficient t value Significance (Intercept) 0.131 0.912 0.331 Sex: Woman 0.216 1.349 0.103 Generation: G2 0.590 4.414 <0.001 Generation: G3 1.009 7.546 <0.001 Generation: G4 1.066 8.337 <0.001 Collar: Blue 0.400 2.820 <0.001 Preceding /d/ -0.115 -1.145 0.265 Preceding /n/ -0.015 -0.157 0.892 Preceding /s/ -0.052 -0.431 0.745 Preceding /ʃ/ 0.209 1.953 0.049 Preceding /t/ 0.001 0.010 0.997 Preceding /z/ -0.102 -0.533 0.623 Following /b/ 0.192 1.107 0.344 Following /d/ -0.110 -2.374 0.032 Following /f/ 0.375 1.507 0.138 Following /h/ -0.111 -0.450 0.685 Following /k/ 0.235 0.970 0.351 Following /m/ -0.225 -0.917 0.316 Following /p/ -0.050 -0.599 0.592 Following /s/ 0.014 0.095 0.949 Following /t/ -0.026 -0.410 0.692 Following /ð/ 0.038 0.202 0.827 Following /z/ 0.038 0.793 0.345 Woman*G2 -0.176 -0.850 0.292 Woman*G3 -0.210 -1.061 0.187 Woman*G4 -0.389 -2.009 0.013 Woman*Blue -0.066 -0.293 0.729 G2*Blue -0.297 -1.530 0.057 G3*Blue -0.353 -1.925 0.018 G4*Blue -0.370 -2.058 0.014 Woman*G2*Blue 0.071 0.238 0.758 Woman*G3*Blue 0.153 0.557 0.493 Woman*G4*Blue 0.350 1.242 0.131 Table 4.1: Coefficients in the Linear Mixed Effects Model Predicting the F2 of SHOES from the Social Variables (Sex, Generation, Class), Their Interactions, and Following Phonological Environment. P-values were obtained using Markov Chain Monte Carlo simulations (Baayen et al.., 2008).

149 SHOES is increasing over time in Columbus, as shown by significant positive coefficients for all 3 generational groups. Note, however, that the coefficients for G3 and

G4 are very similar, suggesting that fronting may have reached its maximum limit. This increase is also impacting all social class groups in Columbus, regardless of sex of the speaker, as suggested by the independent significant coefficients for age and social class.

However, as the independent results for social class also reveal, although both groups are showing increases, Blue Collar speakers show a significant lead over White

Collar speakers across generational groups (and regardless of sex), with Blue Collar speakers leading in fronting within generational group. This lead is reduced among Blue

Collar speakers in G3 and G4, as suggested by the significant negative coefficients for the interaction of G3*Blue and G4*Blue. In addition, the values suggest that although women led the change in G1, G2, G3 in each class group, in G4, men have moved into the lead, and they generally lead over women independent of class group among the youngest generation of speakers, as revealed by the significant negative coefficient for the interaction of Woman*G4. These results suggest further that the leaders of linguistic change overall for fronting of the F2 of the SHOES subclass is ultimately young men, regardless of social class group membership.

Turning to linguistic factors, the lmer analysis reveals one preceding segment, /ʃ/, and one following segment, /d/, have significant coefficients. (The baseline segments for this analysis are /ʒ/ for preceding segment and “pause” for following segment.) However,

150 the directionality of the conditioning is in opposite directions for the two segments, as revealed by the values of their coefficients. /ʃ/, with a positive coefficient of 1.953, conditions a significantly fronter average realization of /uw/ than other preceding segments, while /d/, with a negative coefficient of -2.374, conditions a significantly backer /uw/ realization, on average, than other following segments.

4.4.2 F2 of BOOT

Turning next to the results of our lmer analysis for the nuclear F2 of BOOT, we find several similar results to those we found for nuclear F2 of SHOES. These results are presented via table 4.2, with statistically significant results again presented in bold. As with SHOES, older White Collar men are again used as the baseline for the analysis, although this time, /b/ was used as the baseline segment for both the preceding and following segment analyses. /b/ was chosen in part because it was alphabetically first, but also because, in both positions, /b/ is essentially a neutral segment. It strongly conditions neither fronting nor backing of /uw/, and thus, seems a good choice as a baseline segment as a result.

For nuclear F2 of BOOT, our model reveals the significant independent social factors influencing fronted realizations of BOOT tokens include age (as indicated by significant coefficients for G2, G3, and G4) and social class (specifically Blue Collar).

What these results suggest is that BOOT fronting has increased over time in the

151 community (as demonstrated by the G2, G3, and G4 results), regardless of speaker sex or social class background, and that Blue Collar speakers tend to use significantly fronter realizations than White Collar speakers, regardless of age or sex. Interestingly, a significant interaction of sex and age occurs in G4 (Woman*G4), although it is important to note that the negative coefficient value for this factor means it is actually men, not women, who more robustly front BOOT tokens among this youngest group of speakers.

This trend appears to be a recent change, given the changes in coefficient value trends demonstrated by the interaction of sex and age in our analysis. Taking all of these findings into account, our analysis suggests that, overall, BOOT fronting has been on the increase over time in Columbus, with working class speakers generally leading in fronter

F2 realization, and women leading over men until fairly recently, when men have now taken over as the leaders of linguistic change for fronting of the BOOT class among G4 speakers.

For linguistic factors, our results suggest both preceding and following environment have a significant impact on the conditioning of BOOT realizations. In both cases (preceding and following environment), the significant segment conditioning fronting is /k/. Again, the distinction in directionality of segmental conditioning influence is revealed by the difference in coefficient values for the different type of segments-- positive values indicating a fronter BOOT and negative values indicating a backer

BOOT. Thus, preceding /k/ conditions significantly fronter BOOT realizations, while

152 Estimate t value pMCMC (Intercept) -1.434 -3.866 0.000 Sex: Woman 0.368 1.126 0.142 Generation: G2 0.700 2.624 0.000 Generation: G3 0.981 3.602 0.000 Generation: G4 1.438 5.655 0.000 Collar: Blue 0.469 1.649 0.035 Preceding /f/ 0.551 1.704 0.142 Preceding /g/ 0.460 0.883 0.398 Preceding /h/ 0.553 1.854 0.095 Preceding /k/ 0.904 3.526 0.000 Preceding /m/ 0.416 1.280 0.274 Preceding /w/ 0.244 0.826 0.460 Following /f/ 0.260 0.393 0.663 Following /g/ 0.451 0.674 0.518 Following /k/ -1.321 -2.734 0.008 Following /p/ -0.444 -1.385 0.136 Following pause 0.139 0.645 0.640 Following /s/ 0.284 0.952 0.423 Following /ʃ/ -0.133 -0.269 0.582 Following /t/ 0.343 1.075 0.399 Followth 0.250 0.738 0.539 Followth2 0.327 0.847 0.360 Following /v/ -0.051 -0.198 0.846 Following /z/ 0.393 1.745 0.129 Woman*G2 -0.190 -0.457 0.541 Woman*G3 -0.211 -0.525 0.446 Woman:* G4 -0.781 -2.016 0.011 Woman:*Blue -0.619 -1.338 0.115 G2*Blue -0.481 -1.205 0.132 G3*Blue -0.310 -0.846 0.245 G4*Blue -0.216 -0.606 0.458 Woman*G2*Blue 0.887 1.460 0.082 Woman*G4*Blue 0.452 0.821 0.361 Table 4.2: Coefficients in the Linear Mixed Effects Model Predicting the F2 of BOOT from the Social Variables (Sex, Generation, Class), Their Interactions, and Following Phonological Environment. P-values were obtained using Markov Chain Monte Carlo simulations (Baayen et al.., 2008). 153 Woman*G3*Blue 0.550 0.987 0.228 Woman*G4*Blue 0.452 0.821 0.361 Table 4.2 (Cont): Coefficients in the Linear Mixed Effects Model Predicting the F2 of BOOT from the Social Variables (Sex, Generation, Class), Their Interactions, and Following Phonological Environment. P-values were obtained using Markov Chain Monte Carlo simulations (Baayen et al.., 2008).

following /k/ conditions significantly backer BOOT realizations. Beyond these segments, no other preceding or following segments appear to have an influence on conditioning either significantly fronted or backed variants of BOOT subclass tokens in our data set.

4.4.3 F2 of BOAT

For our lmer analysis of the nuclear F2 of BOAT, we find a more extensive set of significant results for preceding environment than we found in our analyses of BOOT and

SHOES in sections 4.4.1 and 4.4.2. As with SHOES and BOOT, older White Collar men are used for the baseline for the analysis. Results are presented in table 5.3, and statistically significant results are again presented in bold.

Turning first to a consideration of the significant social factors, our model reveals

sex, age (generation groups 2, 3, and 4), and social class are all significant independent

social factors influencing fronted realization of BOAT nuclei in our data. These results

suggest BOAT-fronting has increased across generational groups, such that younger

speakers realize BOAT tokens on average with fronter F2 values than older speakers,

regardless of sex or social class of the speaker. In addition, these results suggest Blue

Collar speakers realize fronter BOAT tokens on average than White Collar speakers,

154 Coefficient t value Significance (Intercept) -1.153 -7.497 0.000 Sex: Woman 0.312 2.554 0.007 Generation: G2 0.416 4.088 <.001 Generation: G3 0.678 6.644 <.001 Generation: G4 0.848 8.655 <.001 Collar: Blue 0.223 2.029 0.025 Preceding /b/ -0.303 -1.684 0.117 Preceding /ʒ/ 0.36 1.254 0.191 Preceding /d/ 0.264 1.851 0.050 Preceding /f/ -0.297 -1.583 0.13 Preceding /g/ 0.157 1.143 0.258 Preceding /h/ -0.199 -1.556 0.129 Preceding /dʒ/ 0.063 0.213 0.88 Preceding /k/ 0.01 0.075 0.913 Preceding /m/ -0.272 -1.899 0.07 Preceding /n/ 0.424 2.002 0.049 Preceding /p/ -0.111 -0.782 0.469 Preceding /s/ 0.33 2.485 0.013 Preceding /ʃ/ 0.405 2.705 0.007 Preceding /t/ 0.6 3.066 0.003 Preceding /θ/ 0.526 3.06 0.003 Preceding /ð/ 0.229 1.642 0.09 Preceding /v/ -0.202 -1.165 0.262 Preceding /w/ -0.648 -1.678 0.111 Following /b/ -0.032 -0.088 0.978 Following /ʒ/ 0.791 2.161 0.037 Following /d/ 0.081 0.88 0.369 Following /k/ 0.024 0.26 0.809 Following /p/ 0.106 1.014 0.32 Following /s/ 0.136 2.141 0.036 Following /ʃ/ -0.277 -1.057 0.268 Following /t/ 0.187 2.713 0.011 Following /θ/ 0.088 0.634 0.586 Following /v/ 0.038 0.264 0.78 Table 4.3: Coefficients in the Linear Mixed Effects Model Predicting the F2 of BOAT from the Social Variables (Sex, Generation, Class), Their Interactions, and Following Phonological Environment. P-values were obtained using Markov Chain Monte Carlo simulations (Baayen et al.., 2008). 155 Following /z/ 0.077 1.372 0.194 Woman*G2 -0.091 -0.569 0.522 Woman*G3 -0.240 -1.590 0.079 Woman*G4 -0.460 -3.103 0.002 Woman*Blue -0.117 -0.676 0.436 G2*Blue -0.044 -0.290 0.722 G3*Blue -0.018 -0.129 0.861 G4*Blue -0.011 -0.078 0.891 Woman*G2*Blue 0.043 0.189 0.806 Woman*G3*Blue 0.086 0.406 0.630 Woman*G4*Blue 0.114 0.544 0.525 Table 4.3 (Cont.): Coefficients in the Linear Mixed Effects Model Predicting the F2 of BOAT from the Social Variables (Sex, Generation, Class), Their Interactions, and Following Phonological Environment. P-values were obtained using Markov Chain Monte Carlo simulations (Baayen et al.., 2008).

regardless of sex or age, and that women have tended to front BOAT more than males in

most of the generational groups analyzed in this study, regardless of age or social class

background. We argue these results suggest this has been the case only for most, rather

than all, generational groups, because there is also an interaction between age and sex

for G4 speakers, and in this group, the coefficient is actually negative rather than

positive. This further detail suggests that women were fronting F2 more than men in the

oldest 3 groups, but that in the youngest group, men have now over taken women as

leading users of fronted realizations of BOAT. As there is no class interaction effect, we

further argue this trend should be extended to speakers of both class groups. That is,

Blue Collar, as well as White Collar, men appear to have pulled into the lead as

linguistic leaders of change for BOAT fronting among the G4 speaker cohort.

Turning to linguistic environments, our results reveal that a large set of preceding

156 segments, all of them coronal, are statistically significant conditioners of fronted BOAT token realizations in our data. (For both preceding and following environment, “pause” serves as the baseline condition for the purposes of the lmer analysis.) Among these significant segment groups are /d/, /n/, /s/, /ʃ/, /t/, and /θ/. This finding is of some interest, considering the overwhelming impact that coronals have had historically in Columbus on conditioning generally fronter variants of /uw/, as discussed earlier in this chapter, as well as previously in Chapter 3. This is a matter that will be discussed in more detail in the next section, below. Also interesting to note is that several following segments also appear to influence fronting of BOAT tokens in Columbus. These include /s/, /t/, and /ʒ/.

4.4.4 A Summary of the lmers

Across the analyses, the general pattern that emerges is that working class speakers stand out as the leaders of linguistic change across all 4 generational groups analyzed in our study for all three vowel classes analyzed in this section. In G1, G2, and

G3, women led over men in both class groups, making Blue Collar women the linguistic leaders of change in G1, G2, and G3, followed by Blue Collar men, then White Collar women, and finally, White Collar men. In G4, however, men have overtaken women as the leaders in both class groups, so that Blue Collar men are now the leaders, followed by

White Collar men, then Blue Collar women, and finally White Collar women. Regarding phonetic conditioning, preceding coronals as a group clearly more robustly condition

157 fronting of both /uw/ and /ow/ than other preceding segments, while different following segments significantly condition SHOES, BOOT, and BOAT-fronting, with no specific consonants overlapping between any of the classes. Given the relative differences in significance between the following segments and preceding segments, preceding segment is by far the much more robust conditioner, making the pronounced similarity of coronals for both /ow/ and /uw/ quite interesting. Given the significance influence of this segment group not only in our data, but also in all regions of the US where /uw/ and /ow/ fronting were found by Labov, et al. (2006) in the ANAE, we believe this result requires more detailed investigation, particularly given what we believe the result tells us about the nature of parallel shifts and the role of phonetic generalization in the advancement and propagation of parallel shifts. This investigation follows in section 4.5.

4.5 Phonetic Analogy and the Fronting of the Nuclei of SHOES, BOOT, and BOAT

As discussed in section 4.3, as described in early Labovian research, parallel movement was viewed as a form of rule generalization/extension or rule simplification

(depending on how these concepts were used by analysts), with rule being defined in these contexts within the framework of generative phonology (e.g., Kiparsky, 1968;

King, 1969; Labov, et al., 1972; Labov, 1994). In more recent discussions, Labov and his associates (e.g., Labov, et al., 1972; Labov, 2010) have discussed parallel movement as a process of phonetic generalization, but without attaching it to any specific theoretical

158 framework for discussion. However, it should be noted that, although this does seem to frame the material in a framework-neutral perspective, aspects of Labov's discussion on phonetic generalization and children's dialect acquisition in his Principles of Linguistic

Change series (Labov, 1994, 2001, 2010) do suggest a kinship with recent models of linguistic grammar often invoked in discussions of phonetics, phonology, and the phonetics/phonology interface, such as exemplar theory (regardless of flavor, e.g.,

Skousen, 1989; Johnson, 1997; Pierrehumbert, 2002; 2003; Tomasello, 2003; Goldberg,

2006). Specifically his take on the role of pattern matching in the drawing of generalizations by speakers of vowel shifting behavior (cf, Labov, 1994), and also his perspective on the intricacies of how children appear to be subconsciously aware of adult and peer native-speakers patterns and levels of use of linguistic variables in their speech community (Labov, 2001, 2010), reflect this perspective. However, as Labov has pointed out in recent publications (Labov, 2006; 2010), he does not agree with the views of exemplar theorists like Pierrehumbert that frequency plays a fundamental role in the regularity of sound change, and in fact, uses data on the parallel fronting of /ow/ and

SHOES/BOOT in ANAE to show why he disagrees with their view. In the ANAE data,

Labov does not find strong evidence of the frequency effects linguists like Pierrehumbert posit are integral to the versions of the exemplar theory model they adhere to (Labov,

2006, 2010). Thus, we suspect findings such as these play a large role is his not

=grounding his recent findings in a theoretical framework such as exemplar theory.

159 In our own approach to discussing the process of generalization that occurs in parallel shifting, we take a somewhat different perspective from either Labov, generative phonologists as represented by Kiparsky (1968) and King (1969), or exemplar theorists like Pierrehumbert (2002; 2003) or Johnson (1997).23 Our perspective, we first articulated in Durian & Joseph (2011), is that the process of phonetic generalization that occurs during parallel shifts is phonetic analogy, specifically, phonetic analogy in the sense of

Schuchardt (1885) and Vennemann (1972). As defined by these scholars, phonetic analogy refers to a process in which a likening between two entities, along some definable dimension, in this case phonetic (and/or possibly phonological) is drawn between two entities, by a speaker (Durian & Joseph, 2011). As the chronology of the source dates indicates, Schuchardt (1885) was obviously the initiator of the concept, but

Vennemann's (1972) re-investigation of the concept during the era of generative linguistics nearly 100 years later allowed the concept to be reconceptualized for use in the modern linguistic era. Given that both of these conceptualizations prove useful for our understanding and discussion of phonetic analogy in our Columbus data, we draw on both linguist's interpretations of the concept in our discussion to follow.

As discussed by Schuchardt (1885), who wrote long before the field had conceived of modern formalisms such as generative rules, the likening could conceivably

23 Here, by different, we mean simply "in constrast" to perspectives discussed previously. Vennemann (1972) argues that phonetic analogy is compatable with a generative perspective, while we also imagine an exemplar theorist like Pierrehumbert would likely argue phonetic analogy is compatable with an exemplar theoretic perspective. Here, our contrast is meant to be read from a theory neutral perspective, hence our discussion of it "in contrast" to both the generative and exemplar theoretic perspectives discussed earlier. 160 be drawn by a speaker between two sound segments--say two different consonant sounds-

-or between sound clusters in words, such as consonant clusters (Durian & Joseph, 2011).

During the process of the spread of a sound change, the inception, or to use the term suggested by Weinreich, et al. (1968), the actuation of change, occurs in a very restricted set of phonetic environments. This is what Schuchardt (1885) calls a sporadic change. As the change spreads, it becomes less sporadic, affecting more and more phonetic environments. As time goes on, the change begins to become quite regular, until all segmental candidates that could potentially undergo the change do so.

For Vennemann (1972), who wrote during the era of generative linguistics, the likening could be made not only between segments, but also between different linguistic rules (Durian & Joseph, 2011). This likening allows for speakers to make generalizations across different linguistic rules, such as in rule generalization, or to reorder rules based on their phonetic similarity, such as in rule reordering, as well (Vennemann, 1972:185-187).

In addition, rule simplification, as conceptualized by generative phonologists like King

(1969) and Kirparsky (1968) would also be a case of phonetic analogy in Vennemann's

(1972) interpretation of the concept. For Vennemann, all of the types of rule generalization/extension and simplification that Labov, et al. (1972) associate with parallel movement and the generalization process that accompanies it typify phonetic analogy.

As applied to a sound change or sound change in progress affecting vowel classes,

161 the sporadic change of Schuchardt (1885) would first occur in the highly restricted environment of some specific consonantal segment, or perhaps a set of segments--say only voiceless fricatives following a vowel--with only those segments conditioning a change in realization of the vowel. In our hypothetical example, let us say these segments condition raising of the nucleus of the vowel. As time goes on, the spread of the sound change would occur, impacting an ever-increasing number of consonantal segments and segment groups, expanding from voiceless fricatives only to say, front nasals and voiceless fricatives, then front nasals, voiceless fricatives and voiced stops, and then finally front nasals, as well as all fricatives and stops, regardless of voicing. One could also imagine that the same thing could happen with segments organized by place features, starting perhaps with coronals, then spreading to labials and then finally dorsals, as well.

The end stages of the change in these examples represent the point at which the change would have affected all possible candidates for the change.

When generalized across vowel classes, with one vowel class undergoing the process of environment by environment change just discussed, and another vowel class beginning to show the same changes, with some delay, but also mirroring the changes of the first class at a later time, parallel movement shows the process of phonetic analogy spreading not just within a vowel class, but also across vowel classes. This cross-vowel class spread showcases phonetic analogy the way Vennemann (1972) envisions its role in spreading sound change, particularly his view of how phonetic analogy influences rule

162 generalization/rule extension within the generative framework in which he casts phonetic analogy. From this perspective, the generalization of the rules is what would lead to cross-vowel class phonetic generalization of the vowel shift from one vowel class to another.

Although this all sounds good in theory, what actually matters is whether adopting this perspective does anything for us in terms of better accounting for phonetic generalization in actual data. In our data, we argue that it does do something for us in terms of better explaining phonetic generalization, as we will now demonstrate. In our data, the process of phonetic analogy is clearly at work due to the way the process of generalization occurs during the inception, incrementation, and development of parallel shifts, such as the fronting of the back diphthongs in Columbus. The kind of segment by segment, and segment group by segment group, type of spread of innovation that

Schuchardt (1885) describes is quite evident in our data. Meanwhile the kind of general expansion of systematically regular change across vowel classes, envisioned by

Vennemann (1972) in his rule-based reinterpretation, can also be found in our data.

This process of the spread of change just described can be seen when we compare the results of our lmer analyses for SHOES, BOOT, and BOAT and the vowel systems of representative speakers in our data. Turning first to our lmer analyses, we first need to compare the results of our lmer analysis of BOAT in Table 4.3 with the results of our analysis of SHOES and BOOT. Actually, we want to make a comparison of our

163 combined SHOES and BOOT for the best comparison between BOAT and /uw/ as full vowel classes. Table 4.4 provides this more direct comparison, via a new lmer analysis of the combination of the SHOES and BOOT subclasses into one unified vowel class. This analysis allows us to compare the results for this combined /uw/ category with the Table

4.3 results for BOAT. This side-by-side comparison is most easy to see when we compare just the significant segments in Tables 4.3 and 4.4, as in Figures 4.1 (for /uw/

(Combined SHOES+BOOT)) and 4.2 (for BOAT).

Based on this more direct comparison, we see that the robust force of preceding coronal segments compared to non-coronals we discussed in section 4.3 for SHOES and

BOAT is still quite transparent, even when a mix of the SHOES and BOOT preceding segments is used for SHOES/BOOT. In Figure 4.1 for SHOES/BOOT, we see that the preceding segments /dʒ/, /d/, /n/, /s/, /ʃ/, /t/, /z/, and /tʃ/ all of which are coronals, are among the most statistically significant preceding segments which condition fronting.

Other segments are also significant, of course, but these are clearly the leading segments, as revealed by their high positive coefficient and t values. Comparing coefficient and t values for these segments with those occurring in preceding position for the BOAT class in Figure 4.2 reveals a strong degree of parity between results. That is, coronals preceding BOAT show quite similar coefficient and t values to the SHOES/BOOT preceding coronals.

Ranked from highest to lowest value, the coronal segments which are the same

164 Coefficient t value Significance (Intercept) -0.826 -3.773 0.000 Sex: Woman 0.293 1.572 0.065 Generation: G2 0.602 3.909 <0.001 Generation: G3 1.012 6.548 <0.001 Generation: G4 1.187 8.071 <0.001 Collar: Blue 0.410 2.497 0.004 Preceding /ʒ/ 0.954 3.197 <0.001 Preceding /d/ 0.830 4.088 <0.001 Preceding /f/ 0.666 2.818 <0.001 Preceding /g/ 0.222 0.641 0.614 Preceding /h/ 0.362 1.644 0.057 Preceding /k/ 0.548 2.140 0.013 Preceding /m/ 0.252 0.903 0.322 Preceding /n/ 0.950 4.241 <0.001 Preceding /s/ 0.994 4.878 <0.001 Preceding /ʃ/ 1.157 7.039 <0.001 Preceding /t/ 1.018 5.284 <0.001 Preceding /w/ 0.181 0.836 0.566 Preceding /z/ 0.873 2.413 0.007 Following /b/ 0.156 0.485 0.610 Following /d/ -0.239 -1.790 0.005 Following /dʒ/ 0.793 2.316 0.002 Following /f/ 0.380 1.070 0.182 Following /g/ 0.608 2.179 0.001 Following /w/ -0.338 -0.742 0.422 Following /k/ -0.785 -2.444 0.010 Following /m/ -0.363 -0.796 0.318 Following /p/ -0.297 -1.633 0.067 Following /s/ -0.214 -1.251 0.364 Following /ʃ/ -0.781 -1.957 0.035 Following /t/ -0.132 -0.999 0.267 Following /θ/ -0.153 -0.524 0.513 Following /ð/ 0.020 0.084 0.847 Following /v/ -0.359 -1.568 0.089 Table 4.4: Coefficients in the Linear Mixed Effects Model Predicting the F2 of /uw/ (SHOES/BOOT) from the Social Variables (Sex, Generation, Class), Their Interactions, and Following Phonological Environment. P-values were obtained using Markov Chain Monte Carlo simulations (Baayen et al.., 2008). 165 Following /z/ -0.006 -0.049 0.798 Woman*G2 -0.160 -0.665 0.368 Woman*G3 -0.273 -1.188 0.162 Woman*G4 -0.536 -2.393 0.004 Woman*Blue -0.255 -0.974 0.252 G2*Blue -0.301 -1.332 0.102 G3*Blue -0.380 -1.801 0.042 G4*Blue -0.260 -1.255 0.143 Woman*G2*Blue 0.310 0.889 0.283 Woman*G3*Blue 0.330 1.049 0.242 Woman*G4*Blue 0.280 0.873 0.314 Table 4.4: Coefficients in the Linear Mixed Effects Model Predicting the F2 of /uw/ (SHOES/BOOT) from the Social Variables (Sex, Generation, Class), Their Interactions, and Following Phonological Environment. P-values were obtained using Markov Chain Monte Carlo simulations (Baayen et al.., 2008).

before SHOES/BOOT and before BOAT demonstrating this parity are as follows. For

SHOES/BOOT, the segments rank /ʃ/, /t/, /s/, /tʃ/, /n/, and then /d/. For BOAT, the segments rank /t/, /ʃ/ /s/, /tʃ/, /d/, and then /n/. There is also parity for non-statistically significant segments, suggesting a similar type of parity typifies non-coronal segments, as well. For the non-coronal segments, those preceding SHOES/BOOT rank as follows: /f/,

/k/, /h/,/m/, and /b/. For BOAT, they rank /k/, /h/, /f/, /b/, /m/. Here the similarity is not as strong, but then again, perhaps this should not be a surprise given that these segments also less significantly condition fronting of SHOES/BOOT and BOAT overall. If the fronting influence were stronger, based on the suggestive evidence provided by the relative conditioning behaviors of individual coronal segments to each other, it seems quite plausible that we should expect that, if the significance level of individual non- coronal segments were also higher, there should also be stronger numerical parity

166

Figure 4.1: Following Segments Conditioning /uw/-Fronting. All segments significant at p<.05 or greater. P-values obtained using Markov Chain Monte Carlo simulations (Baayen, et al.., 2008). Standard error is indicated by error bars. Baseline segment is /ʒ/.

Figure 4.2: Following Segments Conditioning /ow/-Fronting. All segments significant at p<.05 or greater. P-values obtained using Markov Chain Monte Carlo simulations (Baayen, et al.., 2008). Standard error is indicated by error bars. Baseline segment is /b/.

167 between coefficient and t values for these segment groups.

The overwhelming similarity in conditioning of SHOES/BOOT and BOAT shown by the segments in Figures 4.1 and 4.2, we argue, provides statistical confirmation of the fact that speakers do appear to generalize patterns of phonetic conditioning from a smaller set of very specific segments which are the initial conditioners of vowel variation, initially for SHOES/BOOT, and then later for BOAT, to an increasingly larger set of segments over the course of a vocalic sound change. In particular, the expansive pattern appears to spread across segment groups that are initially quite similar to one another

(coronals), and then to those which are increasingly more different (the non-coronals). In the case of the vowel classes exhibiting parallel movement--such as SHOES/BOOT and

BOAT--this generalization pattern affects more than one vowel class simultaneously, with the pattern of spread across segment groups of one vowel class, in this case

SHOES/BOOT, being mimicked by a second vowel class, in this case BOAT.

Further evidence of the mimicking of the conditioning behavior of segments before SHOES/BOOT by segments before BOAT can be seen when we observe firsthand the movement and clustering of individual tokens within each vowel class in the vowel systems of individual speakers in our data. Such observations can be made of the tokens plotted for the various speaker systems presented in the vowel plots presented in figures

4.3 and 4.4, which show the speakers who tend to be the most innovative among the oldest 3 generations of speakers in our study (Blue Collar women) contrasted with the

168

Figure 4.3: Progression of SHOES, BOOT, and BOAT Fronting in Blue Collar Women's Speech (Blue Collar women are the most innovative social class group overall for these vowel classes (exccept among G4 speakers) in all 4 generational cohorts in our study, as demonstrated by select exemplary individual speaker vowel systems. Note how alveopaltals clearly lead as a conditioning environment for /uw/ (as shown by SHOES (blue) versus BOOT (green)) and /ow/ (BOAT (red)) in older women's speech, but then show a less prounced lead in younger women's speech, where non-aleveopaltal segments appear to condition fronting much like alevepalatal segements did in older women's speech. This is seen more clearly by contrasting the G1 and G2 women in the top row with G3 and G4 women in the bottom row of this figure. These plots illustrate the gradual nature of phonetic analogy as it impacts the realizations of these vowel classes in the vowel systems of Columbus speakers.)

169

Figure 4.4: Progression of SHOES, BOOT, and BOAT Fronting in White Collar Men's Speech (White Collar men are the most conservative social class group overall for these vowel classes (except among G4 speakers) in all 4 generational cohorts in our study, as demonstrated by select exemplary individual speaker vowel systems. Note how alveopaltals clearly lead as a conditioning environment for /uw/ (as shown by SHOES (blue) versus BOOT (green)) and /ow/ (BOAT (red)) in older men's speech, but then show a less prounced lead in younger men's speech, where non-aleveopaltal segments appear to condition fronting much like alevepalatal segements did in older men's speech. This is seen more clearly by contrasting the G1 and G2 men in the top row with G3 and G4 men in the bottom row of this figure. This demonstrates that even among the most conservative speakers in the study, the effects of phonetic analogy for these vowel classes can be seen farily clearly in our data.)

170 speakers who tend to be the most conservative among the oldest 3 generations (White

Collar men). Regardless of group, we see the generalization process within vowel class groups is the same. Coronal consonants lead in conditioning innovation, while non- coronal consonants follow. In older speaker systems, we see more robust fronting conditioned just by the coronal consonants, since coronal consonants naturally condition fronter variants of back vowels (Thomas, 2001; Labov, et al.., 2006). But then among increasingly younger speakers, we see the non-coronal consonants begin to act like the coronal consonants in older speakers vowel systems, with the non-coronal consonants now also beginning to condition fronting. The coronal consonants still condition the frontest realizations, but the gap between differing segment types decreases as time goes on and the non-coronals begin to act more and more like the coronal segments in terms of conditioning fronting.

In addition, in these plots, we can clearly see the analogization of phonetic conditioning impacting fronting from one vowel class to another vowel class. One of the two classes, in this case SHOES/BOOT, serves as the innovating class, and the pattern of segmental spread is initiated first across segments preceding only SHOES/BOOT. As time goes on, however, the pattern is generalized to a second vowel class, with the behaviors evidenced by innovative conditioning segments affiliated with the first vowel class being "copied" or “cloned” by the second class, in this case BOAT. The "cloning" involved is a pattern matching behavior, with the end result of this behavior being that

171 innovative conditioning segments affiliated with the second class end up approximating quite closely the exact same innovative behaviors shown by conditioning segments affiliated with the first class. In this way, changes impacting the segments conditioning variation in the first vowel class serve as a model for the conditioning of variation of the second class. This can be seen most clearly by contrasting the movement of the classes in the systems of older speakers, who appear in the top row of the figures, with those of younger speakers, who appear in the bottom row.

One final comment we wish to make is that the distinction that we've been discussing here, between analogy that impacts segments within vowel classes and an analogy that impacts vowels across classes suggests that two types of phonetic analogy occur in the process of parallel shift. The first we believe should be called internal vowel class analogy, which leads a vowel class to undergo a sound change on a segment by segment basis, as per our description above. The second we believe should be called external vowel class analogy, which leads to the spread of the internal analogy behaviors shown by one class to a second class, as per our discussion above.

172 Chapter 5: The Inception and Development of the Third Dialect Shift in Columbus24

5.1 Introduction

In this chapter, we investigate the occurrence of the Canadian/California Shift, the covariant shift of BOT, BAT, BET, and BIT, which we find in the vowel systems of many Columbus speakers in our data set. As discussed in Chapter 3, this vowel shift is typically defined as involving a low vowel chain shift, in which the retraction of BAT is linked to the merger of BOT and BOUGHT, and a front vowel shift involving the retraction of BET and BIT. In some definitions of the Shift, BET and BIT are seen as being chain linked to the backing and lowering of BAT, while in other definitions, BET and BIT are linked to the backing of BAT only as a parallel shift. In the second half of this chapter, we will consider which of these groupings makes more sense in our data, as the grouping of these vowels in dialect areas where the Canadian/California Shift is found is a point of significant interest in the field at present.

First, however, we will discuss previous studies of the Canadian and California

Shifts, as well as similar shift behavior in the US Midland, to provide some context for

24 Aspects of this chapter were first presented in Durian (2008b), Durian (2009) and Durian (2012). Aspects of Section 5.5 reflect our take on similar argumentation which was also used by Jacewicz, et al. (2011a) in describing aspects of the Third Dialect Shift. (as the North American Shift) We presented this argumentation earlier, in Durian (2009), which is a source cited by Jacewicz, et al. (2011a), but this aspect of our presentation is not discussed in that article. We are uncertain why this is the case, since they do discuss other findings of Durian (2009) in the article. Given that we presented them earlier in time, we attribute the arguments solely to us in the text of this chapter. This chapter owes a debt of gratitude to Charles Boberg and Matt Gordon for influencing our views on the Third Dialect Shift, both through their personal conversations with us and through their writing, as our views are presented in this chapter. 173 our analysis of what appears to be the occurrence of the same shift here in Columbus. We then move on to a consideration of the results of lmer analyses for each individual vowel class typically argued to be included in the larger series of shifts usually grouped as both the Canadian and the California Shift in the literature. That is, not only BIT, BET, and

BAT, but also BOT. Along the way, we will also consider lmer analysis results for

BOUGHT, since, as we also noted in Chapter 3, many of the speakers in our data set appear to be engaging in either merger, or more often, close realization of the BOUGHT and BOT classes. Following these analyses, we will consider whether the merger of

BOUGHT and BOT is actually the triggering event for the version of the Canadian and

California Shift we find in Columbus or not, since many of our speakers show only close realization, yet many of our young speakers also engage in the Canadian/California Shift- like vowel shift in our data. We then explore the evidence suggesting BOT, BAT, BET, and BIT are involved in one unified chain shift in Columbus, as suggested in the standard model of the Canadian and California Shifts (as per Labov, et al., 2006; Clarke, at al.

1995; Eckert, 2004). We also consider the evidence suggesting BOT and BAT are involved in a chain shift, while BAT, BET, and BIT are involved in a separate parallel vowel shift, as suggested in the "alternative" model of the Canadian Shift, presented by

Boberg (2005) in his analysis of the Canadian Shift as it occurs in Montreal, Canada.

Before doing this, however, we first wish to introduce the term Third Dialect Shift as a way of referring to the vowel shift pattern(s) that will be discussed throughout this

174 chapter. We do this for several reasons. First, we do so in light of recent research, including our own pilot work presented previously as Durian (2008/2009), which demonstrates that the same general vowel shift(s) found in California and Canada and discussed above are also found in the US Midland. Combining the three areas we have just discussed where the pattern of vowel shifts we will be discussing have now been confirmed to have been found--Canada, the US West, and the US Midland- -we can see that this set of shift(s) typifies a large stretch of North America that was originally clustered together by Labov in (1991) as the Third Dialect area of North American

English. Given this earlier cluster-grouping by Labov, and the fact that it aptly applies to the areas showing overlapping shifting tendencies for these vowel shift(s), we thus choose to use Labov's designation as a way of labeling them. Hence, we shall from this point forward use the term Third Dialect Shift in place of Canadian Shift or California

Shift for describing the backing of BOT, BAT, BET, and BIT as a grouped set of shifts.

5.2 Previous Studies of the Third Dialect Shift in North American English

The earliest reports of the pattern of vowel shifts we describe in this chapter as the

Third Dialect Shift actually were presented by Luthin (1987) for California. At the time, the shift was not given a specific name, although the major movements of the shift were reported in some detail. Among the patterns of shift Luthin reported were the retraction of BAT, BET, and BIT, as well as the merger of BOT and BOUGHT. Clarke, et al..

175 (1995) reported this same pattern for Canada among speakers living in Ontario. As

Clarke and her associates noted at the time, the version of the shift they were observing in

Ontario at the time did in fact seem to bear a striking resemblance to the California version discussed by Luthin. However, one significant difference noted by Clarke and her associates between the versions of the shift in California and Canada was the difference in the merger outcome for BOT and BOUGHT in the two different geographic locales. In

Ontario, the outcome of the BOT and BOUGHT merger more closely resembles a higher, rounded back vowel similar to BOUGHT, while in California, the outcome of the merger more closely resembles a lower, centralized vowel that is fairly similar to BOT. Given this difference, as well as the more obvious geographic difference between locales,

Clarke and her associates named the version of the shift they observed in Ontario the

"Canadian Shift," deciding to play up the differences between California and Canada, rather than the similarities, which they might have done using a term like the Third

Dialect Shift themselves, if they had chosen to do so.

Initially, the retraction behavior shown by BAT, BET, and BIT in the versions of the Third Dialect Shift noted in both Luthin (1987) and Clarke, et al. (1995) was primarily downward in nature, rather than backwards (or backwards and downwards), although BAT was said to move somewhat backwards as well, even though BET and BIT did not. Meanwhile BOT and BOUGHT were simply noted to merge in both dialects, with the difference in merger outcomes noted by Clarke, et al. (1995) between the

176 locales. Towards the end of the study, Clarke, et al. note that further exploration of the similarities between California and Canada should be conducted, but then leave the issue of conducting a more involved comparison between the areas based on their data and

Luthin's to the side.

Following these initial studies, several additional studies in each locale confirmed the patterns found initially by Luthin (1987) and Clarke, et al. (1995). DeDecker and

Mackenzie (2000) confirmed the operation of the shift in Ontario, while Eckert (2004) and Hagiwara (2005) confirmed its operation in California. In addition, others began to find other areas in Canada and the US West showing speakers with vowel systems that demonstrated the operation of the Third Dialect Shift, such as St John's, Newfoundland

(D'Arcy, 2002), Halifax, Nova Scotia and Vancouver (Sadlier-Brown & Tamminga,

2008), Montreal (Boberg, 2005), Winnipeg (Hagiwara, 2006), Toronto (Roeder &

Jarmasz, 2008), and Northern Arizona (Hall-Lew, 2004). These studies confirmed the downward, chain-shift like movement of BIT, BET, and BAT, although some studies, such as Boberg (2005), found a different pattern of shift for the front vowels. In Boberg's

(2005) study of Montreal, he found BIT and BET to show only significant backward movement, which led him to posit that BIT and BET might actually be a parallel shift, rather than a chain shift, in Montreal, since BIT and BET were not showing any "hole filling" behaviors by only moving backwards.

In Labov, et al. (2006), more documentation of the Third Dialect Shift in cities in

177 the US West and throughout Canada was provided, although only the Canadian version was described using the formal term "Canadian Shift" in the study. Like Boberg (2005), they found a different pattern of shift for BET than Clarke, et al. (1995), although the pattern they found was also different from what Boberg (2005) found. Instead, they found

BET to both lower and back, with backing being shown first, and then later lowering.

BIT was not found to show significant movement behavior in either direction, however.

In addition, like the other follow up studies discussed above, Labov, et al. (2006) also found BAT backing and lowering, and the low back merger, to be a part of the Third

Dialect Shift in the areas they noted it to occur (as the Canadian Shift).

Throughout the time period that these follow up studies were conducted, researchers seemed content to focus on documenting the Third Dialect Shift in only the

US West and Canada, and also only to look at the occurrence of the Shift in isolation of other areas. No follow up study was conducted to confirm the similarities between

California and Canada directly. Meanwhile 4 studies noted BAT backing and lowering to occur along with BOT/BOUGHT merger among Midland speakers in Johnstown and

Worthington, OH, as well as Indianapolis, IN (Thomas, 1996; Dodsworth, 2003, 2005;

Fogle, 2008). However, no researchers moved to studying the US Midland25 to see if the

Shift might also be occurring in the other part of North America that Labov (1991) billed

25 Note however, that both Boberg (2000) and Dodsworth (2003) present speaker data from Central Ohio speakers hinting at the possibility that the Third Dialect Shift might be found quite soon. The Boberg (2000) speaker shows hints of retraction in her BET class, while Dodsworth (2003) had speakers who we now know do Third Dialect Shift, but in 2003, the analysis seemed a bit of a stretch, so Dodsworth ended up leaning towards a Northern Cities Shift influence in her analysis of the patterns instead. 178 the Third Dialect.

In the late 2000s, however, two studies of cities in the US Midland provided confirmation of the operation of the Third Dialect Shift in that area: Durian (2008, 2009), an early pilot study presentation of data from this dissertation, found the Third Dialect

Shift among speakers in Columbus, while Bigham (2009) found it among young adults living in several small cities in central and southern, Illinois. In both studies, all of the elements of the Third Dialect Shift were found to occur among speakers born after 1980, although Durian, having older speakers than Bigham, actually traced it in the systems of speakers born as early as 1950. Both studies also found the retraction of BET and BIT in the US Midland to be of the backwards, parallel shift variety of Boberg's (2005) Montreal data rather than the ANAE (2006) backwards and downwards variety or the Clarke, et al.

(1995) downwards variety. (The analysis of this aspect of the shift as a parallel or chain shift will be something we will return to in section 5.5.) Durian (2008, 2009) drew a connection between the Canadian Shift and the US Midland pattern, while Bigham drew a connection between the California Shift and the US Midland pattern. In addition,

Bigham (2009), using ANAE's normalization and vowel scaling procedure, found that the

US Midland version of the shift seemed to match with that of California speakers in the study who also showed the Third Dialect Shift, providing the first hard evidence of the similarity of the Third Dialect Shift in two of the three dialect areas where it has been found. Meanwhile, in California, Grama and Kennedy (2009) (later published as

179 Kennedy & Grama, 2012) followed a similar procedure to Bigham, with data collected in various cities in California, using ANAE's normalization and scaling procedure to make a comparison between their speakers and ANAE's Canadian speakers, to determine if

Clarke, et al.'s (1995) hypothesis that California and Canada have the same shift was valid. Although they found the outcome of the BOT/BOUGHT merger to be quite different in their data from the outcome in Canada, they did generally find the Third

Dialect Shift to be quite similar overall in both geographic locales.

Following both of these studies, Jacewicz (2011a, b) further confirmed the operation of the Shift among Central Ohioans born after 1958, although only the low vowel component (the BOT/BOUGHT merger and BAT-backing/lowering), as the front vowel component was not found in their data. Given that their study covered North

Carolina, Central Ohio, and Wisconsin, they also confirmed that the front vowel component appears to operate as a parallel shift, but only in their North Carolina data, given that front vowel shifting was not found in their Central Ohio data. However, as we discussed in Chapter 3, this may be because they used citation form data rather than natural conversational data. It may also be, however, that their speakers are also less advanced in showing it than our speakers, since, as we will see, not all younger speakers show all components in our data either. Finally, they confirmed instrumentally that the

North Carolina and Central Ohio versions of the Shift are the same. (However, it remains to be determined if Central Ohio and Canada or California are truly the same, given they

180 did not have data for either locale.)

At the time of the writing of this dissertation, given the research results we just discussed, it might seem like a detailed discussion of the occurrence of the Third Dialect

Shift in Columbus is less necessary. However, our discussion here provides a wealth of new detail on the inception and historical development of the Third Dialect Shift in

Columbus and the US Midland more generally that has not yet been discussed previously in Jacewicz, et al. (2011a,b), Durian (2008/2009), or Bigham (2008/2009). As well, we also provide new information that helps to resolve the controversy of whether the Shift, as it is been observed in California, Canada, and the US Midland, is in fact the same shift or not, which is a question that has been asked since Clarke, et al.'s (1995) analysis of the

Third Dialect Shift in Ontario, and yet, as of the time of this writing, has still not been answered definitively. For these reasons, we now present our analysis of the Third

Dialect Shift in Columbus, beginning with lmer analysis of the vowel classes that comprise the Third Dialect Shift in Columbus.

5.3 Linear Mixed Effects Regression (lmer) Analysis of Nuclear F1 and F2 of the Third Dialect Shift Series Vowels (BAT, BET, BIT, BOT (and BOUGHT))

We now turn to lmer analyses of changes in the value of F2 for all of the vowel classes involved in the Third Dialect Shift. We will also analyze changes in the value of

F1, when applicable, as well, since F1 variation is lingusitically and socially significant for several of the Third Dialect Shift vowels. In addition, just as in Chapter 4, we will 181 investigate specifically the impact that the linguistic (phonetic) factors of preceding and following segment, as well as the social factors we have chosen to stratify our data by

(age, social class, and sex), have had on variation in F1 and F2 values for the nuclei of all vowel subclasses investigated among speakers born during the course of the 20th

Century. To do so, we will investigate the values of specific coefficients that allow us to predict the F2 or F1 of each subclass in a lmer model from the chosen social and linguistic factors. Unlike the vowel classes in the last chapter, however, in our analyses here, we will focus in most cases on backing rather than fronting behaviors exhibited by most of the vowels on F2. The exception is fronting among older speakers and younger

Blue Collar men for F2 of BAT (although note that backing is the change of interest among our other younger speakers), and the fronting of the nucleus of BOUGHT among some older speakers.

As in Chapter 4, we again will use one specific social group as a baseline group for all lmer analyses presented. However, unlike the previous chapter, the group used in this chapter will be much more variable, and thus, we will specify who the social group used is for each analysis. In each case, the social group selected will be chosen as such because their vowel system is the most conservative for the given vowel class under analysis. We also again include speaker and token as random effects in the lmer analyses.

Both factors were included unless log likelihood tests showed only one should be included. If one of the factors was excluded, this fact will be referenced in the text. In

182 addition, we also use the same generational cohort age groupings here to quantify age that we used in Chapter 4.

5.3.1 F1 of BAT (+BAG)

Turning first to the lmer analysis of F1 of BAT, our model reveals a number of coefficients which are significant predictors of F1 realizations among speakers. These significant coefficients are listed in table 5.1 in bold. For the analysis of F1 variation, positive coefficients indicate a less raised /ae/ production, whereas negative coefficients indicate a more raised /ae/ production. The model used for this analysis uses the vowel systems of the G1 Blue Collar men as the baseline for the analysis. This was done, since, for this particular vowel class, the vowel systems of these men were actually more conservative than those of the G1 White Collar men. /b/ is used as the baseline for the phonetic segment analysis. The model used for this analysis includes both speaker and token as a random effect, and it accounts for 39% of the data.

As discussed in Chapter 3, in Columbus, since at least the beginning of the 20th

Century, speakers in the area show a robust tendency toward realizing /ae/ before velars and nasals significantly differently than how they realize /ae/ before other types of segments. This tendency mirrors a more general tendency among most US English speakers since about the same time period to realize /ae/ significantly higher before nasals than before other segments, while also realizing it somewhat lower and backer

183 before velars than before other segments, as demonstrated in data analyzed in dialect atlas studies since the 1930s (e.g., Kurath, et al., 1939; Kurath & McDavid, 1961; Thomas,

2001; Labov, et al., 2006) for many parts of the contential US. Given this fact, in our descriptive analyses of /ae/ here in Chapter 5, we have chosen to maintain a subdivision of /ae/ into three subclasses, just as we did earlier in Chapter 3--those before nasals, those before velars, and those occuring before any other segment types. Following similar conventions, we again refer to these subgroups as BAN, BAG, and BAT in the analyses featured below. However, given the tendency of BAG and BAT to differ much less significantly from one another than either of these subclasses differ from BAN in the statistical analyses, for the purposes of the lmer analyses presented in this chapter the

BAT and BAG subclasses will be grouped together (although the BAN subgroup remains distinct).

Turning first to linguistic factors, only one--following segment, specifically, the voiced following segments /d/ and /ð/--is statistically significant. Both of these segments significantly condition nucleus raising. Moving on to social factors, we can see that age

(quantified as G2, G3, and G4) is a significant independent factor, while significant interactions of social factors include Woman*G3, Woman*G4, White*2,

Woman*G3*White, and Woman*G4*White. Generally, these results reveal that /ae/ lowering has increased over time across all social groups in Columbus, as shown by the increase in positive coefficient and t-values as age decreases. However, as the significant

184 interaction of age*social class (G2*White), as well as a general consideration of the coefficient values shown for the non-significant interactions sex*age*social class and the remaining, also non-significant interactions for age*social class, reveal, the increase in lowering shows an interesting social distribution among the age groups, with Blue Collar men and women generally leading the change, and White Collar men and women lagging somewhat in its adoption.

Specifically, a consideration of the values of the coefficients in table 5.1 for each social group reveals that, although both men and women initially raise in G1 and then begin to lower over time, the reversal to a realization pattern of lowering among the groups occurs within different generational groups for each sex group. Among women, the values show that men raised more than women among G1 speakers, regardless of class background, but then in G2, White Collar women continued to show some raising, while Blue Collar women begin to show lowering. In G3, White Collar women also begin to show lowering, a trend that continues into G4. Meanwhile, Blue Collar women also show continued lowering in G3 and G4, leading White Collar women by about one generation in regards to the incrementation of the change.

Among men, Blue Collar speakers in G2, G3, and G4 show a steady lowering trend as age decreases. As we noted earlier, given the increasing positive coefficient values of Blue Collar males across all generation groups, we argue this trend reveals these men lead over White Collar men on lowering. Overall, Blue Collar Women show

185 Estimate t value pMCMC (Intercept) 0.898 3.984 0.000 Sex: Woman 0.048 0.471 0.641 Generation: G2 0.247 2.739 0.006 Generation: G3 0.175 2.176 0.027 Generation: G4 0.392 4.982 0.000 Collar: White 0.096 1.016 0.326 Following /d/ -0.384 -1.650 0.044 Following /dʒ/ 0.165 0.266 0.778 Following /f/ -0.243 -0.977 0.339 Following /g/ -0.236 -0.884 0.382 Following /k/ 0.056 0.250 0.885 Followning /g/ -0.400 -1.591 0.065 Following /p/ -0.126 -0.522 0.654 Following /s/ -0.208 -0.909 0.297 Following /ʃ/ -0.232 -0.785 0.393 Following /t/ -0.277 -1.207 0.119 Following /θ/ -0.089 -0.331 0.577 Following /ð/ -0.669 -1.656 0.074 Following /v/ -0.226 -0.845 0.302 Following /z/ -0.555 -0.943 0.304 Woman*G2 0.145 1.066 0.254 Woman*G3 0.356 2.874 0.005 Woman*G4 0.283 2.298 0.020 Woman*White 0.103 0.651 0.387 G2*White -0.296 -2.340 0.023 G3*White 0.101 0.826 0.441 G4*White 0.072 0.618 0.495 Woman*G2*White -0.324 -1.625 0.063 Woman*G3*White -0.460 -2.437 0.010 Woman*G4*White -0.352 -1.894 0.034

Table 5.1: Coefficients in the Linear Mixed Effects Model Predicting the F1 of BAT (+BAG) from the Social Variables (Sex, Generation, Class), Their Interactions, and Following Phonological Environment. P-values were obtained using Markov Chain Monte Carlo simulations (Baayen et al.., 2008).

186 the strongest lowering trend across age groups, making them the leaders of linguistic change for BAT-lowering in Columbus.

5.3.2 F2 of BAT (+BAG)

For F2 of BAT, our lmer model reveals a number of coefficients which are significant predictors of F2 realizations among speakers. For this model, the vowel systems of G1 White Collar men are used as the baseline for the analysis, as their systems are most conservative for F2 of BAT. /b/ is used as the baseline for the phonetic segment analysis. The model used for this analysis also includes speaker as a random effect.

Turning first to significant social factors in the lmer analysis, age (listed as G2,

G3, and G4) sex (listed as Woman), and social class (listed as Blue) are all significant as an independent factors. In addition, interactions of sex*age (listed as Woman*G4), age*social class (Blue*G4) and sex*age*social class (listed as Woman*G4*Blue) are also all significant statistically. Considered together, these results suggest that, as our initial analysis in Chapter 3 also suggested, backing of the nucleus of BAT is increasing over time in Columbus, as shown by significant negative coefficients for F2 of

BAT+BAG for the three youngest generations in the study.

At first blush, this increase appears to be impacting both men and women of all social groups. However, as the results for the interactions also reveal, although no group seems to be a non-participant in the overall backing trend revealed by the data across

187 Estimate t value pMCMC (Intercept) 1.173 9.298 0.000 Sex:Woman -0.150 -2.120 0.023 Generation: G2 -0.188 -3.614 0.000 Generation: G3 -0.183 -3.457 0.001 Generation: G4 -0.544 -10.739 0.000 Collar: Blue -0.132 -2.398 0.012 Following /d/ -0.183 -1.433 0.138 Following /dʒ/ -0.518 -1.463 0.173 Following /f/ -0.250 -1.838 0.040 Following /g/ -0.357 -2.441 0.011 Following /k/ -0.545 -4.363 0.000 Following /ŋ/ -0.014 -0.101 0.959 Following /p/ -0.126 -0.944 0.376 Following /s/ -0.312 -2.475 0.008 Following /ʃ/ -0.273 -1.659 0.100 Following /t/ -0.241 -1.899 0.033 Following /θ/ -0.290 -2.007 0.055 Following /ð/ 0.063 0.277 0.736 Following /v/ -0.347 -2.435 0.008 Following /z/ -0.653 -1.905 0.056 Woman*G2 0.148 1.735 0.079 Woman*G3 -0.047 -0.562 0.673 Woman*G4 0.271 3.328 0.001 Woman*Blue 0.119 1.299 0.157 G2*Blue 0.100 1.354 0.156 G3*Blue 0.079 1.112 0.193 G4*Blue 0.502 7.365 0.000 Woman*G2*Blue -0.077 -0.666 0.490 Woman*G3*Blue -0.097 -0.880 0.291 Woman*G4*Blue -0.692 -6.376 0.000

Table 5.2: Coefficients in the Linear Mixed Effects Model Predicting the F2 of BAT (+BAG) from the Social Variables (Sex, Generation, Class), Their Interactions, and Following Phonological Environment. P-values were obtained using Markov Chain Monte Carlo simulations (Baayen et al.., 2008).

188 time, two groups show strong patterns of variation that distinguish them from other groups. The interaction of sex*age suggests that the youngest men in the study seem to be diverging rather quickly from G2 and G3 speakers by showing a renewed fronting trend, as revealed by the rather large positive coefficient value constrasted with the negative value shown by the 3-way interaction of Woman*G4*Blue (which suggests G4 working class women show instead a strong backing trend). This is an interesting juxtaposition which we will discuss in more detail later in this chapter.

In opposition to these speakers, young White Collar men in G4 actually show a rather strong backing trend for BAT, as revealed by the extremly large negative coefficient value for G4 and the fact that woman*G4 is positive, which suggests G4 white collar women lag behind White Collar men and Blue Collar women in showing a backing trend. In fact, G4 White Collar men stand out in contrast to women of both social class backgrounds in the study, who also back, by showing the strongest backing trend of all speakers. Overall, G4 White Collar men appear to be the linguistic leaders of change for the backing of BAT in Columbus. Turning to linguistic factors, the lmer analysis reveals five following segments, /g/, /k/, /f/, /v/, and /t/, have significant coefficient values. (The baseline following segment for this analysis is /b/.) All of these segments condition signficiantly backer variants of BAT than other segments.

189 5.3.3 F1 of BET

For our lmer analysis of the F1 of BET, we again used the vowel systems of G1

Blue Collar women as a baseline rather than the systems of the G1 White Collar men, just as we did for BAT, since again, the G1 Blue Collar women had the most conservative values for this vowel class. Again, /b/ is used as the baseline segment in this analysis, and again, speaker is included as a random effect.

As the results in table 5.3 reveal, no independant social factors were significant, and there were no significant interactions among social factors. We argue this is the case because, although there does appear to be a general lowering trend among all groups over time, the trend is never strong enough among any given speaker group to rise to the level of significance. This is likely the case because some degree of lowering usually occurs, simply as a byproduct of front vowel backing, since backing is typically accompanied by laxing of vowel tenseness. But as we saw earlier in our discussion of back vowel fronting, not all vowel movement turns out to be socially or linguistically significant when it occurs, and this appears to be another example where this is the case.

Turning to linguistic factors, one--following segment--is signficant. Specifically the following segments /d/, /g/, /v/, and /z/. Values for all 4 are negative, indicating that the segments significantly condition raising of the nucleus of BET, a finding that is perhaps not surprising, considering all 4 segements are voiced, and voiced segments often condition raised realizations of short front vowels (Thomas, 2001).

190 Coefficient t value Significance (Intercept) 0.653 2.220 0.027 Sex: Male 0.184 1.026 0.249 Generation: G2 -0.003 -0.019 0.989 Generation: G3 0.173 1.045 0.235 Generation: G4 0.257 1.570 0.082 Collar: White 0.208 1.064 0.260 Following /d/ -0.721 -2.733 0.006 Following /f/ -0.317 -0.865 0.463 Following /g/ -0.832 -2.423 0.023 Following /k/ -0.299 -1.128 0.306 Following /p/ -0.326 -1.193 0.242 Following /s/ -0.407 -1.540 0.137 Following /t/ -0.432 -1.635 0.133 Following /θ/ -0.122 -0.416 0.702 Following /v/ -0.662 -2.186 0.042 Following /z/ -0.986 -3.436 0.001 Male*G2 -0.171 -0.709 0.391 Male*G3 -0.220 -1.008 0.258 Male*G4 -0.299 -1.372 0.122 Male*White -0.157 -0.619 0.506 G2*White -0.109 -0.431 0.626 G3*White -0.042 -0.181 0.895 G4*White -0.371 -1.607 0.082 Male*G2*White 0.249 0.741 0.359 Male*G3*White 0.080 0.255 0.813 Male*G4*White 0.343 1.114 0.218

Table 5.3: Coefficients in the Linear Mixed Effects Model Predicting the F1 of BET from the Social Variables (Sex, Generation, Class), Their Interactions, and Following Phonological Environment. P-values were obtained using Markov Chain Monte Carlo simulations (Baayen et al.., 2008).

5.3.4 F2 of BET

For our lmer analysis of the F2 of BET, we used the vowel systems of G1 White

Collar women as a baseline since these women have the most conservative values for this

191 Coefficient t value Significance (Intercept) 0.783 3.844 0.000 Sex: Male -0.072 -0.883 0.400 Generation: G2 -0.198 -2.483 0.016 Generation: G3 -0.262 -3.561 0.000 Generation: G4 -0.178 -2.434 0.013 Collar: Blue -0.045 -0.510 0.619 Following /d/ -0.031 -0.154 0.718 Following /f/ -0.255 -0.924 0.268 Following /g/ 0.280 1.099 0.200 Following /k/ -0.165 -0.821 0.404 Following /p/ -0.293 -1.394 0.177 Following /s/ -0.253 -1.264 0.121 Following /t/ -0.180 -0.897 0.272 Following /θ/ -0.319 -1.384 0.125 Following /v/ -0.427 -1.849 0.041 Following /z/ -0.293 -1.270 0.122 Male*G2 0.072 0.682 0.595 Male*G3 0.054 0.533 0.546 Male*G4 -0.304 -3.088 0.003 Male*Blue 0.005 0.047 0.994 G2*Blue 0.170 1.496 0.199 G3*Blue 0.103 0.981 0.352 G4*Blue 0.081 0.786 0.447 Male*G2*Blue -0.050 -0.326 0.880 Male*G3*Blue 0.051 0.361 0.759 Male*G4*Blue 0.307 2.183 0.025 Table 5.4: Coefficients in the Linear Mixed Effects Model Predicting the F2 of BET from the Social Variables (Sex, Generation, Class), Their Interactions, and Following Phonological Environment. P-values were obtained using Markov Chain Monte Carlo simulations (Baayen et al.., 2008).

vowel class. The model used for this analysis also includes speaker as a random effect, but not token because it did not contribute significantly to model fit. As the results in table 5.4 reveal, only one independent social factor--age--is significant, while only one

192 linguistic factor--following segment, specifically /v/--is also significant. In addition, two interactions among social factors was also signficant--Male*G4 and Male*G4*White.

As revealed by further study of the significant independent social factor--age--we can see that speakers in G2, G3, and G4 all show significant backing trends. The interaction Male*G4 suggests both G4 Blue Collar and White Collar men show the significant backing patterns, while the interaction Male*G4*White indicates the most robust backing trend of all is shown by G4 White Collar Men. In constrast to then men,

G3 and G4 Blue Collar women also appear to be showing some reversal of backing tendencies, given the positive values shown in values for age*social class for both groups. White Collar Women continue to show some backing in G3 and G4, as well, although it is not as strong of a trend as either Blue Collar Men or White Collar Men.

5.3.5 F1 of BIT

For our lmer analysis of the F1 of BIT, we again used the vowel systems of G1

Blue Collar women as a baseline rather than the systems of the G1 White Collar men, just as we did for BET, since again, the G1 Blue Collar women had the most conservative values for this vowel class. The model used for this analysis also includes speaker as a random effect, but not token because it did not contribute significantly to model fit.

In contrast to F1 of BET, we find that a variety of social factors are significant for

F1 of BIT. In particular, the independent social factors of sex and social class

193 background, as well as the interaction of sex*social class, age*social class (specifically,

G2*class and G4*class) are significant. In addition, the linguistic factor of following segment is also significant, with /d/, /f/, /k/, /p/, /s/, /t/, and /θ/ showing significant, positive coefficient values. Positive values indicate that these segments signficantly condition BIT lowering.

Regarding the significant social factors, the main effects and interactions make for a somewhat more complex picture of variation in the community for F1 of BIT than other vowel classes. As a scan of table 5.5 reveals, in G1, Blue Collar females, who serve as the baseline, actually made use of the most raised realizations of BIT, followed by White

Collar men, then Blue Collar men, and finally White Collar women (who made use of the most lowered realizations). However, as time goes on and speaker cohorts get younger, differences between the sex and class subgroups of G1 more or less become flattened out, as the Blue Collar speakers (both men and women) show an overall trend toward lowering, while White Collar speakers (both men and women) show an overall trend toward raising. This is shown by the significant interactions for sex*class and age*social class, which are all negative, indicating the significant raising trend is found in the speech of White Collar speakers (both men and women) across generation groups, while the significant positive coefficients for the independent social factors of sex and social class indicate the Blue Collar speakers (both men and women) as a group making use of lowered BIT realizations in comparison.Within G4, this leads to White Collar speakers

194 showing a significant raising trend overall, while Blue Collar speakers make use of BIT realizations that are more or less the same as the “standard” BIT realization initially used

Coefficient t value Significance (Intercept) -1.181 -6.982 0.000 Sex: Male 0.214 1.990 0.038 Generation: G2 0.120 1.128 0.235 Generation: G3 0.100 1.018 0.312 Generation: G4 0.085 0.867 0.342 Collar: White 0.267 2.326 0.024 Following /d/ 0.330 2.049 0.015 Following /f/ 0.400 2.405 0.005 Following /g/ 0.248 1.470 0.107 Following /k/ 0.419 2.732 0.002 Following /p/ 0.368 1.812 0.045 Following /s/ 0.269 1.680 0.037 Following /ʃ/ 0.132 0.652 0.483 Following /t/ 0.459 2.965 0.001 Following /θ/ 0.573 3.066 <0.001 Following /v/ 0.184 0.703 0.429 Following /z/ 0.148 0.791 0.453 Male*G2 -0.118 -0.833 0.332 Male*G3 -0.155 -1.207 0.230 Male*G4 -0.164 -1.253 0.178 Male*White -0.324 -2.203 0.029 G2*White -0.291 -1.981 0.050 G3*White -0.247 -1.832 0.083 G4*White -0.278 -2.057 0.040 Male*G2*White 0.306 1.581 0.120 Male*G3*White 0.357 1.972 0.062 Male*G4*White 0.246 1.368 0.182

Table 5.5.:Coefficients in the Linear Mixed Effects Model Predicting the F1 of BIT from the Social Variables (Sex, Generation, Class), Their Interactions, and Following Phonological Environment. P-values were obtained using Markov Chain Monte Carlo simulations (Baayen et al.., 2008).

195 by the baseline group (G1 Blue Collar women).

Given the significance of the raising trend for White Collar speakers, it is important to consider this trend in context with the results of our earlier lmer analysis for

F1 of BET, where we also found a raising trend among these speakers--specifically White

Collar males. Both of these results (raising for F1 of BIT and F1 of BET) are suprising when we also consider that these same men show backing trends for F2 of both BET and

BIT (BIT is discussed below), given that previous studies of front vowel retraction in US and Canadian dialects of North American English have typically argued that the retraction is a result of a chain shift, with BET lowering and backing chain-linked to BAT lowering and backing, and BIT lowering and backing chain-linked to BET lowering and backing. Under this view, the lowering and backing of BAT opens a “hole” in the front of the vowel system, into which BET ultimately backs and lowers into, so that the “hole” becomes filled. This in turns opens a second “hole,” which is why BIT then also lowers and backs, to fill the open “hole” now left by BET.

Our data here suggests such a chain-link connection among the front vowels does not typify the vowel shifting process in Columbus among the leaders of linguistic change--our White Collar men--since robust lowering is not found to accompany backing of BET and BIT, even though BAT does lower and back in their vowel systems. Instead, the connection among the three vowels in Columbus seems to show a parallel movement process which appears to be similar to the kind of movement behavior shown by the back

196 diphthongs. However, in this case, as mentioned at the outset of this chapter, the parallel movement involves backing rather than fronting, like the back diphthongs show. This pattern becomes even more clear when we also look at the F2 movement trends shown among speakers for BIT.

5.3.6 F2 of BIT

As with F1 of BIT, the vowel systems of G1 Blue Collar women were used as a baseline for our lmer analysis of F2 of BIT, since these speakers have the most conservative values for this vowel class, as well. The model also includes speaker as a random effect, but not token because it did not contribute significantly to model fit.

As with F2 of BET, no social factors are significant for F2 of BIT. As mentioned earlier, coefficient values for BIT suggest backing typifies the speech of several groups in the study, most notably Blue Collar men and women, and G4 White Collar males.

However, the backing trends shown by these groups do not rise to the level of statistical significance due to the relatively low coefficient values of each social group. (For instance, although mild, G4 White Collar males show some degree of backing versus G3

White Collar males, as revealed by the decrease in coefficient values for

Male*G3*White, .212, versus Male*G4*White, .149.)

As was also the case with BET, one linguistic factor--following segment--is also significant for F2 of BIT, albeit this time, several segements are in fact significant, rather

197 than just /v/, as with BET. For BIT, the significant segements conditioning fronting are

/d/, /g/, and /k/. The significant segments conditioning backing are /v/ (as with BET) and

Estimate t value pMCMC (Intercept) 0.597 3.946 0.000 Sex: Male 0.103 1.052 0.277 Generation: G2 -0.026 -0.265 0.766 Generation: G3 -0.150 -1.679 0.102 Generation: G4 -0.013 -0.142 0.929 Collar: White 0.032 0.308 0.822 Following /d/ 0.345 2.411 0.008 Following /f/ 0.108 0.727 0.486 Following /g/ 0.460 3.067 <0.001 Following /k/ 0.244 1.783 0.039 Following /p/ -0.004 -0.020 0.991 Following /s/ -0.199 -1.398 0.092 Following /ʃ/ 0.254 1.398 0.132 Following /t/ 0.013 0.098 0.952 Following /θ/ -0.402 -2.417 0.007 Following /v/ -0.419 -1.787 0.050 Following /z/ -0.116 -0.693 0.406 Male*G2 -0.045 -0.351 0.720 Male*G3 -0.064 -0.550 0.550 Male*G4 -0.117 -0.981 0.293 Male*White -0.108 -0.805 0.453 G2*White -0.064 -0.476 0.690 G3*White -0.048 -0.392 0.748 G4*White -0.097 -0.788 0.443 Male*G2*White 0.126 0.719 0.497 Male*G3*White 0.212 1.291 0.209 Male*G4*White 0.149 0.912 0.365 Table 5.6: Coefficients in the Linear Mixed Effects Model Predicting the F2 of BIT from the Social Variables (Sex, Generation, Class), Their Interactions, and Following Phonological Environment. P-values were obtained using Markov Chain Monte Carlo simulations (Baayen et al.., 2008).

198 /θ/. Note, however, that, except for one token of “pithy,” all tokens of /θ/ included in our study are tokens of “with.” Because this is the case, we argue the significance of /θ/ may be related solely to the w_/th phonetic frame that all of these tokens thus feature, rather than /θ/ being a truly significant conditioning segment per se.

Following /v/ as a conditioning segment for BIT mirrors its significance for BET and most likely does so for BIT because, as a voiced fricative, /v/ typically encourages backing of short vowels phonetically. In addition, it may have risen to the level of statistical significance rather than simply occurring with a higher coefficient value than other segments for BIT via phonetic analogy with BET, in much the same way that coronals appear to more robustly condition fronting of BOAT than they might otherwise, due to phonetic analogy of /ow/ with /uw/, as discussed in Chapter 4. A detailed discussion of this possibility (and its implications) will be taken up in section 5.5. Before turning to this discussion however, we first need to discuss the remaining two vowel classes we have yet to touch on that participate in the Third Dialect Shift as it occurs in

Columbus: the low back vowels BOT and BOUGHT.

5.3.7 F1 of BOT

For F1 of BOT, our lmer model reveals one significant independent social factor-- age--along with no significant linguistic factors and also no signficant interaction effects.

For this analysis, G1 White Collar women are used as the baseline group for analysis,

199 since for BOT, White Collar women generally are the most conservative group for realzation of this vowel class. The model used for this analysis again also includes

Coefficient t value Significance (Intercept) 1.705 8.677 0.000 Sex: Male -0.215 -1.075 0.268 Generation: G2 -0.364 -1.865 0.057 Generation: G3 -0.467 -2.554 0.010 Generation: G4 -0.326 -1.797 0.062 Collar: Blue -0.153 -0.729 0.431 Following /ʒ/ 0.066 0.107 0.889 Following /d/ -0.092 -0.533 0.616 Following /f/ -0.554 -1.197 0.220 Following /g/ -0.307 -0.492 0.630 Following /k/ -0.003 -0.022 0.907 Following /p/ 0.090 0.638 0.379 Following pause -0.218 -0.351 0.739 Following /s/ -0.056 -0.307 0.908 Following /ʃ/ -0.249 -1.014 0.313 Following /t/ 0.022 0.156 0.881 Following /ð/ 0.003 0.010 0.968 Following /z/ 0.143 0.456 0.580 Male*G2 0.436 1.740 0.080 Male*G3 0.212 0.887 0.362 Male*G4 0.173 0.740 0.440 Male*Blue 0.092 0.343 0.716 G2*Blue 0.133 0.505 0.621 G3*Blue 0.200 0.823 0.387 G4*Blue 0.047 0.194 0.804 Male*G2*Blue -0.029 0.084 0.921 Male*G3*Blue -0.093 -0.290 0.737 Male*G4*Blue -0.035 -0.117 0.904 Table 5.7: Coefficients in the Linear Mixed Effects Model Predicting the F1 of BOT from the Social Variables (Sex, Generation, Class), Their Interactions, and Following Phonological Environment. P-values were obtained using Markov Chain Monte Carlo simulations (Baayen et al.., 2008).

200 speaker as a random effect, but not token because it did not contribute significantly to model fit.

As shown in table 5.7, age is a significant independant social factor, but only G3 has realization patterns different enough from G1 to rise to the level of significance. As a scan of the coefficient values for all generational groups reveal, it is actually the case thatall groups do show a trend toward raising for F1

5.3.8 F2 of BOT

For F2 of BOT, our model again uses G1 White Collar women as the baseline for analysis, this group being the most conservative for realzation of this vowel class. The model used for this analysis again also includes speaker as a random effect, but not token because it did not contribute significantly to model fit.

For F2 of BOT, no social factors are statistically significant, and only one linguistic factor--following environment--is. Specifically, following /d/, /k/, /p/, /s/, /z/, and pause are the significant segments, each conditioning the backing of BOT. As mentioned in Chapter 3, all generational groups beyond G1 show backed BOT realizations, with an increase in backed realizations as age decreases. This is reflected in the increasingly negative coefficient values contained in table 5.8, but none of these values rises to the level of significance. This should not be surprising, perhaps, given the subtley of BOT backing as a process, as also discussed earlier in Chapter 3. Something

201 new the added information here brings to the table, however, is some actual quantitative data showing the slow and subtle, yet steady, backwards movement of BOT across time

Coefficient t value Significance (Intercept) -0.447 -4.331 0.000 Sex: Male -0.003 -0.031 0.966 Generation: G2 -0.038 -0.436 0.630 Generation: G3 -0.106 -1.298 0.183 Generation: G4 -0.113 -1.396 0.143 Collar: Blue 0.066 0.695 0.561 Following /ʒ/ -0.066 -0.229 0.802 Following /d/ -0.200 -1.976 0.021 Following /f/ 0.111 0.478 0.644 Following /g/ 0.363 1.225 0.207 Following /k/ -0.139 -1.500 0.050 Following /p/ -0.133 -1.514 0.050 Following pause -0.747 -2.535 0.009 Following /s/ -0.315 -3.000 0.001 Following /ʃ/ -0.051 -0.350 0.854 Following /t/ -0.020 -0.226 0.771 Following /ð/ -0.158 -1.083 0.215 Following /z/ -0.416 -2.311 0.005 Male*G2 0.015 0.137 0.874 Male*G3 0.121 1.139 0.252 Male*G4 0.069 0.664 0.501 Male*Blue -0.050 -0.423 0.757 G2*Blue -0.036 -0.310 0.830 G3*Blue -0.018 -0.164 0.936 G4*Blue 0.005 0.043 0.935 Male*G2*Blue 0.039 0.255 0.921 Male*G3*Blue -0.121 -0.849 0.349 Male*G4*Blue -0.149 -1.053 0.240 Table 5.8: Coefficients in the Linear Mixed Effects Model Predicting the F2 of BOT from the Social Variables (Sex, Generation, Class), Their Interactions, and Following Phonological Environment. P-values were obtained using Markov Chain Monte Carlo simulations (Baayen et al.., 2008).

202 in Columbus. In addition, when coupled with the information from the overlap analysis of BOT and BOUGHT, which will be presented in section 5.4, and the correlation analysis of BOT and BAT, to be presented in section 5.5, the information presented here also provides further confirmation of the fact that BOT backing, and not BOT backing as a part of the merger process with BOUGHT per se, is not only the triggering stage of the

Third Dialect Shift in Columbus, but also a contributing factor in the ongoing incrementation of the Third Dialect Shift among the youngest generational groups.

5.3.9 F1 of BOUGHT

For F1 of BOUGHT, our lmer model reveals one significant independent social factor, one significant linguistic factor, and several significant interactions among social factors. As with BOT, our model again uses G1 White Collar women as the baseline for analysis, this group being the most conservative for realization of this vowel class. The model used for this analysis again also includes speaker as a random effect, but not token because it did not contribute significantly to model fit.

Turning first to the significant linguistic factor--following environment--three following segments all significantly condition BOUGHT realization, albeit raising, rather than lowering, which is the direction of innovation of interest shown by BOUGHT in our analyses of its variation. These segments are /g/, /s/, and pause. Given that lowering is the direct of innovation for BOUGHT on F1, we argue these segments are really best viewed

203 as significant inhibitors to innovation, given the large negative coefficient values for each segment in the lmer analysis.

Turning to the significant social factors, sex is a significant independent factor, while the significant interactions include sex*age and sex*age*social class. Generally speaking, the significance of sex as an independent social factor suggests that men as a group tend to lag behind women in showing more raised realizations of BOUGHT than women. However, the clarification of the significant interactions of sex*age and sex*age*social class reveal that it is actually only Blue Collar men who show this lag.

The coefficient values for G2 and G3 for the interaction of sex*age*social class shows that Blue Collar men stand out within these age group cohorts as having higher

BOUGHT realizations than other speakers. Meanwhile, the robustly positive coefficients for G2 and G3 for the interaction of sex*age reveals that White Collar men show a converse pattern--they tend to have some lowering for BOUGHT. Rather interestingly, although the patterns for G4 men in either group are not statistically significant, the coefficients for these groups, as again revealed by the same interaction sets, reveal that

White Collar men show somewhat higher realizations in G4, and Blue Collar men show much lower realizations of BOUGHT than speakers of previous generations. Regardless of these trends across groups, however, the overall changes shown by speakers in Columbus reveal that most speakers tend to show less lowering of BOUGHT than we would expect, given that Columbus has often been described as a "low back

204 merger strong" area in a number of previous studies (e.g., Labov, et al., 2006; Ash, 2006;

Thomas, [1989]/1993, 2001; Campbell-Kibler, et al., 2008). (For US dialect areas, including the Midland, the low back merger is usually described as involving BOUGH lowering as a principal part of the merger process, hence our expression of expectation of

Coefficient t value pMCMC (Intercept) 1.229 7.247 0.000 Sex: Male -0.442 -2.466 0.015 Generation: G2 -0.308 -1.751 0.092 Generation: G3 -0.217 -1.302 0.217 Generation: G4 -0.045 -0.273 0.821 Collar: Blue -0.184 -0.960 0.357 Following /g/ -0.308 -2.022 0.028 Following /k/ -0.141 -1.333 0.143 Following pause -0.346 -2.130 0.021 Following /s/ -0.287 -2.074 0.029 Following /ʃ/ -0.183 -0.784 0.412 Following /t/ 0.042 0.406 0.749 Following /z/ -0.221 -1.127 0.234 Male*G2 0.567 2.611 0.010 Male*G3 0.569 2.721 0.005 Male*G4 0.329 1.598 0.112 Male*Blue 0.328 1.396 0.167 G2*Blue 0.341 1.469 0.159 G3*Blue 0.414 1.895 0.064 G4*Blue 0.006 0.028 0.994 Male*G2*Blue -0.643 -2.164 0.031 Male*G3*Blue -0.537 -1.934 0.050 Male*G4*Blue -0.076 -0.275 0.780 Table 5.9: Coefficients in the Linear Mixed Effects Model Predicting the F1 of BOUGHT from the Social Variables (Sex, Generation, Class), Their Interactions, and Following Phonological Environment. P-values were obtained using Markov Chain Monte Carlo simulations (Baayen et al.., 2008).

205 lowering in our data.) Why this is the case, we are not totally certain, but we believe this may be one reason why the BOUGHT and BOT merger seems to be less pervasive across

G2, G3, and G4 in our data than previous studies of the Midland and of Columbus specifically, would have us expect. This is an issue we will explore in more detail in sections 5.3 and 5.4.

5.3.10 F2 of BOUGHT

The final lmer analysis we present in this chapter is an analysis of speaker realizations of the F2 of BOUGHT. For this analysis, we change from using G1 White

Collar women to using G1 Blue Collar men, as this is the most conservative group on F2 for BOUGHT. As discussed in Chapter 3, we believe the more conservative values for F2 found among older Blue Collar men is directly influenced by late 19th Century/early 20th

Century Upper Southern/South Midland speech norms, such as those found among speakers with strong family background influences from West Virginia, Kentucky, and

Southeastern, OH (Markwardt, 1940, 1942; Nobbelin, 1980). Blue Collar men in both G1 and G2 of our study have such family background influences in several cases, and even the Blue Collar women in our study seem to show some possible residual influence of a backer BOUGHT in their own productions, as well, as revealed by the more pronounced negative coefficient value for G2 women found in the appropriate sex*age row for this group.

206 As shown in table 5.10, no linguistic factors were statistically significant in the lmer analyses for F2 of BOUGHT, but two independent social factors were--sex and age.

Taking the coefficient information just discussed above into account with the significant coefficient values for each of these independent social factors, we see that the significance of sex here actually means women tend to have significantly fronter

BOUGHTs than men, regardless of age. Meanwhile age as a significant independent factor tells us that all groups, regardless of social class background, show fronting of

BOUGHT versus the G1 Blue Collar men. However, the decrease in the positive coefficient value of G4 as an age group also reveals that G4 as a group tends to show backer realizations of BOUGHT than members of either G2 or G3. This, again, is a surprising finding in our data, since again, a number of previous studies of Midland and specifically Columbus speech suggest we should find fronter realizations of BOUGHT among the youngest speakers, again as a result of movement patterns affecting BOUGHT as it undergoes the low back merger with BOT (Nobbelin, 1980; Thomas, 2001; Labov, et al., 2006; Ash, 2006). In this case, we expect that BOUGHT should front more as it lowers more, since the two patterns usually co-occur in the process of US-style low back merger.) We believe the backer realization of BOUGHT among G4 speakers generally is another reason explaining why there is less BOT/BOUGHT merger in our data than a number of previous Midland and Columbus studies would suggest there should be.

Again, this is a topic we will return to and discuss in more detail in sections 5.4 and 5.5.

207 Estimate t value pMCMC (Intercept) -1.247 -13.076 0.000 Sex: Woman 0.239 1.956 0.022 Generation: G2 0.276 2.529 0.003 Generation: G3 0.255 2.654 0.003 Generation: G4 0.182 1.889 0.029 Collar: White 0.148 1.352 0.126 Following /g/ -0.075 -0.682 0.318 Following /k/ -0.026 -0.350 0.652 Following pause -0.140 -1.145 0.168 Following /s/ 0.044 0.494 0.687 Following /ʃ/ 0.171 1.268 0.226 Following /t/ 0.081 1.120 0.193 Following /z/ -0.081 -0.685 0.431 Woman*G2 -0.233 -1.426 0.095 Woman*G3 -0.048 -0.324 0.716 Woman*G4 0.025 0.167 0.857 Woman*White -0.094 -0.527 0.575 G2*White -0.195 -1.296 0.141 G3*White -0.055 -0.394 0.677 G4*White -0.005 -0.037 0.978 Woman*G2*White 0.018 0.076 0.931 Woman*G3*White 0.014 0.064 0.969 Woman*G4*White -0.024 -0.112 0.844 Table 5.10: Coefficients in the Linear Mixed Effects Model Predicting the F2 of BOUGHT from the Social Variables (Sex, Generation, Class), Their Interactions, and Following Phonological Environment. P-values were obtained using Markov Chain Monte Carlo simulations (Baayen et al.., 2008).

5.4 The Merger of BOT and BOUGHT

As discussed earlier in section 5.2. in earlier accounts of the Third Dialect Shift,

BOT/BOUGHT merger was seen as an integral part of the vowel shift, serving as the triggering event in the accounts provided for the California Shift by Eckert (2004),

Hagiwara (2005), and Gordon (2006), as well as the accounts provided for the Canadian

208 Shift by Clarke, et al. (1995), Boberg (2005), and Labov, et al. (2006). Yet in the accounts of the Third Dialect Shift provided by Durian (2008, 2009) and Bigham (2008,

2009) in the Midland, BOT/BOUGHT merger is quite often not found, yet other stages of the Third Dialect Shift, namely BAT backing and lowering, are. This suggests that

BOT/BOUGHT merger may not play an integral role, or at least, not as integral of a role in triggering the Third Dialect Shift as previous accounts have argued.

Given the emphasis placed on BOT/BOUGHT merger in previous studies for triggering the Third Dialect Shift, we provide a detailed analysis of the merger here, to see what role it plays in triggering the Third Dialect Shift in Columbus. In addition, we have decided to include a detailed analysis, given the mixed nature of previous studies of

Columbus, which have provided varying results on how robust the merger actually is in

Columbus (regardless of its status as a trigger for the Third Dialect Shift). As discussed in

Chapter 3, these results have varied from reports of more or less full merger among informants born after World War II in the community (e.g., Thomas, [1989]/1993;

Campell-Kibler, et al., 2008) to mixed reports of many cases of only partial merger, in addition to some cases of full merger among informants born during the time period (e.g.,

Labov, et al., 2006; Ash, 2006).

As noted in the previous section, the results of our lmer analyses for F1 and F2 of

BOT and BOUGHT reveal a lack of evidence of full-scale merger of BOT and BOUGHT in our data on these individual formant parameters among speakers born during the

209 period 1896-1991. The principal reason, as our results suggest, is not due to a lack of movement of BOT, which in fact does exhibit some subtle movement backwards and upwards among speakers, with younger speakers showing stronger trends in these directions than older speakers. Rather, in our data, it is the lack of significant movement of BOUGHT in the direction of BOT, a key movement in the merger of BOT and

BOUGHT, at least as it has previously been studied in US English speaking communities in previous studies (e.g., Thomas, 2001; Labov, et al., 2006).

This collection of trends becomes more obvious when we consider the degree of overlap that tokens of BOT and tokens of BOUGHT show in each individual vowel system of the 62 speakers in our corpus. As shown in table 5.11 below, when we analyze the individual vowel systems of our speakers, the trends just predicted do seem to hold up quite robustly. For this analysis, we generated individual vowel plots for all speakers, looking at the degree of general overlap between classes for each speaker. In clear cases where the vowel classes were distinct, the speaker's system was labeled as "distinct" for the purposes of our analysis. In cases of clear overlap between the classes, then a system was labeled "merged" for our analysis. In cases where a clear distinction is not visible, but clear overlap is also not visible, then we labeled the system as "partially merged"

("partial" in the table), further noting which tokens showed the partial merger behavior, so that an understanding of the following environments conditioning merger or partial merger in each system could be obtained.

210 This approach to the quantification of merger in our data was used rather than a more complex quantitative statistical method of analysis, because of the type of data for

BOT and BOUGHT that we were able to obtain. Unlike many other studies of BOT and

BOUGHT merger, we have not used data elicited using reading lists or minimal pair tests. As a result, we often do not have exact minimal pairs to compare and contrast.

Instead, we more typically have near-minimal pairs, which, although useful for a broad general comparison such as the one we make here, are not exact enough of a comparison to get meaningful results from a quantitative statistical analysis of individual vowel system distinctions, such as the use of t-tests, which are often employed in more detailed studies of BOT and BOUGHT merger than we are able to conduct here (e.g., Nycz, 2005;

Johnson, 2007; Fogle, 2008).

In our data, to distinguish between the categories of distinct, close, and merged, we have judged those speakers who show only 0-20% overlap of their tokens as being distinct, while those who show between 20% and 80% overlap of their tokens are partially merged, and those who show 80% or more overlap of their tokens are merged.

Example plots demonstrating different degrees of merger are shown in Figures 5.1 and

5.2. As these plots demonstrate26, ratings of vowel systems as distinct, partially merged, or merged is actually a relatively straight-forward process, given the clarity with which most speaker systems present themselves.

26 This is an additional reason why t-tests have not been used, as they are not really necessary given how clearly most systems fall into one of the three categories based on our visual analysis. 211 As table 5.11 reveals, only 17 of our speakers (roughly 27%) show what we would argue are truly merged BOT and BOUGHT classes, while another 6 (roughly

10%) show what we would argue are essentially distinct BOT and BOUGHT vowel classes. The other 39 speakers (roughly 63%) show only partial merger of BOT and

BOUGHT in our data. Most often, this overlap occurs with words that end in /t/ or /k/. As has been found in many previous studies of BOT and BOUGHT merger in US English, this is typically the order in which segments increasingly influence the conditioning of the merger, so in this regard, our data conforms with previous studies. However, the general pattern revealed by our age data shows a different trend overall than we've come to expect in previous studies of the US Midland.

When broken down by age cohort group, we see that the G1 speakers show no evidence of full merger, although they do show a good deal of partial merger before /t/, just as earlier studies would have us expect. 70% of the speakers have partial merger (7 of 10), while 30% (3 of 10) have distinct systems, with no particularly strong lead either way for social class group. In G2, 66% of the speakers (8 of 12) have partial merger, 17%

(2 of 12) have distinct systems, and 17% have fully merged systems (2 of 12). Again, as with G1 speakers, G2 speakers typically have partial merger only before /t/, although a couple of the speakers also show some evidence of partial merger before /k/. Turning to

G3, we find 45% of our speakers (9 of 20) have full merger of BOT and BOUGHT, while

50% (10 of 20) have only partial merger, and 5% (1 of 20) have a distinct BOT and

212

Figure 5.1: Examples of Distinct and Fully Merged BOT and BOUGHT Classes Among Speakers (The two plots in the top row represent speakers with distinct vowel systems. The two plots in the bottom row present speaker systems where BOT and BOUGHT are more or less fully merged. Note that Tony's plot shows a system where some tokens of BOT and BOUGHT can be close, but they still do not actually constitute overlapping instances of token realization. It should be noted all tokens plotted in each of each of these individual system plots were obtained from conversational speech, not read or minimal pair elicited speech, as in many studies, which may in part account for some of the differences in our findings versus those of other studies of Columbus and/or US Midland speech conducted previously.)

213

Figure 5.2: Examples of Partially Merged BOT and BOUGHT Classes among Speakers (The two plots in the top row represent speakers with less than 50% partial merger in their vowel systems. The two plots in the bottom row represent speakers with more than 50% partial merger of BOT and BOUGHT. As in Plots 5.1-5.4, we again wish to note that all tokens plotted in each of each of these individual system plots were obtained from conversational speech, not read or minimal pair elicited speech, as in many studies. We believe this may in part account for some of the differences in our findings versus those of other studies of Columbus and/or US Midland speech conducted previously, where full merger of BOT and BOUGHT is often found more extensively than we find it in our data. )

214 Speaker Characteristics BOT/BOUGHT Overlap Generation Collar Sex Name Degree of Overlap Subtotals 1 (b. 1896- Blue Female Ann Distinct Distinct : 2 1913) Louise Partial Partial : 3 Male ColumbusMK Partial Merged : 0 Frank Partial Arthur Distinct White Female DAREOH098 Distinct Distinct: 1 Thea Partial Partial: 4 Male LANCSOH28C Partial Merged: 0 Tobias Partial William Partial 2 (b. 1924- Blue Female Marie Partial Distinct: 1 1938) Mary Merged Partial: 4 Helen Partial Merged: 1 Male Joe Partial Tony Distinct Jim S Partial White Female Joyce Partial Distinct: 1 Nancy C Partial Partial: 4 Barbara Merged Merged: 1 Male Tom C Partial Don Distinct Tom W Partial 3 (b. 1945- Blue Female Susan Partial Distinct: 0 1968) Michelle Partial Partial: 5 Linda G Partial Merged: 5 Paula Merged Molly Merged Male Terry Partial Michael Merged Ted Merged Bill Partial Jim Merged White Female Marsha Merged Distinct: 1 Sheri Partial Partial: 5 Margaret Merged Merged: 4 Nancy Partial Toni Partial Male Steve Partial Ed Merged Peter Partial Ben Distinct David Merged Table 5.11: Analysis of the Extent of BOT/BOUGHT Merger in Our Data

215 Speaker Characteristics BOT/BOUGHT Overlap Generation Collar Sex Name Degree of Overlap Subtotals 4 (b. 1976- Blue Female Angela Partial Distinct: 0 1991) Kelly Partial Partial: 6 Julia Merged Merged: 4 Rachael Partial Mollie Merged Male Justin R Partial Vincent Partial Travis Partial Jason Merged Louie Merged White Female Janna Partial Distinct: 0 Jayne Partial Partial: 8 Liz Partial Merged: 2 Mandy Merged Kaitlyn Merged Male Abe Partial Bryan Partial Justin L Partial Matt M Partial Meesha Partial Table 5.11 (Continued): Analysis of the Extent of BOT/BOUGHT Merger in Our Data

BOUGHT. Finally, in G4, 70% of our speakers (14 of 20) have partially merged BOT and BOUGHT, while 30% have fully merged sets (6 of 20). Among G3 and G4 speakers, partial merger usually occurs before /t/, with many cases of /k/ also occurring. No strong class differences seem to differentiate our speakers, although in both G3 and G4, slightly more Blue Collar speakers merge than White Collar speakers.

Overall, what our analysis reveals is that although a good number of speakers in our data show full merger of BOT and BOUGHT, the vast majority show only partial merger of BOT and BOUGHT. Of particular interest are two trends. The first is that our

G4 speakers do not show an increasing trend toward BOT and BOUGHT merger, which

216 several previous studies of Columbus have suggested should be what we should find

(Thomas, 2001; Labov, et al., 2006; Ash, 2006). Although we are hesitant to say the decline in our speakers represents a decline in the general Columbus population, because it could well be that its only our speakers who show this trend, and if we had a larger speaker group, perhaps we would find an increase. But at the same time, we also cannot claim that BOT and BOUGHT merger is on the increase, either, based on our data, regardless of previous study findings.

A second trend of interest is that our G4 White Collar speakers who realize other vowel variables according to the parameters of the Third Dialect Shift are among the speakers whom we do not see extensive BOT/BOUGHT merger. Among these speakers, all of the men, and more than half of the women, have only partially merged BOT and

BOUGHT classes. Based on previous studies of Third Dialect Shift, this is rather strange, given that BOT/BOUGHT merger has been argued to be the triggering event for BAT backing and lowering. This finding, along with our earlier finding that although this is the case, our speakers showing backing and raising of BOT compared to G3 speakers, leads us to conclude that the triggering event for Third Dialect Shift in Columbus is not in fact

BOT/BOUGHT merger. Instead, it is simply the backing, and possibly the raising, of

BOT that is the triggering event. This is a hypothesis we now investigate in section 5.5.

217 5.5 The Inception and Development of the Third Dialect Shift in Columbus

Moving now to a full consideration of the implications of the results of sections

5.3 and 5.4, we can now begin to understand fully the story of the inception and development of the Third Dialect Shift during the course of the 20th Century in

Columbus. As we saw in section 5.4, in Columbus, speakers can often show all or most elements of the Third Dialect Shift in their vowel systems without having full merger of

BOT and BOUGHT. Thus, although this has often been seen as the triggering event for the development of the Third Dialect Shift in previous studies in other geographic locales in Canada and the US West, we argue that our data show it is not actually a requirement for Third Dialect Shift, nor is it in fact the actual triggering event. Instead, we argue that the triggering event is the backing and raising of BOT.

We argue for this analysis for several reasons. First and foremost, this is the triggering event suggested by our data. All of our G2 speakers, and several of our G1 speakers, show the retraction of BOT t an earlier stage than they show any other stages of the Third Dialect Shift. In G2, only 2 of our speakers show full merger of BOT and

BOUGHT, while the rest show either distinct BOT and BOUGHT classes or only partial merger of the classes, typically only before /t/. In G3, we find speakers showing backer and more raised productions, on average, for BOT than G2, and we find statistically significant backer values for F2 of BAT/BAG in the vowel systems of G3 speakers as a group than G2 speakers (although note that some G2 speakers also show some backing of

218 BAT/BAG, versus G1). Many of these same G3 speakers also have only partial merger of

BOT and BOUGHT, most often only before /t/, or in some cases, /t/ and /k/, but not other consonants. This event ordering suggests that although BOT and BOUGHT merger can certainly be found along with BAT/BAG retraction, it is not a prerequisite.

Although this analysis is clearly geared towards explaining the occurrence of the

Third Dialect Shift in our data, a second reason we argue for this analysis is because

BOT-backing and raising is still compatible with the findings that low back merger is found to precede /ae/-retraction in those areas more often than we find in Columbus. In both areas, BOT backing and raising accompanies the merger of BOT with BOUGHT. In

Canada, BOT is actually the main "moving vowel" during the merger of the two classes.

This is why the outcome of BOT and BOUGHT merger is much more likely to be quite close to BOUGHT in Canada rather than BOT (Thomas, 2001; Labov, et al.. 2006).

At the same, in Western US cities, BOT also moves back and raises during the process of merger with BOUGHT, although here the movement is much more like

Columbus or other cities in the US Midland, where the movement can be rather minimal

(Labov, et al., 2006; Kennedy & Grama, 2012). This is because, typically in the US West version of low back merger, it is most often the case than the "moving vowel" is actually

BOUGHT, not BOT (Thomas, 2001). This is also why the outcome is most often rather close to BOT, as Kennedy and Grama (2012) note in their analysis of the Third Dialect

Shift in California. In fact, the movement of BOT is often minimal enough, it is often not

219 commented on by analysts in studies of vowel variation, since analysts, at least in most modern sociolinguistic or sociophonetic studies, tend to only comment on vowel movement trends if they are statistically significant (e.g., Clopper, et al.., 2006; Labov, et al.., 2006). Yet, as our analysis shows, BOT still does move in non-Canadian areas that show low back merger tendencies, even if the movement is not significant statistically

(For evidence of this trend in California/US West vowel systems, see figure 1 of

Kennedy & Grama, 2012). And as our analysis also suggests, this "non significant" movement can also sometimes be "just" significant "enough" to matter, even if it has no significant p value attached to it. (Again, the reader is directed to Kennedy & Grama

(2012) for their commentary on BOT's moving "just enough" in California to potentially trigger the Third Dialect Shift there, as well.)

A third reason we argue for BOT as the triggering event is because, again, our data provides us with information that says we should. Although BOT movement was not statistically significant on its own in our lmer analyses, an analysis of the correlation of movement of the BOT class with movement of the combined BAT/BAG class we also used for analysis of /ae/-movement in sections 5.3.1 and 5.3.2 reveals that there is a significant correlation between the two classes across all speakers in our data. The correlation is significant with an adjusted r2 of 0.050 (at p <.05), as shown in table 5.12 and Figure 5.3. Obviously, this is a very weak correlation, and if we had not given our data extensive descriptive analysis treatment like we utilized in Chapter 3, and just

220 approached the data like a statistician might, we might have just passed over this finding as "noise." We believe, however, it is not just noise, but instead, information that points in the direction of the actual triggering event for Third Dialect Shift, not only in

Columbus, but, as our analysis here suggests, for all areas in North America which exhibit the Third Dialect Shift.

We also believe the relationship is rather weak, because there is usually not a one- to-one formant value correspondence between the degree of backness of BAT/BAG and the backness of BOT in our data. Some speakers show more backing of BAT/BAG as a vowel class, as indicated by their mean vowel class average, relative to BOT as a vowel class than other speakers, who show more backing of BOT as a vowel class, again as indicated by their mean vowel class average, relative to BAT/BAG as a vowel class. This would not be the case if a one-to-one formant value correspondence relationship existed in our data. If that were the case, then we would always expect a backer BOT vowel class

Residuals: Min 1Q Median 3Q Max -0.200 -0.066 0.0125 0.065 0.196 Coefficients: Estimate Std. Error t value Significance (Intercept) -0.584 0.034 -17.28 <0.001 BATbySpeaker$Meanf2t25 0.116 0.057 2.045 0.045 Multiple R-squared: 0.065 Adjusted R-squared 0.050 F-statistic: 4.18 on 1 and 60 DF, p-value: 0.0453 Residual standard error: 0.09343 on 60 degrees of freedom

Table 5.12: Correlation of BOT and BAT/BAG Backing across All Speakers in Our 62 Speaker Data Set 221

Figure 5.3: Correlation of BOT Backing and BAT/BAG Backing among All 62 Speakers in Our Data Set

mean formant value to correspond to a backer BAT/BAG mean formant value.

This is why we argue that ultimately, what matters in regards to BOT serving as

222 the trigger for the Third Dialect Shift is not that it must show robust backing, or a strong correlation with BAT/BAG backing. It also does not actually matter whether BOT is merged with BOUGHT, as our data show. What matters is that BOT is "back enough" to allow room for BAT/BAG to retract. In Canada, there is much more room for BAT to back because BOT almost always moves further back during the merger process with

BOUGHT there than in the US West or US Midland. Meanwhile, in the US West and US

Midland, we also might expect to find somewhat different degrees of shift, depending on how much BOT moves while undergoing full or partial merger with BOUGHT. We believe these observations will ultimately prove to be clearer in data drawn from a variety of locales in the US West, US Midland, and Canada as more comparative studies of these areas are conducted, and most importantly, studies involving speakers of multiple age groups are used.

Thus far, one recent study we are aware of where this has been found to be the case where speakers of multiple age groups have been observed are Jacewicz, et al.

(2011a,b). As discussed earlier, Jacewicz, et al. (2011a,b) studied speakers living in

North Carolina, Wisconsin, and Central Ohio born circa 1919-1998. Significant BOT- backing with significant BAT-backing which both increase as speakers get younger is found in speaker systems in both Central Ohio and North Carolina in their study with only partial merger of BOT and BOUGHT.

Turning back to our analysis of G3 vowel shift patterns, and moving on to an

223 analysis of G4 patterns, many G3 speakers show some backing of BAT/BAG coupled with BOT backing. In addition, a good many speakers also show some BET backing tendencies. However, these appear to be shown by the speakers for somewhat different reasons, depending on the social group involved, as mentioned earlier in section 5.3.

Among White Collar men, BET-backing appears to be a parallel movement, linked to

BAT/BAG backing, and showing the emergence of the Third Dialect Shift as an attribute of White Collar men's speech among younger speakers. All of the Third Dialect Shift trends shown by White Collar G3 men are also shown by White Collar G4 men, and in fact, all of them are found more robustly in the speech of the G4 men. We argue here that

BET-backing is a parallel shift with BAT-backing, since the movement of BET is only backwards, as Boberg (2005) found for the Third Dialect Shift in Montreal, and not downward, or downward and backward, like the chain-shift-like behaviors reported for

Third Dialect Shift in Ontario, by Clarke, et al. (1995), and other areas of Canada more generally, by Labov, et al. (2006). The spatial elements of the shift can be seen by comparing the realziations of the 4 white collar men's vowel systems that we have selected as example systems showing the stages of the Third Dialect Shift in Figure 5.4.

For White Collar women, we argue that G3 BET-backing is actually more likely

"residue" remaining from a shift that actually began in G2 White Collar women's systems for BET. That is, BET-backing resulted perhaps as a "collision avoidance" strategy with

BAT, which showed fairly robust raising, coupled with some simultaneous backing,

224

Figure 5.4: Stages of the Third Dialect Shift in White Collar Men's Vowel Systems

(White Collar men are the leaders of linguistic change for the Third Dialect Shift Vowel Series. In the top two vowel plots, we can see the beginnings of the Third Dialect Shift, via the backing of BOT and some backing of BAT in the vowel system of Tom W versus William. Peter's vowel system shows lowering of BAT backing of BET, as well as additional backing of BOT. Abe's system shows BOT, BAT, and BET backing, all further advanced than Peter's. Abe's system provides a useful exemplar of the Third Dialect Shift in full bloom in Columbus vowel systems. Note, however, that his BIT is conservative compared to some other Third Dialect Shifters.)

225 which itself is involved in a chain shift with BOT, as it backs, in these women's systems, as section 5.3 discussed. In G2 women's systems, BET-backing thus appears to be more like a chain shift than a parallel shift, which we could depict graphically as in Figure 5.5.

As Figure 5.5 shows, the chain shift like relationship of BOT, BAT, and BET in these women's systems appears to involve BAT moving into the space of BET, "crowding" the acoustical space occupied by BET, as BAT moves both up and back at the same time.

This suggests the relationship between BAT and BET is a push chain, showing classic A-

>B-> chain-link movement. Actually, it is really B-> C-> in this case, since BOT backing perhaps, since less than a handful of White Collar women born before 1945 have been analyzed in prior studies. (See Chapter 3 for more discussion.) Our data suggest this shift is no longer active by G4, however, as women appear to be showing BET-backing that shows alignment with BAT backing as a parrallel shift, just like BET-backing in G4

White Collar men's speech.

Among Blue Collar G3 speakers, we also find BET-backing, although it is not

BET

BAT BOT

Figure 5.5: Possible Chain Shift of BOT, BAT, and BET in G2 White Collar Women's Vowel Systems

226 completely clear what this backing is indicative of. G4 Blue Collar women show a reversal of BET-backing compared to G3 women, while G4 Blue Collar men show continued BET-backing, like G4 White Collar Men. G4 Blue Collar women continue to show BAT/BAG-backing and BOT-backing, which makes them continue to participate in the low vowel component of the Third Dialect Shift. Meanwhile, G4 Blue collar men show signs of a possible new BAT/BAG-fronting trend, as they are no longer backing

BAT/BAG lke G3 blue collar men. This may mean that BET-backing has not yet reversed, but perhaps may begin to follow BAT/BAG, if BAT/BAG continues to showing a fronting trend in the next generation of G4 speakers. This will need to be determined in a later study.

Regardless of the differences we find in each group's patterns, the trend that unifies all the vowel realization patterns in our data is a strong correlation between the

Residuals: Min 1Q Median 3Q Max -0.356 -0.059 0.010 0.073 0.420

Coefficients: Estimate Std. Error t value Significance (Intercept) 0.226 0.047 4.760 p<.0001 BATbySpeaker$Meanf2t25 0.358 0.080 4.465 p<.0001

Residual standard error: 0.1313 on 60 degrees of freedom Multiple R-squared: 0.249 Adjusted R-squared: 0.237 F-statistic: 19.94 on 1 and 60 DF, p-value: 3.596e-05

Table 5.13: Correlation of BET and BAT/BAG Backing across All Speakers in Our 62 Speaker Data Set

227

Figure 5.6: Correlation of BET Backing and BAT/BAG Backing among All 62 Speakers in Our Data Set

movement of BET and the movement of BAT/BAG across all speaker vowel systems in our data. We argue that this finding makes for a stronger case for a clear relationship

228 between the vowel movement trends for BET and BAT/BAG than for BOT and

BAT/BAG in regards to showing a statistically significant link between the former two.

This link can be seen in the results presented in table 5.13 and Figure 5.5, which show the strength of the correlation of BET backing and BAT/BAG backing in our data. Here, we

2 find the correlation is statistically significant with an adjusted r of 0.24 (p<.001).

As our analyses suggest, although it is the case that G4 Blue Collar women differ from their G4 White Collar male and female counterparts in not showing BET-backing coupled with BAT backing as a parallel shift, we still argue that the continued BAT- backing and BOT-backing trends shown by these women indicates that they, too, are still participants in the Third Dialect Shift, but they participate differently than the other speakers, in only having one of the two possible components of the Shift.

As mentioned earlier, Boberg's (2005) analysis of the Third Dialect Shift in

BIT

BET

BOUGHT (BOT~BOUGHT)

BAT BOT

Figure 5.7: The Third Dialect Shift, depicted as the Canadian Shift in Clarke, et al. (1995)

229 Montreal suggests a different vowel shift organization for the Third Dialect Shift than the original formulation of Clarke, et al. (1995), the formulation of Figure 5.7. Clarke, et al.

(1995) is a chain shift, with the low vowel component of the shift connected to the front vowel component, with all elements involved in the chain shift, as in Figure 5.7. In

Boberg's (2005) configuration, based on his Montreal data, the Third Dialect Shift has two components: a) a low vowel chain shift component, and b) a front vowel parallel shift component. The two component version is depicted in Figure 5.8. Here, BAT backing and lowering and BOT/BOUGHT merger are grouped as a chain shift forming the chain shift component [1], while BET and BIT group with BAT as a backing parallel shift, component [2] of the Shift. In our data, we argue that the Third Dialect Shift operates more like Figure 5.8, although without the merger of BOT/BOUGHT being a prerequisite, as we discussed above. Instead, as we have argued, BOT backing is the

BIT

[2]

BET

[1] BOUGHT (BOT~BOUGHT)

BAT BOT

Figure 5.8: The Third Dialect Shift, depicted as the Canadian Shift in Boberg (2005)

230 prerequisite.This makes for the version of the Third Dialect Shift shown in Figure 5.9.

The two component version of Figure 5.9 is the version of the Third Dialect Shift found among our G3 and G4 White Collar men, who have the most advanced version of the shift, including signs of the beginning of BIT-backing (as revealed by our analysis in section 5.3) along with more robust BET-backing, BAT-backing, and BOT backing and raising, and our G4 White Collar women, who have BET-backing coupled with BAT- backing and BOT backing and raising, but not much in the way of signs of the beginnings of BIT-backing. In this characterization of the shift, speakers have both the parallel backing shift of BET and BAT (and variably BIT, as among White Collar men), along with the low vowel chain shift invovling BAT-backing and BOT backing and raising.

When speakers have only the low vowel, chain shift component, as with our G4 Blue

Collar women, however, then the shift looks like figure 5.8, but without the component 2

BIT

[2]

BET

[1] BOUGHT

BAT BOT

Figure 5.9: The Third Dialect Shift in Columbus, with Low Vowel Chain Shift and Front Vowel Parallel Shift Noted

231 vowel classes shifting.

One final note about the parallel shift of BET-backing and its link to BAT- backing is that we note that, again, the process of parallel shifting appearts to be one of phonetic analogy. Here, we again see segmental conditoning similarities between the vowel classes, somewhat similar to what we saw with SHOES/BOOT-fronting and

BOAT fronting in Chapter 4. Here, the similarities between the classes are that /v/ is a significant conditioner in the case of both vowel classes. Given this fact, we argue /v/ may serve as a kind of natuarally conditioned "leading environment" that lays the foundation for an analogy to be drawn between the two classes, since /v/ tends to condition backer variants of front vowels naturally to be begin with (Thomas, 2001).

Among other segments conditioning the backing of BAT are /s/, /θ/, /z/, and /t/, while among the other segments conditioning the backing of BET are /p/, /θ/, /z/, and /s/.

Thus, like with the back diphthongs, there is some parody of conditioning segments for

BET and BAT, as well, although the relationship does not seem to be as strong, either in terms of the influence on the conditioning of backing of the vowels shown by the segments (only /v/ and /s/ are statistically significant for BAT, while only /v/ is significant for BET), nor in terms of the degree of analogization of the relationship between the vowels (fewer segments show similar alignment of conditioning tendencies between BET and BAT than between BOAT and SHOES/BOOT). This link may be less strong between BAT and BET than the link we saw between BOAT and SHOES/BOOT,

232 but it also seems plausible this might be so given that the parallel shifting shown by BAT and BET appears to be a newer change in progress than the parallel shift shown by

BOAT and SHOES/BOOT. At least, this is what the mix of apparent time data provided by the comparison of G2, G3, and G4, and the real time data provided by the comparison of G1 to G2 suggests, anyhow. Given that there may be a weaker link shown by the classes, it is perhaps not surprising then that BET-backing does not occur as pervasively in the data among speakers as BOAT-fronting does.

233 Chapter 6: The Rise and Fall of Short-a in Columbus27

6.1 Introduction

In this chapter, we discuss the emergence and development of several different types of short-a systems in the Columbus vowel system during the second half of the 19th

Century, as well as over the course of the 20th Century, as we have observed them in our data. As was discussed briefly in Chapter 3, the Columbus short-a system underwent a change progression during the course of the second half of the 19th Century and during the 20th Century from one in which short-a was commonly realized by speakers as a phoneme showing conditioned tensing and raising in specific phonetically-conditioned environments (before front nasals, front voiceless fricatives, and variably before voiced stops, as well as the voiceless stop /t/)--a kind of split short-a system--to one that was realized as raised and tensed only before front nasals among all informants, but then also sometimes before voiced stops among others and before /ŋ/ among others. These latter two types of systems are traditionally referred to as the nasal and the continuous respectively in studies of short-a systems in US English (e.g., Labov, et al.., 2006; Labov,

2010; Dinkin, 2009).

Thus far, we have only considered the brief discussion of the changes to the system outlined in Chapter 3. But as we will now see, the change process across

27 Aspects of this analysis first appeared in Durian (2010). This chapter owes a debt of gratitude to Daniel Ezra Johnson, Aaron Dinkin, and Michael Friesner, whose conversations with us about the analysis we present here helped us to better sort out and clarify our thoughts. 234 generational groups in Columbus has been somewhat more complicated than this initial analysis suggests. In addition, given the contribution this information can provide to our deeper understanding of the historical occurrence of split short-a systems in US English generally, as well as other types of short-a- systems in US English, we feel a more detailed exploration of this change process is useful to present in its own distinct chapter.

To provide context for our analysis of short-a systems in Columbus, we begin in section 6.2, with a discussion of previous studies of historical split short-a systems in US

English. Here, we frame the discussion as "historical," because in almost all areas of the

US, including New York City (Becker & Wong, 2009; Becker, 2010), split short-a systems have begun to recede, diminishing as a dialect feature of the areas in which they were once found in the speech of speakers born after 1971, who show either continuous or nasal systems in place of the split system, just as our Columbus informants do (Boberg

& Strasel, 1995, 2000; Johnson, 1998; Ash, 2002; Labov, 2007; Dinkin, 2009; Becker &

Wong, 2009; Becker, 2010). As of the time of this writing, in fact, the only community in which split short-a systems appear to continue to be found consistently in the speech of subjects born after 1971 is Philadelphia (Labov, et al.., 2006), and even then, recent research suggests this foothold may be beginning to slip (Prichard & Tamminga, 2012).

Although split short-a systems are typically associated in much of the discussion in the literature with speakers living in New York City and Philadelphia (Trager, 1940;

Ferguson, 1972; Cohen, 1970; Labov, et al., 1972; Labov, 1981, 1989, 1994, 2007;

235 Labov, et al., 2006), a growing body of recent research (Boberg & Strassel, 1995, 2000;

Johnson, 1998; Ash, 2002; Thomas, 2006; Labov, 2007; Dinkin, 2009) has found that split short-a systems may have actually been more widely spread in US English during the 19th Century, as well as the early part of the 20th Century, than the field previously realized. The areas in which split short-a systems have been found to occur includes, as of the time of this writing: New Haven, CT (Johnson, 1998, expanding significantly on earlier results); the Hudson Valley area of New York (Dinkin, 2009); additional cities along the East Coast in the area between and surrounding Philadelphia and New York

City (Ash, 2002), such as Newark, DE and Trenton, Brick, and Bridgeton, NJ; New

Orleans, LA (Labov, 2007); Cincinnati (Boberg & Strassel, 1995, 2000); and various other cities in Central and Southern Ohio (Thomas, 2006).

Although the discovery of split short-a systems in a wider variety of geographic locations in these recent studies is certainly interesting on its own, an equally interesting aspect of research on split short-a systems we will discuss in section 6.2 is the story of the

"lost" series of early studies in the field that also revealed split short-a systems being used in a wider geographic area than much more recent, but also previous, more recent, research, has discussed (for instance, the "classic studies" on short-a in New York City and Philadelphia of Trager, 1940, Cohen, 1970; Ferguson, 1972, Labov, et al.., 1972,

Labov, 1981, 1989, 1994, among others). This series of older studies appears to be have become "lost to the sands of time," not having been consistently cited in work on split

236 short-a systems since they were first published 80 to 110 years ago. These studies include

Emerson (1891), Grandgent (1892), and Tuttle (1902), and to a lesser extent Babbitt

(1896), Kurath (1928a), and Trager (1930). It is not completely clear why some of these earlier details were overlooked in many more recent studies, such as the "go to" studies cited above, but they unfortunately have been. We suspect it may have been due to a combination of lack of awareness of the existence of the earlier studies and/or possible lack of access to the original studies (Tuttle (1902) being available only in a limited pressing Yale "working papers"-style publication).

However this situation developed, the end result was that several early studies of split short-a in US cities essentially became "lost" for a number of years until fairly recently, when Johnson (1998), in an undergraduate thesis on split short-a in New Haven,

CT, "found" them again. In our discussion, we will include these additional studies so that the largest net of information is available for access in our discussion throughout this chapter. As we will see in section 6.4, this expanded set of discussions will also prove to be vital in our discussion of the contribution of our data to developing a novel new perspective on the historical development, occurrence, and diffusion of split short-a systems in US English, especially 19th Century US English.

Before this, however, in section 6.3, we turn to discussing the results of our more detailed analysis of the occurrence of split short-a systems as they occurred historically in

Columbus during the second half of the 19th Century, as well as the early part of the 20th

237 Century, as demonstrated by speaker vowel systems in our data. In section 6.4, we turn to a discussion of what we believe our data on the historical occurrence of split short-a systems adds to the field's understanding of the development of split short-a systems in

US English generally during the 19th and 20th Centuries, with particular emphasis on what these data tell us about the development of these systems in light of the expanded set of "lost" early split short-a studies we present in section 6.2, and the results of recent studies of communities in New York City (Becker, 2010), New York State (Dinkin,

2009), New Haven (Johnson, 1998), and Cincinnati (Boberg & Strassel, 2000).

6.2 Previous Studies of Split Short-A Systems in US English

As discussed in Chapter 3, in previous studies of split short-a systems in US

English, split systems have typically been associated most strongly with the vowel systems of 4 locations in the US, all of which are located on the East Coast: New York

City, Philadelphia, Baltimore, MD, and Newark, NJ (Trager, 1940; Cohen, 1970;

Ferguson, 1972; Labov, et al.., 1972; Labov, 1981, 1989, 1994, 2007; Labov, et al.,

2006). As these studies established, New York City and Newark differ from Philadelphia and Baltimore in having one version of the split, while Philadelphia and Baltimore have a second type. Among these two types, the Philadelphia/Baltimore version of the system has been shown to show stronger constraints on raising than the New York City and New

Jersey version, with more restricted conditioning environments and more complex extra-

238 phonetic conditioning, as well (Labov, 1981; Ferguson, 1972).

Although both of these areas have been found to have different versions of the split, it seems likely they share a common source. This source is the historical lengthening of /ae/ before fricatives and nasals (Dobson, 1957; Ekwall, 1946).

Lengething of /ae/ before fricatives is a process which represents an innovation going back in English to at least the 17th Century (Dobson, 1957; Lass, 1976), and possibly as far back as the 15th Century (Ekwall, 1946; Wyld, 1920). At some point later, raising before nasals began, first in nasal cluters, where a nasal consontant precedes a non-nasal consonant (such as in dance or branch), then later before simple nasals (ham, can). It is not known for certain when raising before nasal clusters began, but it was likely sometime during the 18th Century (Ekwall, 1945: 169-170). Raising before simple nasals appears to be datable as beginning in the mid-19th Century, as suggested by data presented in Grandgent (1892) and Babbitt (1896). Concurrent with raising before simple nasals, raising before voiced stops also appear to have began, again, in the middle of the

19th Century (Babbit, 1896; Grandgent, 1892).

In New York City and Newark, NJ, the general phonetic parameters of the split are that a variety of consonants following /ae/ condition tensing and raising of /ae/, while other consonants following /ae/ do not. This is why the system is said to be "split"--the tokens of /ae/ showing conditioned raising and tensing all cluster into one raised group, while the tokens of /ae/ not showing conditioned raising all cluster in to a second, non-

239 raised/non-tensed group. In addition, /ae/ in this situation is said to be literally split as a phoneme--into two distinct phonemes--a high tense, ingliding phoneme /aeɛ/ (or possibly

/ae:/ or even /aeeI/ ), which is the underlying phoneme in the raised and tensed tokens, and a lower phoneme /ae/, which is the underlying phoneme in the non-raised and lax tokens. The consonants conditioning the tense phoneme in New York City and Newark, shown in Figure 6.1, are following voiceless fricatives, front nasals, and all voiced stops.

Beyond the phonetic conditioning of the following segment groups noted above, a number of additional extra-phonetic constraints, including lexical, grammatical, and morphological constraints, also condition tensing and raising in New York City and New

Jersey. Among these are the following (listing taken directly from Labov, 2007:354-355):

a) Function-word constraint: Function words with simple codas (an, I can, had) have lax short-a, while corresponding content words have the tense variant (tin can, hand, add); can't, with a complex coda, has the tense vowel, however, which preserves the contrast of tense can't vs. lax can in environments where the /t/ is elided or neutralized. b) Open-syllable constraint: Short-a is lax in open syllables, yielding tense ham, plan, cash, but lax hammer, planet, cashew. There is considerable variation before voiced fricatives and affricates (magic, imagine, jazz). c) Inflectional-boundary closing: Inflectional boundaries close syllables, so that tensing occurs in planning as well as plan, staffer as well as staff. d) Initial condition: Initial short-a with a coda that normally produces tensing is lax (aspirin, asterisk) except for the most common words (ask, after). e) Abbreviations: Short-a is often lax in abbreviated personal names (Cass, Babs) f) Lexical exceptions: There are a number of lexical exceptions: for example, avenue is normally tense as opposed to lax: average, savage, gavel. g) Learned words: Many learned or late-learned words have lax short-a in environments where tensing would normally occur: alas, carafe.

In contrast to the New York City and New Jersey system, the Philadelphia and Baltimore 240 version of the system has tensing and raising of short-a conditioned by a more limited set of following consonant groups, as well as several more extra-phonetic restrictions. As shown in Figure 6.2, the phonetic consonants which condition tensing and raising in

Philadelphia and Baltimore are: following front voiceless fricatives, front nasals, and the voiced stop /d/, but only in three words--bad, mad, and glad. In addition, short-a is lax before nasals in irregular verbs (e.g., swam, began).

Although split short-a systems are most commonly associated with these 4 areas in modern US English, studies of regional variation across the United States have established that split short-a systems were actually found in a number of areas within the

US at earlier points in time, in particular during the 19th Century, but also the early to

p t tʃ k b d dʒ g m n ŋ f θ, s ʃ v ð, z ʒ l Figure 6.1: "Classic" New York City, NY Split System (Speakers described as possessing this system born c. 1890-1971 (Labov, 1981, 1994; Labov, et al., 2006; Becker & Wong, 2009; Becker, 2010) Solid lines indicate tensing/raising environment; dotted lines indicate variable tensing/raising environment p t tʃ k b d dʒ g m n ŋ f θ, s ʃ v ð, z ʒ l

Figure 6.2: "Classic" Philadelphia, PA Split System (Speakers described as possessing this system born c. 1888 and later (Labov, 1989, 1994; Labov, et al., 2006)) Solid lines indicate tensing/raising environment; dotted lines indicate variable tensing/raising environment

241 middle part of the 20th Century in some areas. In early studies, some of which have not seen their results discussed in much detail since their initial publication, split short-a systems were observed not only in New York City (Babbitt, 1896), but also Ithaca, NY

(Emerson, 1891) and New Haven, CT (Tuttle, 1902), as well as among speakers living in parts of Pennsylvania, Maryland, Virginia ("the Valley of Virginia"), and Western

Tennessee in the dialect survey results of Grandgent (1892). Actually, he suggests the split system may have been even more widely used than just these areas:

"In Pennsylvania, Maryland, the Valley of Virginia, western Tennessee, and doubtless in many other parts of the North, South, and West, ae before the spirants f, s, and θ is drawled and formed very high, so that it becomes almost or quite a long e: 'half', 'grass,' 'path'=heef, grees, peth" (271).

In somewhat later studies, the geographic scope of split-a systems was said to include

Newark, NJ, as well as "parts of New York State, New York City, eastern Nebraska, and

Rhode Island" by Trager (1930:399)28, and even possibly a good part of "the Middle

Atlantic States (New York, New Jersey, and Pennsylvania), the Middle West (Ohio,

Indiana, Illinois, Wisconsin, Minnesota, Iowa, and Northern Missouri), and 'Further

West' to the Pacific Coast29" by Kurath (1928a:286).

Beyond these early studies, a growing batch of recent studies published since

1995, beginning with Boberg & Strassel's (1995) paper on the occurrence of split short-a

28 Contra Labov (1981) and (1994), who states Trager's comments in his 1930 and 1940 articles represent only his Newark, New Jersey accent, the locations cited here are Trager's own words describing where he observed the pronunciations he himself notes in his article in his own words. 29 Unfortunately, no further definition of "Further West" is provided by Kurath than this statement. 242 systems among their informants living in Cincinnati, OH, have begun to find split short-a systems in a variety of other geographic locales, albeit "historically." That is, the systems in these studies are not found among young informants, but rather older informants, typically speakers born before World War II. Boberg and Strassel (1995) found the oldest speakers they analyzed (born c. 1920) had split-a systems that looked very much like the

"classic New York City pattern" usually described for New York City speakers in earlier quantitative sociolinguistics studies (e.g., Labov, et al., 1972; Labov, 1981), although with a key difference being that some speakers also raise /ae/ before voiced fricatives and

/ŋ/, in addition to before voiceless fricatives, voiced stops (including /g/), and front nasals, as in New York City In addition, the systems in Cincinnati lacked the extra- phonetic conditioning of the New York City system. Among increasingly younger speakers in their data, however, they found the split system to replaced increasingly by continuous or nasal systems, with the youngest speakers (born c. 1965-1975) showing predominately nasal systems (that is, raising of /ae/ only before nasals).

Following the publication of Boberg & Strassel (1995, and later 2000), other scholars also began to report evidence of split short-a system among older speakers in

New Haven, CT (Johnson, 1998, expanding significantly on earlier results); the Hudson

Valley area of New York (Dinkin, 2009); additional cities along the East Coast in the area between and surrounding Philadelphia and New York City (Ash, 2002); New

Orleans, LA (Labov, 2007); and various other cities in Central and Southern Ohio beyond

243 Cincinnati (Thomas, 2006). In all of these studies, speakers born before circa 1971 were found to show some evidence of having split short-a systems, with the bulk of the speakers showing split systems actually born before roughly 1940. Among speakers born after 1971, split short-a systems are rarely found, and instead, either nasal or continuous short-a systems are usually found. Speakers born in these areas who were born in between these periods (1940 and 1971) show transitional realization behavior, with an increasing number of continuous and nasal systems. Although some speakers still show split short-a systems, this number decreases sharply as speaker age decreases (Johnson,

1998; Boberg & Strassel, 2000; Ash, 2002; Dinkin, 2009; Becker, 2010).

Unlike in Boberg & Strassel (1995, 2000), however, the kinds of split short-a systems found among informants born before 1971 in recent studies (Johnson, 1998; Ash,

2002; Thomas, 2006; Labov, 2007; Dinkin, 2009); show a fair amount of diversity with regard to the segments conditioning /ae/ raising. This is in part due to differences in geographic location of the study, and also in part due to the birthdate of the speakers involved in the study. In Thomas (2006), all speakers were born between 1885 and 1910, while in Johnson (1998), speakers were born in three distinct periods: 1865-1912, 1936-

1959, and 1978. In Ash (2002), Labov (2007), and Dinkin (2009), speakers were born c.

1920 and later. In all of these studies, the split short-a systems shown by informants have in common that they do not show evidence of being impacted by the extra-phonetic conditioning affiliated with the New York City system. However, where they differ is

244 how the segmental phonetic conditioning they do show impacts the raising of /ae/.

In Labov (2007) and Dinkin (2009), speakers in the Hudson Valley, New Orleans, and Cincinnati, show conditioned raising before front nasals, voiceless fricatives, and voiced stops—basically all the environments that condition raising in New York City except /g/. In Ash's (2002) study of cities in the greater Philadelphia and New York metropolitan areas, depending on the city in question, speakers in the cities studied can show a variety of tensing behaviors, ranging from the full list just given to only the full range usually provided for Philadelphia (front nasals, front voiceless fricatives, and before /d/) to just before nasals, with behaviors being motivated by the proximity of the city in question to either New York City or Philadelphia. Meanwhile, in Johnson (1998), the 1865-1912 born informants show raising in all of the classic New York City environments, including /g/, as well as before /ŋ/, while the informants born later than

1936 show only either continuous or nasal systems. In Thomas's (2006) study of cities located in southern and central Ohio, on the other hand, speakers show raising consistently before front nasals and voiceless fricatives, while variably before voiced stops (excluding /g/) and voiced fricatives, making them like Cincinnati speakers, except without raising before /ŋ / and /g/.

Turning back to the early studies we mentioned earlier in this section (Emerson,

1891; Grandgent, 1892; Babbitt, 1896; Tuttle, 1902; Kurath, 1928a, and Trager, 1930), we find an even more diverse variety of conditioning environments for /ae/ than the

245 modern-era studies we just discussed. In Grandgent (1892) and Kurath (1928a), the possible tensing and raising of /ae/ includes /ae/ before nasals (in nasal clusters) and before front fricatives only (and not before voiced stops, back fricatives, or additional segments such as /g/ or /ŋ/). This same set of environments is also described among older speakers (born before c 1840) in New York City in Babbitt (1896). Emerson (1891) describes tensing and raising of /ae/ before front fricatives, front nasals, and voiced stops

(except /g/), as well as before /ŋ/ among speakers in Ithaca, NY, while Trager (1930) describes the classic system of Figure 6.1 among the speakers he claims he has heard use the system living in parts of New York State, New York City, eastern Nebraska, and

Rhode Island. Tuttle (1902), describing what appears to be his own native-speech pattern

(New Haven, CT), describes /ae/ as raised and tensed before front voiceless fricatives, front nasals, and voiced stops (including /g/), as well as the back voiceless fricative /ʃ/ and the velar nasal /ŋ/. Finally, Babbitt describes the classic New York City pattern of

Figure 6.1, minus raising before /ʃ/ and /g/, among his younger New York City speakers, all of whom were born between circa 1860 and 1870.

At first blush, the variety of subtle differences between conditioning environments presented, and the differences in geographic locations covered by these additional studies just mentioned can be a bit overwhelming, given their diversity.

However, just like we saw earlier, when one also takes into account the birth dates (or potential birth dates) of the informants mentioned in these studies, and when one also

246 attempts to group them based on both similarities in phonetic conditioning noted in the study, as well as geographic location of the speakers surveyed, the complexity becomes easier to manage. Figures 6.3 and 6.4 provide an organization of these studies, as well as in fact all of the other studies discussed up to this point, since we believe our discussion will benefit from this presentation, which attempts to manage the complexity along these lines. Figure 6.3 presents data from speakers born 1819-1860, while Figure 6.4 presents

p t tʃ k b d dʒ g m n ŋ f θ, s ʃ v ð, z ʒ l A) New York City, NY Split System among Speakers Born c. 1830-1840 (Babbitt, 1896)

p t tʃ k b d dʒ g m n ŋ f θ, s ʃ v ð, z ʒ l B) Split System Reported for Pennsylvania, Maryland, "Valley of Virginia," and Western Tennessee among Speakers Born c. 1810-1870 (Grandgent, 1892)

p t tʃ k b d dʒ g m n ŋ f θ, s ʃ v ð, z ʒ l C) Ithaca, NY Split System among Speakers Born 1819-1840 (Emerson, 1891)

Figure 6.3: Split Short-a Systems among Speakers Born c.1810-1865 (Note that speaker age in Babbitt (1896) and Grandgent (1892) are based on hypothetical estimates. See text for description of the estimation process.) Solid lines indicate tensing/raising environment; dotted lines indicate variable tensing/raising environment

247 data from speakers born 1850-1913. In addition, Figure 6.5 presents data from speakers born 1888-1971, with special emphasis placed on speakers born 1920-1971.

One thing we should note about the arrangement of data into these three figures

p t tʃ k b d dʒ g m n ŋ f θ, s ʃ v ð, z ʒ l A) New York City, NY Split System among Speakers Born c. 1860-1870 (Babbitt, 1896)

p t tʃ k b d dʒ g m n ŋ f θ, s ʃ v ð, z ʒ l B) Split System Reported for New York City, New York State, Newark, NJ, Rhode Island, and Eastern Tennessee among Speakers Born c. 1850-1910 (Trager, 1930) (See also Trager, 1934, 1940)

p t tʃ k b d dʒ g m n ŋ f θ, s ʃ v ð, z ʒ l C) New Haven, CT Split System among Speakers Born 1865- 1912 (Tuttle, 1903; Johnson, 1998)

p t tʃ k b d dʒ g m n ŋ f θ, s ʃ v ð, z ʒ l D) Central/Southern Ohio Split System among Speakers Born c. 1885-1910 (Thomas, 2006)

Figure 6.4: Split Short-a Systems among Speakers Born c. 1850-1912 Reported in Studies Published before Ferguson (1945) and Labov's Philadelphia and New York City work, not Counting Kurath (1928a), and Supplemented by Johnson (1998) and Thomas (2006) (Note that speaker age in Babbitt (1896) and Trager (1930) are based on hypothetical estimates. See text for description of the estimation process.) Solid lines indicate tensing/raising environment; dotted lines indicate variable tensing/raising environment 248 by birthdate is that, in the case of several studies, we have had to estimate the birthdates we have assigned to speakers based either on descriptions of the speakers provided in the study or on "educated guesses" of what the age group would have been based on when the study was conducted. In studies conducted before the 1960s, particularly those conducted within descriptive dialectology or structural linguistics, it was not always the case that authors would provide systematic information, such as the birth dates of

p t tʃ k b d dʒ g m n ŋ f θ, s ʃ v ð, z ʒ l A). "Classic" New York City, NY Split System (Speakers described as possessing this system born c. 1890-1971 (Labov, 1981, 1994; Labov, et al., 2006; Becker & Wong, 2009; Becker, 2010)

p t tʃ k b d dʒ g m n ŋ f θ, s ʃ v ð, z ʒ l B) Split System Found among Speakers Born c. 1926-1970 in the Hudson Valley (Dinkin, 2009) and c. 1927 in New Orleans (Labov, 2007)

p t tʃ k b d dʒ g m n ŋ f θ, s ʃ v ð, z ʒ l C) Cincinnati Short-a System among Speakers Born c. 1920-1955 (Boberg & Strassel, 2000)

Figure 6.5: Split Short-a Systems Among Speakers Born 1890-1971 (Ash, 2002 speakers excluded) (Note Speakers in New York and Philadelphia Were Actually Born c. 1888-1920. Also Note-A) Replicates Figure 6.1.) Solid lines indicate tensing/raising environment; dotted lines indicate variable tensing/raising environment

249 informants, whether informants were male or female, or in some cases, even the number of informants the report of a dialect trend or set of dialect trends was based on. Another issue was also that analysts would report trends that were based purely on observation of a speech pattern, regardless of geographic location of the pattern, which would sometimes lead to somewhat strange groupings of locations, as seen in Trager's (1930) grouping of Eastern Tennessee with New York State, New York City, and Newark, or

Grandgent's (1892) grouping of Western Tennessee with Maryland, Pennsylvania, and

Virginia.

Thus, in Figures 6.3, 6.4, and 6.5, we maintain the original groupings presented by authors in their studies, even if they are odd on geographic grounds, and we have done our best to estimate informant age groupings for those studies that required us to do so.

These include Babbitt (1896), Grandgent (1892), and Trager (1930). Babbitt (1896) indicates only that he has for analysis data from an "an older group" and a "younger group," and also that all speakers were of adult age in the study, leading us to hypothesize his informants could have been feasibly born any time between c. 1810 and 1870, which are the age groups we have listed here in Figures 6.3 and 6.4. (We should note that

Johnson (1998), who also used Babbitt's information to estimate the age of Babbitt's informants, chose similar groupings based on the information provided, which makes us feel somewhat more confident about our choices.)

Grandgent (1892), in contrast, reviews the results of a series of 186 postal surveys

250 he received from academics located throughout the United States shortly before the publication date of his article. Given this rather diverse set of respondents, we hypothesize that his informants would have been adults who could have hypothetically been anywhere between the age of 22 and 82, if they were academics who might have been employed in teaching or research positions at the time. This means his informants were probably born somewhere around the birth years we have decided to assign to his subjects in Figure 6.3.

For the potential informants of Trager (1930), here we are making a less well- informed guess on who his informants might have been. In this initial article of his article series, he discusses split short-a as if its occurrence in US English would have been fairly well-known to readers at the time, having also been discussed in contemporary pronunciation dictionaries of the era, such as Palmer (1926), Krapp (1925), and Kenyon

(1924), but that the specifics of its phonetic and geographic parameters were simply not well-enough known, and thus, needed to be specified by him in his article. As such, he provides a rather matter of fact report of its phonetic parameters, and then states that his formulation of the parameters are based not only on his own speech, but also the speech of other speakers he has observed living in parts of New York State, New York City,

Rhode Island, and Eastern Nebraska. As with our estimates with Grandgent (1892), we again argue that the generalness of the description of the informant data suggests that

Trager could be potentially talking about any number of adult informants he might have

251 known in his professional working life, so again, this suggests fellow academics who, if employed at the university level, might have feasibly been anywhere between the ages of

25 and 80 at the time of Trager's observations of their speech. Or they could have been students of his, which would mean an even younger speaker of, say, 20 years old. Given this potential time window, we have suggested the possible birthdates of Trager's informants could have thus been between 1850 and 1910 if he observed their speech in

1930 when he reported the observations, and so these are the dates we use in Figure

6.4.30.

One additional thing we should mention about the data displayed in Figure 6.4 is that it does not include any of the information presented by Kurath (1928a). The main reason for this exclusion is that Kurath is simply too vague in his discussion of what age groups his discussion refers to in this article. We do know that Kurath means his discussion to reflect contemporary adult speech patterns in the US at the time it was written, so using a similar rationale to what we did with Trager's (1930) discussion, we could perhaps estimate the youngest speakers as being born c. 1905 or so. However, we have no idea where to set the "starting date" for his discussion, given that his comments at times discuss vowel realization patterns as early as the beginning of the 1800s, but then other times, they describe vowel realization patterns from the mid-to-late 1800s. A second reason is also the vagueness of his geographic category "Further West," as

30 Here we use slightly different age estimates than we used for Grandgent (1892) only because they allow us to stick to using nicely rounded off birth dates for comparison across the different studies. 252 discussed earlier. Given these concerns, we thus thought it better in this case not to include his remarks here in our figures.

We also left Ash's (2002) speakers out of Figure 6.5 to conserve space. Given the diversity of the results she reports for the cities involved in her study, it would take several more displays to show all of the results accurately. In addition, since a good deal of the cities in her study show systems that are more like Philadelphia than New York

City, we also left out her results in our discussion. This is so because all other systems we discuss throughout the remainder of this chapter are more like the New York City system overall than Philadelphia.

Using the natural divisions provided by the estimated and actual birthdates of the speakers in this set of early studies, all of which were published before the well-known set of "classic studies" mentioned at the outset of this chapter--Trager (1940), Ferguson

(1972), and Labov's work (namely, Labov, [1966]/2006, 1981, 1989; Labov, et al.,

1972)--we can see that a rather detailed set of data is available for a number of cities outside of New York City, Newark, Baltimore, and Philadelphia, with data being represented for speakers born quite a bit earlier than Labov's New York City and

Philadelphia informants in many cases, or else around the same time as the oldest speakers in his studies.

These additional data show us several things about split short-a systems in US

English. First, they show us that, among the earliest speakers for whom we have data, the

253 types of conditioned raising they showed may have been quite a bit more restricted than the conditioned raising shown among the oldest New Yorkers and Philadelphians in

Labov's work. Among all of the speaker groups who were either born or quite likely born between c. 1810 and 1870, highlighted in Figure 6.3, save the Ithaca speakers, we see that the conditioned raising of short-a appears to have occurred only before front nasals (and only in nasal clusters) and voiceless fricatives. This is a trend found not only in Babbitt's

(1896) New York City data, but also in Grandgent's (1892) mixed area results.

Second, these data also show us that, during the middle of the 19th Century, a time period we can say is represented by the speaker data included here because of the birthdates of the speakers, split short-a systems appear to have already been in use outside of New York City and Philadelphia. As suggested by the geographic range of the speakers represented, the area of coverage includes Maryland, Pennsylvania, New York

City, Ithaca, NY, parts of Virginia, and parts of Tennessee. If we include Kurath's data in this estimation, cities in the Midwest (Ohio, Indiana, Illinois, Wisconsin, Minnesota,

Iowa, and Northern Missouri) and West would also be included. This is an important fact to note, for as we will shortly see, it will impact the perspective we take on the occurrence of split short-a systems in our Columbus data.

Third, among speakers born after 1850, we see an increase in the types of conditioning segments and types of segment groups that condition raising in the geographic locations listed. This expanded list suggests an expansion in the types of

254 phonetic conditioning that occurred in the areas shown in Figure 6.4 contrasts with the types of phonetic conditioning found among speakers in the areas shown in Figure 6.3.

This is another important point to note, for as we will see in section 6.4, this information will also impact the perspective we take on the occurrence of split short-a systems in our

Columbus data.

A final point to note about the non-New York City and non-Philadelphia systems documented in Figure 6.5 is that these systems may also differ from the New York City system in a much more significant way than the segmental conditioning difference we have just discussed. Labov (2007) and Dinkin (2009) argue that a fundamental difference between these systems and the New York City system is that in the New York system,

/ae/ is biphonemic, while in the other systems they discuss--New Orleans, the Hudson

Valley, and Cincinnati--/ae/ is monophonemic. That is, they argue that in New York City,

/ae/ is split into two underlying phonemes--/ae/ and /ae:/--whereas in the other areas, it is underlyingly one phoneme--/ae/--and the split in this version is allophonic rather than phonemic. That is, here, we are dealing with two allophones of one underlying phoneme, rather than two underlying phonemes, as in the New York City system.

A principal reason they argue this is so is because, even when the systems in the areas outside of New York City show a more robust pattern of split, one never sees a totally crisp break of the tokens into two distinct groups with a noticeable gap in phonetic space between them like one finds among older speakers in New York City, at least as

255 they appear in instrumental analyses of New Yorkers born 1890 and later who have clearly split systems. Instead, in the split systems in the other areas, what one tends to find is a kind of "clumping" of tokens into two groups, instead, with the tokens showing conditioned raising tending to cluster into a clump that is generally quite a bit higher than the non-raised group, which form a much lower, second clump of tokens. This is another important point to note, as this information also impacts the perspective we take on the analysis we present in section 6.4.

Before we move on to discussing how and why this information will impact the perspective we take on our data, however, we first need to introduce our data in the context of the more detailed discussion and analysis we intend to provide via this chapter.

Thus, we now move to discussing our own data, drawn from our Columbus speaker vowel systems, in section 6.3. We will then return to this discussion in section 6.4.

6.3 Our Analysis of Split Short-a Systems in Columbus

Turning to Columbus, and our Columbus data, we note, as we also did earlier in

Chapter 3, that no previous study has found or has commented specifically on the occurrence of split short-a systems in speaker vowel systems in Columbus. However, it does turn out, knowing what we now know of usage patterns in the area, that one previous study that focuses specifically on the short-a system did have data that show evidence of the residue of a split system, but the analysts of that study thought their data

256 showed evidence of Northern Cites Shift-style raising, and not residue of split short-a raising, as we now argue it did. This study was Keiser, et al. (1997), who found evidence of residue in the vowel system of a Blue Collar radio announcer who was born in the area in 1950.

However, they did not know it was residue at the time of the study, because, although they had the data at hand, they did not have earlier data of Columbus speakers showing split short-a systems with which to compare the man's system and make the determination. Based on the data we now have on hand, we can say that what they actually observed was what we would call a continuous short-a system in our data. We can also see why they thought it might have been Northern Cities Shift-like without this information to contextualize their finding. Continuous raising, particularly among systems that show signs of residue of earlier systems, where the more prototypical split short-a pattern that Columbus systems is more readily observed, can sometimes look quite similar to the Northern Cities Shift.

This is so because continuous systems often show raised tokens that do not show the systematic environment-specific raising of a split system. However, prototypical

Northern Cities raising typically involves all tokens of /ae/ being raised to mid-front or high-mid-front position. In contrast, continous sytems involve not only some raised tokens occuring in mid-front or high-mid-front position, but also non-raised tokens which can be realized in more of a low central position.

257 As our discussion will now show, and other data observed in other parts of Ohio among speakers of similar age and/or older also show, the pattern of raising observed by

Hartman-Keiser et al. (1997) is definitely not evidence of the Northern Cities Shift in

Columbus. As we saw earlier in Chapter 3, in Columbus, although it has not been noted to occur in the area in earlier studies, our (re)analysis of 19th Century patterns of vowel variation in the greater Columbus area suggests that split short-a systems have been found in the vowel systems of Columbus speakers at least as far back as the second half of the 19th Century. In fact, it seems plausible an early form of split short-a may have even been found in the speech of the original Columbus settlers, given information we now have on hand, both for mid-19th Century born Central Ohioans, and speakers born and raised in other parts of the country, as discussed in section 6.2. Before we explain why we think this may be the case, however, we must first discuss in more detail the data

LANCS 28A (b. 1846) LANCS 28B (b. 1854) BAT (3/3) æ⋅(1:mad) æ (3: mad, Saturday, æ (2: Saturday, catty) catty) BATH (9/9) æ⋅ (2: mass, calf) æə (1:mass) æ (7: half, basket, past, æ⋅ (2: calf, class) pasture, ashes, class, æ (6: half, basket, past, casket ) pasture, ashes, casket) BAG (2/3) æ (2: sack, actress) æ (3: bag, sack, actress) BAN (5/6) æə̃ (1:dance) æ⋅̃ (2: aunt, ma'am) æ⋅̃ (1:ram) æ⋅(1:ram) æ ̃ (1:aunt) æ̃ (1: dance) æ (2:handy, hammer) æ (2:handy, hammer) Table 6.1: Raw IPA Transcriptions from Reynard's LANCS Field Notes for Short-a Words

258 on split short-a systems we have found among our speaker vowel systems in the present study.

Turning first to the two oldest subjects for whom we have data--the two Central

Ohioans who were born in 1846 and 1854 and interviewed by Cassil Reynard for LANCS in 1933, we can see from the transcriptions noted in his field records (reproduced here from the compilation we initially presented in Chapter 3 as table 6.1) that tensing and raising of /ae/ is noted to occur in the speech of both men when /ae/ is followed by the following consonants: /n/ (as in "dance" and "aunt"), /m/ (as in "ram"), /s/ (as in "mass" and "class"), and /f/ ("calf"). In addition, one of the two men--the 1846 born speaker with a Pennsylvania born and raised father--raised /ae/ before /d/ in "mad."

Although tokens of /ae/ realization were not observed in other classic split short-a raising environments such as before /b/, /dʒ/, and /θ/ for either men (since no tokens of words in these environments were elicited), we hypothesize that if tokens had actually been observed, raising may likely have been found before at least /θ/, if not actually all 3 environments. We argue that raising at least before /θ/ would be found here because tensing/lengthening before /θ/, along with /m/. /n/, /s/, and /f/, is noted to occur in the speech of speakers living in two other communities--Ithaca, NY and New York City,

NY--where speakers were born either somewhat earlier or around the same time as the

Columbus-area men, as shown earlier in Figure 6.3. As discussed by Babbitt (1892), speakers he observed in New York City who were born between c. 1830 and 1840

259 showed tensing and raising before front nasals (/n/ and /m/) (although in nsasl clusters, and not including simple nasals) and front voiceless fricatives (/f/, /s/, and /θ/), but not voiced stops (/d/, /b/, /dʒ/), the other "classic" split-a raising environment. Meanwhile, in

Ithaca, speakers born between c. 1819-1840 were observed to tense and lengthen /ae/ before front nasals (including simple nasals as well as nasal clusters) and front voiceless fricatives, as well as before voiced stops and the back nasal /ŋ/ by Emerson (1891). Given that there is overlap between these communities in showing raising of /ae/ before front nasals (nasal clusters and at least some simple nasals) and front fricatives among these speakers of similar age with what we are able to observe in the Columbus area, this is why we argue that the non-observed segment /θ/ at least would likely condition tensing/lengthening in Columbus in the speech of men born between 1846 and 1854.

Since raising before /b/ and /dʒ/ was also found in Ithaca, we argue further it might also be plausible that raising in these environments might have been found in Columbus had tokens been elicited by Reynard, given that raising before /d/ does occur in one of the two men's speech.

Turning to the speakers in our instrumentally recorded data, among our G1 informants (speakers born c. 1896-1913), we find many speakers in this birthyear cohort who show tendencies towards raising when /ae/ occurs in monosyllabic words before a word final consonant when the final consonant is not only a front nasal (that is, /m/ or /n/) or a voiceless fricative (/f/, /s/, /θ/, /ʃ/), like we argue the 19th Century born Central

260 Ohioans fairly certainly do, but also a voiced stop (/b/, /d/, /dʒ/). In addition, among these speakers, we argue raising appears to occur variably when /ae/ occurs before the voiced fricatives /v/, /z/, and /ð/ in our data, as well. Before voiceless stops, /ae/ usually does not raise, just like in other areas in the country where split short-a systems are found, although note that one exception to this generalization is that some speakers sometimes raise /ae/ before /t/. Labov (2007) argues that the occasional raising before voiced fricatives seems to occur as an over-generalization of raising before voiceless fricatives and provides evidence that some versions of the split short-a system show evidence of being phonetically simplified. We argue that the raising before /t/ in our data shows similar evidence of overgeneralization, albeit with stops rather than fricatives.

As Labov (2007) also argues, further evidence of the non-New York City systems being a more generalized system than the New York City system beyond segmental generalization is also demonstrated by the observable lack of complex lexical and morphological conditioning in many of the other split short-a systems. What this means is that lexical exceptions do not appear in the systems of these geographic locations, and the locations of syllabic or morphological boundaries do not seem to matter in the conditioning of raising either. The lack of these additional exceptions appears to hold true for the Columbus system, at least as we observe its occurrence in our data.

Although many G1 speakers show split short-a system usage we have just described, we should note that not all of our G1 speakers have a split short-a system.

261 Several of our speakers (3 of the 10 we analyzed) show a continuous system rather than a split system. As defined by Labov, et al. (2006), and as demonstrated by these G1 speakers in our data, a continuous system is a short-a system typified by:

"the highest and frontest tokens of /ae/ occurring before nasals, but there is no robust difference between prenasal and non-prenasal tokens of /ae/, as in the nasal system. In the continuous system, tokens occur in a more or less uninterrupted smear from mid-front or high-mid-front position on down to low central position. In some cases, tokens before /d/ or other voiced stops are typically higher than before other non-nasal environments, while in other cases, tokens before the voiced velars /g/ and /ŋ/ are higher, but both of these trends are subject to speaker-by-speaker variation" (180-181).

The contrast between the two systems is highlighted by the G1 speaker systems shown in Figures 6.6 and 6.7. In Figure 6.6, we see the split short-a of Frank, a Blue

Collar G1 male born in 1913. In Figure 6.7, we see the continuous short-a system of Ann, a Blue Collar G1 woman born in 1912. As we can see from comparing the vowel systems side-by-side, Frank's system shows a stronger contrast between /ae/ realizations spoken before voiceless fricatives, front nasals, and voiced stops than tokens of /ae/ occurring before other consonants, save for /t/, than Ann's. This can be seen by looking at the types of tokens that occur above the dark dividing lines drawn in each plot. In Frank's plot, with the exception of one token of "back," all tokens which occur before voiced stops or the velar stops /k/, /g/, and /ŋ/ occur below the black line drawn horizontally through the plot, and all of these same types of tokens other than "hanging" and "back" occur to the right of the black line drawn vertically through the plot. In contrast, Ann's plot shows a

262 stronger mix of these token types with the others in the plot, making for less differentiation overall between token types than Frank's system shows. In addition, some tokens of words with /ae/ occurring before voiceless fricatives and voiced stops, as well

Figure 6.6: A Vowel System Showing an Allophonic Split Short-a System

263 as before /t/, occur below and to the right of the black lines.

This differentiation between speakers is important because in these plots, the black lines mark off points for both F1 and F2 that are typically used by Labov and his

Figure 6.7: A Vowel System Showing a Continuous Short-a System

264 associates (e.g., Labov, et al.., 2006; Dinkin, 2009; Labov, 2010) in quantitative studies

=of split short-a systems to mark tokens that clearly show a nucleus which has been conditioned to raise versus those that do not show a raised nucleus. In the studies where this convention is used, F1 is typically marked off at between 650 to 700 Hz, while F2 is typically marked off at between roughly 1550 and 1650 Hz (with F1 and F2 here being quantified using "Labov normalization" as per the ANAE protocols and the data then being scaled to normalized Hz).31 Similar mark off points were also used by Thomas

(2006) in his study of split short-a realization in Central and Southern Ohio among

DARE speakers and by Becker (2010) in her follow up study to Labov ([1966]/2006) of short-a systems in New York City.

In addition, another important indicator of differentiation between the patterns shown in the plots is how the tokens cluster based on token color. In these plots, red tokens symbolize /ae/ realizations occurring before the "classic New York City" conditioning environments of following voiceless fricatives, following voiced stops

(except /g/), and front nasals. Blue tokens symbolize environments that do not typically condition /ae/ raising in the "classic New York City" split short-a system, but have been observed to condition raising in some split systems found outside of New York City in past studies of Ohio cities, including Cincinnati (Boberg & Strassel, 1995, 2000),

31 This determination was made by matching up raw Hz plots of 12 of our speakers, selected randomly from our data, with plots in these other studies, and then determining by visual confirmation where to draw the line in our plots. We then compared the raw Hz plots of these 12 speakers with their normalized plots, and determined the coordinates to redraw the lines in the normalized plots (.03 for F1, and .08 for F2). We then programmed these coordinates into our R scripts and generated plots for all 63 speakers using them. 265 Westfield Center, Johnstown, Belleville, Dover, and Clarington (Thomas, 2006). These include the voiced fricatives, and also the velar stops /k/, /g/, and /ŋ/. Finally, green tokens symbolize following voiceless stops, an environment that normally does not condition raising in any of the previous studies of split a systems in US English we have discussed. Our reasoning for including this color-coding, and also its significance in our analysis of short-a systems, will be discussed in more detail in section 6.4, when it becomes relevant to our analysis. Thus, for now, we simply note how tokens are coded, so as to help the reader make sense of the plots as they appear here for our current analysis purposes.

One other attribute to notice about the differences between speakers in these two plots is the way the tokens showing raising as opposed to those not showing raising generally group together. As we pointed out earlier, in Frank's plot, almost all of his red tokens are above and to the left of the black dividing lines, whereas in Ann's plot, more of the red tokens are mixed in with tokens of other colors and more occur below and to the right of the black dividing lines. If we look more closely, what we can also see is that the tokens in Frank's plot more robustly cluster into two "clumps" of tokens--a raised clump and a lowered clump--than Ann's tokens do. This pattern of "clumping" suggests that, for speakers with split systems in Columbus, their systems are more like those of the non-

New York City split system users we discussed in section 6.2, specifically Dinkin (2009) and Labov (2007), showing evidence of a split that looks more allophonic, than those of

266 New York City speakers showing the classic biphonemic split of Labov ([1966]/2006) and Labov, et al. (2006).

For now, we will formalize this difference in clumping patterns of New York City and non-New York City split system users as a difference between speakers using the

"classic NYC (New York City)" split and what we will call the allophonic split short-a system. The specifics of this difference will be addressed in more detail in section 6.4.

For the moment, however, we want to keep our discussion focused on only the basic patterns in our Columbus data, and so we will hold off on following this second train of thought until section 6.4 for the sake of coherence of discussion.

As the plots in Figures 6.6 and 6.7 demonstrate, older speakers in our data tend to have one of two types of short-a system--allophonic split short-a and continuous short-a.

In G1, the allophonic split system is usually conditioned by following front nasals following voiceless fricatives, and voiced stops (excluding /g/). Several speakers also show raising conditioned by some, but not all voiced fricatives (specifically /v/, /z/, and

/ð/), while others show raising conditioned by a following /t/ (but not other voiceless stops). In G2 and G3, the allophonic split system is usually conditioned by the same following segments already discussed for G1, in addition to the voiced fricative /ʒ/, and, among some speakers, the velar nasal /ŋ/. The difference in conditioning environments for G1 versus G2 and G3 is highlighted in Figure 6.8. The more conservative version

(including /θ/, but not /b/ or dʒ/) of the conditioning environments conditioning raising

267

p t tʃ k b d* dʒ g m n ŋ f θ, s ʃ v ð, z ʒ l

Alterative 1: Columbus Split Short-a System among Speakers Born c. 1846-1854 (* found only in the speech of the 1846 born man, whose father hails from Pennsylvania)

p t tʃ k b d* dʒ g m n ŋ f θ, s ʃ v ð, z ʒ l

Alterative 2: Columbus Split Short-a System among Speakers Born c. 1846-1854 (* found only in the speech of the 1846 born man, whose father hails from Pennsylvania)

) p t tʃ k b d dʒ g m n ŋ f θ, s ʃ v ð, z ʒ l

Columbus Allophonic Split Short-a System among Speakers Born c. 1896-1913 (Generation 1)

p t tʃ k b d dʒ g m n ŋ f θ, s ʃ v ð, z ʒ l

Columbus Allophonic Split Short-a System among Speakers Born c. 1924-1950

Figure 6.8: The Columbus Split Short-a System, in its Initial Form (Alternative 1 and 2), and in Its Expanded Form in G1, G2, and G3 Vowel Systems, among Speakers in Our Full 62 Speaker Data Set Solid lines indicate tensing/raising environment; dotted lines indicate variable tensing/raising environment.

268 among the 1846 and 1854 born LANCS men, discussed earlier, are also included, so that the contrast between the three is highlighted.

Note that, unlike in G1, where the allophonic split system dominates in speaker vowel systems, this is not the case in G2 and G3. In G2, only roughly half (7 of 12) of our speakers make use of allophonic split systems, while in G3, only 2 speakers (both born in the 1950s) have allophonic split systems. Other speakers in G2 have continuous systems, while other speakers in G3 have either continuous or nasal short-a systems.

As defined by Labov, et al.., 2006, nasal short-a systems feature "a sharp distinction of allophones of /ae/ [which occur] before nasals versus other phonetic environments. The distinction between pre-nasal and non-prenasal tokens of /ae/ can be quite robust" (174-175). As included in our actual data set, nasal short-a systems begin to appear among some G3 speakers, but they are not found extensively in our data until G4.

In Columbus, nasal short-a systems tend to look like one of the two example vowel systems contained in Figures 6.9 and 6.10.

In Figure 6.9 we see the vowel system of Molly, a Blue Collar woman born in

1952 who is a member of the G3 age cohort. In her vowel system, we can clearly see that only following front nasals consistently condition raising, and that the overall difference between the following nasals and tokens of /ae/ occurring before other consonants is quite robust. In Molly's system, some residue of the older pattern of split short-a realization found in many Blue Collar women's vowel systems is also found, via the

269 occurrence of several mildly raised and fronted tokens of "have," "dad," "mass," and

"sack," but by and large, almost all of her non-pre-front-nasal tokens of /ae/ words occur below the "F1 dividing line," and a number also occur to the right of the "F2 dividing line." This overall shift is indicative of /ae/ lowering and backing, a vowel shifting trend we focus on more extensively for /ae/ in Chapter 5 of this dissertation.

In Figure 6.10, we are presented with a second vowel system showing a nasal short-a system, albeit this time in the vowel system of a G4 White Collar female who was born in 1981. Here, we again see following nasal tokens forming a distinct cluster from non-following-nasal tokens, although this time, the nasal cluster contains not only following front nasals, but following back nasal tokens, as well (that is, tokens ending in

/ŋ/). Among some G3 and G4 speakers, only following front nasal tokens condition raising, while among other G3 and G4 speakers, following front and back nasal tokens condition raising, hence our showcasing both types of pattern here, with two differentspeaker vowel plots. We believe the speakers showing raising of /ae/ before both front and back nasals may be the most advanced speakers in our data for showing a changeover to a nasal system as the dominant system of young speakers, given that the front and back nasals together form a natural class that is then set in opposition to other consonants as the one class showing conditioned raising.

Having now had a chance to showcase each of the major types of short-a system realization patterns we find among Columbus speaker vowel systems in our G1, G2, G3,

270 and G4 speakers born between 1896-1991, we now present the results of an analysis of the 3 possible short-a patterns found in Columbus vowel systems as they occur in the speech of all 62 of the speakers who we have available for instrumental analysis. The

Figure 6.9: A Vowel System Showing a Nasal Short-a System

271 results of this analysis are presented in table 6.2. In this table, short-a systems are categorized as either being allophonic split, continuous, or nasal using the visual analysis criteria and definitions we presented for each type above while introducing the speaker

Figure 6.10: A Vowel System Showing a Nasal Short-a System

272 Speaker Characteristics Short-a Systems Generation Collar Sex Name Type Subtotals 1 (b. 1896- Blue Female Ann Continuous Split: 4 1913) Louise Allophonic split Continuous: 1 Male ColumbusMK Allophonic split Nasal: 0 Frank Allophonic split Arthur Allophonic split White Female DAREOH098 Allophonic split Split: 3 Thea Continuous Continuous: 2 Male LANCSOH28C Allophonic split Nasal: 0 Tobias Allophonic split William Continuous 2 (b. 1924- Blue Female Marie Continuous Split: 2 1938) Mary Continuous Continuous: 4 Helen Allophonic split Nasal: 0 Male Joe Allophonic split Tony Continuous Jim S Continuous White Female Joyce Allophonic split Split: 5 Nancy C Allophonic split Continuous: 1 Barbara Continuous Nasal: 0 Male Tom C Allophonic split Don Allophonic split Tom W Allophonic split 3 (b. 1945- Blue Female Susan Allophonic split Split: 2 1968) Michelle Nasal Continuous: 6 Linda G Continuous Nasal: 2 Paula Continuous Molly Nasal Male Terry Continuous Michael Continuous Ted Continuous Bill Allophonic split Jim Continuous White Female Marsha Nasal Split: 0 Sheri Nasal Continuous: 6 Margaret Nasal Nasal: 4 Nancy Nasal Toni Continuous Male Steve Continuous Ed Continuous Peter Continuous Ben Continuous David Continuous Table 6.2: Columbus Short-a Systems by Type Across All 62 Speakers

273 Speaker Characteristics Short-a Systems Generation Collar Sex Name Type Subtotals 4 (b. 1976- Blue Female Angela Continuous Split: 0 1991) Kelly Nasal Continuous: 6 Julia Nasal Nasal: 4 Rachael Continuous Mollie Nasal Male Justin R Continuous Vincent Nasal Travis Continuous Jason Continuous Louie Continuous White Female Janna Nasal Split: 0 Jayne Nasal Continuous: 1 Liz Nasal Nasal: 9 Mandy Nasal Kaitlyn Nasal Male Abe Nasal Bryan Nasal Justin L Nasal Matt M Continuous Meesha Nasal Table 6.2 (Continued): Columbus Short-a Systems by Type Across All 62 Speakers

systems contained in Figures 6.6, 6.7, 6.9, and 6.10.

As table 6.2 reveals, the pattern of change in Columbus across the 4 generations of speakers in our study is that the split system appears to be eventually changing to a nasal system after going through two stages of change. This can clearly be seen by looking at the patterns of short-a realization shown by all speakers in study, and comparing speakers in G1 with speakers in G4 for the beginning and end states of the change process, and speakers in G2 with speakers in G3 for the two stages of change in between these end states.

As we can see, the initial state within our G1, G2, G3, and G4 data is the

274 allophonic split system, shown clearly by the speech of all G1 speakers with the exception of two White Collar speakers and one Blue Collar speaker. The two stages of change are shown by G2 and G3.The two stages are: a) stage 1, during which primarily continuous systems co-occur with allophonic split systems, as shown by the patterns typifying the speech of G2 (regardless of speaker social class or sex); and b) stage 2, during which the continuous systems which began to show use in stage 1 switch to co- occurring with nasal systems, as shown by the patterns typifying G3 (again, regardless of social class or sex). Turning to G4, we can see from looking at the bulk of the G4 speakers' systems that the nasal system is the predominant system among all speakers save Blue Collar males, who show strong continuous system use (4 out of 5 speakers).

Given other patterns of change we have observed in the study, we argue that the G4 Blue

Collar men show a conservative pattern of realization here that evidences "lag" compared to the other G4 speakers groups. This is most clearly seen when G4 Blue Collar men are compared with G3 Blue Collar men, who also show a similar pattern of strong continuous system realization (4 out of 5 speakers), and also G3 White Collar men, who also show a similar level of continuous system use (5 out of 5 speakers) as the G3 Blue Collar men.

If we compare the results of this analysis with patterns of change in other communities in which split short-a systems (allophonic or classic NYC) were once found, which we discussed earlier, we can see that these patterns are actually fairly similar to patterns of change in those communities. As mentioned earlier, recent studies of

275 Cincinnati (Boberg & Strassel, 2000), New York City (Becker &Wang, 2009; Becker,

2010), New Haven, CT (Johnson, 1998), and cities in New York's Hudson Valley

(Dinkin, 2009) show a common pattern of transition in regards to the types of short-a systems found among speakers, pretty much regardless of the area under discussion or type of split short-a system shown (allophonic or classic NYC). In these studies, speakers born before roughly 1940 tend to show fairly robust split system usage, while speakers born later than 1940, typically until roughly 1971 (the birthdate of the youngest speaker in Becker, 2010, to show a split system), tend to show transitional use of continuous and nasal systems, with some speakers continuing to also show some split system use, but most speakers being continuous or nasal. Among speakers born after 1971, however, speakers primarily switch to having nasal systems, although some continuous systems also continue to found to be among some speakers.

As we can see here, our data conform to this general pattern to a good extent.

Among our G1 and G2 speakers, all of whom were born before 1940, we find the split system to predominate, with 7 of 10 G1 speakers, and 7 of 12 G2 speakers, using the split system. Among speakers born between 1945 and 1968--our G3 speakers--we find a mix of split, continuous, and nasal systems existing side by side, although the continuous system is by far the most dominant system. 2 of 20 speakers have split systems, while 6 speakers have nasal systems. The other 12 of 20 have continuous systems. Finally, among speakers born after 1971--G4 in our data--we find nasal systems used by most speakers

276 (13 of 20), with the non-nasal users being continuous (7 of 20). Becker (2010) finds nasal systems in New York City to be of both the front nasal and front/back nasal varieties, just as we do in Columbus. She also postulates that the front/back version may be an ideal

"end goal" for the system to reach in New York City for the same reason we do. That is, the combination of front and back nasals forms a natural class that becomes opposed to non-nasals in the nasal short-a system.

6.4.Revisting the History and Development of Split Short-A Systems in US English

Via the analysis in section 6.3, we have learned several things. We have now been able to see the change in split-a systems from the more restricted conditioned split short-a system of the 1846 and 1854 born speakers, to the more expanded conditioned raising evidenced by our G1 speakers, in particular, but also some G2 and G3 speakers as well, to the lowered and retracted nasal systems most common among our G4 speakers. We have also been able to see that, when compared to other communities in which comparative data is available for speakers born during similar time frame(s) to our speakers, the general development, occurrence, and later, dissolution of the split short-a system in Columbus and these other areas is quite similar.

Perhaps most interesting is the similarity between Columbus and New York City

(Becker, 2010) in this regard, given the interest by the field historically in the New York

City split short-a system specifically. This similarity suggests that in some ways, the New

277 York City system is ultimately not so different from the Columbus system. During the main period in which split systems were being used in both areas, as well as the other areas in the US we discussed, there clearly were differences, as we discussed earlier. The

New York City system was typified by a variety of additional extra-phonetic constraints that we have not found in Columbus (at least not based on the data we have at hand), nor have researchers in other areas with quite similar systems to the New York systems, such as Cincinnati, New Orleans, North Plainfield, NJ, or the Hudson Valley in New York

State (Boberg & Strassel, 2000; Labov, 2007; Dinkin, 2009). Furthermore, the New York

City version seems to have been biphonemic until fairly recently, whereas the split found in these other areas appears to be monophonemic, or at least, this is what the analyses presented for the New York City system by Labov, et al. (2006), Labov (2007), and

Becker (2010) suggest. However, as the system "dissolved" into a continuous system and then later a nasal system in New York City, it does appear as though a quite similar general dissolution process happened in other, non-New York City, areas as well., as we discussed in section 6.3.

As Becker (2010) describes, a part of this dissolution process in New York City may have actually been that the short-a system transformed into a monophonemic, allophonic system. In her data, she finds reduced evidence among her speakers who do show split systems, of those systems showing the added extra-phonemic constraints that the split systems found by Labov (1966/[2006]) show among his speakers, many of

278 whom were older than Becker's. Given this difference, Becker argues that the New York

City system among her informants may have switched to being an allophonic, monophonemic system. Such a shift would suggest a cause for why it might have ultimately dissolved there just as it has in other areas of the country. As we noted earlier, both Labov (2007) and Dinkin (2009) have argued that the non-New York City systems they observed seemed to be allophonic rather than biphonemic, too. As we discussed earlier in section 6.2, all of those systems have also since dissolved into nasal systems like Becker found in New York City. As we have now found, the dissolution of the system is quite similar in Columbus to these other areas, including New York City. This suggests the Columbus system is also likely allophonic, given the similarity of the patterns in our data to those shown by Dinkin (2009) and Becker (2010).

This also leads to a rather natural question as this point--given this similarity between Columbus and New York City during the dissolution stage, might it be possible that the systems were also related to one another at an earlier stage? Labov (2007) and

Dinkin (2009) have argued previously that the New York City system and split short-a systems found in areas outside of the Mid-Atlantic are related. Specifically, they have argued they are related because the split short-a system in New York City diffused to these other areas. That is, it first developed in New York City and then later diffused to the other areas. Furthermore, they have argued, this explains why systems as they occur in other areas of the country are more generic than the New York City system. In other

279 words, these systems show conditioned raising of /ae/ that is purely phonetically conditioned only, rather than the more esoteric and complex conditioning of the New

York City system, with its series of extra-phonetic raising constraints. According to

Labov (2007), this generalization (through rule simplification) is a hallmark of diffusion of a phonetic feature, since simplification makes learning the system easier for speakers during the dialect feature acquisition process. Diffusion is also why Dinkin (2009) and

Labov (2007) argue the systems in the other areas have ended up being monophonemic,

(while the New York City system has remained biphonemic until fairly recently, or so

Becker (2010)) argues). Given this argument, Labov (2007) and Dinkin (2009) have taken to referring to the more generalized, monophonemic system we began calling the allophonic split short-a system in section 6.3 the diffused split short-a system in their analyses (see also Labov, et al., 2006 for this usage).

However, in Columbus, the historical data on settlement patterns we have at hand suggests that, although New York City certainly did make some contribution of settlers to the initial settlement mix of Columbus during the 19th Century, groups from other parts of the East Coast were more heavily represented in that mix than settlers from New York

City. Specifically, the information suggests that the most dominant group was settlers from the Mid-Atlantic region more generally (Pennsylvania, New Jersey, Maryland, and to a much lesser extent, New York State, including New York City), followed by

Northeasterners (from Connecticut and Massachusetts specifically), and then Upper

280 Southerners (West Virginia, Virginia, and Kentucky) (Chaddock, 1908; Wilhelm, 1976,

1982). Given this mix, and particularly the ratio of New Yorkers to other Mid-Atlantic settlers, it seems more likely that in Columbus, settlers from other Mid-Atlantic states and cities besides New York City would have been more influential on the formation of the earliest stages of the Columbus vowel system than settlers specifically from New

York City. Given this fact, the argument made by Labov (2007) and Dinkin (2009) regarding the influence of New York City as a significant contributor to the spread of the diffused short-a system in the other areas of the country where we see historical split short-a system usage seems like it may not work as an explanation for Columbus. In addition, we also find the diffusion argument to be troublesome given Babbitt's data, which, on our read, suggests that the New York City system appears to have been under development at the same time as other systems were developing in other parts of the country, rather than developing first and then diffusing. As we will now also see, we believe this co-development typified not just the development of the Columbus short-a system, but other non-New York City short-a systems, as well.

Thus, we find ourselves needing to look in other areas for an explanation for the similarity between the New York City system and the Columbus system, and also the

Columbus system and the allophonic split short-a system as it has been found in other areas in the US, such as the Hudson Valley, New Orleans, Cincinnati, and other cities throughout Southern and Central Ohio. Here, the data we painstakingly assembled in

281 section 6.2, and in particular, the reanalysis of aspects of the "forgotten" early studies of split short-a which we used to construct Figures 6.3 and 6.4 have something to contribute, we believe. First, we will begin with the speaker systems shown in Figure 6.3, as some of the speakers in this figure represent people who would have been born before the earliest speakers for whom we have data in Central Ohio. It was likely the Central Ohio men's parents would have been born around the same time as the speakers listed in Figure 6.3, given the average age people tended to have children during the 19th Century. Thus, these other speakers systems can give us an idea of what was possibly going on in the country at the time that our Central Ohio men's parents would have been acquiring their systems.

As we can see from these other systems, it would appear as though at least two alternatives may have been in use by the early 1800s in US English. Many speakers born in the early 1800s often use the first alternative, which is a system that shows a reduced number of conditioning following segments and segment groups than we usually see discussed for split short-a systems in US English. These environments include raising only before front nasals in nasal clusters and before voiceless front fricatives. These systems are found in the speech of the speakers listed in Figure 6.3 as living in New York

City, Pennsylvania, Maryland, the "Valley of Virginia," and Western Tennessee at the time (Babbitt, 1896; Grandgent, 1892). This also appears to the pattern of use described by Kurath (1928a) for speakers living in the Mid-Atlantic States (New York, New Jersey,

282 and Pennsylvania), the Midwest (Ohio, Indiana, Illinois, Wisconsin, Minnesota, Iowa, and Northern Missouri), and "Further West" to the Pacific Coast. Meanwhile, the Ithaca,

NY speakers of Figure 6.3 differ from these other groups, and show the second alternative. This system includes raising of /ae/ not only before front nasals (including nasal clusters and simple nasals) and front fricatives, but also voiced stops and /ŋ/.

Turning to the data we have for Central Ohio, which we compiled for visual display in Figure 6.8, we see that, depending on the interpretation of Reynard's field notes, the two Central Ohio speakers we have data from who were born in 1846 and 1854 show one of two possibilities as split system users. Either both use systems quite similar to the non-Ithaca speakers discussed above, or else one uses a system more like the Ithaca system, while the other uses a system more like the non-Ithaca system. That is, both men raise /ae/ only before front nasals (nasal clusters and at least some simple nasals) and front fricatives, or so the observed data suggests (as indicated by the system labeled

"Alternative 1" in Figure 6.8). But then only one of the men may also raise before voiced stops. It is important to remember here that Reynard lists no tokens with a following /dʒ/,

/b/, or /θ/, so we do not know whether one (or perhaps even both) men might also have raised before those consonants as well. And this is where they may differ. Only one of the two of them raises before /d/ in the field notes, while the other does not.

Comparing the similarities of our men to the systems of the speakers in Figure

6.3, it appears as though, if we take the more conservative view (Alternative 1), that our

283 men only raise before front nasals and front fricatives, then they look most like groups A and B in the Figure 6.3. That is, New York City, Pennsylvania, Maryland, the "Valley of

Virginia," and Western Tennessee. If we go with the one man's /d/ and also what might have been had there been raising before /dʒ/ and /b/ (indicated as "Alternative 2" in

Figure 6.8), then the men's systems would look more like group C in Figure 6.3--the

Ithaca, NY speakers. In this case, speakers raise not only before front nasals and front fricatives, but also voiced stops.

Given this similarity between systems, and also the ambiguity of whether the

Central Ohio men should be grouped more closely with group A/B or group C in Figure

6.3, we argue that it seems possible that any or even all of the systems in Figure 6.3 could have served as influential source dialects from which the Central Ohio men's systems emerged. Although the data for our 1896-1913 born G1 group might further suggest that

Alternative 2 might be more likely, it is also equally important to note the systems of G1 speakers may also simply show generalization of conditioning from a smaller set of environments. This would be the case if the earlier system was that of just the Alternative

1 speakers in Figure 6.3, with the systems of Alternative 2 or our G1 speakers being a more complex version showing the effects of generalization to differing degrees.

We argue this could be possible since a generalization process leading to expanded systems like Alternative 2 could theoretically have happened across the roughly 2 generations of speakers who were born between our early Central Ohio men

284 and our G1 instrumental analysis speakers. We argue this is so, because it also appears such a generalization process may be seen with the New York speakers in Figures 6.3 and

6.4. There, we can see what looks like an expansion of environments, from a basic system with fronting only before front nasals and front fricatives, to a system with fronting before front nasals, front fricatives, and voiced stops, as speakers get younger in age. This becomes clearer when we compare the systems of speakers in Figure 6.3 with those of Figure 6.4 for New York City.

This expansion becomes more pronounced when we go on to compare New York

City speakers born 1 or 2 generations after the speakers displayed in Figure 6.4. This would be Labov's speakers (displayed in Figure 6.5), who were born beginning in 1890 and forward. This makes them born about one generation later, and younger, than

Babbitt's (1892) younger speakers, displayed in Figure 6.4. Compared to the Figure 6.4 speakers, Labov's speakers show further expansion of the system. That is, they now raised /ae/ not just before front nasals, front voiceless fricatives, and voiced stops, but also before the back fricative /ʃ/ and the velar voiced stop /g/.

Not only do the New York City speakers show this kind of expansive realization behavior across generations, but so do New York State speakers more generally, as well as Southern Ohio and Central Ohio speakers, if we more generally compare areas within

New York State and Southern and Central Ohio as they are divided across generation groups in Figures 6.3, 6.4, 6.5, and 6.8 (for Central Ohio, as represented by Columbus).

285 For New York State, the first stage shown is that of the Ithaca, NY system in Figure 6.3.

Here, we see a more expanded system already in place, as we discussed earlier. In this case, the initial stage includes raising of /ae/ before nasals (front as well as back), front voiceless fricatives, and voiced stops. Stage 2, among a group of speakers roughly one generation to two generations younger (and possibly slightly more), is shown in Figure

6.4 by Trager's (1930) New York State speakers. These speakers show an expansion of the Ithaca speakers' system, with raising now expanded to include /ae/ before the back voiceless fricative /ʃ/ and the velar stop /g/ (which replaces raising before /ŋ/). Stage 3 is represented by Dinkin's Hudson Valley speakers in Figure 6.5, who are one generation younger (or more) than most of the younger speakers in Trager (1930) who have basically the same system as Trager's speakers, although with no raising before /g/.

For the Southern and Central Ohio speakers, if we take the Alternative 1 Central

Ohio speakers of Figure 6.8 as the first stage of the system in Ohio, then the first stage involves just raising of /ae/ before front nasals (nasal clusters and some simple nasals) and front fricatives. This makes stage 1 in Ohio like stage 1 in New York City. If we then use the speakers displayed in Figure 6.4 to represent stage 2 in Southern and Central Ohio outside Columbus, and those displayed in the area labeled Generation 1 in Figure 6.8 for

Columbus, we see expansion as well. In this case, the expansion occurs in the systems of speakers roughly 2 generations younger than those of the stage 1 Ohioans. The systems of these speakers feature raising of /ae/ before front nasals and front voiceless fricatives,

286 but now also include raising before the back voiceless fricative /ʃ/, voiced stops (except

/g/), and voiced fricatives. In addition, in Columbus, some speakers occasionally raise before the voiceless stop /t/. Stage 3, represented by Generation 2 and 3 speakers in

Columbus in Figure 6.8, and the Cincinnati speakers of Figure 6.5, again features even further expansion among somewhat younger speakers than stage 2, like we saw previously in New York City and New York State. In Stage 3, the expansion in

Columbus includes raising before /ŋ/, whereas in Cincinnati, the expansion includes raising before /ŋ/ and /g/.

Looking across these examples, the similarities in New York City, New York

State, and Ohio suggests to us that a four stage process of development and generalization seems to typify conditioned raising in split short-a systems in these areas

(as well as possibly in US English more generally). The stages are included in Table

6.3.32 As this diagram shows, the development of this system involves evolution of the system from a basic system type, where raising only occurs before voiceless front fricatives and front nasals in nasal clusters, to a system showing more expansive raising, with raising of /ae/ occurring before front nasals, all voiceless fricatives, and voiced stops, along with voiced fricatives in some cases, as well (as in Cincinnati and

Columbus).

32 Note that our argument here is somewhat similar to the developmental argument for split short-a in English made by Ferguson (1972:270-273). However, ours differs by a) taking into account co-existing stages of development; and b) being based on a larger body of observed dialect evidence than his version of the development of the system. 287 Stage I of this development process is seen generally in most systems displayed in

Figure 6.3 (all areas listed except Ithaca) and Alternative 1 in Figure 6.8 for our Central

Ohio/Columbus speakers (the men born 1846 and 1854). This is the version of the system with raising only before front voiceless fricatives and front nasals in nasal clusters. Stage

II, with raising now also before voiced stops (sometimes including /g/, but not always) and simple nasals, may also include raising before /ŋ/ (but again, not always). This version can be seen in the systems of the Ithaca speakers of Figure 6.3, Alternative 2 in

Figure 6.8 for Columbus, and New York City speakers in Figure 6.4. As suggested by our analysis above for Columbus, and also the data contained in Figure 6.3 from other areas

Stage of Split Short-a System Growth Segments or Segment Groups Involved Stage 0 front voiceless fricatives Stage I front nasals (nasal clusters only) front voiceless fricatives Stage II front nasals (nasal clusters and simple nasals, sometimes also includes /ŋ/) front voiceless fricatives voiced stops (sometimes including /g/) Stage IIIa (Note IIIa and IIIb are alternatives) front nasals voiceless fricatives (now including /ʃ/) voiced stops (sometimes including /g/) Stage IIIb (Note IIIa and IIIb are alternatives) front nasals (sometimes also includes /ŋ/) voiceless fricatives (now including /ʃ/) voiced stops (sometimes including /g/) voiced fricatives Stage IV (Refinement) Stage IIIa and IIIb phonetic outcomes continue to be realized, with the additional optional outcomes which refine the system:

- raising before /ŋ/ and/or /g/ - dropping of raising before /ŋ/ and/or /g/

Table 6.3: Stages of 19th and 20th Century Split Short-a System Development and Growth in US English (Note: Stage 0 is an initial state for the system as it existed before the 19th Century (see Ekwall, 1945)) 288 of the US, it appears as though stage II may have co-existed next to stage I for a period in

US English, with speakers likely interacting with each other that possessed the 2 alternatives. (This possibility is explored in more detail in Columbus further below.) This also suggests that at some early point in time, likely the late 1700s, Ithaca also would have looked like the other Figure 6.3 areas, showing only raising of /ae/ before front nasals in nasal clusters and front voiceless fricatives. Although we do not have data from

Ithaca to confirm or deny this supposition, the Central Ohio data we have on hand certainly suggests this could be possible.

Stage III has two variants. Variant A includes generalization of the split system to include raising before front nasals, front voiceless fricatives, and voiced stops, like stage

II, but it now also includes raising before /ʃ/ and sometimes also /g/. Variant B is just like

Variant A, except raising has also been generalized to include /ae/ before voiced fricatives. Variant A can be seen in the New York State systems of Figure 6.4 and the

New York City systems of Figure 6.5, while Variant B can be seen in the

Southern/Central Ohio systems of Figure 6.4 and the Columbus G1 systems of Figure

6.10. Like Stage I and II, Stage II and the Stage III variants also appear to have co-existed for a time in various parts of the US.

Stage IV involves refinements to the Stage III versions of the system. This can include the loss of conditioning segments, like in the Hudson Valley systems of Figure

6.5 (raising before /g/ is lost), or the addition of individual segments which condition

289 raising, as in the Cincinnati system of Figure 6.5 or the G2/G3 Columbus systems of

Figure 6.10. In Columbus, the addition is raising before /ŋ/, while in Cincinnati, the addition is raising before /ŋ/ and /g/. Stage III and Stage IV systems appear to have co- existed for a period of time in the US, just like Stage II and Stage III. Stage IV also appears to represent the apex of split system realization, because shortly after this point is reached, we begin to see recession in all of the areas depicted in our various figures, with the systems changing to continuous and then to nasal systems, as we discussed earlier.

6.5 An Alternative Perspective on the History and Development of Split Short-A Systems in US English

If we are correct in our analysis in section 6.4, this suggests that all of the short-a systems that were once found in US English historically ultimately descend from one basic system that developed increasing complexity as time went on as it began to develop in different geographic areas in the US during the 19th and early 20th Centuries. As we mentioned at the beginning of this chapter, historical scholars have posited that the earliest stages of the split system included lengthening and tensing/raising before front voiceless fricatives and nasal consonant clusters (Dobson, 1957; Lass, 1976; Ekwall,

1946). These are the same initial environments found in most of the areas and most of the oldest speaker systems we have access to, including the Central Ohioans we first analyzed the systems of in Chapter 3. Thus, we argue the historical scenario we present here, that the systems all go back to this initial system, and later developments represent

290 different forms of generalization of the conditioning behavior of that initial system, is certainly both historically and linguistically plausible.

In addition, our analysis presents an interesting alternative to the "diffusion from

New York City" scenario presented by Labov (2007) and Dinkin (2009). If the scenario we present here is correct, then the split short-a systems found in all these other areas, including Columbus, did not diffuse specifically from New York City to these areas the way they have argued in their analyses. Instead, this analysis suggests those systems were likely always in these other areas and that they developed in these areas simultaneously with the system in New York City. Furthermore, we believe the system looks different in

New York City than it does in Columbus, or the other areas of the US summarized in the

Figure 6.3, because New York City developed different constraints over time when the system was developing there than it did in the other areas, including Columbus. We believe this is so based on the analysis we have presented above, where we find that, at least if Babbitt is correct, a quite similar system co-existed in New York City and at least some other areas of the US throughout much of the 19th Century. This is further supported by the similarities across areas we discussed in our comparative analysis of

New York City, New York State, and Ohio (including Columbus) in Section 6.4.

The systems in Columbus, as well as the other non-New York City areas, may have originally been biphonemic just as Labov claims the older version of the New York

City system was, but then they quickly changed to monophonemic systems. If this is so, it

291 could explain why the non-New York City systems look like more generic versions of the

New York City system from the vantage point of Labov's (2007) and Dinkin's (2009) diffusion analysis. If this was so, the process likely happened rather early, in the systems of speakers born several generations before the speakers who were recorded and analyzed for studies of areas like Cincinnati, New Orleans, the Hudson Valley, and Albany, NY.

Without taking into account the results of the early studies of split short-a that we reviewed in section 6.2, which Labov (2007) and Dinkin (2009) do not, then it certainly does look like the systems in these areas diffused from New York City. However, given the data we have for Columbus, and the data we have been able to assemble from the studies of the speakers born during the 19th Century that we compiled for Figures 6.3 and

6.4, the scenario we present here as an alternative to that presented by Labov (2007) and

Dinkin (2009) for areas outside of New York City showing split short-a systems does seem well-supported by observed data. In addition, the comparative data available for

New York City among speakers born since 1890, as provided by Labov's work in the

1960s and Becker's work in the 2000s, also supports our postulations, as we have discussed above. In Becker's (2010) New York City data, where speakers belong to birth cohorts quite a bit younger than those found in Labov ([1966]/2006), we see a collapse within two generations of the system once it becomes an allophonic system. In cities such as Columbus, Cincinnati, and possibly those in New York State outside of New York

City, we hypothesize that the same thing generally seems to have happened as well.

292 In New York City, the collapse seems to have begun in the speech of speakers born around the same time as our G2 speakers (circa 1920-1940 in Becker's data), and completed around the time our G4 speakers were born (circa 1971-1990 in Becker's data). In cities like Columbus, we argue the collapse may have begun in our G1 and completed around the time our G3 speakers were born. This difference in timing of the collapse in different communities would further explain why the New York City system still appeared biphonemic in Labov's ([1966]/2006) work while it does not appear biphonemic in our Columbus data (or in most of the speakers in Becker's (2010) data) if indeed it was biphonemic at some point earlier in time in Columbus (and/or other non-

New York City communities).

6.6 The Possible Formation and Development of the Columbus Split Short-a System

In addition, our analysis suggests that, in Columbus, the strongest influences on the formation of the split a system were split-a systems that existed in other parts of the country during the earliest days of Columbus's settlement. Taking the data in Table 6.3 as being potentially representative of what those dialects may have looked like at around this time, we argue that the influential systems either had raising only before front nasals and front fricatives, as in Babbitt's (1896) older group data, and Grandgent's (1892)

Pennsylvania, Maryland, "Valley of Virginia" and Western Tennessee speakers, or else raising before front nasals, front fricatives, and voiced stops (except /g/), like Emerson's

293 (1891) Ithaca, NY speakers. Given what we know of the mix of influential settlement groups at the time (as discussed in section 6.3), it is quite likely that speakers with both systems were represented in the settlement groups, and thus, settlers may have been using both systems simultaneously.

As time went on, native-speakers born in the community would have likely been exposed to both of these systems, plus others as well. This would have included exposure to other, somewhat more exotic variants of the split system, like early versions of the

"Classic New York City" system of Labov ([1966]/2006), with its more specific extra- phonetic conditioning, or the Philadelphia system, with its special additional constraints, as settlers from these areas also moved into Columbus. Of course, they also would have been exposed to other types of short-a systems, too, such as the broad-a system used widely in parts of New England at the time, or the backed, raised, and diphthongal

Southern system, found in parts of the South at the time

During this period of contact among adult speakers with various systems, child speakers likely settled on a compromise version of the split short-a system, through the contact process known as focusing (Trudgill, 1986). Focusing in a process where dialect diversity is reduced through the reduction of different types of variants of a dialect feature, with one in particular being specifically focused in upon. This process typically happens during the early stages of dialect formation in a community, such as we describe here for Columbus, making it a plausible possibility.

294 In addition, over time, speakers likely began to generalize the segmental conditioning of a smaller set of environments to a larger set of environments, possibly through a quite similar process of internal class phonetic analogy that we discussed in

Chapter 4 for the back diphthong classes. However, here, the generalization process shows some signs of spreading on a segmental group basis (that is, by types of segments-

-front nasals, voiceless fricatives, voiced stops, etc), rather than a segment-by-segment basis (that is, individual consonants, like /n/ then /m/ then /f/ then /θ/ and so on), like we saw in Chapter 4 for the back diphthongs. Eventually, the system arrived at the point where raising had been generalized to a rather large set of environments--those noted for

Columbus speakers born in G1, G2, and G3 in our data in Figure 6.8--and this became the "norm" for many speakers in Columbus for at least a couple of generations.

In Columbus, as well as Cincinnati, at least, the "norm" eventually became raising before front nasals (among some speakers, also /ŋ/), all voiceless fricatives, voiced stops

(among some speakers, sometimes /g/, among others, not), and then variably for some speakers, before voiced fricatives and sometimes /t/ (but not for all speakers). In other parts of the country, such as New Haven, the Hudson Valley, and New Orleans, raising happened in more or less the same environments as it did in New York City, but without the added non-phonetic constraints found in New York City. It also happened in New

York City, with the added constraints, and in Philadelphia, of course, too, but with different added constraints there than in New York City.

295 6.7 Some Final Thoughts on The Rise and Fall of Split Short-a US English

Clearly, the somewhat esoteric collection of environments that conditioned raising in the split systems were never very phonetically or phonologically natural, regardless of which version speakers used, and so the sustainability of split systems were clearly not going to be very long lasting. The evidence we now have at hand seems to show that it has not been. If our analysis is correct, it suggests that the system in most areas it was once found in, once /ae/ began to show raising in environments other than before front nasals and front voiceless fricatives, developed somewhere around the mid-1800s

(although possibly the early 1800s in some areas), and had worked its way back out in speaker systems by the later quarter of the 20th Century or so, if not somewhat earlier in some areas. From the information we have been able to reconstruct in our analysis here, it would appear that New York City had the most complex and also the most long lasting of the split short-a systems analyzed extensively in this chapter. During the initial development of the more complex system (with raising before other environments besides front nasal and front fricatives), the New York City system possibly developed slightly earlier than the Columbus system, by a generation or two. Later, when the system began to dissolve in these areas, a comparison between our data and Becker's (2010) data suggests the change from the split system to continuous systems and then on to nasal systems in New York City appears to have lagged behind Columbus by a full generation.

Furthermore, the New York City system was the most complex of the systems, with the

296 extra-phonetic conditioning environments commonly associated with the system demonstrating this complexity.

In sum, this means that the time window for the change was probably about roughly 150-200 years. This story might change a bit if we also include data on the history of the Philadelphia version of the split short-a system, of course, since the system still seems to be in place there. But for the sake of argument in the analysis we present here, if we take the time-window presented above as representative, we can see the split short-a system had a long life in US English. But it was a limited life, nonetheless.

297 Chapter 7: “We Speak ‘Normal Midwestern English’…Most of the Time—On The

Social Motivation of the Spread of Vowel Variation and Sound Change in Columbus33

7.1 Introduction

Now that we have had a chance to look in some detail at several of the significant patterns of vowel variation present in our data, we turn next to exploring the impact of the influence of the social factors we have chosen to analyze on the patterns of vowel variation and sound change we have observed in our data. In addition, we will explore the influence of the geographic, demographic, and economic change forces we first discussed in Chapter 2 that have impacted Columbus throughout the 19th and 20th Centuries. As well, we will explore our informants' answers to several of the evaluative and attitudinal questions we asked during the course of our sociolinguistic interviews with them. In doing so, we will delve deeper into our data and attempt to explore possible social causes for aspects of the propagation and spread of the vowel variation and sound change patterns we have found.

As we have seen in Chapters 3, 4, 5, and 6, each of the major patterns of vowel shift we have observed in Columbus is found to some extent in the speech of all speakers, regardless of sex, social class background, and generational cohort membership.

However, not all vowel classes affiliated with each vowel shift are also found to be

33 We wish to acknowledge Lisa Ann Cairns and Barbara Johnstone for influencing the approach to the analysis of discourse we take in this chapter. 298 undergoing variation in realization among all speakers, and often, not all vowel classes have been found to be undergoing variation among all speakers of similar age, even if they are of similar sex, given the influence of social class as a social factor on the results.

As the same time, as we saw with our G3 speakers, there have been times in the history of Columbus vowel variation that speaker systems, regardless of sex or social class background of the speaker, have been, overall, quite similar.

The reasons for these varying patterns of similarity and difference are complex and multifaceted, yet given our understanding of the history of the community and the impact of change forces on it over the course of time, we believe we may be able to determine at least some of these reasons. As we have already discussed, several of the changes we have looked at, including the fronting of SHOES, BOOT, and BOAT, the

Third Dialect Shift, and the historical occurrence of split short-a systems, appear to have a primarily linguistic cause which explains their initial occurrence in Columbus vowel systems. That is, the motivation for these changes in progress appears to be triggered either by instability in certain parts of the vowel system (as in the near merger of BOT and BOUGHT or the fronting of SHOES), linguistic generalization as a result of phonetic analogy (in the case of fronting of BOOT and BOAT, as well as the backing of BET), segment-related phonetic conditioning (the occurrence of the split short-a sytem), or change in realization of certain vowel classes which appear to be linked to other changes as the result of chain shift (such as the backing of /ae/). In each of these cases, we have

299 attempted to explore the triggering events leading to the occurrence of the change in the vowel systems of our informants, as we discussed in Chapters 4, 5, and 6. However, this exploration only takes us part of the way in describing how these linguistic changes in progress have unfolded, as well as continued to unfold, in the community. The continued incrementation and propagation of these changes in the community appear to have social motivations in addition to linguistic motivations. These social motivations will be the emphasis of this chapter.

Specifically, we focus on an analysis of the social motivations of the incrementation and propagation of changes noted previously to have a primarily linguistic cause and motivation in earlier chapters. As with the analyses we presented in

Chapters 4 and 5, we primarily limit our discussion in this chapter to the analysis of the speakers in our data set who were born between 1896 and 1991. In large part, this is due to the fact that we can only clearly observe a number of the issues we look at among speakers born in these generational cohorts. In fact, for some, we were only able to obtain data from the informants we ourselves were able to interview, and so for those issues, we only refer to data from speakers born 1924-1991. Unlike previous chapters, however, in this chapter, we draw much more heavily on our experiences as a resident of the community. In doing so, we combine an understanding of our speakers' views, obtained via their answers to questions in our sociolinguistic interviews with ethnographic observations we have made while living in the community for 7 and 1/2 years as a

300 resident, both in the North Campus neighborhood and the suburb of Grandview Heights.

In addition, we also draw on an understanding of the community we obtained from reading community publications, such as The Columbus Dispatch, and from publications about Columbus drawn from areas outside of linguistics, such as history (Studer, 1873;

Lentz, 2003), geography (King, 1969; Burgess, 1994; Hunker, 2001), sociology

(McKenzie, 1921), and economics (Wright, 1958; Hunker, 1958). In this way, we combine elements of narrative analysis, ethnographic observation, discourse analysis of written discourse, and the qualitative analysis of informant responses in our analysis of the social motivations of vowel variation and sound change in Columbus.

Before turning to our analysis of these issues, however, we wish to first briefly remind the reader about the principal patterns of vowel variation and language change we have found among our 1896-1991 born speakers in Chapters 4, 5, and 6. This summary is provided via the grouping of informants by their sex, social class, and generational cohort membership. It is presented in section 7.2.

7.2 Sex, Social Class, and Generational Cohort Membership

In Chapter 3, we saw that a number of the dialect features of Columbus speech that initially differentiated our working class speakers from our middle class speakers appear to have emerged as a result of long-term settlement dialect influences. For instance, the raising of BIT, BET, BAT, and PUT before /ʃ/ appears to have entered into

301 the vowel systems of our Blue Collar G1 and some G2 speakers via the original dialect usage patterns of speakers of Upper Southern background. The merger of the NORTH and FORTH classes also most likely entered into the Columbus vowel system (whether

Blue Collar or White Collar) from speakers of New England background, as evidenced in the speech of G1 and some G2 speakers as a merger heading toward completion.

Meanwhile, other forms which appear in both White Collar and Blue Collar speech, like the occurrence of the split short-a system, may have emerged as it does in the speech of

G1 speakers as a kind of compromise form that arose as a result of focusing, being influenced by settlers from different cities in the Mid-Atlantic who possibly had somewhat different short-a systems themselves, as we discussed in Chapter 6.

More generally, given that our G1 population is actually the 5th generation of speakers who would have been born in Columbus, and the fourth born since Columbus really began to grow as a city during the late 1830s, we suspect that G1 speakers may be the first or second to truly show general dialect stability, after a period of significant dialect koineization during the course of much of the 1800s. As Trudgill (1986) discusses, the process of dialect formation is usually at its most intense during the formative years of the second and third generations living in an area, and by the fourth generation, the process is significantly settled down, enough so that we can say the dialect has essentially stabilized. In G1, this stabilization expresses itself in the occurrence of two well-formed dialects in our data--that of the Blue Collar speakers and

302 that of the White Collar speakers--which show differentiation, but also a good degree of shared community norms as well. These differences include a stronger degree of raising shown for /ae/, more instances of raising of BAT, BET, BIT, and PUT before /S/, a stonger pattern of backing for BOT, and stronger patterns of fronting for SHOES, BOOT,

BOAT, and BOUT among Blue Collar G1 speakers than White Collar G1 speakers. At the same time, the fact that both groups show some use of each of these patterns of vowel variation also suggests community norms had definitely been settled on by this point in time. That is, White Collar speakers also do back BOT or front SHOES, BOOT, BOAT, and BOUT, they just do it less so than the Blue Collar speakers. In other words, where the groups vary from one another is degree of use of the vowel variation patterns, rather than use or non-use.

7.2.1 Blue Collar Patterns of Vowel Variation in G2, G3, and G4

In G2 and G3, the differences would become further diminished, although we argue for somewhat different reasons, as we will now discuss. Among some G2 speakers, we find increasing similarity among Blue Collar and White Collar speakers, as we just discussed above. Among other G2 speakers, however, as well as later generations of speakers, we see stronger patterns of difference, but seemingly equally as often, similarity among speakers that likely emerged in the vowel system for other reasons. As we saw in the previous chapters, Blue Collar speech has, over time, developed in

303 somewhat different directions from White Collar speech, although the extent to which this difference is found varies depending on whether we are talking about male or female speakers, as well as whether we are comparing our Blue Collar speakers to White Collar men or White Collar women. As our analyses of the vowel systems of the 4 generations of Blue Collar women in our study population reveals, the overall changes in the vowel system, as indicated by apparent time trends, have led to the occurrence of the Third

Dialect Shift among G3 and G4 women (although only the low vowel component among

G4 women). Meanwhile, among men, G3 Blue Collar speakers show a similar tendency towards the Third Dialect Shift as women, but then in G4, things change, with men showing a reversal of the backing tendency for BAT shown by the G3 men. In addition, they also show quite robust fronting of all of the back vowel diphthongs (and also the short back vowels PUT and BUT), including BOUT. This continued fronting stands in opposition to all other G4 speakers, of White or Blue Collar background, who show retraction of the nucleus of BOUT, likely as a result of its being involved in a low vowel reversal chain shift with BAT and BOT in the vowel systems of the other speakers, as we first suggested in Chapter 3.

7.2.2 White Collar Patterns of Vowel Variation in G2, G3, and G4

In contrast to the Blue Collar patterns of vowel variation noted above, White

Collar speakers show some similarities to, but also some differences from, our Blue

304 Collar speakers, particularly G3 and G4 Blue Collar women. The principal similarity among speakers is of course, that G3 and G4 White Collar men and women show the

Third Dialect Shift, just as the G3 and G4 Blue Collar women do. However, White Collar men and women differ from the Blue Collar women by showing BET-backing in addition to BAT-and BOT-backing. In addition, the G3 and G4 White Collar men and women show somehwat more similarity with one another than Blue Collar men and women do.

As already mentioned, among G4 speakers, both our male and female white collar groups show BOUT nucleus-retraction, suggesting this may be a third step in the retraction chain shift (of BOT and BAT) component of the Third Dialect Shift.34 In addition, they show similar trends for SHOES/BOOT-, BOAT-, PUT-, and BUT-fronting.

7.3 The Influence of Econonomic, Demographic, and Geographic Changes on Vowel Variation

As noted in Section 7.2, over time in Columbus, overall, regardless of speaker sex, Blue Collar speakers in our data have shown tendencies towards realizing vowel shifts similarly, as well as differently, from their White Collar contemporaries over the course of time in Columbus. Among older speakers, G1 Blue Collar men and women tended to be more different from White Collar speakers than G2 Blue Collar men and

34 As we first noted in Durian (2010), we suspect BOUT-retraction to be shown only by dialects also showing the Third Dialect Shift that had BOUT-fronting occurring among older speakers. Thus, in Canada, for instance, we would not see BOUT-retraction, since BOUT-fronting never occurred in most areas to begin with. Meanwhile, in cities like Columbus and Pittsburgh, younger speakers now do show BOUT- retraction, as the Third Dialect Shift takes hold. In Pittsburgh, BOUT-retraction accompanies BOUT- lengthening, a process found among speakers of similar age to our G4 speakers in McCarthy (2004). 305 women are from G2 White Collar men and women. In contrast, among younger speakers,

G4 Blue Collar speakers tended to be more different from G4 White Collar speakers than

G3 Blue Collar speaker are from G3 White Collar speakers. We believe this set of contrasts emerged in our data for several reasons. Two of these reasons are possibly the influence of economic changes and the influence of demographic changes in Columbus.

In regards to the influence of demographic changes in the community, we argue that fluctuations in demographic changes during the course of the 20th Century may have actually led to the stronger patterns of social dialect convergence among G2 and G3 Blue

Collar and White Collar speakers, but later, stronger divergence among G3 and G4 Blue

Collar and White Collar speakers. By demographic changes, we mean several things.

First, we mean population increases involving in-migrants and immigrants35, who moved into Columbus from elsewhere. Second, we mean population movement trends within

Columbus itself involving European Americans that have impacted Columbus since its founding. These include "white flight" of residents from the urban core the suburbs beginning in the 1950s, as well as further changes to the socio-economic mix of residents in various neighborhoods and suburbs as a result of changes to the Columbus economy during the later half of the 20th Century. In addition, other economic changes, such as general changes to apsects of the nature of the Columbus economy involving a shift in working class jobs from light-manufacturing-oriented to service-oriented, and changes in

35 An in-migrant is someone who is a US resident and who has moved from one part of the US to another part. An immigrant is someone who is not originally from the US and who has moved to the US from another country. 306 the size and shape of the urban geography of Columbus, have had a further impact on differences in vowel variation patterns among G2, G3, and G4 Blue Collar and White

Collar Columbusites.

Turning first to G2, it is important to keep in mind that, although the population continued to grow in Columbus during the time period 1920-1945, the years that our G2 speakers were either growing up, or spending their formative years, population growth throughout this period was actually slower, on average, that it was a most other periods during the 200 years that Columbus has existed. This can be seen in Figure 7.1, which reprints the population growth information we originally saw in Chapter 2, Figure 2.2. In large part, this was because of the effect of the Great Depression in the 1930s and the country's involvement in World War II during the first half of the 1940s. Meanwhile, a

Figure 7.1: Population Growth in Central City Columbus (1815-2000) (Note: *1815 is the date of the first population census, per Studer, 1873) 307 recession during part of the 1920s that served as a precursor to the Great Depression also slowed growth down during much of that decade (Lentz, 2003; Hunker, 2001; Raphael,

1979; McKenzie, 1921).

During this period of lower population growth, many Columbusites were also struggling more financially than other periods in the 20th Century, given the recession during the 1920s and the depression of the 1930s (Lentz, 2003; Hunker, 2001). This is why there was slower growth in the population, since folks were not moving to

Columbus at as high of a rate, due to there being not as many financial opportunies in the city for in-migrants and immigrants to earn income as there were during earlier years

(Lentz, 2003). Also because of this economic slowdown, folks who already lived in

Columbus moved around less within Columbus, as well, both geographically and in regards to movement on the social mobility ladder (Raphael, 1979; Lentz, 2003). Thus, speakers of different socio-economic backgrounds might be more likely to remain living near one another, since the structure of the city at the time was one in which speakers of different social backgrounds lived in relatively close proximity to one another, and with speakers having less money to pay to move to other parts of the city, they were more likely to remain in the area in which they already lived (Raphael, 1979; Lentz, 2003).

This put speakers of different class backgrounds in more regular contact with one another than they would be in later years and we believe this increase in "face time" among speakers may have led to an increase in similarity between speakers in the two class

308 groups than we saw among G1 speakers.

In regards to patterns of vowel variation in G2, the increase in contact between groups may have led to an increase in similarity of fronting patterns for the back vowel diphthongs SHOES, BOOT, BOAT, and BOUT. As we argued in Chapter 4, we believe fronting of SHOES, BOOT, and BOAT proceeds throgh a process of phonetic analogy within a speaker's vowel system. However, part of what feeds this analogization process is having input material on which to base the analogy inference process for a speaker. For

G2 speakers, growing up in communties where they would have had more exposure to speakers other social class groups, this means children would receive more linguistic input for the acqusition of their patterns of vowel variaiton from care givers of other class groups, but also peers from other class groups. As a result of this increased contact, we believe this may have led to an increase in fronting behaviors shown by White Collar speakers, who converged more closely with the norms of the Blue Collar speakers they were more heavily in contact with at this point in history.

Similar convergence has been reported for Blue Collar African American speakers with Blue Collar European American speakers for the back diphthongs during a period of increased contact between groups, although in that case, it occurred as a result of the desegregation of Columbus public schoools in the 1980s (Durian, et al., 2007;

Durian, et al.., 2010; Thomas, [1989]/1993). Among African American speakers born during that time period, Thomas, ([1989]/1993) as well as Durian, et al. (2010) found

309 increased fronting relative to African American speakers born in the previous generation.

This made their fronting more like that of European Americans of similar age, who they were in increased contact with on a daily basis as a result of attending desegregated schools than when compared to the speakers of the earlier generational group. The similarity in our results here with what we found for African Americans and European

American speakers in Durian, et al. (2010) leads us to suspect something similar is happening in our data here.

During the period 1945-1980, the period during which our G3 speakers would have been spending their childhood or formative years in the community, population growth in Columbus started to again pick up pace, becoming more like it was before the

1920-1945 "slow period" we described above, in all decades except the final one of this period, 1970-1980. At the same time, Columbus began its "explosive geographic growth" phase, discussed ealier in Chapter 2, during which time Columbus grew from 42 square miles in 1950 to 186 square miles by 1980 via annexation of surrounding geographic space. As shown in Figure 7.2 (reprinted from Figure 2.1 of Chapter 2), although

Columbus would continue to grow after 1980, the rate of growth has been much slower since 1980 than it was between 1950 and 1980. Columbus, of course, also had been growing before 1950, but the growth rate was also slower decade by decade in the years before 1950 than the growth that occurred during the "magic window" of 1950- 1980.

This geographic growth brought with it additional population growth, via new members

310 of the population being added from nearby land that was being annexed and added to the scope and size of Columbus as a political, geographic, and metropolitan entity.

In the midst of all this growth, two other change factors began to impact

Columbus as well, and would continue to strongly impact the community during the

Figure 7.2: Annexation in Columbus (1834-2008) (Source: Columbus Dispatch, 2008) 311 1950-1980 time period. One was "white flight," which involved the large movement of

European Americans with the financial means to do so from the urban core to the suburbs surrounding Columbus. The second was the economic change of the movement of business from the urban core area to the suburbs, a change which accompanied the movement of residents out of the core (Jacobs, 1994, Hunker, 2001). The impact of these changes, mixed with the increase in population from outside of Columbus and the increase in the size of Columbus through annexation, had the combined impact of decentralizing residents and businesses. Although the urban core remained the cultural center of the Columbus area during this time (and still does today), the impact of this period of growth and change had the impact of drastically changing the community from one in which community residents resided closely next to one another in residental neighborhoods located within the core to one in which residents lived in a dispersed social space, with the suburb now becoming the "basic block" in which residents reside, rather than the neighborhood.36

For G3 speakers, many of whom were centrally affected by these changes during the period 1950-1980, the impact that this seems to have had on their vowel systems is to make them even more alike, regardless of social class background or sex. Rather interestingly, in terms of the realization of the different vowels of the Third Dialect Shift,

36 Here, we use "basic block" to mean the basic geographic living space with which residents affiliate, mentally, socially, and physically in their daily lives. As Beauregard notes in Chapter 6 of When America Become Suburban (2006), this change in conceptualization of the "basic block" as we use the term here had a profound impact on how American's perceived the American dream as well as the notion of American exceptionalism, during the mid-to-late 20th Century. 312 speakers born during G3 show the most overlap in having more similar realizations of

BOT, BAT, and BET than among other generational groups. The same can also be said of the back diphthong shift of SHOES, BOOT, and BOAT, and also the low back diphthong

BOUT. As we discussed in Chapter 5, we also find the highest percentage of BOT-

BOUGHT full merger during this period among speakers of different social class backgrounds than among other groups.

Given the amount of demographic, economic, and geographic change and turmoil going on through this period, we might expect our G3 speakers to look more different from one another, since the close-knit cohesion that we argued led to more similarity among speakers of different backgrounds in G2. We believe, however, that the dialectal cohesion of G3 is partly a factor of the change and turmoil of the period, but that it emerged ultimately for several reasons. These include: a) the facilitation of continued similarity by the underlying linguistic structure of the Columbus vowel system as it had come to exist by this point in time (as we already discussed in section 7.1); b) a renewed vigor in the quest for upward mobility by the Blue Collar speakers in post-war Columbus; c) a more overt enregisterment of Blue Collar working class speech as "bad English" among speakers; d) the increasing awareness of dialect differences in other parts of Ohio and other parts of the country; and e) the formation of a "symbolic link" to Columbus as a

"Midwestern place" among some G3 White Collar speakers. (This final trend is also pervasive among G4 White Collar speakers, as we shall see.)

313 Given this rather complex list of reasons, we will devote several sections in the following pages to dealing with them. Via these sections, each reason will be addressed.

Where appropriate, we will also turn to discussing, comparing, and contrasting patterns of vowel variation among G4 groups, as we believe the changes in G4 from G3 are also a byproduct of the same reasons, albiet leading to more differences among speakers than similarities, as in G3.

7.4 The Influence of Social Evaluation and Language Attitudes on Vowel Variation

As residents began to "flee" from the urban core to either the suburbs or outlining

"urban fringe" neighborhoods like Clintonville, many did so with the hope of either

"moving upward" socially, or with the hope of providing their children with a better chance at academic and finacial success. For many Americans during the years 1950-

1980, the chance to move upward by moving outward to the suburbs or the urban fringe became a dominant part of the ideology of "The American Dream" (Beaureguard, 2006), the idea that anyone in America can start "from nothing" and become a financial (and perhaps also social) success. By the mid-20th Century, this dream had been somewhat reinvisioned by Americans, to now include living in suburbs, specifically, as a part of this ideology (Huker, 2001; Hayden, 2003; Beauregard, 2006). Columbus residents were certainly no different, as noted by members of both G2 and G3 in our study, who expressed, in their own words and own ways, that the pursuit of the American Dream as

314 formulated above was explicitly a part of why their families moved from the urban core to the suburbs during the period 1950-1980.

Connected to this aspect of the dream, many of our G3 residents noted that their parents moved their families to the suburbs in the hopes that doing so might open new economic opportunites for their children, since they would be students at suburban schools, which even as early as the 1950s, were already beginning to become perceived as being superior to Columbus Public Schools. This perception would be reinforced over successive years by a combination of real and sometimes imagined differences, such as those revealed by achievement test scores and the winning of academic achievement awards by schools (Foster, 1997).

Said one G2 White Collar man,37 who grew up in Columbus but moved to the

Upper Arlington area later in life after getting married and having a family: "We moved to Upper Arlington because of the better schools and better housing. I thought the kids would have a better chance of succeding in life in the Arlington schools." Said one G3

White Collar woman, who was born in Columbus, but raised for most of her life in

Worthington: "My dad saw better opportunities for us living [in Worthington] than we had [in Columbus]. There were better schools, a safer community, the house was better. I think I was probably more successful as an adult because of the move." Reflecting on his growing up in the "urban fringe" area of Clintonville, a G3 White Collar man said:

37 Throughout this chapter we use a generic reference form for all speakers since, during our interviews, we told our speakers we would protect their anonymity when reporting any direct quotes from them. 315 "Living in Clintonville was prety nice. But Clintonville schools weren't quite as good as the suburban schools, so I went to a Catholic school. That helped to make up the difference between living in Clintonville versus, say Upper Arlington or Worthington."

Along with this emphasis on better education, families often placed some emphasis on better speech use. Several G3 Blue Collar speakers noted several linguistic features of older Blue Collar speech they were discouraged directly from using, either by parents, or more often peers, via mocking of their speech. Among the features of their speech that speakers noted in our interviews with them were the raising of BIT and PUT before /ʃ/, the use of overly diphthongal BOUT realizations, and the use of non-standard syntactic forms like past tense "they come" or "wants VERB+ed." As one Blue Collar G3 man said "My dad use to say things like 'poosh the cart' or 'I'm gonna' go feeshing.'38 I don't talk like that, though, cuz I think it sounded uneducated." A Blue Collar G3 woman said "My grandfather would say things like "poosh" or "boosh"and "feesh." I don't talk like that, but I have friends that do sometimes, probably with their working class friends."

As we can see from these quotes, a typical story told by our Blue Collar G3 informants was that their parents or grandparents would use these forms (and our analysis of G1 and G2 vowel systems supports this observation), but they themselves either would not at all, or would only use them when they were talking to other Blue Collar speakers, likely in more casual speaking situations. Rather interestingly, this was more true of

38 Throughout this chapter, we use orthographic representations of the vowel sounds in an attempt to better simulate the pronunciations of our informants as they attempted to imitate these pronunciations. 316 informants who were raised in Blue Collar homes and were also themselves Blue Collar.

This suggests the ideology of "correctness in language use," or at least sounding more like a White Collar speaker at any rate, was important enough to speakers in this generation that even for those for whom the dream of upward mobility did not become a reality in their own lives, it continued to "stick with them" even into their adult lives.

We suspect part of what also has made this ideology "stick" with these speakers is that, on a more community-wide level, Blue Collar speech, beginning in this time period, and continuing through today, in the community has become marked by community members at varying levels of the social strata as "Bad English." In Columbus, overt social evaluation and commentary regarding "Bad English" often comes in the form of commentary on linguistic forms that tend to be considered "bad" no matter what region one lives in in the US, such as the use of double negation or subject-verb disagreement, but many forms are directed towards a very specific local form--older Blue Collar

Columbus speech. It should be noted here that this form is actually not really local--what is often referred to in the criticisms of these forms are actually historically Upper

Southern/Appalachian speech features that became more widespread when areas like

Columbus were being initially settled, and then later, through the infusion of additional

Appalachian speakers during the 1910s and 1920s, and then again after World War II, into the Blue Collar work force in Columbus. These features include the raising of BIT and PUT before /ʃ/, "intrusive r" in words like "wash," or the use of non-standard

317 syntactic forms like "they come" or "wants VERB+ed," or the use of lexical items such as

"paw paw" or "maw maw." But being used by (often older) Blue Collar locals, and thus, being brought into the perception of Columbusites by Blue Collar locals, the features are perceived and semiotically linked to local Blue Collar speech, thus making Columbusites deal with them in commentary and criticism as local Blue Collar speech forms.

This fact can be seen not just in comments provided by Blue Collar G3 speakers regarding their own vowel realizations, but also in comments provided by G3 White

Collar speakers, and G3 and G4 Blue and White Collar speakers as well. Overt social commentary was provided on overly diphthongal BOUT as being a feature of Blue Collar speech by G3 and G4 speakers of White Collar backgrounds, G4 speakers of Blue Collar backgrounds, and the aforementioned G3 Blue Collar speakers as well. Speakers called the feature variably "hickish," "bad English," "ignorant sounding," "Southern," "hilljack speech," or "a hillbilly way of talking." G3 White Collar speakers also commented explicity on raising of BIT and PUT before /ʃ/ this way, although like their Blue Collar

G3 contemporaries, they did not also comment on raising of BET or BAT before /ʃ/ negatively. (In fact, neither group commented on raising of BET or BAT in this environment at all.)

Among G4 speakers, none commented explicitly on BIT or PUT raising before

/ʃ/, but two speakers did comment on BET raising before /ʃ/ as sounding either "old fashioned" or "Blue Collar." White Collar G4 speakers who mentioned BET-raising

318 tended toward descriptions of the "Blue Collar" variety, while Blue Collar G4 speakers who mentioned it tended toward decriptions of the "old fashioned" variety. This difference seems to accurately reflect its use, in the sense that it would be "old fashioned" for a Blue Collar speaker, in being found mostly in the speech of G2 Blue Collar speakers

(although some G3 speakers still do use it on occasion, as well), while it is accurately a

Blue Collar feature, as noted by White Collar G4 speakers. Beyond mentions of some of the syntactic features of Blue Collar speech mentioned earlier, with the exception of the mention of BAT-raising as being a feature of Cleveland speech, and exaggerated imitations of SHOES and BOOT fronting as being features of California speech, speakers in our interviews did not specifically directly mention any other features of Columbus speech as being notable, nor did they provide social commentary or remarks in which language attiudes regarding Columbus linguistic variables could be assertained.

Beyond informant commentary, one another source of linguistic information indicating a semiotic link being drawn between Blue Collar English in Columbus and both "Bad English" as well as overtly Appalachian speech characteristics comes in the form of discussion of speech forms in local media, such as newspaper columns. During the 1990s and 2000s, Columbus Dispatch columnist Mike Harden wrote a series of articles which provides fairly robust documentation of the expression of these views in the larger community. In them, Harden quite directly points to Appalachian links between the forms he discusses, and he also calls most of them forms of "Bad English" outright.

319 The pieces are written in a humorous tone, with the implication being that the forms would naturally be in Harden's speech if he hadn't worked to eliminate them, and thus, the dialect features are also marked as being "historical" in the sense that they are forms that are best left "behind" in one's speech. Although clearly written for entertainment purposes, Harden's views quite succinctly summarize those of my G3 Blue Collar speakers who expressed negative views about their own use of marked features like BIT and PUT raising before /ʃ/ and overly diphthongal BOUT.

Among Harden's most detailed articles on the subject are "Distinctive Local Lingo

Lives on in 'Clummus" (2006a), "How to Speak NASCAR" (2006b), "Kidspeak Leaves

Paw Paw Smacking His Forehead" (2001), and "Fixin' to Twang" (1995). In each of these articles, Appalachian English features play a prominent role, not just in the discussions in the text, but the titles, as well. Typically, when the dialect features are presented in text,

Harden provides an exaggerated orthographic spelling for the vowel variable, with an absurd example of its use in real speech and some kind of admonishment that using the form makes one sound either stupid, uneducated, poor, foolish, or an idiot, or more than likely, some combination of these things. For instance, when discussing PUT raising before /ʃ/, Harden writes "Boosh: The sitting president of the United States. 'I see that

Boosh's approval rating is down to 37 percent. Where I went to high school, that's an F'"

(2006a). When discussing the use of "fixing to," a syntactic feature of Blue Collar speech, at least as we observed it in the speech of G2 and G3 speakers in our data, he

320 writes:

"To most people, the word "fixing" meant repairing; to me it meant preparing, as in: 'I was fixin' to go to White Castle as soon as I get these clothes warshed.' Actually, I probably would have said: 'I was fixin' to get me some White Castles as soon as I get these clothes warshed'"(1995).

Here, Harden attempts to drive the point home that people are foolish or ignorant who use

"fixin' to" by combining its use with other marked features of Blue Collar speech, such as the use of "warshed" (intrusive "r") and "get me some" in the second, or "enhanced," version of the sentence he gives for "fixin' to." His overt discussion of these language matters in the Dispatch in these articles shows the extent to which language attitudes regarding use of Blue Collar speech features as negative or unappealing are pervasive among many G3 Blue Collar speakers, as well as their G3 White Collar contemporaries.

Given the overall attitude of negativity towards the use of these forms that G3

Blue Collar speakers have experienced, it is perhaps not surprising these speakers have shifted away from the use of realizations that are more overtly marked as Blue Collar.

Further reinforcement of this desire to avoid sounding overly marked has also appeared to come in the form of economic incentives to do so as the speakers continue to be employed in the workforce as adult speakers. During the course of their lifetimes, the shift in Blue Collar jobs from those which were more light-industry- and manufacturing- based to those that are more information-industry and customer-service oriented has been

321 quite pronounced (Hunker, 2001). A number of the older speakers in the G3 group began working in light industry jobs towards the beginning of their careers but have now found themselves having had to switch to working in service-oriented jobs as they have gotten older. These service-oriented jobs often put the speakers in contact with other speakers of both Blue Collar and White Collar backgrounds, and so continuing to sound less Blue

Collar in their realization of at least the overt markers of Blue Collar speech noted by many informants--robustly diphthongal BOUT, raising of BIT and PUT before /ʃ/, or the use of syntactic forms like "fixin' to," is something the speakers likely want to avoid, since sounding "intelligent" or "White Collar" has an economic benefit for them. As one

Blue Collar G3 woman said "You got to sound professional when you work a service- oriented job like the ones that around today. So you have to have no accent--you can't use incorrect grammar or say 'feesh' and 'boosh' of the 'brawn caaw.' You got to talk normal."

Among G4 speakers, however, even though we still find overt discussion among informants of Blue Collar speech features as being negative and/or features of "Bad

English," we also find a change among G4 Blue Collar men to stronger use of marked vocalic features than in G3. As we discussed in earlier chapters, G4 Blue Collar men front BAT rather than backing it, and they continue to front BOUT, rather than beginning to retract it, like other G4 speakers, and realize it more dipthongally than other G4 speakers, as well. These speakers also continue to show a continous realization pattern for short-a, rather than the nasal pattern of most other G4 speakers.

322 We believe the changes in the vowel reaizations of G4 Blue Collar men may have not just linguistic significance, but social significance, as well. Specifically, we belive these changes may may indicate an attempt to symbolically react negatively to the economic changes that impacted Blue Collar jobs in Columbus during the last 20-30 years. Unlike their G3 counterparts, who lived through at least some, if not many or even most, of these changes as working adults, the G4 men have essentially been born directly into a changed culture, or at least have spent all of their adult lives working within it.

Even the oldest speakers in our G4 group would have not started working as adults within

Columbus until the mid-1990s, a time when most of the significant changes in the Blue

Collar job market in Columbus had basically been set into place in full swing (Hunker,

2001). As a result, these men may not see as much direct benefit to converging with

White Collar speech norms as their G3 counterparts, since they have been under less pressure to change career paths (at least so far) from one type of work (light industry) to another (service-oriented work), and they may also see less of a direct link between using less overtly marked Blue Collar speech forms with economic success than their G3 counterparts.

During their interviews, G4 Blue Collar men did not express negative feelings about blue collar speech , and they tended to focus on non-Columbus speech forms when commenting on vowel realizations they have noticed. For instance, one man, when asked if any pronuncations stand out to him in local speech, either in Columbus or another part

323 of Ohio, said: "Not in Columbus, but in Cleveland everyone is nasally. They say things like "AEpple" or "clAEss" [emphasis on the /ae/]." Another man said "People down in

Cincinatti say the "braawn coww" with a stretched out sound. We don't do anything like that around here." The lack of awareness of different realizations for these vowel classes in Blue Collar Columbus speech suggests a lack of awarness by these men. It also suggests possibly a lack of stigma being attached to them, because if they were stigmatized, we might expect either than the men would comment on their use, or, their own use would show a difference; that is, they would not continue to front BOUT and realize it so dipthongally, or perhaps they would not front BAT. (In other words, both men front BAT and BOUT). Beyond economic concerns, then, it may also be that community norms are changing, so that G4 men are not receiving the same number of

"negative" messages regarding Blue Collar language use as G3 men did when they were growing up.

Responses from our G4 Blue Collar women suggest his lack of "negative" messages regarding Blue Collar forms reported by men may be gender-marked. Several

G4 women reported being told, either explicitly or implicitly, by their parents and other care-takers, and/or via interactions with peers in school, that overtly marked Blue Collar forms should be avoided in speech, and that if they wanted to be successful, or have an upwardly mobile career path, they should use "correct speech forms." Although these women gave syntactic examples of what these forms might be, for instance--"don't use

324 double negatives, don't say 'ain't,' don't use profanity, or even don't use 'fixin' to,'" the only vocalic forms they would point to was fronted or overly diphthongal (or both)

BOUT. For example, one woman said "Saying 'bwaan' makes you sound like a hilljack,

Ladies don't talk like hilljacks." Another woman said: "I've always been told don't say

'ain't' because 'aint's not a word," and another said "I was told: 'don't curse' cuz it's unladylike. Don't say "braawn," cuz you sound like a hillbilly." Like their G4 White

Collar counterparts, the women also say they have been told they should not use raised

BAT or fronted BOOT, so as to avoid sounding like a Clevelander or "Valley Girl," but neither of these forms was marked as being either a local form or a Blue Collar form, to be avoided because of social class marking. We believe the admonishments our G4 female speakers received versus the lack of admonishments our G4 males received, has also contributed strongly to the differences between G4 Blue Collar men's and women's patterns of vowel variation in our data.

In sum, we believe the recent divergence for BAT and BOUT shown by G4 Blue

Collar men may ultimately represent a subconscious change in realization patterns among these speakers as a kind of symbol or badge of pride in maintaining a strong traditional

Blue Collar image, even when traditional Blue Collar work opportunities in Columbus are becoming increasingly non-existent. For G4 Blue Collar women, however, we believe they are continuning to show more robust use of "non-Blue Collar" features, such as the low vowel component of the Third Dialect Shift, in part because they have been

325 encouraged to do so, but also in part because we believe they may still see language use being linked to wider economic employment opportunities. As such, this marks their language use ideologies and apparent vowel realization strategies as being more like those of their G3 Blue Collar counterparts than those of their G4 Blue Collar male contemporaries.

7.5 The Influence of Attitudes Regarding Columbus as a "Midwestern Place" on Vowel Variation

As we just discussed, in G3 and G4 speech, we find commentary on Blue Collar forms overly marked as Blue Collar. Specifically, these forms include BOUT-fronting and raising of short vowels before /ʃ/, but as we also saw, we do occassionally run across commentary regarding vowels like BAT and SHOES/BOOT being marked as "non-local" by these younger speakers For BAT, this finding suggests that there is a difference in awareness of the significance of raised /ae/ as a dialect marker of Northern Cities speech, and specifically a marker of Cleveland speech, among speakers of different age groups.

We suspect this difference may have emerged for one of two reasons in our data.

The first is that many of the G3 and G4 speakers admitted to having more contact with

Cleveland speakers than G2 speakers--thus, they had more opporunity to notice a difference between dialect areas than their G2 counterparts. The second is that G3 and G4 speakers have an /ae/ that is significantly lower than G2 speakers, many of whom either still show the older split short-a system in their speeh or else have a continuous system

326 which shows residue of the older split system. In either case, their /ae/s are higher and tenser, and thus, more like the NCS-raised /ae/ of Cleveland, which likely makes it less noticable to G2 speakers as a result, hence their not commenting on it's raised quality in interviews.

For SHOES and BOOT, it seems likely that awareness through movies and television of this "California" or "Valley Girl" stereotype is responsible for the commentary of G3 and G4 speakers. As one G4 White Collar woman said, "In California, they have these drawn out pronucations of 'DUEde' and 'RUde' or, like, 'BUEt.' Totally surfer girl, like in movies." A G3 White Collar woman said "California is definitely different than Columbus. Like you see on tv, they say 'DUEd,' not 'dude' like we do." For

BOOT and SHOES, then, we suspect media, rather than real world speech usage, has led to the awareness, since a) some speakers refer specifically to media stereotypes, and b) those that did not admitted to not having native-Californians as friends.

More generally, the contrast between the variables our speakers are aware of-- those marked as Blue Collar and local, on the one hand, and those marked as regional

(and classless, or least not overtly class-marked by any of our speakers), also suggests to us that our speakers' notions of non-standardness do not often extend to White Collar speech in Columbus. As we discussed in section 7.4, both Blue Collar speakers and

White Collar speakers identifed the features of Blue Collar speech discussed there. And both groups also identified the regional features we have discussed above in this section.

327 But when it came to identifying noticeable features of White Collar speech, neither Blue Collar speakers, nor White Collar speakers themselves, could identify much in the way of notable speech features. A few of our G4 White Collar speakers did mention some syntactic and lexical features of White Collar Columbus speech as being notable--namely the use of constructions such as "needs Verb+ed" and positive anymore or lexical items like the use of pop for a carbonated beverage,. However, all of these items are either non-local, regional variants which tend to enjoy wide notablility in discussions of regional variation--like pop--or they were features that are marked enough to speakers of other dialects to warrant notice by those speakers in the speech of our

Columbus speakers. If those other speakers had not pointed out the use of these features to our Columbus speakers, it seems likely the Columbus speakers might have continued to remain unaware of their own usage of them in natural speech.

What speakers, both White Collar and Blue Collar, alike, did notice about White

Collar speech in Columbus, particularly among G3 and G4 speakers, was just how standard they perceive the speech to be. Most speakers, when I asked to interview them, thought it was almost an absurb idea that I would want to interview them for a study on the use of a "Columbus dialect," and seemed more interested in doing the interview to see what kind of "crazy ideas" I had about this than for any other reason. In most interviews I conducted, speakers would again and again discuss just how standard they believe

Columbus speech to be. As one G3 White Collar woman said, "middle class folks in

328 Columbus speak standard, Midwestern speech." Another woman, a G4 White Collar speaker who had had her use of positive anymore pointed out to her, said just after mentioning this fact in the interview: "middle class folks in Columbus speak standard,

Midwestern speech most of the time...". Among many White Collar speakers, statements like these on language standardness are often accompanied by a sense of pride in the fact that what the speaker speaks is this standard variety of American English.

Although the underlying views and perceptions of our speakers on this issue clearly require more investigation in a later study to determine exactly how this ideology regarding their language use has formed and exactly what it means to our informants, one conclusion we believe we can draw from this material at this point is that this ideology may be a contributing factor to the continued use of G3 White Collar vowel shifts among our G4 White Collar speakers without much change or modification, like we saw among our G4 Blue Collar men. How large of a factor, it is difficult to determine given the data we have on hand at present, but we can see it is likely connected. However, we suspect the stronger factor in its continuation is likely simply a linguistic one. The Third Dialect

Shift has naturally evolved in White Collar speech, as we saw in Chapter 5, and so, short of some kind of strong social factor suddently impacting its continued incrementation, its progression in White Collar vowel systems is what we should expect to observe.

329 Chapter 8: Conclusion

8.1 Introduction

Throughout the last 6 chapters, we have attempted to tell the long and complex story of the most robust vowel system changes that have occurred in the vowel systems of White Collar and Blue Collar speakers living in Columbus over the course of the 20th

Century. With the addition of the data from the Central Ohio men born in 1846 and 1854 who were interviewed for the Linguistic Atlas of the North Central States [LANCS], we have also been able to supplement our discussion, such that, at several key points, we have actually been able to reconstruct what we argue is a plausible version of the Central

Ohio vowel system as it likely existed during the middle of the 19th Century. As we have also argued, given what is known about Columbus as a dialect area versus what is known about other cities in the US Midland, many of our results for our 20th Century speakers are likely reflective of vowel variation in these other cities, too.

Specifically, based on the comparative analysis of Labov, Ash, and Boberg

(2006), as well as our own understanding of the comparative results available in Fogle

(2008), Lusk (1976), Boberg & Strassel (1995, 2000), Bigham (2008), and Gordon &

Strelluf (2012), we argue the cities our results most likely generalize to beyond

Columbus are cities such as Cincinnati; Indianapolis; Kansas City, MO; Kansas City, KS; and Springfield, IL (as well as other parts of Central and Southern, IL). However, we

330 wish to note this comparative estimation is based purely on similarities between our data and the results reported that we cite above. It may turn out to be the case that, if historical data similar to what we have acquired for Columbus is found, analyzed, and explored for these areas, that we may in fact find somewhat different patterns of historical development in these other cities. Previous cross-area research for the US Midland has indicated that historically, the Midland has been an area typified by strong local diversity in different subareas and cities, but it should also be mentioned that much of that data is based only on impressionistic analysis of vowels, which is highly sensitive to problems of inter-transcriber reliability (e.g., Marckwardt, 1940, 1942; Frazer, 1978; Nobbelin, 1980), and/or lexical item incidence (e.g., Davis, 1948; Markwardt, 1957; Carver, 1987;

Cassidy, 1985). Just because cities and areas were known historically to show different lexical item distributions from one another does not mean their vowel systems must be as equally different.

8.2 Summation and Closure

Given the length of this work, we think that, as a closing statement, perhaps the best choice for our concluding remarks is to present a chapter-by-chapter review of its most important points. This will allow the reader to also obtain a view on what we thought the most important contributions of our work have been. We begin with the major content points, taking Chapter 2 as our starting point, and working through to the

331 end, with Chapter 7.

In Chapter 2, we provided a profile of Columbus as a speech community. In doing so, we attempted to present the most important facts about the community that are relevant to the present study. This includes a brief discussion of the settlement history and the influence settlement patterns may have had on dialect formation during the early

19th Century in Columbus. In addition, we attempted to present a discussion of the most relevant geographic, demographic, and economic factors we believe have had an impact on the realization of vowel variables, and on the progression of vocalic sound changes and vowel shifts, during the course of the 19th and 20th Centuries, in the community.

In Chapter 3, we provided an overview of the major patterns of vowel variation that have been found in previous studies of Columbus, as well as an overview of the patterns we found in our own study. The discussion focused on our instrumental analysis of the whole vowel system, with data selected for analysis being collected from speakers born between c. 1896 and 1991, and divided into 4 birth-year cohorts. These cohorts served as the basis for the analysis presented not only in this chapter, but all the remaining chapters, as well. Beyond this, we also presented a re-analysis of raw field records originally made by Cassil Reynard for the Linguistic Atlas of the North Central

States [LANCS], obtained from informants born 1846 and 1854 in Central Ohio, so as to establish a "baseline" initial state for our analyses of vowel systems in our instrumental data. All major classes were analyzed in these men's vowel systems, as well.

332 As a result of our analysis of both sets of data, and our comparison of our results with earlier studies, we observed a number of significant patterns of vowel variation. The most systematic among them were the occurrence of fronting of the back diphthongs

SHOES, BOOT, BOAT, and BOUT, fronting of the short back vowels PUT and BUT, raising of the low mid diphthongs BITE and BIDE, backing of the low vowels BOT and

BAT, backing of the front vowel BET, and merger of NORTH and FORCE, as well as merger of BOT and BOUGHT. Among these patterns, two were chosen for detailed focus in individual chapters, where a more detailed statistical analysis of the vowels in question could be completed. These patterns are the fronting of the back diphthongs SHOES,

BOOT, and BOAT, as well as backwards shifting of the low vowels BOT and BAT, and the backwards shifting of the front vowels BET (and sometimes) BIT. Together, the combination of this shift of vowels was identified as The Third Dialect Shift.

Additionally, it was argued in this chapter that BOT and BOUGHT merger should be analyzed in conjunction with the Third Dialect Shift. In addition, the older pattern of raising of BAT, which preceded BAT's involvement in the Third Dialect Shift, was analyzed separately in Chapter 6. This was so because a detailed study of the raising of

BAT in Columbus contributes valuable information on the occurrence of split short-a systems in US English, information that has not been discussed, observed, or analyzed in

Columbus previously.

Chapter 4 built on Chapter 3 by presenting the first of the detailed explorations of

333 vowel shifting patterns chosen for highlighting in the dissertation. The pattern chosen for analysis here is the parallel fronting of the back diphthongs SHOES, BOOT, and BOAT.

Chapter 4 not only discussed the results of linear mixed effects regression analyses for each class, it also discussed possible triggering events for the vowel shift. Ultimately, we argued that Ash's (1996) explanation that backing was triggered by merger of /iw/ and

/uw/ seems most plausible based on the available evidence we have at hand for

Columbus.

Chapter 4 also dealt with some important theoretical contributions our data were able to make to our understanding of vowel variation and vocalic sound change. Chapter

4 discussed parallel shifting and the role phonetic analogy (Schuchardt, 1885;

Vennemann, 1972) appears to play in the generalization process that accompanies the occurrence of parallel shifts and their incrementation in speaker vowel systems, not just in Columbus, but also in varieties of US English more generally. We find that parallel shifts show strong evidence of diffusion through speaker vowel systems via two types of phonetic analogy. The first is vowel class internal analogy, which involves the spread of sound change on a conditioning environment-by-conditioning environment basis. The second is vowel class external analogy, where sound change spreads across vowel classes on a vowel class-by-vowel class basis.

Chapter 5 presented a detailed analysis of the vowel classes involved in the Third

Dialect Shift. Specifically, it deals with documenting the inception and development of

334 the Third Dialect Shift in Columbus, a topic that has not been investigated previously in either a study of Columbus or of the US Midland. Like Chapter 4, Chapter 5 presented the results of linear mixed effects regression analyses for each class, and it also attempted to investigate possible triggers for the shift. Here, we ultimately found that the triggering event for the Third Dialect Shift is not BOT/BOUGHT merger, per se, as has been argued in previous studies, but rather simply BOT-backing. We argue this based on patterns in our own data, where we find speakers who do not merge BOT and BOUGHT engaging in

Third Dialect Shift, but also, speakers who do not merge but who also do not begin to back BAT until BOT first begins to back.

Chapter 6 presented an in-depth analysis of the historical occurrence of split short-a systems in Columbus. This is another pattern of variation not previously discussed in earlier analyses of the Columbus vowel system. Specifically, this chapter dealt with a descriptive analysis focusing on the change of the system from a split short-a system to a continuous system. A second change then follows, one involving a switch from a continuous system, which was found most often among G3 speakers, to a nasal system, which was found among most G4 speakers.

In addition, we also presented a novel analysis in which we argued that the split short-a system, found among most G1 speakers and some G2 and G3 speakers, emerged in the form it did in Columbus as a result of focusing resulting from the period of dialect contact that occurred among the settler dialects of Columbus during the course of much

335 of the 19th Century. Most likely, the most influential dialects in the formation of the system were those of the Mid-Atlantic and the Northeast, given the higher percentage of speakers from those areas among the initial settlers of Columbus. Thus, although it resembles the classic New York City system, it also features a number of key differences-

-namely, it lacks any of the extra-phonetic conditioning found in the New York City system. However, we argue this difference is not the result of loss of constraints via diffusion of the New York City system to Columbus, a la the explanation Labov (2007) provides for the occurrence of the quite similar Cincinnati historical short-a system, but rather is the result of generalization of the phonetic conditioning environments of the split-a systems found in the settler dialect systems, which likely also lacked the non- phonetic constraints of the New York City system.

In Chapter 7, we completed our analysis of vowel variation and sound change in the Columbus vowel system by discussing the impact of demographic change, urban growth, and economic change--three change forces first discussed in Chapter 2 which we believe had an influence on some of the patterns of vowel variation and sound change found in our data. In addition, the role of social evaluation, language attitudes, and conceptions of Columbus as a "Midwestern Place" as contributing factors in the incrementation and propagation of these changes was also explored to some extent.

Among the vowel classes we believe have been impacted by these social forces and attitudinal factors are the convergence of G2 speakers, regardless of class

336 background or sex, on similar degrees of backing of SHOES, BOOT, and BOAT, as well as the convergence of G3 speakers, again regardless of class background or sex, on similar patterns of realization for the majority of the Third Dialect Shift vowels-- specifically BOT, BAT, and BET. In addition, we believe the recent divergence for BAT and BOUT shown by G4 Blue Collar men may represent a conscious change in realization patterns among these speakers as a kind of symbol or badge of pride in maintaining a strong traditional Blue Collar image, even when traditional Blue Collar work opportunities in Columbus are becoming increasingly non-existent. In contrast to them, we believe G4 Blue Collar women are showing stronger signs of continuing convergence with G4 White Collar speakers, who show Third Dialect Shift usage tendencies, because to them, "proper/standard language use" may still be seen as a link to obtaining better economic job opportunities in the current Columbus economy than using more traditional Blue Collar vowel realization patterns.

Overall, in G4, we are left with a snapshot of the most recent patterns of vowel variation in Columbus being that White Collar Men stand out as the leaders of linguistic change for the Third Dialect Shift, with speakers showing backed realization of BET,

BAT, and BOT. G4 White Collar women follow closely behind these men, showing the same patterns, but not quite as robustly. G4 Blue Collar women follow next in line, showing backing only of BAT and BOT, but not BET, meaning they realize only the low vowel chain shift component of the Third Dialect Shift in Columbus. All 3 sets of

337 speakers also show backing of BOUT, which is likely chain linked to the backing of BOT and BAT, and thus, may represent a new stage of the Third Dialect Shift among

Columbusites. Finally, as mentioned above, G4 Blue Collar men stand out as a group which differs from the others, showing non-backing of BAT, a continuous rather than nasal pattern of realization of BAT, as well, and continued fronting of BOUT, rather than the backing shown by other G4 speakers.

These chapters, thus, provide the data and analyses needed for the definitive account of the history of Columbus's vowel systems, and, by extension, much of the development of vowel variation patterns in vowel systems in US English over the course of the 19th and 20th Centuries.

338 Bibliography

Allen, Harold B. 1977. Regional dialects, 1945-1974. American Speech, 52: 163-261.

Anderson, John, Hope C. Dawson, and Brian D. Joseph. 2010. Historical linguistics. In K. Malmkjaer (Ed.), The Routledge Linguistics Encyclopedia (3rd Edition). Oxon: Routledge Publishers. pp. 226-252.

Ash, Sharon. 1996. Freedom of movement: /uw/-fronting in the Midwest. In Jennifer Arnold, Renee Blake, Brad Davidson, Scott Schwenter, and Julie Solomon (eds.), Sociolinguistic variation: Data, theory, and analysis. Stanford, CA: Center for the Study of Language and Information, Stanford University, pp.3-25.

Ash, Sharon. 2002. The distribution of a phonemic split in the Mid-Atlantic region: Yet more on short-a. University of Pennsylvania Working Papers in Linguistics, 8.3:1- 11.

Ash, Sharon. 2006. The North American Midland as dialect area. In Language variation and change in the American Midland: A new look at “Heartland” English, ed. Thomas E. Murray and Beth Lee Simon. Amsterdam/Philadelphia: John Benjamins. 33-56.

Babbit, E. H. 1896. The English of the lower classes in New York City and vicinity. Dialect Notes, 1: 457-464.

Baayen, R.H., Davidson, D.J. and Bates, D.M. (2008) Mixed-effects modeling with crossed random effects for subjects and items. Journal of Memory and Language 59:390-412.

Beauregard, Robert. 2006. When America became suburban. Minneapolis: University of Minnesota Press.

Becker, Kara. 2010. Social conflict and social practice on the Lower East Side: A study of regional dialect features in New York City English. Doctoral Dissertation: New York University.

Becker, Kara, and Amy Wing-mei Wong. 2010. The short-a system of New York City: An update. University of Pennsylvania Working Papers in Linguistics,15.2:11- 20.

339 Benson, Erica J. 2005. Folk perceptions of dialect perceptions in Ohio. In Joseph, Brian D., Carol G. Preston, and Dennis R. Preston (Eds.), Language diversity in Michigan and Detroit. Ann Arbor, MI: Caravan Books. pp. 35-60.

Bigham, Douglas S. 2008. Dialect contact and accommodation among emerging adults in a university setting. Doctoral Dissertation: University of Texas, Austin.

Bigham, Douglas S. 2009a. Northern California Vowels in Southern Illinois. Paper presented at the Annual Meeting of the American Dialect Society, San Francisco, CA.

Bigham, Douglas. S. 2009b. Correlation of the low back merger and TRAP-retraction. University of Pennsylvania Working Papers in Linguistics, 15.2:21-31.

Bloomfield, Leonard. 1933. Language. Chicago: The University of Chicago Press.

Boberg, Charles. 2000. Geo-linguistic diffusion and the U.S. Canada border. Language Variation and Change, 12: 1-24.

Boberg, Charles. 2005. The Canadian shift in Montreal. Language Variation and Change, 17:133-154.

Boberg, Charles. 2009. Divergent and convergent patterns in the phonetics of Canadian English. Paper presented at the Annual Meeting of the American Dialect Society, San Francisco, CA.

Boberg, Charles, and Stephanie Strassel, 1995. Phonological change in Cincinnati. Penn Working Papers in Linguistics, 2.2:25-35.

Boberg, Charles, and Stephanie Strassel. 2000. Short-a in Cincinnati: A change in progress. Journal of English Linguistics, 28: 108-126.

Boryczka, Raymond, and Loren Lee Cary. 1982. No strength without union: An illustrated history of Ohio workers, 1803-1980. Columbus, OH: Ohio Historical Society.

Brown, Philip Leslie. 1967. People on the move: The formation of Ohio's ethnic composition, 1870-1900. MA Thesis, The Ohio State University.

340 Burgess, Patricia. 1994. Planning for the private interest: Land use controls and residential patterns in Columbus, OH, 1900-1970. Columbus, OH: The Ohio State University Press.

Campbell-Kibler, Kathryn, Teresa Pratt, and Victoria Cook. 2008. Vowel systems in Ohio: Reality and perceptions. Paper presented at NWAV 37, Houston, TX.

Carver, Craig. 1987. The midlands: The Upper South and Lower North dialects. American regional dialects. Ann Arbor, MI: University of Michigan Press. pp. 161-204.

Cassidy, Frederic C. (Ed.). 1985. The dictionary of American Regional English. Cambridge, MA: Belknap Press of Harvard University Press.

Chaddock, Robert E. 1908. Ohio before 1850: A study of the early influence of Pennsylvania and Southern populations in Ohio. Studies in History, Economics, and Pubic Law, XXXI, No.2. New York: Columbia University.

Chomsky, Noam, and Morris Halle. 1968. The sound pattern of English. New York: Harper and Row.

Clarke, Sandra, Ford Elms, and Amani Youssef. 1995. The third dialect of English: Some Canadian evidence. Language Variation and Change,7:209-228.

Clopper, Cynthia, David B. Pisoni, and Kenneth de Jong. 2005. Acoustic characteristics of the vowel systems of six regional varieties of American English. Journal of the Acoustical Society of America, 118.3:1661-1676.

Cohen, Paul. 1970. The tensing and raising of short [a] in the Metropolitan Area of New York City. M.A. thesis, Columbia University.

Cole, Charles C., Jr. 2000. A fragile capital: Identity and the early years of Columbus, Ohio. Columbus, OH: Ohio State University Press.

Columbus Planning and Development Division. n.d. Columbus growth map series, 1950-2000. City of Columbus Department of Development Website. Available: http://development.columbus.gov/

D'Arcy, Alex. 2002. Caught in the act: Dialect change in St. John's English. Paper presented at NWAVE 31, Stanford, CA. 341 Davis, Alva L. 1948. A word atlas of the Great Lakes region. Doctoral Dissertation, The University of Michigan.

Davis, Alva L., and Raven I. McDavid, Jr. 1950. Northwestern Ohio: A transition area. Language,26:264-273.

DeDecker, Paul, and Sara Mackenzie. 2000. Slept through the ice: A further look at lax vowel lowering in Canadian English. Toronto Working Papers in Linguistics, 18:1-11.

Dinkin, Aaron. 2009. Dialect boundaries and phonological change in upstate New York. Doctoral dissertation. The University of Pennsylvania.

DiPaolo, Marianna, Malach Yaeger-Dror, and Alicia Beckford Wasnick. 2011. Analyzing vowels. In Marianna DiPaolo and Malcah Yaeger-Dror (Eds.), Sociophonetics: A student's guide. London: Routledge. pp. 87-106.

Dobson, Eric J. 1957. English pronunciation, 1500-1700. Volume II: Phonology. Oxford: Claredon Press.

Dodsworth, Robin. 2003. Ongoing sound change in Worthington: Influence from the North? Unpublished manuscript, The Ohio State University.

Dodsworth, Robin. 2005. Linguistic variation and the sociological imagination. Doctoral Dissertation, The Ohio State University.

Dodsworth, Robin, and Mary Kohn. 2012. Urban rejection of the vernacular: The SVS undone. Language Variation and Change, 24: 1-25.

Dodsworth, Robin, Bartek Plichta, and David Durian. 2006. An acoustic study of Columbus /l/ vocalization. Paper presented at NWAV 35, Columbus, OH.

Donegan, Patricia. 1978. On the Natural Phonology of vowels. Doctoral Dissertation, The Ohio State University. [Reprinted as The Ohio State Working Papers in Linguistics, 23.]

Durian, David. 2007. Getting [S]tronger every day? Urbanization and the sociogeographic diffusion of (str) in Columbus, OH. Penn Working Papers in Linguistics 13.2:65-79.

342 Durian, David. 2008a. The vocalization of /l/ in urban blue collar Columbus, OH African American Vernacular English: A quantitative sociophonetic analysis. The Ohio State University Working Papers in Linguistics, 58:30-51.

Durian, David. 2008b. A new perspective on vowel variation throughout the 20th Century in Columbus, OH. Paper presented at NWAV 37, Houston, TX.

Durian, David. 2009. Purely a chain shift? An exploration of the “Canadian Shift” in the US Midland. Paper presented at NWAV 38, Ottawa, CA.

Durian, David. 2010. Is it Northern Cities or is it split? Reassessing the historical tensing and raising of /ae/ in Columbus in real and apparent time. Paper presented at Changelings (OSU Socio-Historical discussion group), Columbus, OH.

Durian, David. 2011. Durian, David. The impact of the Canadian Shift on /aw/-fronting in Columbus. Annual Meeting of the American Dialect Society, Pittsburgh, PA.

Durian, David. 2012. The inception and development of the Canadian Shift in the US Midland. Paper presented at NWAV 41, Bloomington, IN.

Durian, David, and Matt J. Gordon. 2011.What are we talking about when we talk about vowel shifts? Panel presentation for "New Perspectives on Vowel Shifts" at NWAV 40, Washington, DC.

Durian, David, and Brian D. Joseph. 2011. Getting a handle of shifty changes: The role of phonetic analogy in vowel shifts. Panel presentation for "New Perspectives on Vowel Shifts" at NWAV 40, Washington, DC.

Durian, David, and Andrew Smith. 2005. The fronting of /ow/ and /aw/ in Columbus. Unpublished manuscript.

Durian, David, Robin Dodsworth, and Jennifer Schumacher. 2010. Convergence in urban working class Columbus, OH AAVE and EAE vowel systems? In Thomas, Erik R., and Malcah Yeager-Dror (Eds.), Vowel phonology and ethnicity. Publication of the American Dialect Society, 93.2. Durham: Duke University Press.

Durian, David, Jennifer Schumacher, and Melissa Reynard. 2007. Social class, convergence, and the vowel systems of Columbus AA(V)E and EAE. Paper presented at NWAV 36, Philadelphia, PA. 343 Eckert, Penelope. 2000. Language variation as social practice. Oxford, Blackwell.

Eckert, Penelope. 2004. Vowel shifts in Northern California and the Detroit Suburbs: Northern California Vowels. http://www.stanford.edu/~eckert/vowels.html

Ekwall, Eilert. 1946. American and British pronunciation. Upsala: American Institute of Upsala.

Emerson, Oliver F. 1891. The Ithaca dialect: A study of present English. Dialect Notes, 1: 85-173.

Fogle, Deena. 2008. Indianapolis, Indiana: A prototype of Midland convergence. University of Pennsylvania Working Papers in Linguistics, 14.1:134-148.

Foster, Paul N. 1997. “Which September:” Segregation, busing, and resegregation in the Columbus Public Schools, 1944-1996. BA Honors Thesis, Harvard University.

Ferguson, Charles. 1972. 'Short a' in Philadelphia English. In M. Estellie Smith (Ed.), Studies in honor of George L. Trager. The Hague: Mouton. pp. 259-274.

Frazer, Timothy C. 1978. South Midland pronunciation in the North Central States. American Speech, 53:40-48.

Frazer, Timothy C. (Ed.). 1993. “Heartland English”: Variation and transition in the American Midwest. Tuscaloosa: The University of Alabama Press.

Goldberg, Adele. 2006. Constructions at work: The nature of generalization in language. Oxford University Press.

Grama, James, and Robert Kennedy. 2009. Acoustic analysis of California vowels. Poster presented at the Annual Meeting of the Linguistic Society of America, San Francisco, CA.

Grandgent, Charles H. 1892. "Haf" and "haef." Dialect Notes 1, 269-275.

Gordon, Matthew. 2001.Small-Town Values, Big-City Vowels: A Study of the Northern Cities Shift in Michigan Raleigh: Duke University Press.

Gordon, Matthew, and Christopher Strelluf. 2012. The view from the Western edge; Sound Change in Kansas City. Paper presented at NWAV 41, Bloomington, IN. 344 Habick, Timothy. 1980. Sound change in Farmer City: A sociolinguistic study based on acoustic data. Doctoral dissertation, University of Illinois.

Hagiwara, Robert. 2005. Monophthongs and formant movement in North American English. Poster presented at the 150th meeting of the Acoustical Society of America, Minneapolis, MN.

Hagiwara, Robert. 2006. Vowel production in Winnipeg. Canadian Journal of Linguistics, 51:127-141.

Hall-Lew, Lauren. 2004. The Western Vowel Shift in Northern Arizona. Unpublished manuscript, Stanford University, Stanford, CA.

Hall-Lew, Lauren. 2009. Ethnicity and Phonetic Variation in a San Francisco Neighborhood. Unpublished PhD Dissertation. Stanford University, Stanford, CA.

Harden, Mike. 1995. Fixin' to make. Columbus Dispatch, October 22.

Harden, Mike. 2001. Kidspeak leaves paw paw smacking his forehead. Columbus Dispatch, July 9.

Harden, Mike. 2006a. Distinctive local lingo lives on in 'Clummus.' Columbus Dispatch, April 20.

Harden, Mike. 2006b. How to speak NASCAR: A few tips for y'all. Columbus Dispatch, October 15.

Haudricourt, A.G., and A.G. Juilland. 1949. Essai pour une historie structurale du phonetisme francais. Paris: C Klincksieck.

Haugen, Einer. 1970. The language history of Scandinavia: A profile of problems. In Hreinn Benediktsson. Ed. The Nordic Languages and Modern Linguistics. Reykjavik: Societas Scientarium Islandica.

Hayden, Dolores. 2003. Building suburbia: Green fields and urban growth, 1820-2000. New York: Vintage Books.

Hindle, Donald. 1979. The social and situational conditioning of phonetic variation. Doctoral Dissertation, University of Pennsylvania.

345 Hock, Hans Henrich. 1991. Principles of historical linguistics (2nd edition). Berlin: Mouton de Gruyter.

Hockett, Charles. 1965. Sound change. Language, 41.2:185-204.

Hunker, Henry. 1958. Industrial evolution of Columbus, OH. Columbus, OH: Bureau of Business Research/College of Commerce and Administration, The Ohio State University.

Hunker, Henry L. 1996. Columbus: The capital city. In Peacefull, Leonard (Ed.), A geography of Ohio. Kent, OH: Kent State University Press. p.159-164.

Hunker, Henry L. 2001. Columbus, Ohio: A personal geography. Columbus, OH: The Ohio State University Press.

Jaeger, T. Florian (2008). Categorical data analysis: Away from ANOVAs (transformation and not) and towards logit mixed models. Journal of Memory and Language 59:434-446.

Jacewicz, Ewa, Robert Allen Fox, and Joseph Salmons. 2011a. Cross generational vowel change in American English. Language Variation and Change, 23:45-86.

Jacewicz, Ewa, Robert Allen Fox, and Joseph Salmons. 2011b. Vowel change across three age groups of speakers in three regional varieties of American English. Journal of Phonetics, 39:683-693.

Jacobs, Gregory S. 1994. Getting around ‘Brown’: Desegregation, development, and the Columbus Public Schools, 1954-1994. MA Thesis, The Ohio State University.

Janda, Richard D, and Brian D. Joseph. 2003. Reconsidering the cannons of sound change: Towards a "big bang" theory. In Barry Blake and Kate Burridge, eds., Historical Linguistics 2001. Selected Papers from the 15 International Conference on Historical Linguistics, Melbourne, 13-17 August 2001. Amsterdam: John Benjamins Publishing, pp. 205-219.

Jespersen, Otto. 1908. Modern English Grammar on historical principles. Volume I: Sounds and Spellings. London: Allen & Unwin.

Johnson, Daniel Ezra. 1998. The tensing and laxing of short 'a' in New Haven, Connecticut. B.A. thesis, Yale University. 346 Johnson, Daniel Ezra. 2009. Getting off the GoldVarb Standard: Introducing Rbrul for mixed effects variable rule analysis. Language and Linguistics Compass, 3:359- 383.

Johnson, Daniel Ezra. 2010. Stability and change along a dialect boundary. Publication of American Dialect Society, 95. Raleigh: Duke University Press.

Johnstone, Barbara, Neeta Bhasin, and Denise Wittkofski. 2002. "Dahntahn" Pittsburgh: Monophthongal /aw/ and representations of localness in Southwestern Pennsylvania. American Speech 77.2:148-166.

Joseph, Brian D. 2008. Historical Linguistics in 2008: The state of the art. In P. van Sterkenberg (Ed.), Unity and diversity in languages. Amsterdam: John Benjamins Publishing, pp. 175-187.

Joseph, Brian. D. 2011. Historical linguistics and sociolinguistics-Strange bedfellows or natural friends? In N. Langer, S Davies, and W. Vadenbussche. Eds. Language and History, Linguistics and Histriography. New York: Peter Lang. pp. 67-88.

Keiser, Steve Hartman, Frans Hinskens, Bettina Migge, and Elizabeth Strand. 1997. The Northern Cities Shift in the Heartland? A study of radio speech in Columbus, OH. The Ohio State Working Papers in Linguistics, 50:41-68.

Kennedy, Robert, and James Grama. 2012. Chain shifting and centralization in California vowels: An Acoustic analysis. American Speech, 87.1:39-56.

Kenyon, John Samuel. 1924. American pronunciation. Menasha, WI: George Banta Publishing.

King, Paul E. 1969. The spatial aspects of social area change: Columbus, Ohio 1950- 1960. Masters Thesis, The Ohio State University, Columbus, OH.

King, Robert D. 1969. Historical linguistics and generative grammar. Englewood Cliffs, NJ: Prentice Hall, Inc.

Kiparsky, Paul. 1968. Linguistic universals and linguistic change. In Emmon Bach and Robert T. Harms, eds., Universals in linguistic theory. New York: Holt, Rinehart, and Winston, Inc pp. 171-210.

347 Krapp, George Philip. 1925. The English Language in America. Volumes I and II. New York: The Modern Language Association of America.

Kurth, Hans, 1928a. American Pronunciation. S[ociety for] P[ure] E[nglish] Tract, XXX: 279-297.

Kurath, Hans. 1928b. The origin of the dialectal differences in Spoken American English. Modern Philology 25.4:385-395.

Kurath, Hans. 1930. A specimen of Ohio speech. Language Monograph No. 7: Curme Volume of Linguistic Studies, 92-101.

Kurath, Hans. 1949. A Word Geography of the United States. Ann Arbor, MI: University of Michigan Press.

Kurath, Hans. 1964. A phonology and prosody of Modern English. Ann Arbor: University of Michigan Press.

Kurath, Hans, and Raven I. McDavid, Jr. 1961. Pronunciation of English in the Atlantic States. Ann Arbor, MI: University of Michigan Press.

Kurath, Hans, Marcus Hansen, Julia Bloch, and Bernard Bloch (Eds.). 1939. Handbook of the Linguistic Geography of New England. Providence, RI: Brown University.

Labov, William. [1966]/2006. The social stratification of English in New York City (2nd edition). Cambridge: Cambridge University Press.

Labov, William. 1972. Sociolinguistic patterns. Philadelphia: University of Pennsylvania Press.

Labov, William. 1976. The relative influence of family and peers on the learning of language. In Simone, Raffaele, and Guilianella Ruggiero (Eds.), Aspetti sociolinguistici dell’ Italia contemporanea. Rome: Bulzoni. pp. 11-56.

Labov, William. 1981. Resolving the Neogrammarian controversy. Language, 57,.2: 267- 308.

Labov, William. 1989. An exact description of the speech community: Short a in Philadelphia. In Fasold, Ralph W., and Deborah Schiffrin (eds.), Language change and variation. Amsterdam: John Benjamins. pp. 1-57. 348 Labov, William. 1991. The three dialects of English. In Penelope Eckert (Ed.), New ways of analyzing sound change. Orlando, FL: Academic Press, 1-44.

Labov, William. 1994. Principles of linguistic change: Internal factors. Oxford: Blackwell.

Labov, William. 2001. Principles of linguistic change: Social factors. Oxford: Blackwell.

Labov, William. 2006. A sociolinguistic perspective on sociophonetic research. Journal of Phonetics, 34:500-514.

Labov, William. 2007. Transmission and diffusion. Language, 83.2:344-387.

Labov, William. 2010. Principles of linguistic change: Cognitive and cultural factors. Oxford: Wiley-Blackwell.

Labov, William, Malcah Yeager-Dror, and Richard Steiner. 1972. A quantitative survey of sound change in progress. Philadelphia: US Regional Survey.

Labov, William, Sharon Ash, and Charles Boberg. 2006. The Atlas of North American English: Phonetics, phonology, and sound change. Berlin: Moutin de Gruyter.

Lance, Donald M. 1994. Variation in American English. In American pronunciation (12th edition). Expanded by John S. Kenyon, edited by Donald M. Lance and Stewart A. Kingsbury. Ann Arbor: George Wahr. pp. 345-373.

Lass, Roger. 1976. English phonology and phonological theory. Cambridge: Cambridge University Press.

Lentz, Ed. 2003. Columbus: The story of a city. Charleston, SC: Arcadia Press.

Lobanov, B. M. 1971. Classification of Russian vowels spoken by different speakers. The Journal of the Acoustical Society of America, 49.2:606-608.

Luick, Karl. 1903. Studien zur Englischen Lautgeschichte. Vienna and Leipzig: W. Braunmullier.

349 Luthin, Herbert. 1987. The story of California (ow): The coming of age of English in California. In Keith Denning, Sharon Inkelas, Faye McNair-Knox, and John R. Rickford, Eds., Variation and Language: NWAV XV at Stanford: Proceedings of the Fifteenth Annual Conference on New Ways of Analyzing Variation. Stanford: Department of Linguistics, Stanford University. pp. 312-324.

Marckwardt, Albert. 1940. Middle English ô in American English of the Great Lakes area. In McCartney, Euguene, and Mischa Titiev (Eds.), Papers of the Michigan Academy of Science, Arts, and Letters, Volume XXVI. Ann Arbor: University of Michigan Press. p. 561-571.

Marckwardt, Albert. 1942. Middle English wa in the speech of the Great Lakes region. American Speech, 17.4:226-234.

Marckwardt, Albert. 1957. Principle and subsidiary dialect areas in the North Central States. Publication of the American Dialect Society 27. Tuscaloosa, AL: University of Alabama Press. pp. 3-15.

Martinet, Andre. 1952. Function, structure, and sound change. Word, 8.1: 1-32.

Martinet, Andre. 1955. Économie des changements phonétiques. France: Berne.

McCarthy, Corrine. 2004. Language change in Pittsburgh: The decline of /aw/- monophthongization and the Canadian Shift. Poster presented at NWAVE 33, Ann Arbor, MI.

McCarthy, Corrine. 2011. The Northern Cities Shift in Chicago. Journal of English Linguistics, 39(2).

McDavid, Raven I, Jr., and Richard C. Payne (Eds.). 1976-1978. Linguistic Atlas of the North Central States: Basic materials. Manuscripts of Cultural Anthropology, series 38, nos. 200-208. Chicago, IL: Joseph Regenstein Library, University of Chicago.

McKenzie, R.D. 1921. The neighborhood: A study of local life in the city of Columbus, part II. The American Journal of Sociology, 27.3:344-363.

Nycz, Jennifer. 2005. The dynamics of near-merger: Accounting for gradience in phonology. Proceedings of Con-SOLE XIII. http://www.hum.leiden.edu/lucl/research/sole/proceedings/console13.html 350 Nobbelin, Kent G. 1980. The low-back vowels of the North Central States. Doctoral Dissertation, Illinois Institute of Technology.

Noble, Allen G., and Leonard Peacefull. 1996. Cities in Ohio. In Peacefull, Leonard (Ed.), A geography of Ohio. Kent, OH: Kent State University Press. p. 146-157.

Osthoff, Hermann, and Karl Brugmann. 1878. Morphologische Untersuchgen auf dem Gebiete det indogermanischen Sprachen, I. Leipzig.

Palmer, H. E., J. Victor Martin, and F. G. Blandford. 1927. A dictionary of English pronunciation with American variants. New York: Appleton.

Paul, Hermann. 1889/1920. Principles of language history. (English Trans. of 2nd Ed.). New York: Macmillian.

Payne, Arvilla C. 1980. Factors controlling the acquisition of the Philadelphia dialect by out-of-state children. In Labov, William (Ed.), Locating language in time and space. New York: Academic Press. pp. 143-178.

Pederson, Lee A. 1977. Studies of American pronunciation since 1945. American Speech, 52:262-327.

Pierrehumbert, Janet. 2002. Word-specific phonetics. In C. Gussenhoven, & N. Warner, Eds., Laboratory phonology VII. Berlin: Mouton/de Gruyter. pp. 101–140

Pierrehumbert, Janet. 2003 Phonetic diversity, statistical learning, and acquisition of phonology. Language and Speech, 46(2-3):115-154.

Preston, Dennis. 1996. Where the worst English is spoken. In Schneider, Edgar (Ed.), Focus on the USA. Amsterdam: Benjamins. pp.297-360.

Prichard, Hilary, and Meredith Tammiga. 2012. The impact of education on Philadelphia vowels. University of Pennsylvania Working Papers in Linguistics, 18.2: Article 11. http://repository.upenn.edu/pwpl/vol18/iss2/11

Raphael, Marc Lee. 1979. Jews and Judiasm in a midwestern community: Columbus, Ohio, 1840-1975. Columbus, OH: Ohio Historical Society.

Roeder, Rebecca, and Lidia-Gabriela Jarmasz. 2008. The lax vowel subsystem in Canadian English revisited. Toronto Working Papers in Linguistics, 31. 351 Rusk, David. 2003. Cities without suburbs: A census 2000 update (Third edition). Washington, DC: Woodrow Wilson Research Center.

Schuchardt, Hugo. 1885. Uber die Lautesetze: Gegen die Junggrammatiker. Berlin: Oppenheim. (Reprinted and translated as On Sound Laws: Against the Neogrammarians in Vennemann, Theo, and Terence H. Wilbur (Eds.), Schuchardt, the Neogrammarians, and the Transformational Theory of phonological change: Four essays. Frankfort, Athenaum.)

Shuy, Roger W. 1962. The North-Midland dialect boundary in Illinois. Publication of the American Dialect Society, 38. University, AL: University of Alabama Press.

Shuy, Roger, Walter A. Wolfram, and William Riley. 1967. Linguistic correlates of social stratification in Detroit speech. USOE Final Report, No. 6-1347.

Skousen, Royal. 1989. Analogical modeling of language. Kluwer Academic Publishers.

Speer, Michael S. 1972. Urbanization and reform: Columbus, Ohio, 1870-1900. Doctoral dissertation, The Ohio State University, Columbus, OH.

Stevens, David T. 1996. Population patterns. In Peacefull, Leonard (Ed.), A geography of Ohio. Kent, OH: Kent State University Press. p. 127-145.

Stockwell, Robert P. 1964. On the utility of the overall pattern in historical English phonology. Proceedings on the Ninth International Congress of Linguists. The Hague: Mouton & Co. pp. 663-671.

Stockwell, Robert P. 1972. Problems in the interpretation of the Great English Vowel Shift. In M Estellie Smith (Ed.), Studies in honor of George L. Trager. The Hague: Mouton. pp. 344-362.

Stockwell, Robert P. 1978. Perseverance in the English Vowel Shift. In Werner Winter (Ed.), Recent developments in historical phonology. The Hague: Mouton. pp. 337- 348.

Stockwell, Robert P. 2002. How much shifting occurred in the historical English Vowel Shift? In Donka Minkova and Robert P. Stockwell (Eds.), Studies in the history of the English Language: A millennial perspective. Berlin: Mouton de Gruyter. p. 267-281.

352 Stockwell, Robert P, and Donka Minkova. 1997. On drifts and shifts. Studia Angelica Posnaniensia XXXI:283-303.

Studer, Jacob H. 1873. Columbus, Ohio: Its history, resources, and progress. Salem, MA: Higginson Book Company.

Thomas, Erik R. [1989]/1993. Vowel changes in Columbus, OH. Journal of English Linguistics 22:205-215.

Thomas, Erik R. 1996. A comparison of variation patterns of variables among sixth graders in an Ohio community. In E. W. Schneider (Ed.), Focus on the USA: Varieties of English around the World (General Series, Vol. 16). Amsterdam: John Benjamins Publishing. pp. 309-332.

Thomas, Erik R. 2001. An acoustic analysis of vowel variation in New World English. Publication of the American Dialect Society 85. Durham, NC: Duke University Press.

Thomas, Erik R.2003. Secrets revealed by Southern Vowel Shifting. American Speech, 78:150-170.

Thomas, Erik R. 2004. The perpetuation of historical dialect patterns in Ohio. Manuscript: North Carolina State University.

Thomas, Erik R. 2006. Evidence from Ohio on the evolution of /æ/. In Language variation and change in the American Midland: A new look at “Heartland” English, ed. Thomas E. Murray and Beth Lee Simon. Amsterdam/Philadelphia: John Benjamins. 69-89.

Thomas, Erik R. 2011. Sociophonetics: An introduction. New York: Palgrave Macmillan.

Tomasello, Michael. 2003. Constructing a language: A usage-based theory of language acquisition. Harvard University Press.

Trager, George L. 1930. The pronunciation of "short a" in American Standard English. American Speech, 5:396-400.

Trager, George L. 1934. What conditions limit variants of a phoneme? American Speech, 9:313-315.

353 Trager, George L. 1940. One phonemic entity becomes two: The case of "short a." American Speech, 15:255-258.

Trager, George L., and Henry Lee Smith. 1957. An outline of English structure (4th printing). Washington: American Council of Learned Societies.

Trudgill, Peter. 1974. The social differentiation of English in Norwich. Cambridge: CUP.

Trudgill, Peter. 1986. Dialects in contact. Oxford: Blackwell.

Tuttle, George. 1902. Phonetic notation. In Edward Scripture, Ed., Studies from the Yale Psychology Laboratory X: 96-117.

U.S. Census Bureau. n.d. United States Census Bureau Website. Available: http://www.census.gov

Vennemann, Theo. 1972. Phonetic analogy and conceptual analogy. In Vennemann, Theo, and Terence H. Wilbur (Eds.), Schuchardt, the Neogrammarians, and the Transformational Theory of phonological change: Four essays. Frankfort, Athenaum. p. 181-204.

Watt, Dominic, and Anne Fabricius. 2002. Evaluation of a technique for improving the mapping of multiple speakers’ vowel spaces in the F₁–F₂ plane. Leeds Working Papers in Linguistics and Phonetics 9:159–173.

Weinreich, Uriel, William Labov, and Marvin Herzog. 1968. Empirical foundations of a theory of language change. In Winfred Lehmann and Yakov Malkiel, Eds., Directions for historical linguistics. Austin: University of Texas Press. pp. 97- 195.

Wells, John. 1982. Accents of English. Cambridge: Cambridge University Press.

Wilhelm, Hurbert G.H. 1976. Cultural features. In Anderson, Dennis M., and Charles C. King, (Eds.), Environmental analysis of Central Ohio, Volume III. Columbus, OH: Ohio Biological Survey. O.S.U. Research Foundation Project No. RF-3492 Final Report.

Wilhelm, Hubert G.H. 1982. The origin and distribution of settlement groups: Ohio, 1850. Athens, OH: Dept of Geography, Ohio University.

354 Wilhelm, Hurbert G.H. 1996. Ohio's settlement landscape. In Leonard Peacefull, ed., A geography of Ohio. Kent, OH: The Kent State University Press.

Wolfram, Walter A. 1969. A sociolinguistic study of Detroit Negro Speech. Arlington, VA: Center for Applied Linguistics.

Wright, Alfred J. 1958. Economic geography of Ohio (2nd edition). Columbus, OH: Department of Geography, The Ohio State University.

355 Appendix A: Central Ohio Dialect Survey Interview Protocol

Background (10 mins)

1) Can you please count slowly to ten, so I can make sure my recording levels are set properly?

2) What is your name? Your age?

3) How long have you lived in Columbus, and what sort(s) of work do you do for a living?

4) How long have you worked that job, and what other types of job(s) have you worked before it?

5) Where else have you lived besides Columbus, and how long did you live in each of those places?

6) Have you moved around Columbus at all? (If yes) what other areas have you lived in?

7) Did your parents grow up in Columbus or another part of Ohio, or did they grow up someplace else? Where? (Both mother and father)

8) What did you parents do for a living?

9) What about your grandparents—where did they grow up and what sorts of work did they do for a living?

10) What about your brothers and sisters—what do they do for a living?

356 The Community (15-20 mins)

1) You mentioned that you’ve lived in the Columbus community for the last x years. What was Columbus like when you were growing up? What are 2 or 3 things that make Columbus different today?

2) How do you feel about the “urban expansion” of Columbus over the last twenty-thirty years (for example, the expansion of Columbus into the surrounding areas of Worthington, Dublin, Delaware County, etc.)? a) Do you think it's improved Columbus, or do you feel that it’s been a problem?

b) What about all of the mall and road construction that’s been happening since the late ‘90s. Do you feel that these changes have improved Columbus?

c) In terms of the road construction, do you feel that it's made it easier or harder to get around Columbus? How do you feel about proposals to build a light rail system—do you think it would improve traffic flow/congestion problems, or do you feel that adding this system is unnecessary?

3) Did your children (or you) attend the public schools in Columbus?

a) What do you think of the Columbus school system? b) What changes do you think should be made? c) Is there anything you think works really well about the system?

4) What do you think of the current economic situation in Columbus?

a) Would you say it's gotten better or worse than x years ago? b) What sorts of things do you think have gotten (better/worse) about it?

5) What is your opinion about the local political scene in Columbus?

a) Are you comfortable with the current administration? Do you think they do a good job of representing the people? b) What about the scene more generally in Ohio? What sorts of things have you been happy with? What would you like to see changed?

6) Do you consider yourself to be someone who is “involved” in the Columbus community? What sorts of community groups do you belong to? What inspired you to get involved with these groups—family, friends, co-workers, a church group? 357 7) Do your family and friends share the same sorts of interests in these groups that you do? In what ways?

8—For suburban speakers) Which area do you feel you more strongly identify with— Columbus or x (their suburb)? What do you feel you most strongly identify with in terms of that community?

Linguistic Attitudes (10 - 15 mins)

As this is a dialect study, I now want to ask you some questions about “Columbus speech,” and about accents or differences in the way people talk that you’ve noticed while traveling throughout Central Ohio as well as generally throughout the State.

1) Do you think people sound different or talk differently in different parts of Ohio? a) For instance, do you think people from Cleveland sound different than people in Columbus? In what ways? (Accent, different lexical items, different pronunciations) b) What about Cinncinati or Dayton? Do you think people sound different there, too? In what ways? (Accent, different lexical items, different pronunciations) c) Thinking back to when you were growing up, did folks in those areas sound different back then in terms of how they talked? How so?

2) What about here in Columbus, do you think people of different backgrounds have different ways of talking? For instance, does a janitor sound different than a business person? How so?

3) When you were younger, did you notice anything about Columbus speech that sounded different to you than how it sounds today? In what ways? (Accent, different lexical items, different pronunciations)

4) When you’ve traveled outside of Columbus, do you think people picked you out as being from here based on your speech?

5) What do you think out-of-towners think of Columbus speech? Why?

6) What do you think of Northern speech (Chicago, New York City, Buffalo) as compared to Columbus speech?

358 7) What do you think of Southern speech (Alabama, Texas, Georgia) as compared to Columbus speech?

8) What do you think of your own speech? Do you ever use slang? If you do, who do you use it with?

9) Do you talk differently to your friends and family than you do at work? How so?

10) Have you ever tried to change your speech? What particular things about it?

11) Have you ever taken any courses on language or public speaking? What did the teacher mention in connection with things like pronunciation or diction?

Word Game (10 - 15 mins)

Now I want to record some of your speech, so that a record of the actual speech used by Columbus residents such as yourself can be made. In order to make the recording a little more interesting, I’m going to ask you to make sentences using the words on the cards that I’m about to shuffle here (shuffle the cards while talking). I’ll deal out 4 cards and then I’ll ask you to create either one sentence using those words, if you can. Or you can create a series of sentences, if that that ends up being easier. You can use any other words you want to make up your sentences, as long as you somehow end up using all the words on the cards.

Words appearing in the deck (48):

Anymore Boil Book Boot Booth Booze Bought Bull Caught Coal Cold Cook Cot Duel Fail Father Fault Fell Foil Food Foot Full Hill Mall Meal Mild Milk Nail News Oil Peel Pile Pole Pool Pot Pull Shoes Stall Steal Still Suit Swell Tale Taught Tell Tile Tool These Days

359 Lexical and Syntactic Items (10 mins)

Now I want to ask you about the terms you use for some descriptions I will read off. Please give me the term you use most for the item, and then any other terms you might use. Also, if you can think of any words that folks in Columbus might have used that were different for these terms when you were younger versus today, please also tell me about those words.

1) What do you call a bug that flies around in the summer and lights up at night?

2) What do you call the little gray insect that rolls up into a ball when you touch it?

3) What do you call the large thin spider that skitters along the top of screen doors?

4) What term do you use to refer to something that is directly across from you diagonally when you are standing where both streets meet at an intersection?

5) What do you call the area of grass between the sidewalk and the road?

6) What do you call the action of someone moving in front of you in a line?

7) What do you call the large sugar-covered pastry that you might get at a county fair?

8) What do you call the curdled cheese you might buy in a container at Kroger or Giant Eagle?

9) Please describe what you think of when I say the word “mango”?

USE CUE CARDS—Can you please read the following sentences and fill in the blank with a word or words to complete the sentence?

10) [The table is dirty and requires cleaning.] The table needs ______.

11) [The roof is in need of repair.] The roof needs ______.

12) [The dog is restless and would like to go outside.] The dog wants ______.

13) [The baby is crying and wants to cuddle.] The baby likes ______.

360 14) [You got tired while driving back from Illinois.] Dayton was ______I could drive.

15) [You like to go home early on Fridays, but you used to stay at work later on Fridays.]

______I go home from work early on Fridays.

16) [You used to do your grocery shopping on Wednesday, but now you do it on Tuesdays.]

I only go grocery shopping on Tuesdays ______.

361 Appendix B: Normalized Individual Speaker Vowel Plots

On the following pages are plots containing normalized vowel formant measurements plotted in z-score units for all 62 speakers for whom we have instrumentally analyzed data. Data were normalized using the normalization technique described in Chapter 3. They were then plotted using the program R.

The plots in this section are divided into 4 groups: White Collar Men, Blue Collar

Men, White Collar Women, Blue Collar Men. The oldest speakers in an given class group are presented first, followed by increasingly younger speakers. Generally speaking,

G1 plots are thus presented first, followed by G2 plots, then G3 plots, and finally G4 plots. Men are shown first since the oldest speaker in the whole study happens to be a man. It also happens to be the case that the youngest speaker in the whole study is a woman, so placing women second allows her to also be plotted last.

In each plot, the mean value across 7-10 tokens of each vowel class is presented.

Monophthongs are represented by a single point, whereas diphthongs are represented by a nucleus point and a glide arrow. The direction of the glide for each diphthong is indicated by the arrow. In terms of vowel duration, monopthongs are plotted using 50% of the vowel's duration to represent the nucleus, while for diphthongs, nuclei are plotted using the 25% point of duration the vowel to represent the nucleus. Glide for diphthongs are represented using the 75% point of the duration of the vowel. Vowels are plotted in

Lobanov z-score units

362

Figure B.1: Vowel System of White Collar Male LANCSOH28C, Born c. 1896

363

Figure B.2: Vowel System of White Collar Male Tobias, Born c. 1898

364

Figure B.3: Vowel System of White Collar Male William, Born c. 1905

365

Figure B.4: Vowel System of White Collar Male Don, Born c. 1924

366

Figure B.5: Vowel System of White Collar Male Tom C, Born c. 1928

367

Figure B.6: Vowel System of White Collar Male Tom W, Born c. 1937

368

Figure B.7: Vowel System of White Collar Male Ben, Born c. 1952

369

Figure B.8: Vowel System of White Collar Male David, Born c. 1954

370

Figure B.9: Vowel System of White Collar Male Steve, Born c. 1955

371

Figure B.10: Vowel System of White Collar Male Peter, Born c. 1956

372

Figure B.11: Vowel System of White Collar Male Ed, Born c. 1967

373

Figure B.12: Vowel System of White Collar Male Abe, Born c. 1985

374

Figure B.13: Vowel System of White Collar Male Bryan, Born c. 1987

375

Figure B.14: Vowel System of White Collar Male Matt M, Born c. 1987

376

Figure B.15: Vowel System of White Collar Male Meesha, Born c. 1987

377

Figure B.16: Vowel System of White Collar Male Justin L, Born c. 1990

378

Figure B.17: Vowel System of Blue Collar Male Arthur, Born c. 1898

379

Figure B.18: Vowel System of Blue Collar Male ColumbusMK, Born c. 1908

380

Figure B.19: Vowel System of Blue Collar Male David, Born c. 1913

381

Figure B.20: Vowel System of Blue Collar Male Joe, Born c. 1926

382

Figure B.21: Vowel System of Blue Collar Male Tony, Born c. 1930

383

Figure B.22: Vowel System of Blue Collar Male Joe, Born c. 1932

384

Figure B.23: Vowel System of Blue Collar Male Bill, Born c. 1949

385

Figure B.24: Vowel System of Blue Collar Male Trd, Born c. 1951

386

Figure B.25: Vowel System of Blue Collar Male Terry, Born c. 1959

387

Figure B.26: Vowel System of Blue Collar Male Michael, Born c. 1962

388

Figure B.27: Vowel System of Blue Collar Male Jim H, Born c. 1967

389

Figure B.28: Vowel System of Blue Collar Male Jason, Born c. 1980

390

Figure B.29: Vowel System of Blue Collar Male Justin R, Born c. 1982

391

Figure B.30: Vowel System of Blue Collar Male Voncent, Born c. 1984

392

Figure B.31: Vowel System of Blue Collar Male Travis, Born c. 1985

393

Figure B.32: Vowel System of Blue Collar Male Louie, Born c. 1986

394

Figure B.33: Vowel System of White Collar Female Thea, Born c. 1903

395

Figure B.34: Vowel System of White Collar Female DAREOH98, Born c. 1906

396

Figure B.35: Vowel System of White Collar Female DAREOH98, Born c. 1924

397

Figure B.36: Vowel System of White Collar Female Nancy C, Born c. 1927

398

Figure B.37: Vowel System of White Collar Female Barbara, Born c. 1938

399

Figure B.38: Vowel System of White Collar Female Nancy, Born c. 1947

400

Figure B.39: Vowel System of White Collar Female Toni, Born c. 1948

401

Figure B.40: Vowel System of White Collar Female Sheri, Born c. 1950

402

Figure B.41: Vowel System of White Collar Female Marsha, Born c. 1950

403

Figure B.42: Vowel System of White Collar Female Margaret, Born c. 1962

404

Figure B.43: Vowel System of White Collar Female Janna, Born c. 1978

405

Figure B.44: Vowel System of White Collar Female Mollie, Born c. 1978

406

Figure B.45: Vowel System of White Collar Female Mandy, Born c. 1981

407

Figure B.46: Vowel System of White Collar Female Jayne, Born c. 1984

408

Figure B.47: Vowel System of White Collar Female Liz, Born c. 1988

409

Figure B.48: Vowel System of Blue Collar Female Louise, Born c. 1908

410

Figure B.49: Vowel System of Blue Collar Female Ann, Born c. 1912

411

Figure B.50: Vowel System of Blue Collar Female Marie, Born c. 1925

412

Figure B.51: Vowel System of Blue Collar Female Mary, Born c. 1928

413

Figure B.52: Vowel System of Blue Collar Female Helen, Born c. 1930

414

Figure B.53: Vowel System of Blue Collar Female Linda G, Born c. 1945

415

Figure B.54: Vowel System of Blue Collar Female Susan, Born c. 1950

416

Figure B.55: Vowel System of Blue Collar Female Molly, Born c. 1952

417

Figure B.56: Vowel System of Blue Collar Female Paula, Born c. 1930

418

Figure B.57: Vowel System of Blue Collar Female Michelle, Born c. 1961

419

Figure B.58: Vowel System of Blue Collar Female Rachael, Born c. 1976

420

Figure B.59: Vowel System of Blue Collar Female Angela, Born c. 1977

421

Figure B.60: Vowel System of Blue Collar Female Julia, Born c. 1982

422

Figure B.61: Vowel System of Blue Collar Female Kelly, Born c. 1988

423

Figure B.62: Vowel System of Blue Collar Female Kaitlyn, Born c. 1991

424