THE IMPACT OF INTERN AL SOCIAL CHANGE ON LOCAL PHONOLOGY By Monica Nesbitt A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of Linguistics Doctor of Philosophy 2019 ABSTRACT THE IMPACT OF INTERNAL SOCIAL CHANGE ON LOCAL PHONOLOGY By Monica Nesbitt This dissertation addresses a fundamental question about phonolog y , i.e. how do we account for the appearance of a phonological rule in a community that had previously never posited one? For example, how could we account for someone who posits a phonological difference between /r/ and /l/, i.e. the two are distinct phonemes in their grammar, when no one in their speech community distinguishe s between the two sounds phonologically, i.e. [r] and [l] are simply variants of the same phoneme. The general a rgument in the literature is that phonological change comes about via re - analysis of some phonetic exaggeration (e.g. Hyman 1975; Ohala 1990; Pierrehumbert 2001; Bermúdez - Otero 2007) , but because phonological change is rarely observed, there is a paucity of empirical evidence using production data in a community while a change is underway to support this theory . Recent acoustic analyses of large databases of naturally occurring, ongo ing phonological changes show that the role of phonetic variation is minor and in some cases non - existent when a phonological rule is innovated (Fruehwald 2013; 2016; Berkson, Davis & Strickler 2017) . Therefore, the available empirical evidence for phonological change stands in contrast to what has been proposed in the theoretical literature. This dissertation interven es in the debate through the analysis of a database of naturally occurring speech in Lansing, Michigan where allophonic change is currently underway. In Lansing, /æ/ is being re - organized such that at the turn of the 20 th century, speakers in the communit y did not distinguish between /æ/ in any phonological environments, speakers born in the 1990s, however, do distinguish between pre - nasal and pre - oral /æ/. I utilize a combination of analyses in this dissertation t o account for the initiation and spread o f this change throughout the community . First, I conducted an acoustic analysis of /æ/ in F1/F2 space in a corpus of naturally occurring speech by 36 Lansing natives. In particular, I tracked changes in vowel height (F1 at nucleus), backness (F2 at nucleu s), diphthongal quality (difference in F1 and F2 at nucleus and offset), and relative distributions of pre - nasal and pre - oral /æ/ token clouds for each speaker (Pillai - Bartlett statistic - see Hay, Warren & Drager 2006; Hall - Lew 2010) . I supplemented the acoustic analysis with a sub - phonemic judgement task administered to 107 Lan sing natives via an online survey. During the task, respondents were asked to identify whether the vowels in words like pat and pan are the same. This dissertation finds that phonological change was gradual in Lansing. The measure of speaker - level distributions with an impressionistic investigation of divergent trajectories, and the results of the sub - phonemic judgement task suggest that there is i ndeed an intermediate period between no distinction and phonological di fference in Lansing whereby the difference between pre - nasal and pre - oral /æ/ w as only phonetically implemented . An analysis of the effects of gender and social class on these measures finds that phonetic variation is socially conditioned in Lansing , such that white - collar women are leading the change away from the community norm. I observe that phonetic exaggeration w as promoted by social re - organization in the community , which eventua lly lead to the development and spread of an allophonic rule . In line with the prediction of Baker, Archangeli, and Mielke (2011) , I find that the chance alignment of social and phonetic variability in 20 th century Lansing accounts for the initiation and spread of this phonological change. iv ACKNOWLEDGEMENTS It is amazing what one can accomplish with the support of their village. I cannot begin to describe how Suzanne Evans Wagner, my advisor, has changed my life. She has supported me in so many ways th ese last five years; from securing me funding, to advocating for me when I became a student - parent, to holding my hand while I wrote this dissertation. The list goes on. Never wavering in her support and always pushing me to be better, she is an amazing scholar and the kind of mentor I hope to be one day. The support of my other committee members was far from inconsequential. Anne Violin - Wignet has diligently given me (handwritten!) feedback on this dissertation and on previous work. Yen - Hwei Lin has rem ained enthusiastic in her support of me and my research and has been a constant source of encouragement. I went out on a limb and asked Aaron Dinkin to be on my committee and thank the stars that I did. I am immensely grateful for all of the time that Aar on has put into guiding me through this process. His gentle prodding and extensive feedback have transformed this dissertation immensely. Karthik Durvasula has been an instrumental teacher, friend, and mentor since the very beginning of my PhD program. Always offering sage advice and constantly challenging me to be a better scholar. I am so happy to have learned from and collaborated with him and I look forward to many more research projects together. I have the pleasure of calling other scholars my men tors. Michal Temkin Martinez saw potential in me back when I was an undergraduate student and I am so grateful that she pushed me to attend graduate school and continues to be an outstanding mentor and scholar. It is amazing how the passion in someone els e can spark your own. I knew I wanted to be a linguist v Jodi Tommerdahl and Cindy Kilpatrick my unofficial mentors Betsy Sneller, Sabriya Fisher, and Sharese King they have taken me I am truly indebted to the many Lansing natives who contributed the ir voices and judgements to this project, and to our undergraduate transcribers, and our interviewers, especially Jennifer Pizzo. Her enthusiasm and social network are off the charts. I could not have survived these last five years without my dear friend s and colleagues at Michigan State University. I started my program alongside Cara, Mohammed, Alex, Matt, and ChenChen, and have enjoyed their company and scholarship. I am especially grateful to Cara for her constant positivity and to Alex for our many discussions about Lansing speech, linguistics, and parenting. To the undergrads I have had the pleasure of mentoring, especially Danielle and Jared. May you continue to be fascinated about language variation and change, as am I. I would not have gotten through this last year without my friend, Kaylin. She helped keep me sane and I am a better person for knowing her. Justin, my friend, thank you for the constant intellectual and emotional commiseration. I continue to aspire to be half the scholar you a re. Special thanks to my family Zach, Phylicia, Will, Steve, mom, Sam, dad, River and Riley whose ardent belief in my ability to do this has kept me moving forward. My husband, Zach, followed me across the country for a decade in support of my dream. It h as meant the world to me. I am most grateful for Sophia. She has brought so much joy and purpose to my life. Thank you, again, my village. vi TABLE OF CONTENTS LIST OF TABLES ................................ ................................ ................................ ......................... ix LIST OF FIGURES ................................ ................................ ................................ ....................... xi CHAPTER 1 PHONOLOGICAL CHANGE ................................ ................................ ............... 1 1.1 Introduction ................................ ................................ ................................ .................. 1 1.2 Phonetics vs Phonology ................................ ................................ ................................ 4 1.3 Gradual vs Abrupt Phonological Change ................................ ................................ ..... 7 11 1 1.4 Methods for Distinguishing Phonological from Phonetic Distinction ....................... 16 1.4.1 Divergent Trajectories ................................ ................................ ............................ 17 1.4.2 Sub - Phonemic Judgement Task ................................ ................................ .............. 19 1.4.3 Diagnostics of Phonological Change in this dissertation ................................ ........ 20 1.5 Social/External Aspects of Phonological Change ................................ ...................... 21 1.6 Goals of this Dissertation ................................ ................................ ........................... 26 1.6.1 Main Goals ................................ ................................ ................................ .............. 26 1.6.2 Peripheral Goal ................................ ................................ ................................ ....... 26 1.7 Findings from this Dissertation ................................ ................................ .................. 27 1.8 Organization of this Dissertation ................................ ................................ ................ 27 CHAPTER 2 ALLOPHONIC CHANGE IN LANSING, MICHIGAN ................................ ..... 29 2.1 Introduction ................................ ................................ ................................ ................ 29 2.2 Allophonic Variation in North America ................................ ................................ ..... 30 2.3 Allophonic Change in the Inland North ................................ ................................ ..... 34 2.4 Lansing Speech Community ................................ ................................ ....................... 39 2.4. 1 The Speech Community ................................ ................................ .......................... 39 2.4.2 Preliminary Analyses of /æ/ in Lansing ................................ ................................ .. 48 2.5 Summary ................................ ................................ ................................ ..................... 55 CHAPTER 3 DATA AND METHODS ................................ ................................ ...................... 57 3.1 Introduction ................................ ................................ ................................ ................ 57 3.2 Acoustics of /æ/ in 20 th century Lansing ................................ ................................ .... 57 3.2.1 Participant Selection and Recruitment ................................ ................................ .... 58 3.2.2 Speaker Demographic Information ................................ ................................ ......... 60 3.2.3 Interviewers ................................ ................................ ................................ ............. 66 3.2.4 Interview Methods ................................ ................................ ................................ .. 67 3.2.5 Audi o Processing, Transcription, and FAVE suite ................................ ................. 72 3.2.6 Measurements and Analysis ................................ ................................ ................... 75 3.3 Phonological Change ................................ ................................ ................................ .. 78 vii 3.3.1 Divergent Trajectories ................................ ................................ ............................ 79 3.3. 2 Sub - Phonemic Judgement Task ................................ ................................ .............. 80 CHAPTER 4 / æ / IN 20 TH CENTURY LANSING ................................ ................................ ...... 93 4.1 Introduction ................................ ................................ ................................ ................ 93 4.2 Results ................................ ................................ ................................ ........................ 94 4.3 F1 of Pre - Oral /æ/ ................................ ................................ ................................ ....... 95 4.3.1 F1, Generational Time and Gender ................................ ................................ ......... 97 4.3.2 Summary of F1 Results ................................ ................................ ........................... 99 4.4 F2 of Pre - Oral /æ/ ................................ ................................ ................................ ..... 100 4.4.1 F2, Generational Time and Gender ................................ ................................ ....... 102 4.4.2 Summary of F2 Results ................................ ................................ ......................... 103 4.5 Diphthongal Quality of Pre - Oral /æ/ ................................ ................................ ........ 103 4.5.1 Diphthongal Quality and Gender ................................ ................................ .......... 106 4.5.2 Summary of Diphthongal Quality Results ................................ ............................ 107 4.6 Nasal Allophony ................................ ................................ ................................ ....... 108 4.6.1 Summar y of Nasal Allophony Results ................................ ................................ .. 109 4.7 Summary: /æ/ in 20 th Century Lansing ................................ ................................ .... 110 CHAPTER 5 PHONOLOGIZATION OF / æ / NASAL ALLOPHONY IN LANSING ........... 113 5.1 Introduction ................................ ................................ ................................ .............. 113 5.2 Results ................................ ................................ ................................ ...................... 114 5.2.1 Phonetic and Phonological Allophony ................................ ................................ .. 114 5.2.2 Sub - Phonemic Judgement Task: Actuation and Transition of Phonologization .. 118 5.3 Conc lusion ................................ ................................ ................................ ................ 127 CHAPTER 6 CONCLUSIONS AND DISCUSSION ................................ ............................... 129 6.1 Gender and Allophonic Change in Lansing ................................ ............................. 132 6.2 Internally Motivated Change from Above ................................ ................................ 134 6.3 Contact as a Conditioning Factor ................................ ................................ ............. 135 6.4 Methodological Considerations ................................ ................................ ................ 137 6.4.1 Awareness of Variable Phonologies ................................ ................................ ..... 137 6.4.2 Acoustic Salience vs Representation ................................ ................................ .... 138 6.4.3 Orthography in American English ................................ ................................ ........ 139 6.5 Inland North D ialectology ................................ ................................ ........................ 140 6. 6 Phonological Representations in the Inland North (more questions) ....................... 144 6.7 North American Dialectology ................................ ................................ .................. 145 6.8 Mid - 20 th century and Loss of Regional North American Features .......................... 146 APPENDICES ................................ ................................ ................................ ............................ 147 APPENDIX A Recruitment Flyer for 2018/2019 Sociolinguistic Interviews .................... 148 APPENDIX B Consent Form for 2018/2019 Sociolinguistic Interviews ........................... 149 APPENDIX C Interview Questions for 2018/2019 Sociolinguistic Interviews ................. 150 APPENDIX D Table 15 Speaker Demographics Lansing Speech Corpus plus 21 Millennials from Wagner et al. ( 2016 ) ................................ ................................ ................ 153 APPENDIX E Consent Form for Sub - Phonemic Judgement Task ................................ .... 157 viii APPENDIX F Social Media and Email Recruitment Script for Sub - phonemic Judgement Task ................................ ................................ ................................ ................................ ..... 158 APPENDIX G Table 16 Pairs of Lexical Items for Sub - Phonemic Judgement Task ........ 159 APPENDIX H Table 17 Pillai - Bartlett and Mean F1, F2, and DQ Values for Each Speake r in the Lansing Speech Corpus plus 21 Millennials from Wagner et al. (2016) .................. 161 APPENDIX I Figure 23 Per Token Diagonal Measurement of / æ / in Five Following Phonological Contexts by Speaker Birth Year in the Lansing Speech Corpus ( p lus 21 Millennials from Wagner et al. (2016)) ................................ ................................ .............. 166 REFERENCES ................................ ................................ ................................ ........................... 167 ix LIST OF TABLES Table 1 Process of /g/ - deletion (adapted from Turton 2014: 57) ................................ ................... 9 Table 2 Distribution of the Lansing Speech Corpus by generational cohort, social class, and gender. ................................ ................................ ................................ ................................ ........... 60 Table 3 Number of participants from each data source ................................ ................................ 64 Table 4 Co nditions of Lexical Pairs in the sub - phonemic judgement task ................................ ... 82 Table 5 Distribution of experimental participants by Generational Co hort ................................ . 89 Table 6 Distribution of experimental participants by Generational Cohort and Socio - Economic Status ................................ ................................ ................................ ................................ ............. 90 Table 7 Distribution of experimental participants by Generational Cohort, Socio - Economic Status, and Gender ................................ ................................ ................................ ........................ 91 Table 8 Best fit mixed - effects model of pre - oral /æ/ F1 ................................ ............................... 97 Table 9 Best fit mixed - effects model of pre - oral /æ/ F2 ................................ ............................. 100 Table 10 Best fit mixed - effects model of pre - oral /æ/ Diphthongal Quality (DQ) .................... 104 Table 11 ANOVA results of Pillai - Bartlett score analysis ................................ ......................... 108 Table 12 Distribution of /æ/ features across generational time in 20 th century Lansing ............ 111 Table 13 Summary of social conditions on acoustically salient NCS features in 20 th century Lansing ................................ ................................ ................................ ................................ ........ 111 Table 14 Mixed - effects regression model for responses to CæC pairs. ................................ ..... 120 Table 15 Speaker Demographics Lansing Speech Corpus plus 21 Millennials from Wagner et al. (2016) . 3 Table 16 Pairs of Lexical Items for Sub - Phonemic Judgement Task . 59 x Table 17 Pillai - Bartlett and Mean F1, F2, and DQ Values for Each Speaker in the Lansing Speech Corpu s plus 21 Millennials from Wagner et al. (2016) 1 xi LIST OF FIGURES Figure 1 Reproduced from Fruehwald 2016:382. Trajectory of /a / in Philadelphia in two phonological contexts (before voiced and voiceless consonants) in apparent time. ..................... 18 Figure 2 Northern Cities Shift vowel chain shift configuration . (Adapted from Wolfram & Schilling - Estes 1998: 138) ................................ ................................ ................................ ............ 34 Figure 3 Population rate of change by decade from 1860 to 2010 in Lansing. [ Source: US census 1860 2010]. ................................ ................................ ................................ ................................ 41 Figure 4 Percentage of Non - White population in Lansing. [Source: US census 1960 - 2010]. ..... 43 Figure 5 Rate of change in population by decade from 1860 to 2010 in Lansing compared to two Lansing suburbs; Delta Charter Township and Bath Charter Township. [Source: US census 1860 2010]. ................................ ................................ ................................ ................................ ......... 47 Figure 6 Three distinct /æ/ systems from Greater Lansing: Jack Down, b. 19 24, multiple college degrees, advanced raised continuous system; Michelle Baulch, b. 1971, multiple college degrees, common raised continuous system; Ben Langdon, b. 1994, community college student, nasal system. ................................ ................................ ................................ ................................ .......... 50 Figure 7 Pillai - Bartlett scores for /æ/ by year of birth for Lansing, MI speakers (from Wagner et al 2016:figure 5) ................................ ................................ ................................ ............................ 51 Figure 8 Trajectory of /æ/ diagonal in urban Lansing (blue = blue - collar; red=white collar); dots represent means for individual speakers (n=27). ................................ ................................ .......... 53 Figure 9 Pillai - Bartlett scores in 20th century Lansing by social class (blue - collar = blue). ....... 54 Figure 10 Average F1(Hz) values of pre - oral /æ/ by speaker year of birth in 20 th century Lansing. Lansing Speech Corpus (n=36), plus Millennial speakers (n=21) for reference. ......................... 95 Figure 11 Average F1(Hz) values of pre - oral /æ/ by speaker year of birth and gender in 20th century Lansing. Lansi ng Speech Corpus (n=36), plus Millennial speakers (n=21) for reference. ................................ ................................ ................................ ................................ ....................... 98 Figure 12 Average F2 (Hz) values of pre - oral /æ/ by speake r year of birth in 20th century Lansing. Lansing Speech Corpus (n=36), plus Millennial speakers (n=21) for reference. ........ 101 Figure 13 Average F2(Hz) values of pre - oral /æ/ by speaker year of birth and gender in 20th century Lansing. Lansing Speech Corpus (n=36), plus Millennial speakers (n=21) for reference. ................................ ................................ ................................ ................................ ..................... 102 xii Figure 14 Average Diphthongal Quality (Hz) values of pre - oral /æ/, /e /, and / / by speaker year of birth in 20th century Lansing. Lansing Speech Corpus (n=36), plus Millennial speakers (n=21) for reference. ................................ ................................ ................................ ............................... 105 Figure 15 Average Diphthongal Quality (Hz) values of pre - oral /æ/ by speaker year of birth and gender in 20th century Lansing. Lansing Speech Corpus (n=36), plus Millennial speakers (n=21) for reference. ................................ ................................ ................................ ............................... 106 Figure 16 Per speaker Pillai - Bartlett scor e of pre - nasal and pre - oral /æ/ distribution by speaker year of birth in 20th century Lansing. Lansing Speech Corpus (n=36), plus Millennial speakers (n=21) for reference. ................................ ................................ ................................ ................... 109 Figure 17 Per token diagonal measurement of /æ/ in three following phonological contexts by speaker year of birth in 20 th century Lansing. ................................ ................................ ............ 116 Figure 18 Per token diagonal measurement of / / (left facet) and / / (right facet) in three following phonological contexts by speaker year of birth in 20th century Lansing. .................. 117 Figure 19 Distribution of responses to pre - nasal and pre - oral pairs (left = condition one) compared to pre - oral and pre - oral pai rs (right = condition two) of /æ/ over generational time. 121 Figure 20 Distribution of responses to pre - nasal and pre - oral pairs (l eft = condition one) compared to pre - oral and pre - oral pairs (right = condition two) of /æ/ over generational time and by Social Class (White Collar respondents at the top). ................................ .............................. 123 Figure 21 Distribution of responses to pre - nasal and pre - oral pairs (left = condition one) compared to pre - oral and pre - oral pairs (right = condition two) of / / and / / over generational time. ................................ ................................ ................................ ................................ ............ 126 Figure 22 Distribution of responses to pre - nasal and pre - oral pairs of /æ/ over generational time, by social class (White Collar respondents at the top), and gender (women on the left). ............ 133 Figure 23 Per Token Diagonal Measurement of /æ/ in Five Following Phonological Contexts by Speaker Birth Year in the Lansing Speech Corpus (plus 21 Millennials from Wagner et al. (2016)) 6 1 PHONOLOGICAL CHANGE 1.1 Introduction One of the challenges faced by phonologists is accounting for change, i.e. how do members of a speech community begin to posit a new phonological rule that has never before existed in that speech community? For example, how could we account for someone wh o posits a phonological/categorical difference between / r / and / l / , i.e. the two are distinct phonemes in their grammar, when no one in their speech community distinguishes between the two sounds phonologically, i.e. [ r ] and [ l ] are simply variants of the same phoneme. Though the mechanisms of phonological change have been the subject of many articles, dissertations, and chapters, there is still very little that we actually know about what governs phonological change. This is perhaps because as Hockett (1 958:456) surmises, phonological change is rare and occurs suddenly too soon for anyone to detect by direct observation. Because it is so rare, most of what we "know" about phonological change has been deduced from the results of perception studies, simula tions, synchronic analysis, and historical data. For example, the Big Bang theory developed by Janda and Joseph (2003) utilizes historical and synchronic data to theorize that phonological change is motivated in the ph onetics for a brief amount of time before phonology takes over. Underrepresented in the literature are studies of phonological change in progress using language production data. Sociophonologists have made some inroads in this endeavor with the use of la rge corpora spanning many generations. For example, Fruehwald (2013,2016) investigated Canadian Raising of /a / in Philadelphia through an analysis of spontaneously produced speech in the Pennsylvania Neighborhood Corpus (PNC) (Labov & Rosenfelder 2011) , which is a collection of over 300 hours of recorded interviews with approximately 400 2 Philadelphia natives, conducted by University of Pennsylvania graduate students as part of their course work. In Empirical Foundations for a Theory of Language Change , Weinreich, Labov & Herzog (1968) describe 5 problems to be addressed when attempting to answer this fundamental question about sound change. One , the Constraints P roblem , asks what kinds of innovations are possible and what are not possible. The Transition Problem (or Incrementation Problem) asks how a change progresses in the same directi on over multiple generations. The Maintenance Problem concerns why a change does not revert back to its previous linguistic state . The Evaluation Problem asks how the new and old system s are evaluated in the community. The Embedding Problem asks how the i nnovation is linked to the larger linguistic system and how i t i s connected to different social groups. The Actuation Problem asks what factors can account for the initiation of change : Why do changes in a structural feature take place in a particular lan guage at a given time, but not in other languages with the same feature, or in the same language at other times (Weinreich et al. 1968:102) ? Although many advances have been made in addressing the problems identified by Weinreich et al. in 1968 regarding language change in general, regarding phonological change specifically, some of these problems remain. The present investigation adds to the growing field of sociophonology and addresses what Weinreich et al. termed ge change the Actuation Problem by focusing on one of these rare changes, namely the development of phonological nasal allophony for /æ/ against the backdrop of socio - historical changes underway in Lansing, MI. Whereas / æ/ was previously raised in all ph onological environments for Lansing speakers, e.g. mat and man were both produced with [ æ] and similar realizations, recent work has shown that many speakers born after 1984 raise /æ/ only before nasal consonants. 3 I utilize a combination of analyses, incl uding speaker - level distributions, community level phonetic target analysis, and the results of a judgement task , to investigate this phonological change. The two sources of data for this research are a collection of conversational interviews with natives of Lansing, MI, and responses to a sub - phonemic judgement task. The collection of interviews includes oral histories and sociolinguistic interviews with 36 Lansing natives ranging in date of birth from 1925 to 1984. The sub - phonemic judgement respondent s are 107 Lansing natives who range in date of birth from 1925 to 1999. In addressing the actuation problem of phonological change, I take heed of the following warning in Empirical Foundations : Linguistic and social factors are closely interrelated in the development of language change. Explanations which are confined to one or the other aspect, no matter how well constructed, will fail to account for the rich body of regularities that can be observed in empirical studies of language behavior. Weinreich, L abov, and Herzog (1968:188) Therefore, this work is an investigation into both the linguistic/internal and social/external mechanisms by which speakers in a community posit a phonological rule that never before existed. To do this, I focus on 2 aspects of this change, namely: 1. The ph onetic changes that /æ/ has undergone on its way towards an allophonic split over 3 generations in Lansing, namely lowering in F1, retraction along F2, and less diphthongal quality. 2. The impact of gender and social class on these measures over generational time Lansing is an optimal locus for an investigation into phonological change. Lansing is situated in the Inland North, a regional American English dialect area that has received a lot of attention 4 from sociolinguists over the last few decades . Additional ly, the allophonic change under investigation in this dissertation the development of / æ / nasal allophony is fairly well documented in the sociolinguistics literature in varieties of North American English. I discuss suitability for the present project in Chapter 2 . In the rest of this chapter, I will outline and expand upon the theoretical debates in the field concerning the initiation of phonological change. In section 1.2 , I discuss the theoretical commitments of this dissertation regarding the distinction between phonetics and phonology. In section 0 , I review the existing literature on phonological change as it relates to the A ctuation P roblem described by Weinreich, Labov & Herzog (1968) . Section 1.4 is a review of the methodologies for discerning between phonetic vs phonological allophony within speakers . In this section, I also mo tivate the use of another method : a sub - phonemic judgement task. In section 1.5 , I draw on the sociolinguistics literature to provide a framework that will help me to identify the possible actuators of nasal / æ/ allophony in Lansing. Section 1.6 provides a summary of the main and peripheral goals of this dissertation. S ection 1.7 includes a summar y of the overall findings and section 1.8 provides a layout of each of the chapters in this dissertation . 1.2 Phonetics vs Phonology In this dissertation, I adopt a modular feedforward framework that is rooted in Generative Phonology (Keating 1990; Pi errehumbert 1990; Bermúdez - Otero 2007) . Under this framework, different levels of the grammar occupy different modules. The focus of this dissertation is on the phonetics and phonology modules, which are schematized in (1), below. These modules are a rranged serially, as depicted in (1), so that information flows from one module to another in a feedforward fashion. The feedforward nature of this framework is that one module only has 5 access to the information in the previous module. For example, a pho netic rule will have access to phonological representations but will not have access to underlying representations. (1) Modular feedforward architecture of the phonetics - phonology interface Phonological rules Phonetic rules Sound change is can include any change that occur in either the phonetic or phonological levels represented in (1). Changes in the phonetic module give rise to phonetically grad ient rules which operate over continuous phonetic dimensions (e.g. formant frequency or duration). Many of the changes in the sociolinguistic literature which track change in the phonetic target of a phoneme are examples of this type of change. Changes in the phonology module operate over Underlying representation Phonological representation Phonetic Representation 6 discrete phonological dimensions (e.g. features), which manifest phonetically as abrupt changes between separate regions in phonetic space. Allophonic change, which is the focus of this dissertation, is defined as the emer gence of an alternation between two allophones of the same phoneme where previously there had been only one allophone. In line with the framework just described, I will refer to two types of allophony throughout this dissertation one that is phonetic and one that is phonological. The difference between the two types of allophonic grammars is one of features. Phonetic allophones are those that are distinguished along some phonetic dimension, but in a continuous rather than categorical fashion. Phonologica l allophones are distinguished by features at the level of phonological representation, the middle level in (1) above. The difference between the two types of allophony is exemplified by examining the realization of /æ/ in different dialects of American E nglish (Dinkin 2011a) . The realization of /æ/ as part of the Northern Cities Shift dialect is an example of phonetic allophony. Within this phoneme, tokens of /æ/ in various phonological environments are gradiently different in F1/F2 space, such that by regular phonetic operations, /æ/ tokens before nasal consonants appear at the top of the token cloud while /æ/ tokens before /l/ appear at the bottom of the token cloud. For phonetic allophones, change ov er time would result in the same phonetic target change (e.g. backing along F2) for both the pre - /l/ and pre - nasal allophones. In other dialects of American English, however, /æ/ before nasal codas is phonologically distinct from / æ / in other following co da environments. In these dialects, /æ/ before nasal codas are discretely separated in F1/F2 space from the rest of the token cloud, i.e. there is a distinguishable gap in the vowel space between pre - nasal and pre - other tokens of /æ/, whereby the latter t okens are more retracted along F2 than the former. The discrete nature of this difference suggests that /æ/ before nasal consonants has assumed some feature (perhaps 7 [ peripheral]) that distinguishes it phonologically from the rest of the allophones of th is phoneme. Because they have different representations at the phonological level, phonological allophones can be affected independently of each other by changes in phonetic rules and representations. In many dialects of North America, /æ/ before non - nas al consonants is moving backwards in the vowel space, whereas that before nasal consonants appears to be unchanging. Therefore, the fundamental difference between phonetic and phonological allophony, is whether the two allophones are discretely separated by at least one feature (phonological allophony) or by simply phonetic implementation alone (phonetic allophony). While sound change can occur in any of the modules in (1), this dissertation is concerned with the development of phonological allophony. T he primary question to be addressed is whether change in the phonetics gives rise to the development of this phonological rule or whether the rule was developed without a phonetic precursor. 1.3 Gradual vs Abrupt Phonological Change The prevailing sentiment s urrounding phonological change is that it develops gradually, i.e. some exaggerated difference between two phonetic variants is eventually re - analyzed (by speakers and/or hearers) as a phonological rule (Hyman 1975; Ohala 1981 ; Blevins 2004; Bermudez - Otero 2007; Moreton & Thomas 2007; Baker, Archangeli & Mielke 2011) . Others have argued, however, that there is no phonetic precursor to phonological change change to the grammar is sudden and does not require anything in the phonetics to prompt the change (Fruehwald 2013; Janda and Joseph 2003). In what follows, I discu ss the two arguments in turn. 8 1.3.1 Gradual phonological change A theory of gradual phonological change posits that phonological change proceeds through the grammar incrementally first occurring in the phonetic module and then the phonological module . T he Life - - (Bermúdez - Otero 2007; 2015) is one theory that supports this idea that phonological change is gradual . According to the Life - Cycle, p honological processes first begin as language - independent or mechanical ph onetic effects (stage 1 , Table 1 ). Through a process called phonologization , these universal phonetic effects become cognitively - controlled language - specific phonetic implementation effects (stage 2). They then become stabilized as a phrase - level catego rical/phonological rule (stage 3). Over time, the process advances through the grammar by domain narrowing so that it applies at the word level (stage 4), and then the stem - level (stage 5). In its last stage, the phonological process may then advance to t he lexicon through morphologization and/or lexicalization (stage 6). To illustrate the Life - Cycle, Bermúdez - Otero & Trousdale (2008) make reference to the historical process of /g/ - deletion in Table 1 . Under the theory of the L ife - C ycle, the process o f /g/ - across all environments but perhaps unpredictably deleted in casual conversation on occasion (stage 1). At some point in time, there was a process of gradie nt lenition of the stop but only at the phrase level (stage 2 in Table 1 ). At this point, the gradient nature of lenition means that it must be subject to social conditioning. Later, in stage 3, this gradient process develops into a categorical rule whereby lenition at the phrase - level is categorical within and across speakers in the community. This categorical rule then begins to narrow its domain comi ng under increasing morphological (stage 4), and lexical control (stage 5). The sixth stage of the Life - Cycle is not 9 included in the table but involves the development of lexical exceptions, which for the /g/ - deletion example include words like young er an d long er (Bermúdez - Otero & Trousdale 2012: 10) . Table 1 Process of /g/ - deletion (adapted from Turton 2014: 57) finger sing - er sing it si n g Stage 1 Stage 2 Stage 3 Stage 4 Stage 5 To explain why phonological change proceeds in this fashion, Bermúdez - Otero & Trousdale (2008) appeal to what they call input restructuring . If one thinks about this from the perspective mostly made up of the phonology in the stage prior. Thus, the learner simply re - analyzes the rule based on frequency of occurrence in the previo higher level. For example, if a child is born to parents whose grammars resemble that in stage 3 of Table 1 where the rule only applies at the phrase level, that child is likely to encounter /s 10 more often than /s - deletion as a word - level process, which will result in them having a stage 4 grammar. The transition from stage 1 to stage 2 processes forms the basis of what has been termed the accumulation - of - errors hypothesis, whereby an epiphenomenal phonetic effect is re - an a lyzed as a language - dependent speaker - controlled phonetic effect. Most ardently supported by Ohala (1981; 1989; 1990; 1993) , the accumulation - of - errors hypothesis posits that a listener (or speaker) hea rs tokens that are outside of the distribution they are used to and reinterprets the difference in pronunciation as a new phonetic rule. This new phonetic rule can then become a phonological rule. The classic example of this phenomenon is the hypothetical sound change of /ut/ - fronting (Ohala 1981) . The scenario is as follows. A sp eaker, intending to produce /ut/, produces [yt], with a fronted realization of /u/, as a result of coarticulation with the following [t]. Being aware of this coarticulation effect, a hearer will mentally reconstruct what they heard as an /ut/. Phonetic c the coarticulation into account and they interpret the [+front] feature as belonging to the vowel rather than the /t/, so they hear the /u/ as a /y/. When that listener speaks, they produce what origin ally was /u/ as [ y ] . Crucially, Ohala argues that there has been no phonological rule posited at this point ; the phonological rule can be posited at some later stage. This scenario is a description of the transition between stages 1 and 2 in the Life - Cycle : a mechanical effect is re - analyzed as a speaker - controlled phonetic effect, and at some later point in time, a phonological rule is posited. Thus, the accumulation of error theory is most in line with a gradual rather th an abrupt theory of change. Other research programs subscribe to the same incremental theory of change and push the accumulation - of - errors scenario farther so that re - analysis occurs between stages 1 and 2 and then again between stages 2 and 3, e.g. Hyman (1975; 2013) , Evolutionary Phonology (Blevins 11 2004; 2006) , Exemplar Theory (Pierrehumbert 2001; Garrett & Johnson 2011) . Though the specific details of each theory are different, they all consider phonetic variability to be a significant catalyst to phonological change. 1.3.2 Abrupt phonological change In contrast to a theory of gradual phonological change, the Big Bang theory of sound change (Janda & Joseph 2003) argues that phonetic factors guide sound change only very briefly and that phonological conditioning takes over very quickly. To motivate this argument, Janda and Joseph (2003) re - examine two changes, one of which is Swiss German o - lowering. Preconsonan tal o - lowering originally occurred only before /r/, but there many varieties that have generalized this rule in various ways, e.g. some lower before all obstruents except /b/, others lower before nasals and coronal obstruents but not other obstruents, oth ers lower before obstruents except /b/ and nasals, and so on. Crucially, the central city of Schaffhausen used the simpler rule; lowering before /r/, which while diffusing to different villages was generalized in a variety of ways. Crucial for Janda and Joseph (2003) is that the rule s in these villages do not phonetically favor a lowered o ; therefore, though the initial state was lowering before /r/, these villages developed a rule that was not phonetically driven. The authors argue that if phonetic vari ation did play a role in this change, it must have been very brief and overruled by the phonology in an instant. 1.3.3 Disambiguating abrupt and gradual phonological change Crucially, as Fruehwald (2013) points out, phonological change is rarely observed while the change is underway. Many theories about the process of phonological change therefore suffer from a paucity of empirical evidence . They rest on completed changes (as is the case with (Garrett & Johnson 2011) , p erception 12 studies (Ohala 1981), observations of synchronic data, or prior theories (and see Fruehwald 2016 for an in depth discussion of these methods). Larger databases of naturally occurring, ongoing phonological changes are required to test the theorie s . Sociophonologists have been making inroads in this endeavor, compiling large spoken corpora with enough time depth to address questions about phonological change. One such study is Fruehwald's (2013; 2016) investigation into /a / - raising in Philadelphia. The results of his study suggest that, contrary to popular belief, there is no intermediate (phonetic) stage in phonological chan ge the rule is categorical/phonological from the outset. The literature on change towards a phonemic merger, i.e. the merger of two or more phonemes into one, is also robust in its use of real language data while a change is underway. Though I will not go over the literature in detail here, the general consensus on phonemic mergers is that both types of changes (abrupt and gradual) are possible. Crucially, language/dialect contact predicts which type of that are internally motivated (are not triggered by contact with speakers of other phonological systems) progress gradually (see Guy 1990 and Herold 1997) . Studies on allophonic change, which is the focus of this dissertation, are underrepresented in the literature. There are in fact, only two studies that I am aware of that attempt to disa mbiguate gradual from abrupt phonologization of an allophonic split using natural language data. As I summarize below, these studies show conflicting findings, and thus offer no solution to the current theoretical conundrum. It is still unclear if alloph onic change is abrupt, gradual, or whether like mergers, its development depends on inter - dialect/language contact. 13 1.3.3.1 Fruehwald (2013, 2016) The first investigation of an ongoing phonological split using actual production data is (2013, 2016) study of Canadian Rais ing of /a / in Philadelphia. In North American English, Canadian Raising of /a / is realized in a variety of ways, but the most common realization is that the diphthong raises when it occurs before voiceless codas but remains low before voiced codas. Therefore, raising of the nucleus is observed in words like knife and write , but not in words like knives and ride . Canadian Raising of /a / has been labeled phonological raising because the raising rule also applies before a flapped /t/ such that the /a / in writer su rfaces with the higher variant (Halle 1962; Idsardi 2006; Berkson, Davis & Strickler 2017; Fruehwald 2013; 2016) . The rule must be phonological because although the flapped /t/ in writer is voiced on the surface, it is underlyingl y voiceless; this shows that the rule applies to underlying (phonological) representations, rather than mere surface (phonetic) representations. In his analysis of /a / - raising in Philadelphia, Fruehwald (2013 ; 2016) sought to empirically determine wheth er phonetics played a role in the initial process of phonological change. He asked whether an earlier stage of phonetic variation is what prompts community members to posit a phonological difference between the allophones (gradual phonologization), or if t he allophones were phonologically distinguished from the outset (abrupt phonologization). To do this, he conducted an acoustic investigation of /a / raising in the spontaneous speech of 326 Philadelphia - born speakers whose birthdates range from 1889 to 199 8. He compared three measurements of /a / (nucleus, offset, and duration) over apparent time across four contexts ; faithful /t / ( write ) , faithful / d/ ( ride ) , flapped /t / ( writing ) , and flapped /d/ ( riding ). His analysis showed that there was no intermediate stage whereby phonetic raising (stage 2 in the Life - Cycle) , i.e. the raising of /a / is determined by surface voicing of the following consonant, 14 predated phonological raising (stage 3 of the Life - Cycle) in the community. At the community lev el, the phonetic separation of pre - voiced and pre - voiceless /a / along F1 has been in progress in Philadelphia since about the 1920s (Labov 2001; Fruehwald 2013; Fruehwald 2016) . Fruehwald showed that the difference between pre - voiced and pre - voiceless /a / was phonological from the very outset of this change, i.e. / a / began to raise before flapped /t/ at the same time as before faithful /t/, and not at some later time through re - analysis. Fruehwald argues that in Philadelphia there were two phonological categories of /a / that were phonetically similar and that one of them (the pre - voiceless variant) underwent a phonetica lly gradual change in height while the other remained low. Therefore, phonological change was abrupt in this speech community, i.e. there was no intermediate stage whereby the allophones were differentiated purely phonetically before a phonological rule w version of the Big Bang theory of sound change proposed by Janda and Joseph (2003) who argue that this intermediate/phonetic stage, if ever observed, would be very brief. 1.3.3.2 Berkson, Davis and Strickler (2017) In contrast, Ber kson, Davis and Strickler (2017) appear to have observed this intermediate stage of Canadian Raising in Fort Wayne, Indiana. Berkson et al. analyzed word - list productions of /a / before faithful voiceless, faithful voiced, /t/ flap, and /d/ flap consona nts in the speech of 27 Fort Wayners. They visually inspected F1 trajectories of each token of /a / for every speaker and for each speaker compared the difference in F1 at the 30% mark between the four environments. They show that speakers exhibit differ ent patterns of /a / - raising. 15% of the speakers in this sample do not raise in any environment, two - thirds of the speakers in the sample exhibit phonetic raising, i.e. raising before faithful voiceless consonants but not before flapped /t/ or either /d/ variant, and 22% of the speakers exhibit phonological raising whereby /a / raises before both 15 faithful and flapped /t/ but not before either /d/ variant. Berkson et al. reconcile this result with conclud ing that One unstudied aspect in the Berkson et al. study is change over time. The speakers in this study range in age from 19 to 78. I t is unclear if date of birth conditions the patterning of their sample. There are a few possible scenarios that could be borne out of the data if they were that the individuals who have no raising are older speakers, those who exhibit phonetic raising are middle - aged, and the phonological raisers are all younger speakers. This scenario would support a gradual theory of phonologization, because it proceeds ge neration by generation, with an intermediate stage of phonetic - only conditioning. Alternatively, perhaps those who exhibit no raising are older speakers, middle - aged speakers are a mixture of those who exhibit phonetic raising and those who exhibit phonolo gical raising, and young speakers are phonological raisers. Under this scenario, phonologization would be abrupt because the phonological rule has been posited for some speakers in the middle generation when there was no phonetic raising in the previous g eneration. Thus, to really tease apart whether phonological change is abrupt or gradual in Fort Wayne, we would benefit from a discussion of how year of birth conditions the patterns reported in their sample. Because there is no information regarding how year of birth conditions the patterns they observe, it is not exactly clear whether Canadian Raising in Fort Wayne can be classified as abrupt or gradual phonological change. 1.3.4 Summary In sum, Fruehwald (2013, 2016) provides evidence for the strong view tha t phonological change is abrupt, since there is no transition period between pre - and post - phonologization in his 16 Philadelphia data. Berkson et al. (2017) support a weak view that phonological change is abrupt but there is a very brief transition period, i .e. the period in which allophony is conditioned phonetically. The Fruehwald (2013, 2016) and Berkson et al. (2017) studies thus provide a theoretical conundrum for a description of how allophonic change progresses. The results in Fort Wayne suggest that there is indeed a phonetic precursor for phonological change. The Berkson et al. analysis appears to show that both phonological and phonetic stages are active at the beginning of the phonological change, but this remains to be seen as they have yet to sh ow whether Canadian Raising is a new phenomenon in Fort Wayne or if the patterns they observe are distributed among speaker s of different age groups however, suggests that there is no stage at which allophony is phone tically motivated in the community. With this conflicting evidence, and the conflicting theoretical claims in the literature, it is clear that the addition of more studies tracing allophonic development are needed. As discussed previously, there appear t o be multiple pathways by which other phonological changes, e.g. change towards phonemic merger, are actuated. Because of this, it would not be completely surprising if multiple pathways for allophonic splits are established. With the analysis of allopho nic change in more speech communities, perhaps we can develop a typology of allophonic change. 1.4 Methods for D istinguishing P honological from P honetic D istinction As stated earlier, methods for discerning phonologization from phonetic change are not abunda nt in the literature, especially since phonological change is so rarely observed. In the case of /a / raising, acoustic analysis is fairly straightforward because one can readily contrast phonetic (flapped stops) and phonological (faithful stops) raising. Articulatory analyses have also been 17 employed, e.g. Turton (2014). A method used quite often in the literature on phonemic mergers is a measure of bimodality. This measure is usually a calculation of the distribution of token clouds between two vowel cla sses. A unimodal distribution is evidence for merger, while a bimodal distribution is taken as evidence of two vowel classes. Lately, conclusions based on distributions have been cautioned against, as studies have shown that while bimodality is often an in dication of phonological/categorical difference, the absence of a bimodal distribution does not necessarily entail the absence of a phonological difference (cf. Bermúdez - Otero & Trousdale 2008: 696; Schilling, Watkins & Watkins 2002) . Thus measures of bimodality are best interpreted by supplementing with other types of analyses. Here, I motivate the two that will be employed in this dissertation in addition to a measure of bimodality (in this case, the Pillai - Bartlett measure). Further det ails are given in Chapter 3. 1.4.1 Divergent Trajectories Fruehwald (2013) noted the utility of examining the rate of change of a vowel phoneme in varying phonological environments to determine whether any of the phonetic differences have been phonologi zed in the community, and crucially, when in apparent time the two environments began to diverge. In his analysis of the phonologization of Canadian Raising in Philadelphia, Fruehwald showed that the difference between /a / before voiceless and voiced cons onants was phonological from the outset of the pre - voiceless phonetic target change by inspecting the rate of change of /a / in the two phonological environments. He noted that because /a / in these environments followed different trajectories along F1 fr om the 1930s onwards , they were two phonological allophones rather than differentiated due solely to phonetics. 18 This is captured in Figure 1 . In the figure, are speaker means of F1 at the nucleus for /a / across speaker year of birth in the PNC. /a / means before voiced consonants are indicated with dark circles and those before voiceless are indicated with open circles. Figure 1 Reproduced from Fruehwald 2016:382. Trajectory of /a / in Philadelphia in two phonological contexts (before voiced and voiceless consonants) in apparent time. In F igure 1 , we see that /a / in the two environments proceed along divergent trajectories from the 1930s; that before voiceless consonants moving up the vowel space and that before voiced consonants remaining stable. Thus, he was able to show that phonologization of 19 Canadian Raising in Philadelphia began in the 1930s. To support this claim, he examined the trajectories of other vowel phonemes in different phonological contexts over time in the same community. One such phoneme was / / before fricative, nasal, and other consonants. He noted that unlike pre - voiced and pr e - voiceless /a /, the / / allophones moved in lockstep over the course of the 20 th century and were therefore only phonetically distinguished from one another. 1.4.2 Sub - Phonemic Judgement Task Another diagnostic utilized in the phonemic merger literature is a p honemic judgment task (DiPaolo 1988; Herold 1997; Labov, Ash & Boberg 2006; Johnson & Nycz 2015; Baranowski 2013) . Herold (1997) employed this task in her investigation into whether the low - back merger (merger of / / and / western part of the state or via an independent innovation. For this study she suppl emented production data with a judgement task in which participants were visually presented with pairs of lexical items that differed in incidence of / / and / caught and cot ) and were asked to indicate whether the vowels in the pairs sounded the s ame or different to them. Analysis of the judgement data and production data revealed that the merger in eastern Pennsylvania was an independent innovation, rather than the result of diffusion from western Pennsylvania, since speakers in both parts of the state developed the phon emic distinction at about the same time. Though this judgement task is usually performed to test phonemic merger/distinction, Mellesmoen (2016) has shown that it can be useful for testing whether allophones differ in their phonological representation or whether they merely represent different phonetic implementations of the same phonological features. In her study of /æ/ in British Columbia, Mellesmoen (2016) analyzed speech production and judgement task data to investigate whether there was a phonological distinction between /æ/ before voiced velars ( ) and /æ/ before other 20 conso nants. An analysis of the trajectories in phonetic space of these conditioning environments over apparent time revealed that in BC a three - way allophonic differentiation for /æ/ had developed, such that /æ/ before front nasal consonants, /m/ and /n/, were distinct from /æ/ before voiced velar /g/, and /æ/ before other consonants in F1 x F2 space. The acoustic results for /æ/ /æ/ before /g/ and /æ/ before other na sal consonants. To discern which allophone pre - belonged to, Mellesmoen (2016) conducted a judgement task. For this, she asked participants to indicate whether pre - (e.g. rang ) sounded more like words with a velar coda (e.g. rag ) or words wit h a front nasal coda consonant (e.g. ran ) . She showed that participants classified pre - - /g/ allophone rather than as part of the pre - front nasal allophone. Though this is the only study to utilize the judgement task to determine t he phonological status of an allophonic split, the results are convincing because they align with her analysis of trajectory movement. 1.4.3 Diagnostics of P honological C hange in this d issertation In the present study, I use the two methods described above to address the actuation and transition problems of phonological change in Lansing. I will investigate phonetic vs phonological allophony within the /æ/ phoneme by examining trajectories of pre - na sal and pre - oral /æ/ in F1/F2 space over time, in conjunction with the results of the phonemic judgment task. According to Fruehwald (2013, 2016), phonological allophones follow disparate trajectories over time while allophones that differ in phonetic imp lementation proceed in lockstep. Ultimately, we will want to ask whether at the speaker level, the phonological change was abrupt or gradual, i.e. whether individual speakers had a phonological distinction when the vowel in the two environments began to as sume different phonetic targets (abrupt) or if speakers developed the 21 phonological rule sometime after the two allophones began to move in different directions (gradual). The trajectory diagnostic will allow me to determine when the phonological allophone s began to diverge from the rest of the allophones, i.e. when do the allophones begin to move towards different phonetic targets. Whichever generation in which the relevant allophones begin to differentiate, as determined by this trajectory analysis, is t he generation for which the question of abruptness will be asked. If any speakers in this generation have a phonological rule, I will conclude that phonological change was abrupt. If, however, speakers in this generation do not distinguish pre - nasal and p re - oral /æ/ phonologically, then I will conclude that phonological change was a gradual process. To determine whether an individual in Lansing posits a representational difference between pre - nasal and pre - oral /æ/, I will utilize the sub - phonemic judgemen t task described above. 1.5 Social/External A spects of P honological C hange The Berkson et al. (2017) and Fruehwald (2013, 2016) studies provide fodder for debate about the internal conditions that govern allophonic restructuring. Yet neither study provides e nough information about the social conditions under which these changes have occurred to make it possible to draw conclusions about the social motivations for change towards a phonological split. Berkson et al. (2017) do not address any aspects of contact, awareness, or social conditioning on /a / phonologization. Fruehwald (2013, 2016) argues that contact and social evaluation are not crucial components of his findings. He argues through a statistical analysis of his sample that /a / phonologization in Ph iladelphia is purely internally motivated ; there is no statistical evidence of a sudden increase in speakers with different phonological systems into the community. However, like many other large metropolitan areas, Philadelphia has likely witnessed some i n - migration of non - locals . Indeed, there has been a large amount of 22 i m migration over the last two centuries of speakers of Irish, Italian, Spanish, Vietnamese, and Yiddish linguistic background (Katz et al. 2010) . I suspect that evidence for the appearance of speakers with non - local /a / pronunciations in Philadelphia was not s analysis because (1) - only natives of the city ( Fruehwald 2016:380 ) , and (2) im migration has been a constant trend rather than a sudden event in the city . This, however remains spe culation, as Fruehwald and colleagues do not have a record of population changes in mid - twentieth - century Philadelphia (Labov, Rosenfelder & Fruehwald 2013: 61) . If speakers of /a / p roductions that were different from that in the ambient community immigrated to the city at this crucial period of phonetic target change (1930 - 1940s), then it is possible that contact has played a role in the development of Canadian Raising in Philadelphi a, though this remains to be determined. Social evaluation , however, cannot account for Canadian Raising in Philadelphi a . T hough /a / raising has recently gained status as a marker of masculinity and toughness in the city (Conn 2005; Wagner 2007) , it was not a marker in the earlier half of the 20 th century when Canadian Raising initiated (Labov, Rosenfelder & Fruehwald 2013) . Because these are the only two studies that I am aware of to observe allophonic change in an actual spe ech community, we cannot turn to any other study for examples of the social conditions within which allophonic change is actuated and then propagated throughout a speech community. Baker, Archangeli & Mielke (2011:351) hypothesize that the actuation of important because as Labov (2001: 322) states, echoing his previous sentiments in Empirical to account for the initiation of change by purely internal arguments will fail to a significant 23 degree The impact of social factors on the spread of language change has been acknowledged by every theory of phonological change, including that of the Life - Cycle of Phonological Processes and proponents of the accumulation - of - errors model. Here I quote Ohal a (2013:3) : social and psychological forces induce speakers to copy the pronunciation of the initiator among the social factors that have been proposed are the prestige of the speaker who was the initi ator or how many other speakers s/he comes in contact with Theorists of allophonic restructuring are, in fact, reasonable to assume that actuation and propagation of this kind of change is subject to the same general sociolinguistic principles that gove rn other linguistic changes. Studies of language change at every level of the grammar (syntactic, morphological, phonetic, discourse - pragmatic etc . ) have demonstrated the generalizability of these principles. In this dissertation, I will likewise demonstr ate through the analysis of 36 speakers from the Lansing speech community that the actuation and generational incrementation of nasal / æ/ allophony follows well - attested social pathways. One relevant and crucial distinction in the sociolinguistic literatur e is between language numerous cases). This dichotomy may be strongly associated with whether change towards a phonemic merger is abrupt or gradual, as I will explain below typically linguistic innovations that are introduced from outside of the speech community (via contact with speakers of another variety/language). People are aware of the linguistic feature the level of consciousness) a nd often assign it some social prestige. A notable example is the adoption of post - vocalic /r/ in New York City (Labov 2006 [1966]) , which was used proportionally more frequently by speakers when reading aloud than in their spontaneous speech. Changes from above are often led by the highest social status group . In contrast, c 24 ; they progress rather unconsciously, i.e. the speakers are not aware of the change until late in its diachronic trajectory ; and the change is led by - working - and lower - middle class) socio - economic status groups . Women, somewhat paradoxic ally, lead both types of change . Lansing is situated in the Inland North dialect area, which is defined by a vowel chain shift that has long operated as a change from below: The Northern Cities Shift. Indeed, speakers in this dialect area had been shown to perceive their variety as being identical to standard American English (Preston 1996; Niedzielski 1999, 2002) . Yet I write this description in the past tense because, a s I describe in Chapter 2 , the Northern Cities Shift is being replaced by what m ight be an externally - generated change from above. A key component of the incoming vowel system is nasal / æ/ allophony. Thus a question I ask in this dissertation is: Did the development of nasal allophony enter Lansing from some external source as a change from above? Or did it proceed as a change from below? To answer thi s question, ideally the researcher needs evidence from three diagnostics: (i) awareness of the innovation, (ii) socio - economic patterning, and (iii) degree of contact with speakers beyond the speech community . Recent studies of the regional dialect area su ggest that the change is one from above (see Becker fc for an overview) . Regarding (i) and (ii), speakers have been shown to be aware of nasal allophony, and its incrementation within the region is being led by higher status groups. Regarding (iii), I att empt to address the facts about contact in Chapter 2, section 2.4 , where I argue that there is no evidence that Lansing is a high - contact area. As I mentioned above, th e dichotom y conditions whether change towards a phonemic merger is abrupt or gradual. Phonemic mergers from above occur abruptly, while those that are from below occur gradually (Guy 1990; Herold 25 1997) . For example, a well - known phonemic merger in North America is the low - back merger the phonemic merger of / (2011b) has shown that the low - back merger in upstate New York is progressing gradually, i.e. there was pho netic movement along F2 of / Pennsylvania, however, Herold (1997) showed that the low - back merger was abrupt, i.e. the generation of speakers who identified / / and / is the first in which phonetic similarity is noted, with no gradual trend toward approximation over prior generations . Crucially, as Dinkin (2011a: 341) points out, the low - back merger in upstate New York is motivated by a gradual backing of / internally - motivated resu , while that in eastern Pennsylvania is due to i m migration of speakers who already had the merger. If allophonic changes are subject to the same conditioning, then we would expect abrupt allophonic changes to be above sp eaker awareness, to be led by higher social class groups, and to be introduced via contact. Gradual allophonic change, then, should be below awareness, led by lower social class groups, and generated from within the speech community. It is unclear what t he social facts are regarding the change towards Canadian Raising in Fort Wayne, but i m migration to Philadelphia may account for the abrupt change noted by Fruehwald (2013, 2016) , though this remains to be documented. Therefore, the dichotomy that governs change towards merger (and other linguistic changes) might also govern allophonic change. The awareness and social class facts regarding allophonic change in Lansing lead me to hypothesize that it i s a change from above, and that it will therefore have been actuated abruptly. The facts about contact, however, seem to predict that it should be a gradual change. To resolve this apparent conflict, this dissertation will bring to bear both acoustic and experimental analysis of a mult i - 26 generation, socially stratified speaker sample. The investigation of social factors will also elucidate who the leaders of this change are and how the rule spread throughout the community. 1.6 Goals of this D issertation Because allophonic change is so rarely observed, there is a wealth of information still to be gained about it that will benefit phonology and sociolinguistics. This dissertation will contribute to the overall knowledge base regarding this kind of language change. Additionally, I will show that despite the fact that allophonic change is rarely observed , many of the general principles of language change that have been established in the last fifty years apply to this linguistic phenomenon as well. The work i n this dissertation offers a transdisciplinary approach to age old questions in the field. The goals of the current project are outlined in the next sections. 1.6.1 Main G oals The main goals of this dissertation are two - fold. The first is to address the actuat ion problem of phonological change using data produced by speakers in a speech community while the change is still underway. More specifically, I am interested in whether the development of nasal / æ/ allophony was gradual, with an intermediate stage of ph onetic variation, or if it was abrupt, with no phonetic precursor. To do this, I must first address the embedding problem. Therefore, the second goal of the dissertation is to determine the acoustics of /æ/ in Lansing over the course of the 20 th century, and to examine whether any variants were socially conditioned, i.e. characteristic of particular social groups. 1.6.2 Peripheral G oal Although this dissertation focuses on the internal and social factors that govern phonological innovation, I will also conside r the Transition Problem: How does phonological change propagate throughout the speech community? Because I analyze a speech corpus in which 27 multiple generations and four social status groups are represented, I aim to generate a hypothesis about how the c hange spread. 1.7 Findings from this D issertation This dissertation finds evidence of gradual phonological change for nasal / æ/ allophony in Lansing, MI . The measure of bi impressionistic investigation of div ergent trajectories and the results of an allophonic judgement task suggest that there is indeed an intermediate period between mechanical and categorical phonology. With these measures, I show that there are speakers in the community who exhibit the seco nd stage of phonological change in The Life Cycle of Phonological Change, i .e. one that is phonetically motivated. Through an apparent time investigation of these measures, I show that the community has transitioned from a mixture of the first two stages o f the Life - Cycle to a mixture of the second and third stages of the Life - Cycle. Therefore, in Lansing, allophonic phonological change has progressed gradually such that there was a point in time when speakers did not distinguish between pre - nasal and pre - o ral /æ/. Over time, a phonetic distinction developed between the two allophones, which was then re - analyzed as a phonological distinction. An analysis of the social conditioning of gender and social class on this phonological change finds that it is still spreading throughout the Lansing community, with white - collar women leading the change and blue - collar speakers trailing behind. 1.8 Organization of this D issertation This dissertation is organized as follows. The present chapter introduced the theoretical debates about the role of phonetics in phonological change and reviewed the methodologies for teasing the two apart at the speaker level . Chapter 2 provides a socio - historical profile of the Lansing 28 speech community and a review of the literature on the ol d and new vowel systems in Lansing in particular and the dialect area in general. Chapter 3 describes the methods employed to determine how this change actuated and spread throughout the community. Chapter 4 describes the results of the acoustic analysis of speech production data from 36 Lansing natives. Chapter 5 describes the development of a phonological rule utilizing an impressionistic analysis of the rate of change to pre - nasal and pre - oral /æ/ in apparent time in conjunction with the results of a sub - phonemic judgement task administered to 107 Lansing natives. Chapter 6 provides a summary of the findings in this dissertation and situates them in the context of the literature on phonological change and North American dialectology. 29 ALLOPHONIC C HANGE IN LANSING, MI CHIGAN 2.1 Introduction With respect to the Life - Cycle theory of phonological change described in Chapter 1, Bermúdez - Otero (2015) argues that the theory can be supported via observations of the same phenomenon in multiple dialects of the same language. This is because though various dialects of a language can undergo the same change, they do so at different rates. Turton (2014 , 2017) provides evidence for this argument through her investigation into /l/ - darkening in speakers of different regional dialects of English. Through an ultrasound analysis, she shows that all stages of the Life - Cycle are repre sented across the speaker sample, i.e. each stage of the Life - Cycle describes the /l/ grammar of one of the speakers in the sample. Thus, the theory can account for synchronic variability as well as diachronic change. Fruehwald (2013, 2016), however aptly points out that synchronic evidence alone is not enough to support or refute the Life - Cycle theory. The best evidence would come from data that show a diachronic progression through these stages of the Life - ussed, diachronic data did not support the Life - Cycle theory. Yet there has been no subsequent diachronic test of the Life - Cycle theory using a dataset of speech production . This dissertation will fill this gap. In what follows, I provide a description of /æ/ systems that are characteristic of regional dialects in North America, and show that like /l/, regional variation in /æ/ systems provides synchronic evidence for a gradual theory of phonological change. I then argue that the progression of these system s over time in the Inland North dialect area, where Lansing is located, provides diachronic support for the Life - Cycle, seemingly in contradiction to 30 finding for /a / in Philadelphia. The argument is based on measures of bimodality which are c urrently being debated in the field, therefore the facts here are offered as motivation for further research utilizing additional techniques. The chapter concludes with a description of the Lansing speech community. The socio - historical facts in Lansing ca n perhaps put into perspective when and why the current changes under investigation in this dissertation are underway . This will enhance our understanding of why this change happened at this time and in this place (Weinrei ch, Labov & Herzog 1968) . 2.2 Allophonic V ariation in North America distinguish between regional dialects in North American English is that of /æ/ (Labov 1991; Boberg & Strassel 2000; Labov, Ash & Boberg 2006; Dinkin 2011; inter alia) . The Atlas of North American English (ANAE) (Labov et al 2006) is the most comprehensive analysis of regional dialects in Canada and the United States. Using recorded telephone interviews with 762 individuals, it divides the continent into major dialect regions based on patterns of phonetic and phonolo gical variation in each area . I begin with brief descriptions of the major /æ/ patterns catalogued in the ANAE, before discussing the degree to which they might support the Life - Cycle theory. (2) Raised System The raised /æ/ system is characteristic of speec h in the Inland North dialect area, which includes Chicago and Detroit and encompasses much of the Great Lakes region, including Lansing, MI. This system is described by the ANAE as one in which the /æ/ phoneme sits higher in the vowel space than / raised system in the dialect area. In its more common instantiation, pre - nasal tokens sit at the upper front of the distribution of the /æ/ cloud while pre - oral tokens are dispersed to the lower end of a single c loud of /æ/ tokens . In its more advanced realization, it is 31 described as unconditioned , so that /æ/ cloud (Labov et al 20 06:177) , suggesting that in this system /æ/ tokens before nasals, voiced stops, and after palatals are not distinguished from one another . Therefore, the difference between the realizations of /æ/ in the region is the amount o f overlap between pre - nasal and other tokens; pre - nasal tokens are either distributed at the top of the cloud (common) or interspersed with the rest of the conditioning environments (advanced). Crucially, both instantiations of this system are phoneticall y gradual; the pre - nasal and pre - oral token cloud s phonetic continuum from the least raised to the most raised (Bermúdez - Otero 2007; Dinkin 2011a) . (3) Continuous System The phonetically g radual unbroken chain is also observed in continuous systems of North America. In these systems, unlike the raised system, the /æ/ phoneme does not raise above / continuous system and t he more common raised system are similar ; there is some raising and/or fronting depending on phonological environment, however the /æ/ token cloud remains an unbroken phonetic continuum in this system (Labov, Ash & B oberg 2006: 176) . According to the ANAE (Labov, Ash & Boberg 2006) , the continuous system is traditionally observed in the Midland and Canadian dialect areas. (4) Nasal system The most common system in North America is the nasal system (Labov, Ash & Boberg 2006) . It has traditionally been observed in Eastern New England, and in parts of the W est and M idland dialect areas. In this system, tokens of /æ/ before nasal codas are abruptly separated in F1/F2 space from / æ / before oral codas. (5) Split system Another regional system, the split system, is found principally in Mid - Atlantic cities, e.g. Baltimore, Philadelphia, New York City. The distribution of raised /æ/ tokens is sub ject to complex phonological (and sometimes morphological and/or lexical) conditions that vary from city to city. In Philadelphia , /æ/ is divided into two separate phonemes; a tense vowel class made up of tokens before voiceless fricatives, voiced stops, a nd tautosyllabic nasals, and a lax vowel class elsewhere (including function words), and crucially, there are lexical exceptions, e.g. sad is lax while glad, mad, bad 32 are tense ( Labov 1994: 429 437) . Like the nasal system, the tense and lax vowel classes in the split system are clearly separated in F1/F2 space, whereby the tense class is realized higher and more forward in the vowel space than the lax class. The regional /æ/ systems described in the ANAE and in many other studies (e.g. Boberg & Strassel 2000; Durian 2012; Dinkin 2009; Dinkin 2011 a ) , resemble the stages of the Life - Cycle of Phonological Processes. Therefore, like /l/, these realizations provide synchronic evidence of the Life - Cycle. I am not the first to suggest this; Dinkin (2009; 2011 a ) and Bermúdez - Otero (2007) have both pointed out that the second and third stages of the Life - Cycle can be found in some dialects of North American English. I add, though, that the re is evidence of more than just those two stages. A description of these systems is important for this dissertation because it will provide a methodological jumping off point for which to investigate diachronic evidence of the Life - Cycle in Lansing; i.e. one of the goals of this dissertation is to provide evidence of a diachronic progression from earlier stages described here to a later stage in the community. Here, I will describe these regional systems as they have been described in the literature and h ighlight how they are representative of the stages in the Life - Cycle. When one compares the regional North American / æ/ systems to descriptions of the stages of the Life - Cycle, it becomes clear that these regional realizations of /æ/ are synchronic examp les of the Life - Cycle. Recall that the first stage of the Life - Cycle is that of a mechanical/physiological phenomenon. This appears to be close to a description of an unconditioned system, like the one found in more extreme cases of the raised system, whe reby pre - nasal and pre - oral differentiation is minimal. The second stage of the Life - Cycle is a phonetic implementation stage whereby a phonetic rule operates in a gradient manner, involving ce, such as the frequency of the 33 Bermúdez - Otero 2007). Bermúdez - Otero (2007) and Dinkin (2011 a ) cite the continuous / æ/ system as an example of this stage, in which the /æ/ cloud forms an to most raised, influenced by numerous features of the ne a :78). The third stage of the Life - Cycle is the categorical/phonological rule stage, and is claimed by Bermúdez - Dinkin (2011) notes that this is clearly a description of the nasal system described above. A categorical rule then narrows its domain to the word - level (fourth stage of the Life - Cycle ) and then to the stem - level (fifth stage) and by the sixth stage, lexical exceptions are apparent. Bermúdez - Otero (2013) cites the split system of the mid - Atlantic region as exhibiting these last stages on the grounds that the lax and tense classes are phonetically abrupt in F1/F2 space, the rule applies to a stem - level domain ( word - level affixes are ignored in determing the phonological environment of tensing ), and there are lexical exceptions. Therefore, synchronic variation in these regional / æ/ systems appears to provide support for the theory of the Life - Cycle of Phonological Processes. A study that can show systematic progression of these stages of / æ/ in a community over time will indeed find diachronic evidence of the Life - Cycle. And most crucially, evidence o f gradual phonological change toward / æ/ allophony would have to show a diachronic progression from stage 2 (phonetic distinction) to stage 3 (categorical/phonological distinction) of the Life - Cycle. An abrupt phonological change would not show evidence of stage 2 in the community. Analyses of changes to the phonetic distribution of / æ/ observed in Inland North cities, to be discussed below, suggest (at least at the 34 community level) that there has been a progression from stage 2 to stage 3 of the Life - Cycl e of Phonological Processes. 2.3 Allophonic C hange in the Inland North As mentioned in the previous section, the Inland North dialect area is characterized by the raised /æ/ system . The unconditioned raising of /æ/ is one of the movements in the Northern Citi es Shift, a rotation of 6 short vowels in phonetic space (Labov, Yaeger & Steiner 1972; Eckert 1988; Labov 1994; McCarthy 2011; Dinkin 2009b; Gordon & Strelluf 2017) , displayed in Figure 2 . The arrows display the hypothesized direction of change in phonetic space for each vowel over time. The movement of /æ/ is argued to be the first movements in the NCS chain (although this has been contested more recently, in studies that show / - fronti ng was first (e.g. Gordon & Strelluf 2017) ) . Figure 2 Northern Cities Shift vowel chain shift configuration. ( Adapted from Wolfram & Schilling - Estes 1998: 138) 35 The raising of / æ/ in the Inland North has been so substantial that many NCS speakers have risen /æ/ to mid - front position in the vowel space and often articulate it as a diphthong (Labov 1994; Labov, Ash & Boberg 2006) . This acoustic description has led Labov (1994) to conclude that /æ/ in the Inland North behaves as a tense vowe l : it is produced with a n offglide and diachronically raises along the periphery of the vowel space . Raising and fronting are well documented i n the literature on the NCS (see e .g Ito 2001; G ordon 2001; Dinkin 2009; Dinkin 2011 a ) possibilities. Additionally, although it has not often been discussed in the literature on the NCS (although see e.g. Labov 1991 fo r an exception), I will include diphthongal quality , i.e. change in as a diagnostic of NCS / æ/ . The comparison point for NCS / (Becker fc ) is replacing the Northern Cities Shift in the Inland North in what appears to be a supralocal change . As a result, the NCS /æ/ raised system is being replaced with the nasal system. Along with the lowering and retraction of / / and / / and the progress towards merger of / - oral / æ / such that younger generations of LBMS speakers almost exclusively exhibit a nasal system. The LBMS has been documented with ever greater frequency across North America since the 1990s. First observed in The West and Canada (e.g. Clarke, Elms & Youssef (1995) , Hagiwara (1997) , Labov (1991), Boberg (2008) , and see Becker ( f orthcoming ) ) , it has more recently been observed in the Midland dialect area (see for example Boberg and Strassel (1995) , Bigham (2008) , Durian (2012) , Strelluf (2014) , Holland (2008) , Kohn & Stithem (2015) ), in New England (Stanford 2019) and surprisingly, in the Inland North: Syracuse, NY (Driscoll & Lape 2015) ; O gdensberg, NY (Thiel & Dinkin 2017) ; Buffalo, NY (Milholla nd 2018) ; Rochester, NY (King 2017) ; 36 Detroit, MI (Morgan et al. 2017) ; Grand Rapids, MI (Rankinen, Albin & Neuhaus 2019) ; Chicago, IL . Notably for the present study, it has also been recorded in Lansing, MI (Wagner et al. 2016; Nesbitt, Wagner & Mason fc; Nesbitt 2018) . Section 2.4.2 below provides more details on the Lansing findings. Movement towards a nasal system has been observed at the community level in each of these analyses. Thus, as Nesbitt & Mason (2016) point out, though the adoption of the nasal system and loss of the NCS is being observed in the Inland North, they appear to be part of a larger trend of regional dialect levelling in North America. The social circumstances surrounding NCS decline and LBMS adoption in Lansing are as yet unclear. However, so cial conditioning in the Inland North ern studies cited above, combined with preliminary analyses of social distribution in Lansing, lead to hypotheses that can be tested in this dissertation . I discuss the prelimi nary findings regarding social conditioning of / æ/ in the Inland North below . The results of preliminary analyses of Lansing speech are presented in s ection 2.4 . Section 2.5 provides a summary and sets up the hypotheses to be tested in Chapter 4 . Labov (1971) suggested a basic three - w ay taxonomy for distinguishing sociolinguistic variables according to their level of awareness in the community and the social meanings they index. Linguistic stereotypes are well above the level of awareness, i.e. speakers use them stylistically and they are often commented on overtly by members of the speech community. One example of a stereotype is eh in Canad ian English (Gold & Tremblay 2006) . Linguistic markers are one level down on the awareness continuum. These variabl es exhibit both interspeaker variation across social groups and style - shifting, but they are not mentioned in metalinguistic discourse. The last type of variable are indicators they may index group affiliation at the 37 interspeaker level but are not subject to intraspeaker stylistic variation. The entire NCS chain shift appeared to be a classic example of a linguistic indicator (Labov 2001:196) operating well below the level of awareness (Preston 1996; Niedzielski 2002) . Until recently, the NCS showed no signs of style - shifting or social class stratific ation, i.e. speakers of all social classes were participating in this change. In fact, many studies on the dialect have utilized data from word list speech because its features appeared to be prevalent even when speakers were talking in their more formal speech styles. The distribution of raised versus nasal /æ/ systems across social class groups in the Inland North has not been subject to much analysis thus far , though three studies note its potential relevance . In Syracuse, NY, Driscoll (2016) found a no n - statistically significant association between mean F1 and F2 of /æ/ and speaker education and occupation. She analyzed wordlist data from 50 Syracuse natives aged 18 to 89. In her figures, she shows a slight correlation between higher paying jobs and low er education and /æ/ - lowering and /æ/ - backing (as well as for / / - lowering, and / - backing), but these relationships were not significant in her statistical modeling. One of the issues with the analysis is that the sample was not evenly balanced for educa tion and occupation, i.e. many of the younger speaker s were skewed towards white - collar status. The other issue for this analysis was that social class was analyzed across all speakers in the sample rather than as an interaction with date of birth. Driscoll and Lape (2015) in their preliminary study, and Driscoll (2016) , observe that the decline of NCS participation in Syracuse was initiated around speakers born in the 1980s, therefore I speculate that an investigation into social class would benefit from excluding those born prior to 1980. P erhaps with this methodological revisions, the analysis would have resulted in a significant effect of social class on the realization of NCS features in their sample, with the loss of NCS features being led by 38 speakers with more education and more prestig ious occupations. Thiel and Dinkin (2017) find a similar pattern in Ogdensburg, where /æ/ - lowering appears to b e driven by white - collar speakers, although the education effect does not reach the level of statistical significance. In a larger - scale study of Chicago, there is clear evidence that social class is a factor in NCS recession. In their acoustic analysis of NCS features in the spontaneous speech of 50 Chicago natives born between 1875 - 1990, Durian and Cameron (2018) show that NCS recession began among speakers born in the 1970s and is thereafter always more advanced in white - collar speech than in blue - collar speech. Thus, in Chicago, it appears to be predominantly white - collar speakers who are reversing NCS features. Over Syracuse. Younger Syracusans associate raised /æ/ with older speakers, referencing grandparents or older school teachers as having this nasally, harsh/hard A accent . One res pondent described 2016:81). The qualitative comments in Syracuse are supported elsewhere in the Inland North by quantitative evidence of a shift in the directio n of style - shifting over apparent time. In their analysis of /æ/ - raising in Ogdensburg, NY, Thiel and Dinkin (2017) compared the means of 4 3 speakers while reading from a word list and during conversational speech. They showed that older speakers (born b efore the 19 7 0s) raised /æ/ more in their word list speech than in their conversational speech, suggesting that for them, raised /æ/ is the more prestigious pronunciation, while those born after the 1980s were more likely to lower /æ/ while reading from a word list than in their conversational speech. 39 Thiel and Dinkin (2017) also administered an attitudes survey to their participants. For this, participants listened to a male talker and rated him on Likert scales of localness and educatedness. Crucially, t here were multiple talkers who had similar acoustic characteristics but who differed in whether their sound clip was digitally edited to have a raised or unraised /æ/ . The researchers compared ratings of the two types of voices (raised and unraised). They found that listeners born before 1960 did not rate raised and unraised /æ/ differently on these affective scales. Listeners born after 1960, however, rated raised / æ / as more local but less educated than unraised / æ /. Thus, in Ogdensburg, raised /æ/ has apparently risen from an indicator to a marker. In sum, based on attitudinal, style - shifting, and social class data in the dialect area, there is evidence that the NCS (or at least /æ/) has risen in markedness. As Nesbitt and Mason (2017) argue, this has likely motivated younger, mostly white - collar, speakers to reverse NCS features ; a scenario which is expected when an indicator rises to a marker (Labov 2001) . Therefore, the change towards the LBMS appears to be one towards a more prestigious form. Considering this, I hypothesize that the propagation of LBMS features throughout the Lansing community will be led by the middle - class and especially by women in that social class, as it appears to be in other Inland North communities and in other changes from above. 2.4 Lansing Speech Community Lansing, Michigan is a particularly appropriate site for an investigation into a llophonic change. Here I provide a sociohistorical description of the Lansing area. A discussion of preliminary findings regarding Lansing speech and its social conditioning is included in section 2.4.2 2.4.1 The Speech Community Incorporated in 1859, Lansing is the capital of Michigan. It is in the center of the lower part of the state , located approximately ninety minutes drive west of Detroit, one hour east of 40 Grand Rapids, and three hours east of Chicago, Illinois. Lansing is in Ingham County, which together with Eaton and Clinton counties comprise the metropolitan area of Greater Lansing or - Although it is called a ansing is situated in the middle of low - population - density farming land (130 to 500 people per square mile) , dotted with small towns rarely exceeding 8,000 inhabitants. The closest city to Lansing with a population of more than 50,000 is Battle Creek (pop. 51,286: U.S. Census 2017), which is 50 miles south - west of Lansing. Today, Lansing is a mid - size city of 114,000 inhabitants (US Census 2010), and like many cities in North America, its population is more ethnically diverse than its surrounding neighborhoods and towns; 60% White, 24% Black and 12% Latinx. The population of Ingham County, which includes the college town of East Lansing, is 80% White. Consequently, the city of Lansing, where the speakers in this dissertation were born and raised , is best described as a mid - sized city that is surrounded by less diverse suburban towns which are in turn surrounded by sparsely populated farmland on all four sides. Lansing, like most other Inland North cities, is part of the Rust Belt (see e.g McCle lland 2013 ), a collection of manufacturing communities in the north and north eastern United States that witnessed significant economic and population increases after World War II, but whose economies suffered as the manufacturing industry gave way to the s ervice industry in the 1990s. Lansing is an automotive town. It is where R.E. Olds set up shop and established the Oldsmobile brand, which was a staple brand in Lansing from 1901 to 2004. It is also where General Motors , a world leader in auto manufactu ring at the turn of the 20 th century, has housed many of its assembly plants, its headquarters from 1965 to 1998 , and its body plant Fisher Body (Coase 2000) . 41 The auto industry boom in Lansing from the early to mid - 20 th century prompted a rapid increase in population. The city of Lansing was rather small at the turn of the century (pop. 16,485: U.S. Census 1910) but reached its peak in the latter part of the century (pop. 131,403: U.S. Census 1970) due to jobs created b y these companies. This boom, however, was followed by general population decline due to the dissolution of auto industry jobs in the city starting in the 1980s. Figure 3 Population r ate of change by decade from 1860 to 2010 in Lansing. [ Source: US census 1860 2010]. Figure 3 displays the change in population by decade from 1860 to 2010 in Lansing as reported by the US Census. From 1850 to 1930, the city of Lansing experienced significant 42 population growth, with an average of 71.55% increase per decade (26.75% more than the state average). From 1930 to 1940, population growth plateaued to .45%, reflecting the large - scale economic stagnation of the national Great Depression that affected areas with prominent manufact uring industries with particular intensity. Population rose during the 1950s, 1960s, and 1970s by 18.26% on average until it fell by 3.18% on average in each of the following decades. Because one of the aspects of this dissertation is to discern whether allophonic change is population growth and was their speech different from the existing (white) Lansing population? There was no substantial in - migration from outs ide the Great Lakes area in the 20 th century (Fine 2004) . Therefore, three other groups with potentially non - local speech patterns must be considered in turn: White rural Michiganders, non - White minorities , Midland dialect speakers, and those moving to the area to attend the local university (Michigan State University) . Much of the population growth in the first half of the 20 th century was due to the in - farming towns (Fin e 2004). It is conceivable that the concomitant dialect contact between rural and urban speakers actuated the development of the nasal system in Lansing. This, however, is an unlikely scenario since rural and urban speakers in Michigan likely did not hav e significantly different accents during this time, as sociolinguistic research in Michigan rural towns has shown that rural Michiganders also participate in the NCS (Ito 2001; Gordon 2001 ) . Additionally, it is unclear why nasal system adoption would be led by high socio - economic status community members in the city, since rural speech is not usually positively socially evaluated (Preston 1996; Seale & Mallinson 201 8) . Another thing to consider is that larger Inland North cities like Detroit 43 and Chicago, which saw proportionally less rural in - migration in the early 20 th century than Lansing (Fine 2004), are also exhibiting this change toward nasal / æ/ allophony . A dditionally, like many northern cities, Lansing received some minority in - migrants many of whom were African Americans . But unlike in those other cities, the number of non - White residents in Lansing remained relatively low. Figure 4 displays the percentage of non - White Lansingites from 1960 to 2010 , as reported by the US census . Though there is a steady incline from 10% to 40% over the course o f fifty years, the non - white population stays below 20% up until the 1980s. And although Lansing was a less residentially and educationally segregated city than many others in the 20 th century (Fine 2004), it is unlikely that the minority non - White popula tion exerted significant linguistic influence over the White majority . Figure 4 Percentage of Non - White population in Lansing. [Source: US census 1960 - 2010]. 44 In - migration from Midland states into Lansing is also another possibility, since many Michigan cities experienced in - migration from Ohio, Indiana, and Illinois. According to the US census, the percentage of Lansing residents born in other states rose from around 16 to 18% at the beginning of the 20 th century, to 25% in 1950, before slowly declining again to early 20 th century levels by the end of the century. In - migration into Lansing from other states was never substantial, however . From 1950 to 2010, le ss than 10% of Lansing residents per decade reported that they lived in a different state in the previous decade . Lastly, I consider whether Michigan State University (MSU) might have contributed to a scenario in which large amounts of non - local speakers might interact with the speakers living in neighboring Lansing and therefore provide enough variability to induce change . Two crucial po ints about the MSU student body raise doubt about this possibility . First, the international student body constitute s a very small proportion of the MSU population ( 13.7% in 2017) (Michigan State University 2018) . The se students often reside in East Lansing , which is a neighboring town of Lansing. They leave the area upon finishing their degree and are less likely t o interact with speakers born in the actual city of Lansing since East Lansing is not only where they live and where the university is, but it is also a center for social events, shopping, etc. for its residence . Another source of potential contact are non - local college students coming to MSU for school. Crucially, however, 72.7% of the domestic students at MSU ar e indeed Michigan natives themse lves (Michigan State University 2018) . Therefore, th ough Lansing is just miles away from a state university , the university is not a likely locust of large amounts of contact that are needed to promote phonological change. In summary, Lansing does not appear to have experienced a dialect contact scenario of the type described for linguistic changes from above, whereby supralocal (or simply non - local) 45 sociolinguistic features are adopted from a dialect w ith higher overt prestige than the local dialect. Nor does dialect contact appear to have been sufficient to trigger leveling toward a non - prestigious majority norm (cf . Kerswill & Williams 2000). In the absence of a clear external model for the /æ/ nasal system, therefore, what community - internal social motivation might there have been for its adoption? That is, if we cannot clearly establish that White Lansing speakers were actively adopting the nasal system as an exogenous feature, can we provide some su pport for the idea that they were at least rejecting the raised /æ/ system as an endogenous feature? There are numerous examples in the sociolinguistic literature of speech communities losing features of their dialects in response to changes in the economy and education system (e.g. Schilling - Estes & Wolfram 1999) . Here I trace the major socioeconomic developments in 20 th century Lansing, before discussing the likelihood of those changes being motivating forces for the shift from raised to nasal /æ/ patterns in this speech community. In the U nited States , although manufacturing industries before the 1930s offered low - payin g working - class jobs, the development of labor unions led to relatively high - paying working - class jobs from the 1940s to the 1970s (Knox and Pinch 2006). This is illustrated in interviews with some of the older blue - collar speakers in the current sample ( Chapte r 3 ) who discuss how their salaries rose to $70,000 after the 1970s and 1980s labor strikes in Lansing. Accounting for inflation, this equates to $384,687.55 today (see CPI calculator at Bureau of Labor Statistics) , which would place one well above working - class blue - collar status even today considering the US median income in 2017 was $60,336 (U.S. Department of Commerce 2017) . During this period of economic growth for factory workers , White residents who were now equipped with higher salaries accelerated 46 communities in America a desire to raise families away from cities, where crime, drug use, and the in - migration of minority groups was on the rise (or was perceived to be on the rise). According to the US census, th e city of Lansing was 93.51% White in 1960 and as of White population is only 61.23% . The largest occurred between 1970 and 1980 (i.e. during the auto manufacturing crisis in Michigan) . Figure 5 shows population change in the city of Lansing compared to Delta and Bath throughout the 20 th century. average of 76.82% per decade at a maximum of 127.19% from 1960 to 1970. Although Delta has become less White over time, it is significantly more White than Lansing. In 1980, 94.9% of Delta was White - White populat ion, particularly its African American population, has steadily increased since the 1960s. The differential access to capital allowed the White laborers to move away from the urban center when it became less desirable whereas the non - White populace was lef t to choose from whatever jobs that did not require a college education (mostly in lower paying sectors). 47 Figure 5 Rate of change in population by decade from 1860 to 2010 in Lansing compared to two Lansing suburbs; Delta Charter Township and Bath Charter Township. [Source: US census 1860 2010]. Economic decline began in the 1980s when General Motors rgest employer at the time, slowly began to shift its plants and offices to Detroit (Coase 2000; McClelland 2013) . The first sign of economic shift was in 1984 when General Motors moved its decision making to Detroit , and eventually mov ed its headquarters there in 1998 (Coase 2000) . In 2000, , Oldsmobile , would be phased out and that they would be closing many its factories in the city. With the dissolution of many auto - manufacturing plants in the city came an increase in service sector jobs. At the time of writing, 48 the largest employers in the Lansing are the state government and the hospital system (Lansing Economic Area Partnership). Crucially, the jobs in the service sector require at least some college education (Rosen 2007) . Therefore, working - class autoworkers in Lansing have either had to choose from whatever jobs that did not require a college education (mostly in lower paying se ctors) or leave the city. Many of these workers were forced to go back to their family farms, while some stayed to navigate this new economy (Fine 2003). The economic impact of this transformation has had many social consequences for these communities and for the Since 2000, the unemployment rate in Lansing has increased by 8% and as described above, the population has decreased by 15% (U.S. Census 2010). The Baby Boomer generation (born 1 946 1960) was most directly affected by this industry transition (McClelland 2013; Knox and Pinch 2006). They would have entered the work force in the 1970s and 1980s, inheriting the traditional manufacturing lifestyle of their parents, but they would h ave experienced the brunt of economic decline in the 1990s and 2000s when they were in their 40s and 50s. Generation X (born 1964 1985) would have in turn inherited dismal socioeconomic circumstances, with less economic opportunity in the city. As I no te below, my preliminary analysis finds that many of the linguistic changes that occurred in Lansing took place in this latter half of the 20 th century. Therefore, this dissertation is especially interested in the social conditioning of these changes in th ese two generations. 2.4.2 Preliminary A nalyses of /æ/ in Lansing Even though the White population constitutes the majority ethnic group in Greater Lansing, there has been no substantial investigation of local White phonology. Previous studies have reported on the speech of African American (Jones 2003) and Mexican Ameri can speakers 49 (Roeder 2006, 2010) in Lansing. Only a small number of White speakers were included for in collaboration with my colleagues at Michigan State University and our advisor, Dr. Suzanne Evans Wagn er, and for this dissertation therefore constitute the first large - scale sociolinguistic studies of speech in the White population of Greater Lansing. Here I present the findings from one collaborative and one single - authored analysis of conversational da ta in Lansing. Both analyses utilize data from contemporary and historical speech as part of the Impact of Higher Education on Local Phonology project (henceforward IHELP), funded by US National Science Foundation grant #BCS - 1251437. For this project, we trained undergraduate students who were born and raised in Greater Lansing to conduct sociolinguistic interviews with their Lansing - native friends and family. Sociolinguistic interviews with 30 Lansing natives born 1989 to 1999 were conducted. To get a view of speech in the past, we obtained oral histories (n=21) of Lansing - native auto plant workers recorded in the late 1990s and early 2000s (G. Robert Vincent Voice Library). Speakers in this collection 1961. T he range of subjects and birth dates allowed us to identify changes in local speech from the NCS to the LBMS pattern over the course of 100 years. Our preliminary analysis found that / / is lowering in apparent time, cons istent with both the NCS and the L BMS, while / Regarding /æ/, we found a progression from a n advanced raised continuous system to a common raised c ontinuous system to a nasal system over apparent time . Figure 6 shows three different - configurations in conversational speech (sociolingui stic interview or oral history), arranged left to right by year of birth. Red circles represent tokens with /æ/ realized before nasal 50 consonants; blue triangles represent tokens of /æ/ realized before oral consonants. On the left is the token cloud for Ja ck Down, born in 1924. He has the expected raised system (Labov, Ash and Boberg 2006, Dinkin 2011a), which shows no phonetic tendency for raising pre - nasal above pre - oral tokens. As we move forward in apparent time, the more common raised continuous /æ/ s ystem become more common, like the configuration of Michelle Baulch, born in 1971, in the center facet of Figure 6 . In this system, although there is sti ll a large degree of overlap between pre - nasal and pre - oral token clusters, there is a tendency for pre - nasal tokens to be more raised than pre - oral tokens in the /æ/ cloud. Moving ahead even further in apparent time to speakers born in the 1990s, we encou nter the nasal system, exhibited on the right in Figure 6 by the distribution for Ben Langdon, born in 1994. His pre - nasal and pre - oral token clouds barely overlap. Figure 6 Three distinct /æ/ systems from Greater Lansing: Jack Down, b. 1924, multiple college degrees , advanced raised continuous system; Michelle Baulch, b. 1971, multiple college degree s, common raised continuous system; Ben Langdon, b. 1994, community college student , nasal system. 51 Figure 7 shows the apparent time shift to this more nasal - like /æ/ - system using Pillai - Bartlett scores. The Pillai statistic measures the difference between two clusters of vowels; low scores (approach ing zero) indicate a large degree of overlap between two clusters, while higher scores (approaching 1) indicate a very small degree of overlap (see also Chapter 3). Using this statistic, we measured the overlap between pre - nasal and pre - oral token clouds f or each speaker. Figure 7 Pillai - Bartlett scores for /æ/ by year of birth for Lansing, MI speakers (from Wagner et al 2016:figure 5) The degree of overlap between pre - nasal and pre - oral token clouds de creases as year of birth increases , indicating that Greater Lansing speakers have gradually shifted from a n 52 advanced raised (stage 1 of the Life - Cycle) to a common raised continuous (stage 2 of the Life - Cycle) to a nasal (stage 3 of the Life - Cycle) /æ/ system. Recall that Jack Down (Figure 4) showed virtually no difference between pre - oral and pre - nasal TRAP. He has an extremely low Pillai score of 0.07. Ben Langdon, by comparison, has a high Pillai score of 0.73, indicating very little overlap. Judging from the appare nt time trends of these vowels and the configurations of the /æ/ systems of white - collar speakers born in the Millennial generation we surmised that Lansing was moving toward the nasal system. In Nesbitt (2018), I expanded upon the analysis of /æ/ in th e Wagner et al (2016) study. First, I wanted to focus on the speakers in the sample who were born and raised in the city of Lansing. This was because the NCS has been found to be less advanced outside of larger cities (see Ito 2001, Gordon 2001). Second , because of the emerging evidence of a change in evaluation of the NCS in other Inland North cities, I wanted to investigate whether the Lansing speech community had begun to assign negative social meaning to NCS /æ/ - raising and/or fronting. My third mot ivation was to provide a more specific time window for NCS adoption and decline in the community. I therefore conducted an analysis of /æ/ on a subset of the Wagner et al (2016) sample which consisted of the 27 speakers who reported that they were born an d raised in the city of Lansing rather than in any of the surrounding rural towns (Nesbitt in press). This analysis identified that the change away from the NCS - like raising/fronting of /æ/ and the raised system in the city started with speakers born in t he Baby Boomer generation (date of birth 1946 1 964). Before this generation, urban Lansingites of all social class backgrounds had /æ/ systems that were more raised than that of their rural peers each had higher and fronter realizations of /æ/ as compare d to / - nasal and pre - oral /æ/ in acoustic space. In the Baby Boomer generation, however, only blue - collar 53 speakers continued advancing the NCS pattern while white - collar speakers did not. Figure 3 displays h ow /æ/ has moved along the front diagonal over time across social class. The front diagonal measurement captures movement along F1 and F2 simultaneously, with a higher number indicating a higher and more forward realization (normalized F2 2* normalized F 1, see Labov, Rosenfelder and Fruehwald 2013). In Figure 8 , we see that /æ/ remains relatively stable in the white - collar community (red) until it decr eases in the Millennial generation. Though there is no data from blue - collar speakers born after 1970, I observed that /æ/ raising and fronting notably peaks only for blue - collar speakers in the Baby Boomer generation where values are outside of the range of any other generation in the community. Figure 8 Trajectory of /æ/ diagonal in urban Lansing (blue = blue - collar; red=white collar); dots represent means for individual speakers (n=27). 54 Phonological contrast as measured by Pillai - Bartlett statistic (Hay, Warren and Drager 2006; Hall - Lew 2010 and see Chapter 3) is displayed in Figure 9 wi th social class indicated by color (red = white - collar, blue=blue collar). From the figure, it appears that movement toward nasal separation in the white - collar sub - sample occurred in the middle of the 20th century, where we find Pillai scores increase fr om almost 0 in earlier years to 0.35. Blue - collar speakers, however, appear to remain well below 0.1 over time, save one blue - collar speaker in the Boomer generation who seems to pattern with the white - collar speakers in her generational cohort. Without a more robust and socially stratified sample, however, I was not able to say with confidence whether social class conditions nasal separation in Lansing. Figure 9 Pillai - Bartlett scores in 20th century Lansing by social class (blue - collar = blue). 55 More evidence of social motivation comes from a quantitative analysis of style - shifting within the younger white - collar sample of the Wagner et al (2016) participants. In collaboration with Alexander Mason, I reported on a rise in dialect awareness in Mic higan (Nesbitt & Mason (2016) sample while they were speaking spontaneously during their interview to that when they were reading a word list. Since there is no word list data from the speakers in the AutoTown oral histories collection, t his analysis included only a subset of 29 speakers who were all born in the 1990s. This style - shifting analysis showed that young speakers in Lansing were more likely to lower and retract pre - oral /æ/ but raise and front pre - nasal /æ/ while reading from a word list than in their conversational speech. Though attitudinal data in Lansing is only really robust for younger speakers, the addition of the social class analysis from Ne sbitt (2018) suggests that in Lansing, /æ/ has risen as a linguistic marker as it has in other Inland North communities. These two pieces of information combined with the apparent stratification in /æ / - raising/fronting suggest that the development of nasa l allophony will be socially constrained. There is also supporting qualitative evidence from some of the speakers themselves, particularly those aged 20 4 0 . I n the IHELP interviews and in additional interviews conducted for this dissertation , young speak ers she travels. 2.5 Summary To summarize, the Wagner et al. (2016), Nesbitt and Mason (201 6), and Nesbitt (2018) analyses allow for the generation of hypotheses about the actuation and propagation of allophonic change in Lansing. First, the change has likely occurred gradually. Through the use 56 of the Pillai - Bartlett analyses, we showed that sp eakers exhibit the first three stages in the Life - Cycle of Phonological Processes and crucially, there appears to be progression from stage 1 to stage 3 in apparent time. Second, the change appears to be socially motivated, as white - collar speakers almost exclusively exhibit stages 2 and 3, while no blue - collar speakers in the sample exhibit stage 3. Therefore, this dissertation will extend the analyses in these preliminary studies and include social class and gender as conditioning factors in this change. The Pillai - Bartlett score analysis was a good initial step in exploring /æ/ systems in Lansing. Yet researchers have shown that while bimodality is often an indication of phonological/categorical differences, the absence of a bimodal distribution does n ot entail the absence of a phonological difference (cf. Bermúdez - Otero & Trousdale 2008: 696; Schilling, Watkins & Watkins 2002). Therefore, for this dissertation, an analysis of Pillai - Bartlett scores over time and across social groups will be supplement ed with the impressionistic analysis of rate of change (Chapter 4) and the results of the sub - phonemic judgement task (Chapter 5). The inclusion of these analyses will allow a more definitive answer as to whether phonologization was abrupt or gradual and to address the broader question of Why changes in a structural feature take place in a particular language at a given time, but not in other languages with the same feature, or in the same language at other times. 57 DATA AND METHODS 3.1 Introduction The goals of this dissertation are to address the Actuation problem of allophonic change in Lansing, i.e. why does change occur at a particular place in a particular time. In order to do this , I also address the Embeddin g and Transition Prob lems, i.e. what are the internal and external conditions surrounding the chang e and how did the phonological change spread throughout the community. The Embedding P roblem is addressed in Chapter 4 where I examine four diagnostics of NCS /æ/ in the spontaneous speech of thirty - six Lansing natives : raising and fronting in F1/F2 space, advanced raised system, and diphthongal quality , again st generational time, social class, and gender . I report on the methodology for this component of the dissertation in section 3.2 of this chapter . Along with the Actuation and Transition problems, t h e E mbedding P roblem is further addresse d in Chapter 5 where I examine phonological change via an analysis of acoustic trajectories of pre - nasal and pre - oral /æ/ allophones and the results of a sub - phonemic judgement task. I report on the methodology for this component of the dissertation in section 3.3 . The findings from both of these chapters contribute to a discussion of how and why this change may have occurred , and their implications for our theories of North Ameri can dialectology and phonological change will be discussed in Chapter 6 . 3.2 Acoustics of /æ/ in 20 th century Lansing My previous research in Lansing suggests that the most significant changes to /æ/ occurred among individuals born in the middle of the 20 th century , at which point gender and /or social class condition its acoustic trajectory (Wagner et al. 2016; Nesbitt 2018a; Nesbitt, Wagner & 58 Mason to appear) . In that prior work, the paucity of speakers in the database who were born in the 1970s and 1980s , and the imbalance of gender and social cl ass in the Wagner et al. (2016) sample , made it hard to conclude with certainty when and for whom NCS features have begun to subside. Therefore, the goal of the acoustic analysis in this dissertation is to examine the four selected diagnostics of NCS /æ/ in Lansing with a n expanded speaker sample that is balanced for gender, social class, and generational cohort . Throughout, I will refer to this sample of thirty - six speakers as the Lansing Speech Corpus . Section 3.2.1 describes how I recruited participants for this study ; the demographic distribution of speakers in th e Lansing Speech Corpus is detailed in section 3.2.2 . I nterview methods are described in section 3.2.4 . T he procedures I followed for processing and transcribing the interviews appear in section 3.2.5 , while those for vowel measurement and s tatistical analysis are in section 3.2.6 . 3.2.1 Participant Selection and Recruitment The data for the Lansing Speech Corpus comes from three sources: oral histories recorded in the late 1990s and early 2000s, sociolinguistic interviews conducted in 2010, and sociolinguistic interviews conducted in the summer of 2018 and spring of 2019 . So me (N=12) of the oral histories data in this dissertation come from the Wagner et al. (2016) and Nesbitt (2018) studies and 4 others are newly added; the other sources represent new sociolinguistic information about Lansing speech. I first sampled from the oral histories c ollection . Any speaker in th at collection who met the demographic criteria described below in section 3.2.2 was included in the corpus (n=16) . I then filled in the gaps in the corpus with twenty sociolinguistic interviews , for a total of thirty - six speakers . 59 The oral histories come from a collection of interviews with Lansing auto industry workers, to be described in more detail in section 3.2.4.1 . More information about recruitment methods for this collection can be found on the Michigan State University Vincen t Voice Library website where the collection is hosted and in Fine (2004) . The two 2010 sociolinguistic interviews were conducted by an undergraduate student for (Fraser 2010) . One of these interviewees was recruited at a senior coffee hour in the neighborhood. The second was a friend (and neighbor) of the first , who recommended her as a participant. For the 2017 2018 interviews, some (n= 12 ) were conducted by a local fieldworker who is a longtime Lansing resident, and some (n= 6 ) were conducted by me. Ten participants were recruited by the fieldworker through personal social network s . The participants a re friends, co - workers, children of friends, and neighbors of th is i nterviewer. The remaining eight interviewees responded to my recruitment flyer , which is provided in A ppendix 0 . I distributed this flyer at local businesses and on social media groups of various Lansing neighborhoods. Because the neighborhood can post to other members of the group, I contacted the moderator (s) of these groups and asked them to post my flyer on their page. Some ignored my request but others were happy to help. Each participant was compensated $15 for participating in an hour - long sociolinguistic interview. The demographic profiles of the th irty - six speakers in the Lansing Speech Corpus are provided below. A complete table of speakers in the sample with their demographic information and source of interview is provided in appendix 0 . 60 3.2.2 Speaker Demographic Information Speakers included in the Lansing Speech Corpus are all native s of Lansing. For this dissertation, Lansing is defined as encompassing all residences located wit hin a 10 - mile radius of the state capit o l building, if they had lived in Lansing from at least five years old and had not left the area for more than 3 years. Another requirement o f native status is that at least one parent be born and raised in the Lansing metropolitan area and the other parent be from another lower Michigan town 1 . Table 2 provides a summary of the distribution of the social parameters in the Lansing Speech Corpus. A description of how the levels for each demographic category were determined is provided thereaf ter . Table 2 Distribution of the Lansing Speech Corpus by generational cohort, social class, and gender. Generational cohort Blue C ollar White Collar Silent (birth years 1925 1945) 3f 3m 3f 3m Baby Boomer (birth years 1946 1964) 3f 3m 3f 3m Generation X (birth years 1965 1984) 3f 3m 3f 3m 1 I would have ideally liked to limit the pool of speakers to those for whom both parents were born and raised in Lansing . This was impossible because of (1) the limited demographic information provided by oral history interviewees , and (2) the fluctuation i n residency due to the in - migration of Michiganders into Lansing in the first half of the 20 th century followed by mass exodus in the latter half (Fine 2004) . I do not think the absence of two Lansing native parents is an issue for my current analysis because recent analyse s in lower Michigan show similar trends of NCS participation (followed by recession) to that in Lansing (Morgan et al. 2017; Rankinen, Albin & Neuhaus 2019) . Therefore, the parents of my interviewees would probably have had similar /æ/ productions. 61 . For this dissertation, I use cultural generations rather than birth year or decade of birth, as is common in socio linguistics research, so as to capture how cultural change in the US ( which as discussed in Chapter 2 also applies to Lansing ) has impacted language use. Though this is a less common technique, many sociolinguists have used cultural generations to divide their sample (cf. Dubois & Horvath 2000; Durian 2012 ; Dodsworth & Kohn 2012; Dodsworth & Forrest 2016) . There is no consensus in social science research or in public discourse about the precise limits on popular generational labels, however (indeed, generational gro (Pew Research Center) ). I have employed the date range for Baby Boomers in Knox and Pinch (2006) and made the generations on either side uniform 20 - year periods. The three generational cohorts in this dissertation are the Silent generation (born 1925 1945), the Baby Boomer genera tion (born 1946 1964), and Generation X (born 1965 1985). In this dissertation, I also use data from the 21 Millennial generation speakers (born 1985 199 9 ) in Wagner et al (2016) as a temporal reference. I do not include them in the statistical analyses below because, as discussed in Chapter 2 , this g roup is not balanced for social class or gender. The majority of Millennial speakers here are women (N= 18 ), and they are all middle - class. I include them in the figures below to provide a trajectory of (female) white - collar speech after generation X and as a comparison point to the white - collar female women who participated in the sub - phonemic judgement task . Working - and middle - class Lansingites were targeted for this dissertation. I made the decision to focus on these two groups because (1) Lansing does not have a traditional old money upper class (cf Kroch 1996 for the speech of the Philadelphia upper class) and (2) the se groups, a s opposed to the upper and under class, are often the innovators of linguistic change in 62 urban areas (Labov 2001; Baranow ski 2013) . Classification of soci al class has been and still is highly debated in sociolinguistics and sociology more generally. Because there is no unifying methodology, studies using social class as a variable have utilized metrics that include some combination of median income of a neighborhood, occupation, education level, income, median occupation and/or education level, or job - economic status . O ccupation appears to be the most - economic status in more recent work (Labov 2001; Forrest & Dodsworth 2016; Baranowski & Turton 2018) . Furthermore, Holley, Dodsworth & Wagner ( to appear) argue that occupation is predictive because of the way it tends to capture social prestige and Lastly, occupation seems to be the most salient indicator of social class in the Lansing community , as sociolinguistic intervi ew ees were more likely to discuss occupation than education or material wealth when topics of social class came up. As such, I use occupation as my measure of social class for this dissertation. Speakers in manual labor occupations (e.g. car body assembl y, construction worker, machine operator), or in low - level service and care jobs (e.g. wait staff, daycare worker, receptionist) that require less than 2 years of training after high school were classified as blue - collar . Those in managerial positions and in occupations that require at least 2 years of technical training outside of high school (e.g. chemical engineer, financial manager, information technology supervisor, plant manager) were classified as white - collar . Two speakers in the sample were stay - a t - home parents, so I classified them according to for both was white - collar. It is worth mentioning here that the social class status of most of the participants in this study matched that of their parents, i.e. those 63 who grew up blue - collar were themselves blue - collar. The one exception is a female Generation X speaker whose parents were blue - collar but who is themselves white - collar . Participants were classified as either male or female. Though I recognize that gender is a gradient variable rather than dichotomous, I classified gender here as binary in line with how participants referred to themselves in the demographic part of the in terview. Speakers in the sample were often interviewed by a stranger, so it may very well be the case that some did not feel comfortable enough to disclose intimate details of their identity. So, I cannot say definitively that there was indeed a binary sp lit of gender in the sample, but the categories reflect what participants reported during their interview. The choice to limit the corpus to just thirty - six speakers is the result of various methodological limitations related to recruitment and speaker cl assification. First, the oral histories speakers in general provided very little demographic information during their interviews. The only consistent information gathered by the interviewers was hiring date, ethnicity, marital status, current address, ed ucation level, and job title. Because other demographic information was not readily reported for each speaker, I did my best to gain this information from the interview recording ; oftentimes doing some calculation, e.g. discerning year of birth from report ed year of high school graduation and age at graduation. So, eliminated from the sample were speakers for whom age, education, gender, residency background of their parents, and their own residency status were not readily available in the interview. Also excluded from the pool of speakers were those who reported that they were born and/or raised in a n eighboring rural town outside of the Lansing metropolitan area. A third major complication in sampling was a very common issue in linguistics research th e recruitment of men. The fieldworker and I recruit ed all eleven female participants and conduct ed sociolinguistic 64 interviews with them in less than two months. By contrast, it took nine months to recruit and interview the seven men. While this disserta tion would benefit from the inclusion of more speakers in each demographic group, time constraints did not permit a larger demographically balanced sample. Future researchers working on Lansing speech are encouraged to increase the size of the corpus. Table 3 Number of participants from each data source Year Recorded Silent Boomer Gen X Total Oral History 90s/00s 10 6 0 16 Sociolinguistic Interview 2010 2 0 0 2 Sociolinguistic Interview 2018/9 0 6 12 18 Total 12 12 12 36 The second thing to consider is how change over time can be interpreted , since the speakers were recorded during three different time periods of data collection. Table 3 displays the distribution of speakers in the Lansing Speech Corpus by the interview collection method (sociolinguistic interview or oral history). Recall that the sixteen oral histories inter views were recorded in the late 1990s and early 2000s, the two ethnographic interviews were conducted in 2010, and the eighteen sociolinguistic interviews were conducted in 2018 and 2019. All of the oral histories speakers are in the Silent and Baby Boome r generations, while the sociolinguistic interview speakers are in the Baby Boomer and Generation X cohorts. Crucially, the oral 65 histories interviews were conducted almost 20 years before the sociolinguistic interviews were conducted in 2018 and 2019. T here are two avenues for exploring language change in this sample, and any sample in which speakers differ in both year of birth and year of recording . An Apparent Time analysis compares data by date of birth to show change over generational time . A Real T ime analysis would compare data by date of recording and thus display change . In th e Lansing Speech Corpus , a real time analysis would compare data from the speakers interviewed in the early 2000s to those interviewed in 2018/2 019 to show change over the last 20 years. I t could very well be the case that most of the changes explored in this chapter have taken place in the last 20 years rather than over the course of multiple generations . I use the apparent time construct in thi s dissertation because while there is some evidence that speakers exhibit slight variations in their speech after adolescence (cf Labov 2001; 20 09) , c hanges in (phonological) representation have not been established as changes that occur in adulthood (Labov 1994: 102 107) 2 . 3 2 If we look at Table 3 , it appears that I can investigate real versus apparent time change in this sample. There is indeed an even split in the Baby Boomer generation between the sociolinguistic interviews conducted in 2018 and 2019 and the oral histories conducted almost twenty years prior. I will not do this, however, because much of the social conditioning on these changes occurs in this Baby Boomer generation. Though the interview sources are evenly distributed in this generation, social cla ss and gender are unevenly distributed within these two groups. Therefore, I would not be able to conclude whether any differences between the groups is indeed evidence of real time change or simply an imbalance in the social makeup of these groups. I re cognize that I have not convincingly ruled out a real time conflation in this analysis and I leave it for future research to consider how recent these changes are. 3 Fruehwald ( 2017) offers a solution for disambiguating between apparent and real time in a sample using Generalized Additive Models a nd Tensor Product Smooths to model the data. I do not use these methods here because the Lansing Speech Corpus is relatively small. However, any analysis with a potential real and apparent time confound and a good amount of speakers would benefit from us ing this technique. 66 3.2.3 Interviewers Most of the sociolinguistic interviews (n = 13 ) utilized in this dissertation were conducted by a Lansing community member . I conducted five of them. This interviewer, who I will refer to throughout this dissertation as the Lansing interviewer, was born and raised 4 hours north of Lansing and has resid ed in Lansing since they were 20 years old. They identify as white, female, and are in their mid - 40s. I selected this interviewer because they were a longtime Lansing resident and they have a socioeconomically diverse social network. I recruited this in terviewe r through the social networks of a Michigan State University professor. They were tasked with recruiting, scheduling, and interviewing the participants that they recruited . They were paid $20 per speaker that they interviewed , and I trained them for two hours before they carried out any interviews . During this training session, I introduced them to the study and my objectives regarding speaker demographics. I also trained them to use the recording equipment and introduced them to best practices f or conducting a sociolinguistic intervie w (to be described in 3.2.4 ) . I provided this interviewer with all the materials necessary to conduct the int erview, including consent forms, the script for introducing the formal tasks, the list of interview questions, an interview flow - chart, the recording equipment, and the money for participant pay ment . After the interviewer conducted three interviews in the field, I listened to the interviews and had a second thirty - minute session where I provided feedback and offered some suggestions for answering/deflecting questions that were aimed at learning more about the initial intervi ews, the participant had inquired about the purpose of the study and the interviewer had revealed that we were interested in the way that people pronounce vowels in Michigan . Disclosure of this information would almost certainly guarantee that a participant would monitor their speech and likely produce more careful speech, which is not the objective of these interviews . This was not an issue in this particular case , since 67 th is exchange occurred towards the end of the interview. Neither this part of the conversation nor any following parts were included in the current analysis, in any case. For subsequent interviews, I trained the interviewer to deflect conversations about t he purpose of the study until after the formal tasks were administered. The Lansing interviewer conducted all their interviews between July and September of 2018. After their interviews were over, she and I had a three - hour debriefing session during which I gathered some valuable information not readily apparent in the interviews themselves. From this discussion, I was able to get a better sense of interviewees socio - economic background by asking more questions about the occupations of family members and the neighborhoods that they grew up in. The two sociolinguistic interviews from 2010 were conducted separately by a Michigan State University undergraduate student for a senior capstone ethnographic project on the NCS in the speech of older women (aged 65+) in the Eastside neighborhood of Lansing . Th is interviewer w as white , female and aged between 19 and 21 at the time of the project . She was from Florida, and did not have a distinctive regional accent of any kind (as indicated by an interviewee and impressionistically determined by me) . 3.2.4 Interview Methods Because the goals of the oral histories projects , senior capstone project , and the new sociolinguistic interview projects were different, the procedures for conducting interviews were also somewhat different. The most important difference is that the aim of the oral histories project was to discuss factory life with former auto plant employ ees in Lansing, whereas the goal of the sociolinguistic interviews was to obtain speech for acoustic analysis. Thus, the topics of conversation were a bit different. While the oral histories interviewers focused much of their 68 time on topics related to fac tory life with little time spent on topics outside of that, the other there is some discussion about work - life in the sociolinguistic interviews, topics in thes e interview are broad er and cover the entire lifespan, including childhood and adulthood. I do not think this is a problem for the current analysis as the interviewers of all types of interviews tended to let the interviewee discuss whatever topic they wa nted to . For transparency, I describe the methods used for each of the three sources of data in what follows . 3.2.4.1 Oral Histories The sixteen oral histories recordings analyzed for this dissertation were gathered from a collection of audio recordings housed at the Michigan State University library. This collection, G. Robert Vincent Voice Library 4 includes spoken word recordings dating back to 1888. I utilized recordi ngs from the Lansing Auto Town Gallery, which features audio - recorded oral B o dy and Diamond REO automobile assembly plants. The o ral histories with REO workers wer e recorded in the late 1990s by Dr. Lisa Fine, Professor of History at Michigan State University (Fine 2004) , and the Fisher Body oral histories were recored in the early 2000s by historians of the United Auto Workers Local 602 . Though these interviews took place in a variety of locations, the interviews analyzed for this dissertation were restricted to those union conference room. Likewise, t hough some plant workers were interviewed in groups, interviews analyzed for this dissertation include only those in which a single person was 4 The archive of G. Robert Vincent Voice Library Collection is available to the public through the Michigan State University library (https://vvl.lib.msu.edu/). 69 interviewed. Some interviewees were interviewed by a panel of UAW historians ( usually 2 3 ) and others were interviewed one - on - one by a Lansing area native . Each interview lasted about 1 hour. Topics ranged from tensions between laborers and managers in the plants to racism, sexism, community building, and union activism. Though the purpose of these interviews was to talk about life in the pl ant, many interviews included lengthy in painting the picture of Lansing in the middle of the 20 th century that I included in my description of the speech communi ty in Chapter 2 . Though oral histories do not follow the pattern of sociolinguistic interviews, and oral historians are not concerned with the preocc upations of sociolinguists, e.g. whether a speaker is monitoring their speech , these interviewers did a good job of asking questions that would prompt the interviewee to speak for at least a few minutes at a time, so that much of the interview was the inte rviewee speaking freely about a topic that interested them. This provided a good amount of natural language data from each participant in the sample for which to conduct an acoustic analysis. 3.2.4.2 Sociolinguistic Interviews The 2017 2018 s ociolinguistic interviews were recorded using a Zoom H4N Handy Recorder. They were recorded as uncompressed WAV files at a 44.10 Hz sampling rate in 16 - bit resolution. We used a Sony ECM - CS10 Tie - Clip Omnidirectional Business microphone rather than the mi crophone on the recorder. The microphone was clipped to the lapel or the upper neckline of the the interview. The recorder often sat on a surface (usually a table) next to th e interviewee , whenever possible out of their line of vision . During the interviews that I conducted, I had my computer opened with a map of Michigan showing. Whenever the interviewee talked about 70 geography or geographical mobility, I would indicate the map on the computer and have them identify various places throughout the interview, e.g. where their neighborhood was, where the good and bad parts of town were, where people vacationed in the summer, etc. My main motivation for including the computer as a prop under the guise of assisting in showing a non - native Michigander the lay of the land was to help the interviewee be more comfortable around geography as a result. goals (Lansing and Michigan life, local practices) rather than on its linguistic goals, thus reducing the likelihood of highly self - monitored speech. Other materials employed during interview sessions were the consent form, word list and sub - phonemic judgement task items, which we kept out of view until they were needed. Interviews were conducted in various places throughout the Lansing area: most were conducted in an a Lansing community center, in the Michigan State University Sociolinguistics Lab, or in my home. Though none of the interviews were conducted in a sound attenuated space, inter views were conducted in quiet rooms and there was no discernable background noise in any of the resulting audio files. Two interviews were interrupted by a child or spouse who briefly asked for something. Another interviewee answered their phone in the mi ddle of an interview. These interruptions only lasted a few minutes after which the interview resumed. Other than these minor interruptions, each interview ran continuously in one sitting. Each conversation began with the participant reading the consen t form, consenting to be recorded, choosing a pseudonym, and attaching the microphone to their clothing. The recording device was then switched on. No ne of the recruited individuals declined to consent to a recorded 71 interview . The recording began with the interviewer asking the participant to say their pseudonym . The interviewer then proceeded to ask the interviewee questions about themselves. The full list of possible questions is included in the A ppendix 0 . Following the usual practice in sociolinguistic interviews (Labov 1984) , the interview questions in this list were a guide rather than a fixed list, and interview t opics were mostly determined by the participant. They were encouraged to discuss in detail any topic that was of interest to them. So, while not every question in the list was asked during each interview, the interviewers made it a point to ask at least one question within a given topic in the list. Topics ranged from family history to formative school years, work history, current and past neighborhood demographics, Lansing, and winter in Michigan. This portion of the interview lasted between 30 and 45 minutes. After the c onversational portion of the interview, interviewees participated in two formal tasks which were aimed at testing two phonological phenomen a related to NCS re cession in Michigan . The first was not analyzed for this dissertation but re quired participants to judge syllable boundaries (Nesbitt 2018b) . The second formal task was the phonological judgement task which will be described in detail i n Chapter 5 . After these tasks, participants were asked to read aloud from a list of 100 words adapted from Evans, Ito, Jones, and Preston (2000), Ito (2001); Jones (2003), Roeder (2006) , inter alia . For all words, the nucleus of the primary stres sed syllable was one of the six Northern Cities Shift vowels in various phonetic environments. Following these three formal tasks, the interview concluded with a discussion of attitudes toward Lansing as a place compared to other metropolitan areas , and attitudes t o the local dialect. Because I am interested in speech that is as close to unmonitored as possible to provide a description of everyday speech in Lansing , only the speech produced during the casual conversation portion of each interview was included in the quantitative analysis for this 72 dissertation . The qualitative content of the post - task discussion of Lansing and local speech is discussed where relevan t. S peech produced during the word - list reading, the syllable boundary task and sub - phonemic judgement tasks, and the discussion of attitudes is not included in the analysis for this dissertation, though I intend to include it in future work. The two 2010 sociolinguistic interviews were conducted in the same manner as described above for the conversational portion of the sociolinguistic interview . Participants were audio - recorded in their homes in quiet rooms while wearing a lavalier microphone. Though the specific interview question s were a little different, they covered the same range of topics. Because these interviews were conducted as part of a different project, they did not include the formal tasks mentioned above . 3.2.5 Audio Processing, Transcription, and FAVE suite Twelve of the oral history recordings were obtained from the Michigan State University library in .wav format. The remaining 4 were obtained from the library website in .mp3 format and converted to .wav in Praat (Boersma 2001) . Three of these recordings 5 were con ducted in a room with considerable low frequency noise perhaps produced by a fan or because of the recording equipment. These three recordings were passed through a high - pass filter in Praat to reduce the low - frequency noise. Frequencies above 2000 Hz were removed from the audio. The entirety of each recording was transcribed in ELAN (Max Planck Institute for Psycholinguistics 2018) , a linguistic annotation software provided by the Max Planck Institute. Four annotation tiers were included in the transcription; one for the interviewee, one for the interviewer, one to indicate interview tasks, and the final tier for background noise. Speech was 5 These speakers were Mabel McQueen, Marvin Grinstern, and Vernon Cook in Appendices 0 and 0 . 73 transcribed in breath groups , so that segments were about four seconds each and roughly corresponded to when a speaker took a breath or paused. All sounds produced were transcribed using US English orthography and special symbols for non - lingu istic sounds (e.g. coughs) were employed according to the transcription guidelines put forth by the Linguistic Data Consortium at the University of Pennsylvania (Linguistic Data Consortium 20 13) . After transcription, each interview was passed through the Forced Alignment and Vowel Extraction (FAVE) suite (Rosenfelder et al. 2014) for forced alignment and vowel measurement. FAVE is available as downloadable scripts which can be run locally. 6 I will provide a brief description of the two s tages of FAVE below but see Evanini (2009) and Fru ehwald ( 2013 ) for a more thorough description of the mechanics of FAVE. FAVE uses the Carnegie - Mellon (CMU) pronouncing dictionary to align orthographic transcriptions with sound files at the level of the phoneme . The CMU dictionary is an open - access dictionary of over 134,000 North American English words and their pronunciations. To do the alignment, FAVE must first review a transcription to ensure that every word in it is an actual word in the CMU dictionary. - in a given file is not found in the CMU dictionary, FAVE requires the user to provide an ARPA b et transcription pronunciation symbols utilized by the CMU dictionary of that word and temporarily add it to the dictionary. Once a file has been dictionary checked, FAVE aligns the orthographic transcription with the acoustic signal in the corresponding sound file. 6 The FAVE scripts are currently being maintained by Jo sef Fruehwald on G it H ub ( https://github.com/JoFrhwld/FAVE ) 74 The FAVE - extract stage measures and normalizes all the vowel phonemes in the recording. FAVE measures vowel phonemes according to the parameters set by the user. I used the default settings specified by the FAVE creators, which are based on best practices established by the ANAE (Labov, Ash & Boberg 2006) . The first default set ting was to set the maximum formant measurements for each gender at 5000 Hz for males and 5500 Hz for females. The second was to model formant predictions after those from the ANAE. With this setting, when FAVE encounters a phoneme, it will generate a few possible measurements by changing some parameters (e.g. the number of formants it identifies) and compare this pool of measurements to the distribution of measurements of that phoneme in the ANAE. Based on the ANAE distribution of that phoneme, FAVE will determine which measurement is most likely valid and discard all other measurements for that phoneme. After going through the entire file, FAVE then eliminates outliers. For this, FAVE looks at each measuremen overall distribution of that vowel class. If a given token is found to be extreme for that speaker, i.e. sits outside of the distribution of that vowel class, the token is discarded. The resulting measurements in the ou tput are vowel duration, and formant measurements (F1, F2, F3) at the 20%, 50%, and 80% time points along the vowel for every vowel in the file. Other valuable information provided by FAVE includes stress (stressed or unstressed) and phonetic environment of each phoneme (following manner, following place, following voicing, preceding segment, and number of syllables occurring after the vowel). FAVE outputs a raw data file and a file with normalized data. This dissertation analyzes the normalized data. Nor malization is meant to eliminate differences in measurements that arise due to physiological factors (e.g. age) while maintaining differences that are the result of sociolinguistic variation (Adank, Smits & van Hout 2004; Labov, Ash & Bobe rg 2006; Thomas 75 2011) . Though there are various normalization methods, the Lobanov ( 1971) normalization method i s preferred by sociophoneticians ( for a discussion, see Thomas & Kendall 2007 and Adank, Smits & van Hout 2004 ) over the other methods and it is the one FAVE utilizes. T he Lobanov method is a vowel - extrinsic method of normaliza tion, meaning that it uses measurements of multiple vowels spoken by an individual to calculate a grand mean by which each vowel may be normalized. FAVE then rescales the Lobanov normalized values back to H ertz (Hz). 3.2.6 Measurements and Analysis The thirty - si x interviews in the Lansing Speech Corpus produced 76,723 FAVE - measured vowels. I report in this dissertation on the /æ/ tokens alone. 7 To avoid measuring vowels that are commonly reduced in casual speech, I eliminated any token that did not occur in with /æ/ excluded from the analysis were the following: a, am, an, an, and, as, at, had, has, have, than, . Tokens of /æ/ pre ceding /r/ were excluded from the analysis, as /æ/ is merged with American English dialects, including the NCS (see Labov, Ash & Boberg 2006:56) . Finally, I omitted tokens that were not produced in the casual speech portion of the interview, i.e. any tokens that were produced during the formal tasks or the subsequent attitudes portion of the sociolinguisti c interview. The final token count for /æ/ analyzed in this dissertation is 4,158. Throughout this dissertation I will refer to two allophonic environments; pre - nasal and pre - oral . Pre - nasal /æ/ tokens are those that precede the three nasal consonants in American English; /m/, 7 76 Those that precede a non - nasal consonant are pre - oral. Of the token count, 1,590 tokens are pre - nasal and 2,568 are in pre - oral position. The ggplot2 (Wickham 2009), dplyr (Wickham and Francois 2016), gridExtra (Wi ckham 2017), and tidyr (Augie 2016) packages in R (R core team) were used for all data manipulation and visualizations. To employ the four diagnostics of NCS /æ/ mentioned in section 3.2 at the start of this chapter, four acoustic measures of /æ/ were used: F1, F2, and diphthongal quality of pre - oral /æ/, and Pillai - Bartlett score. The first (F1) and second (F2) vowel formant measures were taken at the mid - point of the vowel in Hertz (Hz), as supplied in the FAVE output. F1 measurements correspond to tongue height during the articulation of the vowel; a higher F1 value indicates a lowered tongue position and a lower F1 value i ndicates a higher tongue position. F2 measurements correspond to tongue back - ness; a higher number indicates a further forward articulation and a lower number indicates a farther back articulation. Diphthongal quality (DQ) is assessed by a measurement of trajectory length (Fox & Jacewicz 2009; Jacewicz, Fox & Salmons 2011) As noted in previous studies on /æ/, the two points fo r diphthongal quality of this phoneme are at 50% and 80% of the vowel (Labov, Ash & Boberg 2006; Mielke, Carignan & Thomas 2017) . DQ is operationa lized (see below) as the square root of the sum of the squares of the difference s in F1 and F2 of the relevant t ime points in the vowel . A higher DQ value indicates more of a difference in F1/F2 space (80%) . (6) DQ = 77 Separation between pre - nasal and pre - oral tokens was measured using the Pillai - Bartlett statistic (Hay, Warren & Drager 2006; Hall - Lew 2010) . In measuring separation, the statistic utilizes MANOVA to evaluate the distance between two distributions and their variance. Pillai - Bartlett scores range from 0 to 1, where 0 indicates no difference be tween distributions and 1 indicates complete separation of the distributions. Thus, in the discussion of /æ/ systems in section 4.6 , speakers with an NCS unconditioned system should have a Pillai - Bartlett score closer to 0, those with a normal raised continuous system would have a score closer to .5, and those with a nasal system should have Pillai - Bartlett scores closer to 1. Linear mixed - effects regression models were utilized to determine the effects of social class, gender, and generational cohort on the four dependent acoustic measures (Pillai - Bartlett score, F1, F2, duration, and DQ). The fixed effects were operationalized as fol lows: Social Class (blue collar, white collar), Gender (female, male), and Generation (Silent, Boomer, Gen X). The linear models included random intercepts for speaker and word to account for any variability caused by any individual speaker producing a lar ge amount of speech or a lexical item occurring disproportionately more often than others in the corpus. An ANOVA analysis was performed on Pillai - Bartlett scores, as each speaker in the sample only had one score. This analysis included a random intercept of speaker as a random effect. For all four analyses, all possible i nteractions between the fixed effects were included with the fixed effects in a full model. Fixed effects and any interactions that were not significant in the full model were removed. Throughout the discussion of results, I will indicate when an effect or interaction was not included in the resulting model. Results of the models are provided for reference in each section, below. I conducted all linear mixed - effects regression modeling with the lme4 package in R (Bates et al. 2015) . W elch two - sample f ollow - up t - tests were performed whenever there was a significant 78 three - way interaction and whenever an effect with more than two levels came out as significant. These statistical analyses will test the hypothesis that generational cohort, gender, and social class condition movement in F1 and F2, diphthongization, and nasal allophony in 20 th century Lansing. 3.3 Phonolog ical Change As briefly summarized at the start of this chapter (section 3.1 ), the goals of this dissertation are to address the A ctuation , E mbedding, and T ransition P roblems of phon ological change , i.e. how did it begin , what are the internal and external conditions surrounding the change, and how did it spread throughout the community ? The E mbedding problem is first addressed in Chapter 4 where I investigate the acoustic changes to /æ/ and the social factors that condition them. To further address th e E mbedding problem and to address the A ctuation and T ransition problems , I will investigate in Chapter 5 the temporal and social mechanisms by which /æ/ nasal allophony developed and spread in Lansing. I ask when speakers in Lansing developed an allophonic distinction between pre - nasal and pre - oral /æ/ and who the leaders and laggar d s of th is change are. To do this, I utilize two techniques; (1) an impressionistic examination of pre - nasal and pre - oral trajectories in the vowel space over time , a la Fruehwald (2013, 2016), and (2) a sub - phonemic judgement task , a la Mellesmoen (2016) . The trajectories analysis includes the conversational speech data from the Lansing Speech Cor pus described above and the results are provided in Chapter 5 (section 5.2.1 ) . The judgement task includes participants of the 2017 2018 sociolinguistic interviews in the Lansing Speech Corpus, and other Lansing natives who completed the task online . Those results are presented in Chapter 5 (section 5.2.2 ) . I n what follows, I describe the methods for conducting both analyses. 79 3.3.1 Divergent Trajectories According to Fruehwald (2013, 2016), phonological allophones follow disparate trajectories over time while allophones that differ only in phonetic realization, not phonological representation, proceed in lockstep with one another. Therefore, I will investigate phonetic vs phonological allophony within the /æ/ phoneme by examining trajectories of pre - nasal and pre - oral /æ/ in F1/F2 space over apparent time. As a point of comparison with /æ/, I include a visual inspection of phonetic trajectories for two other short - front vowel classes in the same phonological environments : / / and / /. Here I will determine whether the phonological profile of these classes is similar to that of /æ/. The hypothesis is that they are different, as the analysis in Nesbitt (2018) suggests that no a llophones within these classes are phonologically distinct from the others. Therefore, the allophones within these phonemes must only be phonetically distinguished and should move in lockstep over time . If the allophones within these phonemes are indeed phonetically distinguished th e n I would expect participants to be more likely to judge them as the same during the judgement task because acoustic difference may not be enough to prompt a different response. If this hypothesis is correct, I can leverage t his difference between /æ/ and these other classes to ascertain whether respondents are making judgements based on acoustic (dis)similarity or on differences in representation. I discuss this in further detail in 5.2.1 . The second purpose of the trajectory diagnostic is to determine when the phonological allophone ( s ) of /æ/ began to diverge from the rest of the allophones. This latter question is impo rtant for the subsequent analysis of the actuation (initiation) and propagation (spread) of phonological change. For t he actuation question , w e will want to ask whether at the speaker level, the phonological change was abrupt or gradual, i.e. whether indi vidual speakers posited a 80 representational difference between pre - nasal and pre - oral /æ/ a t the time the vowel in the two environments began to assume different phonetic targets (abrupt) or if speakers developed the phonological rule sometime after the two allophones began to move in different directions (gradual). To determine this, I must first discern when the relevant allophones began to move towards different targets. Whichever generation in which the relevant allophones begin to differentiate, as determined by this trajectory analysis, is the generation for which the question of abrupt phonologizati on will be relevant. If any speaker s in this generation posits a phonological rule, I will conclude that phonologization was abrupt. If, however, no speakers in this generation posit a phonological difference between pre - nasal and pre - oral /æ/ but speakers in subsequent generations do then I will conclude that phonologization was a gradual process , occurring only after a phonetic ally implement ed distinction . 3.3.2 Sub - P honemic J udgement T ask Though the investigation into trajectories of conditioning environment s can tell us whether and when phonologization has occurred at the community level, it cannot tell us whether individual speakers exhibit phonological or phonetic allophony. Likewise, Pillai - Bartlett scores are not useful for teasing the two apart on thei r own, as studies have shown that while bimodality is often an indication of phonological/categorical differences, the absence of a bimodal distribution does not entail the absence of a phonological difference (cf. Schilling, Watkins & Watkins 2002; Bermúdez - Otero & Trousdale 2008: 696) . Therefore, I supplement the Pillai - Bartlett analysis with a s ub - phonemic judgement task (Mellesmoen 2016) to discern whether for each respondent, phonetics alone is what distinguishes pre - nasal from pre - oral /æ/ or if the two allophones are phonological ly distinct . A comparison of responses between generations will address an aspect of the A ctuation P roblem of phonological change : was phonologization 81 gradual or abrupt in Lansing, i.e. at the beginning of the acoustic change, are there any speakers that consider pre - nasal and pre - oral /æ/ to have different phonological representations ? An investigation into the interaction of change over time against social class will address the T ransition Problem of phonological change; how has nasal allophony spread throughout th e community, i.e. who are the leaders and laggards of the change? 3.3.2.1 Experimental Stimuli The judgement task for this dissertation required participants to judge whether the vowels in thirty pairs of CVC lexical items were the same or different. The complete list of lexical pairs is included in A ppendix 0 . These pairs were grouped into six conditions of five pairs. The differences between the condition s are illustrated in Table 4 and described in detail below. Pairs in Condition 1, which I will refer to throughout this chapter as the target condition , were CVC lexical items in American English that had the same onset and vowel /æ/ but differed in whether their coda consonant was nasal or oral. One of the target pairs in this study was fan and fat ; these words share the same onset /f/ and nucleus /æ/ but differ in whether their coda consonant is nasal /n/ or oral /t/. CVC pairs in Condition 2 shared the same onset and vowel /æ/ but had different oral coda consonants. An example of this type of pair is pass and pack ; they both have an onset /p/ follow ed by an /æ/ but they differ in that one has a coda stop /k/ and the other a fricative /s/. I included this type of pair to be able to test whether phonetically implemented differences might prompt an individual to judge these pairs to be different, thus making this methodology unsuitable for drawing conclusions about phonologization 82 Table 4 Conditions of Lexical Pairs in the sub - phonemic judgement task CONDITION Vowel Phoneme Final Consonant Nasality Example Pair 1 /æ/ nasal - oral FAN - FAT 2 /æ/ oral - oral PASS - PAT 3 / / nasal - oral PEN - PET 4 / / oral - oral PET - PECK 5 / nasal - nasal or oral - oral BOT - BOUGHT 6 mix Mix BIT - BOMB . As is the case in other studies on /æ/ in the dialect area, speakers in Lansing show clear conditioning within the pre - oral token cloud such that there is a gradient difference in vowel space. For example, many speakers produced /æ/ before fricatives l ower than the rest of the pre - oral tokens , others produced /æ/ before affricates higher than the rest of the conditioning environments in the pre - oral cloud, and often /æ/ before /l/ was produced lower than the rest of the pre - oral tokens . Crucially, in t he analysis of trajectory , the difference between /æ/ before these various consonant types is the result of low - level co - articulation rather than any phonological difference, i.e. they moved in lockstep with each other towards a separate phonetic target th an pre - nasal /æ/. Thus, in the present judgement task, if phonologization is what is driving different responses, respondents should judge pairs of different pre - oral tokens as the same whether they produce them with different acoustics or not. Take for example one of the 83 C ondition 2 pairs pa ck and pa ss . Although /æ/ before /k/ is gradiently different than /æ/ before /s/ in the vowel space, this difference is not expected to be meaningful to respondents and thus the expectation is that they will be le ss likely to judge /æ/ in these two words to be different. This is the expectation for responses to all lexical pairs in this condition. If a respondent judges the pairs in this condition to be different, then this is an indication that phonetic implemen tation differences are what the respondent is relying on. In this case, a claim that C ondition 1 (target pairs) judgements are an indication of phonological representations would be unsubstantiated. If a respondent judges Condition 1 pairs (target pairs) to be different but C ondition 2 pairs to be the same, it is more likely that their judgements about C ondition 1 pairs are representative of a perceived phonological difference rather than phonetic differences. If a respondent were to judge both condition one and two pairs to be the same then this might suggest that they are relying on some other cue, i.e. orthography. Two more conditions, Condition 3 and Condition 4 , were included in this experiment to (1) determine whether speakers are making judgements based on phonological representation rather than acoustic differences, and (2) to distract the participant from focusing on /æ/. Condition 3 CVC pairs were like the target pairs but had a vowel other than /æ/ - either / / or / /. An example of this type of pair is pen and pet ; they share the same onset /p/ and vowel / / but differ in whether the final consonant is nasal /n/ or oral /t/. Aside from distracting the respondent from the focus of the study (/æ/), the primary reason for ad ding these pairs was to determine whether judgements for the target pairs those with /æ/ in oral and nasal contexts are indeed based on phonological representation and not just acoustic difference. The logic here is the same as that for condition two pairs (e.g. pass - pat ); because the difference between / / and / / in pre - nasal and pre - oral environments in Lansing is not due to a phonological representation difference 84 but rather a low - level co - articulatory effect, respondents are not expected to judge pre - nasal and pre - oral pairs as different. Participants are also not expected to judge various combinations of pre - oral / / as different because as is the case with pairs of pre - oral /æ/, the difference between them are only the result of phonetic implemen tation rather than an inherent difference in phonological representation. The same is true for / /. Thus, I balanced the pool of stimuli to include a fourth condition , Condition 4, similar to Condition 2 pairs. These pairs had the same onset , and nucleus / / or / / and differed in their following oral consonant. An example Condition 3 pair is pet and pe n , and Condition 4 pair is pet and pe ck . A fifth condition of CVC pairs was included as filler pairs in the experiment. These Condition 5 pairs tested another on - going change in Lansing the merger of / / and / pairs in this condition shared the same onset and coda consonant but di ffered in whether the vowel was / / or / th condition is bot and bought . Judgements for these pairs were not included in the analysis for this dissertation. The final condition , Condition 6, was composed of f iller CVC pairs. Some of these pairs had vowels and final consonants that did not match orthographically or phonemically and should have been judged to be different by every participant in the study. An example pair is bit and bomb , where / / and / different phonemes and are represented with different orthographic symbols, as are /t/ and /m/. The vowels and consonants of other pairs matched orthographically but differed in phonemic representation, .e.g. make and mack. A subset of the pairs in this condition had different vowel phonemes that were represented with the same grapheme, , e.g. make and mack . While the vowels in these words are represented with the same orthographic symbol they map onto different phonemes, /e / and /æ/, respectively. These were included to ensure that participants were not making judgements based on orthography alone. One of the 85 purposes of including this condition was to further distract the participant from focusing on /æ/ pairs by including lexical items with other vowel phonemes in them. The other motivation for including these pairs was to eliminate participants who were not paying attention to the task or who were making judgements based on orthography alone. If a participant judged mak e and mack to be the same then it is likely that they were making judgements based on orthography alone. If, however, a participant judged the pairs in Conditions 1 through 4 (pan - pass, pass - pack, pen - pet, pet - peck) to be the same but those in Condition 6 to be different, it is more likely the case that they are taking into consideration their own pronunciation of these phonemes rather than relying on orthography alone. 3.3.2.2 Experimental Task This experiment was presented via two methods; an online survey and on a piece of paper. The online survey was disseminated via Qualtrics (Qualtrics 2005) . Fourteen of the 2017/8 sociol inguistic interview participants did the task on a piece of paper, while four completed the survey online. Because of technical difficulties during some of the sociolinguistic interviews, ten participants completed a paper version of this survey. The pap er version was identical to that of the Qualtrics survey. Upon entering the survey on Qualtrics, participants read a consent form, which is provided in Appendix 0 . Consent to having their data used for analysis was granted by clicking a button at the bottom of the screen which took the participant to the first block of questions. For those who did the paper version, consent was given at the beginning of the sociolinguistic interview. The first block of questions in the survey was a set of demographic questions. These questions were used to ensure that respondents fit the demographic criteria of this project. Sociolinguistic Interview participants di d not have to respond to this block of questions, as their 86 demographic information was already provided. A discussion of participant demographics is provided below in section 3.3.2.4 . A second block of questions in the survey was part of an experiment for another study (Nesbitt 201 8b) . For this task, participants were presented with CVCVC American English words and asked to indicate which syllable the medial consonant belonged to the second or first. Analysis of the data from this block is not included in this dissertation. The third block of questions made up the judgement task, which is the focus of this dissertation . In this block, participants were visually presented with pairs of words and were instructed to judge whether the vowels in the pairs sounded the same or di fferent to them. The following instructions were positioned at the top of their screen/paper throughout this block of the experiment, so that every time a pair of words was presented on the screen /paper , so too were the instructions. For each pair of wor ds, read aloud each word and indicate whether the vowels in those two words sound the same or different to you. Try not to think too hard about your response. Provide the first response that comes to mind. There are no right or wrong answers. Participan ts indicated whether the vowels in a given pair sounded the same or different by clicking (or circling on their piece of paper) same or different . After judging one pair of words, they went on to the next pair of words, and so on. About five pairs appeared on the same page before a participant had to press a button to navigate to the next page. Pairs were presented 87 in random order for each parti cipant. To ensure more statistical power, each pair of words was presented twice. In total, participants judged whether the vowels in sixty pairs (6 conditions x 5 pairs x 2 trials) of CVC lexical items were the same or different to them. Upon completi ng the survey, participants were asked if they were taking the survey for for participating and notified them that they were finished with the survey. Those who nickname in order to receive the class credit . Upon entering their name, they were informed that they could receive more credit for forwarding the survey to an o lder family member. They were then thanked for participating and notified that they were finished with the survey. 3.3.2.3 Participant Recruitment and Selection This experiment was disseminated on Qualtrics at various times between April 2018 and February 2019. The first group of participants were undergraduate students at Michigan State University who agreed to participate in exchange for class credit, along with any family members to whom they forwarded the survey. Other participants were those who did the so ciolinguistic interview discussed in the previous chapter. I also disseminated the survey via social media and email. From these various recruitment methods, 252 people participated in this experiment. Many participants were eliminated from the study bec ause they did not meet the demographic criteria for this dissertation; they either were not a native of Lansing, MI, did not identify as white/Caucasian, answered stay - at - home parent , or had considerable experience speaki ng a language other than English. For the latter criteri on , I 88 language other than English. Native Lansing status was defined as: (1) having been born and lived in t to enter the zip code of the place they were raised. I identified t hem as native if this zip code was within 10 miles of the Lansing city center (the capit o l). Participants in this study were required to have been born in Lansing , which is a different criterion than that defined for the Lansing Speech Corpus . R ecall that r ather than requiring that the Lansing Speech Corpus speakers be born in Lansing, I set the cutoff at age 5 . This was merely a methodological decision because many of the oral histories speakers did not explicitly state that they were born in Lansing but did make reference to living there since before age 5. I also eliminated anyone who did not respond to the questions about education, occupation, and/or year of birth, and anyone who reported that they were not 18 or older because I was not approved to t est minors. Sixty - five participants were excluded from the analysis due to various performance related issues. First, participants were eliminated if they did not answer at least 90% of the experimental questions. Participants were also eliminated from t he analysis if they provided the same response (either same or different ) to 75% or more of the pairs in the survey. Participants were also excluded from the analysis if they judged more than one of the make/mack or bit/bomb type pairs in Condition 6 to be the same. In theory, every participant should have judged C ondition 6 pairs to be different but because I know that the survey is repetitive and perhaps some participants could have misread one of these pairs, I was more conservative with this elim ination criteri on . Rather than eliminating everyone who responded same for any Condition 6 pairs, I eliminated anyone who responded same to more than one of t hese pairs . 89 3.3.2.4 Demographic distribution After the above demographic and performance eliminations, the resulting pool of participants for this experiment was 107 white monolingual English - speaking Lansing natives. The demographic distribution of these participants is given below. The distribution of participants according to generational cohort is prov ided below in Table 5 Participants were given the following age rang es to choose from: 17 or younger, 18 34, 35 54, 55 74, and 75 and older. Those who responded 17 or younger were eliminated from the analysis because they are considered minors. Table 5 Distribution of experimental participants by Generational Cohort Silent Baby Boomer Generation X Millennial Total 16 35 28 28 107 The remaining respondents were classified according to the generational cohorts employed for the Lansing Speech Corpus construction: the Millennial generation is made up of those who responded 18 34 , those who responded 35 54 are Generation X, those who r esponded 55 74 made up the Baby Boomer generation, and those 75 and older are the Silent generation respondents. As with the sociolinguistic interview participants, socio - economic status for this experiment was determined by occupation and education level . In the survey, participants W hat level of education is necessary to do your type of job? and were given five 90 choices; some high school, high s or post . I followed the same procedure for classifying these participants as I did for the sociolinguistic interview participants. Those in manual labor, low - level service, and care jobs that require less than 2 years of training after high - school were classified as blue - collar . Those in managerial and professional positions whose occupations required at least 2 years of technical training outside of high - school were classified as white - c ollar . For participants who indicated that they were a student or retired, I classified them according to their response to the question about education; those who chose some college/vocational or above were classified as white - collar. The distribution of participants according to generational cohort and social class is provided below in Table 6 . Table 6 Distribution of experimental participants by Generational Cohort and Socio - Economic Status Generation Blue - Collar White - Collar Total Silent 9 7 16 Boomer 17 18 35 Gen X 14 14 28 Millennial 13 15 28 Total 53 54 107 The effect of gender on this change in progress was not explored in this analysis . Though it would have been ideal to consider this social factor, especially because many of the acoustic changes to /æ/ in Lansing are conditioned by gender ( Chapter 4 ) , gender was not evenly 91 distributed across generational cohorts and social class groups. In Table 7 , I display the distribution of gender in this sample for transparency and because, as I note in Chapter 4 , gender might be an important factor for t h e propagation of this change. Table 7 Distribution of experimental participants by Generational Cohort, Socio - Economic Status, and Gender Blue Collar White Collar Silent 6f 3m 5f 2m Baby Boomer 14f 3m 10f 8m Gen X 11f 3m 11f 3m Millennial 10f 3m 9f 6m 3.3.2.5 Statistical Analysis For this task, phonological allophony between pre - nasal and pre - oral /æ/ will be identified if Condition 1 pairs elicit significantly more different responses than Condition 2 pairs, e.g. if participants judge pre - nasal and pre - oral /æ/ to be different but pre - oral and pre - oral pairs to be the same. The primary goal of this experiment was to ascertain whether phonologization was abrupt or gradual in Lansing, i.e. is it the case that the g eneration of speakers in which pre - nasal and pre - oral /æ/ begin to diverge in phonetic space are the same generation of participants to judge /æ/ to be different allophones? As such, abruptness will be investigated by comparing responses between C ondition s 1 and 2 in whichever generation pre - nasal and pre - oral /æ/ diverge from one another in F1/F2 space, as determined by the trajectories analysis . Phonologization 92 will be interpreted as abrupt if this generation of participants judge /æ/ before nasal conso nants to be different from /æ/ before oral consonants but pre - oral /æ/ pairs to be the same. Phonologization, however, will be identified as gradual in Lansing if only subsequent generations judge /æ/ before nasal consonants to be different from /æ/ before oral consonants. As with the production data, I will examine social class as it conditions same/different judgements in each condition. This analysis will be utilized to continue the discussion regarding who the linguistic innovators are and who are the likely retainers of the old mechanical/co - articulatory allophonic system. Responses to this task were analyzed using a binomial mixed - effects logistic regression model in R (R Core Team 2013) . Response was entered as the dependent variable, while Condition, Generational Cohort, Social Class, and their interactions were entered as predictor/independent variables, and Participant and Pair as random intercepts . Response had two levels (same, different), Condition had two levels (one, two), Gene rational Cohort had four levels (Silent, Boomer, Gen X, Millennial), and Social Class had two levels (Blue - collar, White - collar). To conclude, this dissertation will examine the phonetics and phonology of /æ/ over generational time in 20 th century Lansing. Chapter 4 will address the embedding problem of phonological change by examin ing the NCS diagnostics: fronting, raising, diphthong al quality, and /æ/ system as they are related to generational time, social class, and gender in the community. Chapter 5 will examine the actuation , social embedding, and propagation of nasal allophony through the apparent time analysis of allophone trajectory and sub - phonemic judgement task res ponses . 93 / æ / IN 20 TH CENTURY LANSING 4.1 Introduction This chapter is concerned with the influence of generational time, gender, and social class on the realization of /æ/ in 20 th century Lansing. I will examine these factors with respect to 4 of the 5 NCS diagnostics which were preliminarily examined in Wagner e t al. (2016) and Nesbitt (2018), and described in Chapter 2 . that have been described in the literature. These diagnostics include (i) movement alon g F1 of pre - oral /æ/ , (ii) movement along F2 of pre - oral /æ/ , (iii) diphthongal quality of pre - oral /æ/ , and (iv) the separation of pre - nasal and pre - oral allophones in F1/F2 space. The diagnostic being left out of the analysis is the position of /æ/ relative to / / in the vowel space. Because / / lowers as part of both the NCS and LBMS, the relative position of these phonemes is less informative of NCS participation (or rejection) in the current sample. By examining how these diagnostics change over time, we can get a more thorough view of the time window for the adoption and rejection of the NCS in Lansing. The inclusion of social class and gender will allow for a more nuanced picture of which community members are on the leading end of the change, which can in turn allow us to determine whether the movements of /æ/ are changes from above or below in the community. The analysis of Pillai - Bartlett scores will either confirm or deny the preliminary findings in Wagner et al . (2016) and Nesbitt (2018) that there is a diachronic progression from systems that are advanced rais ed to those that are common raised (or less advanced) to nasal systems . This will contribute to the discussion in Cha pter 5 about whether phonologization of the nasal split occurred gradually or 94 abruptly in the community . I investigate these measures against the hypothesis that the emergence of non - NCS features is driven by social aspiration. 4.2 Results The methods for creating the Lansing Speech Corpus (selecting speakers from existing corpora, and interviewing additional speakers to fill demographic gaps) are provided in Chapter 3 (section 3.2 ), along with details of the coding and statistical analysis procedures. The following is an analysis of the four diagnostics with respect to generational time, social class, and gender in the Lansing Speech Corpus. Below, results of the mixed effects regression and ANOVA analyses of the social patterning of /æ/ in Lansing acr oss the 20 th century are presented for each of the four dependent diagnostic variables (see 3.2.6 for full details of statistical methods). Section 4.3 reports on the height of the nucleus (measured by F1 at the midpoint); section 4.4 reports on the backness of the nucleus (measured by F2 at the midpoint); diphthongal quality (trajectory length measurement) is examined in section 4.5 ; and in section 4.6 , the relative conditioning of pre - nasal and pre - oral allophones (measured by Pillai - Bartlett score) is modeled. Section 4.7 language change, chain shifting, the NCS, and North American dialectology are p rovided in Chapter 6 . For consistency across sections, I report the generational change (or lack thereof) for the dependent variable, followed by an analysis of other main effects and interactions that came out as significant in the analyses. In all of the figures below, dots represent speaker means accompanied by a loess curve with a 95% confidence interval. For ease of refer ence, the generational cohort groups used for the present study are as follows: Silent generation (born 1925 1945), the Baby Boomer generation (born 1946 1964), and Generation X (born 1965 95 1985). I also note that data from the Millennial generation speaker s (born 1985 199 9 ) in Wagner et al . (2016) are utilized in the figures as a temporal reference but are excluded from the statistical analyses and general discussion of the changes described. 4.3 F1 of P re - O ral /æ/ Figure 10 displays average F1 values across generational time in the sample. Looking at the trend line, it appears that F1 values remain stable from speakers born in the 1920s to the 1960s that this is followed by a n increase (decline in the vowel space) in the 1990s . Figure 10 Average F1(Hz) values of pre - ora l /æ/ by speaker year of birth in 20 th century Lansing. Lansing Speech Corpus (n=36), plus Millennial speakers (n= 21 ) for reference. This visual interpretation is supported by follow - up t - tests where there is no significant effect of generational time in the transition from Silent ( M = 781 Hz, SD = 37 Hz) to Baby 96 Boomer generation speakers ( M = 793.7 Hz, SD = 43.56 Hz), t (20.8) = - 0.47, p > 0.05, nor in the transition into Generation X ( M = 777.4 Hz, SD = 26.8 Hz), t (18) = - 0.08, p > 0.05. Looking at the distribution of mean values over time in Figure 10 , however, there is a clearer apparent time picture : F rom the Si lent to Baby Boomer generation , the range o f F1 values increases such that some Baby Boomer speakers have higher mean s in the vowel space than Silent generation speakers and others have much lower means . In Generation X , mean F1 values converge at the midpoint of the Baby Boomer range and subsequently increase (lower in the vowel space as shown in Figure 10 ). So, although the trend line and follo w - up t - tests appear to show stability in the first half of the 20 th century, there is movement along F1 in the Baby Boomer generation , albeit in different directions for different speaker s . This is confirmed by the regression analysis . Table 8 displays the results of the mixed effects regression analysis on F1 of pre - oral /æ/ . According to the mixed effects regression analysis ( Table 8 ), the apparent time picture of - oral /æ/ in the height dimension is one of raising in the vowel space (decreasing F1) from Silent t o Baby Boomer generation, followed by lowering (increasing F1) into Generation X. Below, I show that raising from Silent to Baby Boomer generation is conditioned by gender. The main effect of gender is significant in the model (see Table 8 ). There is also a significant interaction of gender with generational cohort, such that the direction of the difference depends on generational cohort. I will expl ore this change via follow - up Welch two sample t - tests in 4.3.1 . Though social class is not significant in the model, it does contribute to a signific ant three - way interaction with generational cohort and gender such that social class is only significant for men. The results of the follow - up t - tests for this interaction are investigated in 0 . 97 Table 8 Best fit m ixed - e ffects m odel of pre - oral /æ / F1 Predictor Value(s) Estimate Std. Error df t - value Significance 747.46 13.5 45.12 55.369 *** GEN GenX 46.69 18.94 41.46 2.465 ** GEN Silent 45.78 18.86 39.72 2.428 ** GENDER male 78.41 17.46 31.48 4.491 *** SES WhiteCollar 30.59 17.5 31.86 1.748 [0.090] GEN GENDER GenX male - 90.69 25.08 32.69 - 3.616 *** GEN GENDER Silent male - 96.57 25.14 32.42 - 3.841 *** GEN GENDER SES Boomer female WhiteCollar - 27.74 25.83 37.18 - 1.074 [0.290] GEN: GENDER:SES GenX female WhiteCollar - 38.82 25.3 33.65 - 1.534 [0.134] GEN GENDER SES Silent female WhiteCollar - 45.72 25.61 35.23 - 1.785 [0.083] GEN GENDER SES Boomer male WhiteCollar - 64.84 23.45 25.83 - 2.765 ** GEN GENDER SES GenX male WhiteCollar - 52.1 24.75 31.14 - 2.105 * Si gnificant difference (p < 0.001) is reported as *** ; significant difference (p < 0.01) is reported as ** ; significant difference (p < 0.05) is reported as *, and non - significant appears as [] . 4.3.1 F1, Generational Time and Gender Fi gure 11 displays the interacting effect of generational cohort and gender on F1 values. As can be seen from the different trajectories for female speakers (red line) and male speakers (blue line), the F1 target is different for each gender group. This gender difference is especially evident for speakers born in the Boomer generation, 1940s 1960s . The wide range of F1 values exhibited in Figure 10 is due to a gender split in this generation, whereby women have lower F1 values (higher in the vowel space) and men have higher F1 values (lower in the vowel space) than any other speakers in the sample. 98 Fi gure 11 Average F1(Hz) values of pre - oral /æ/ by speaker year of birth and gender in 20th century Lansing. Lansing Speech Corpus (n=36), plus Millennial speakers ( n=21 ) for reference. Let us examine women and men separately first. Fi gure 11 shows and follow - up Welch two sample t - tests confirm that within the female group, there are on average higher realizations (lower F1 values) of pre - oral /æ/ in the Baby Boomer generation ( M = 741.3 Hz, SD = 30 Hz) compared to the Silent generation ( M = 777.2 Hz, SD = 38.3 Hz), t ( 9.5 ) = 2.22 , p<0.0 5 . This is followed by a significant decrease in realizations (higher F1 values) in Generation X ( M = 786 Hz, SD = 23.1 Hz) compared with Baby Boomer women, t ( 9.5 ) = 3.3 , p<0.0 1 . Thus, for women in Lansing, the nucleus of pre - oral /æ/ rose in the first half of the 20 th century, and fell in 99 Generatio n X . Recall from Table 8 that social class did not have a significant effect on F1 for women. In the male group, however, there is a different temporal pattern, as well as an effect of social class. Overall, as shown in Fi gure 11 æ/ falls from the Silent ( M = 785.2 Hz , SD = 3 9.4) to the Boomer generation ( M = 808 Hz , SD = 24.5) (higher F1 values), before rising in Generation X ( M = 760.8 Hz , SD = 25.7) (lower F1 values). This is the opposite of the pattern for women, as illustrated previously in Fi gure 11 , whose mean / æ/ nucleus rose and then fell. These different traj ectories have led to the following effect of gender across generational time: No significant effect of gender in the Silent generation; women ( M = 777 Hz, SD = 38.3 Hz) and men ( M = 785.2 Hz, SD = 39 Hz) exhibit similar F1 values, t( 9.99 ) = 0.11 , p > 0.05 . In the Baby Boomer generation, there is a significant effect of gender such that women ( M = 741.3 Hz, SD = 30 Hz) exhibit higher realizations of /æ/ (lower F1) than men ( M = 808 Hz, SD = 24.5 Hz), t ( 9.74 ) = 4.61 , p<0.001. Finally, in Generation X , men ( M = 760.8 Hz, SD = 25.7 Hz) exhibit higher realizations of /æ/ (lower F1) than women ( M = 786 Hz, SD = 23.1 Hz), t( 9.87 )= 1.83 , p<0.0 5 . In sum, there was no gender effect at the turn of the 20 th century on raisin g , women then had higher realizations in the middle of the 20 th century, which was followed by male speakers exhibiting higher realizations later. 4.3.2 Summary of F1 R esults At the community level there was stability of the nucleus of pre - oral /æ/ in the F1 dim ension in the beginning and middle of the 20 th century , which was followed by a rise in F1 values in the 1990s . This is m itig ated by the fact that change was gender - conditioned, i.e. each gender exhibited opposite changes over time . In the Silent to Baby Boomer generational 100 transition, women raised /æ/ while men lowered . In Generation X , wo men lowered /æ/ and men raised /æ/. In the subsequent Millennial generation, women continued to lower /æ/ and men proceeded to do so also. Th erefore, while at the community level there appeared to be stability in the first half of the twentieth century, change over time is evident when gender is taken into account. 4.4 F2 of P re - O ral /æ/ Table 9 displays the results of the mixed effects regression analysis for F2. The reference/intercept group are Baby Boomer s , as before . As shown in the table, there is a main effect of generational cohort, with Generation X speakers having significantly more backed /æ/ (lower F2 values) than the cohort reference group, Baby Boomers. Table 9 Best fit m ixed - e ffec ts m odel of pre - oral /æ/ F2 Predictor Value(s) Estimate Std. Error df t - value Significance 1979.902 38.966 37.454 50.812 *** GEN GenX - 236.107 54.828 36.567 - 4.306 *** GEN Silent - 12.144 54.893 36.727 - 0.221 [0.826] GENDER male - 88.613 53.633 33.645 - 1.652 [0.108] GEN GENDER GenX male 198.338 76.378 34.509 2.597 * GEN GENDER Silent male 6.431 76.521 34.737 0.084 [0.934] Significant difference (p < 0.001) is reported as ***; significant difference (p < 0.01) is reported as **; significant difference (p < 0.05) is reported as *, and non - significant appears as []. Unlike for F1, there is no significant difference in F2 realization between Boomers and Generation X in the regression, and there is no main effect of gender. There is, however, a significant interaction between generation and gender. Social class and its interaction with the 101 other main effects did not come out as significant in the model and will not be explored in what follows Figure 12 displays average F2 values across apparent time in the sample. This, along with the regression output in Table 9 , show s that in Lansing, the general pi cture is one of stability in the transition from Silent to Baby Boomer generation, which is followed by significant backing (lower F2 values) in Generation X. This is the same trend that was observed for F1 values in the follow - up t - tests. Figure 12 Average F2 (Hz) values of pre - oral /æ/ by speaker year of birth in 20th century Lansing. Lansing Speech Corpus (n=36), plus Millennial speakers ( n=21 ) for reference. 102 Since neither the main effect of social class or its interaction with other social variables was significant in the model, I will not explore the effects of social class on F2 values in what follows. Gender was not a significant main effect on F2, but as w e saw for F1, its interaction with generational cohort was significant, whereby the effect of gender changes over generational time. 4.4.1 F2, Generational Time and Gender Figure 13 displays the interacting effect of generational cohort and gender (red = female) on F2 values in the corpus. Figure 13 Average F2(Hz) values of pre - oral /æ/ by speaker year of birth and gender in 20th century Lansing. Lansing Speech Corpus (n=36), plus Millennial speakers (n=21) for reference. 103 As shown in the figure (and see Table 1 ), for women, stability in F2 between the Silent ( M = 1977 Hz, SD = 131.5) and Boomer generations ( M = 1996.5 Hz, SD = 127.5), t (9.99) = 0.285, p > 0.05) is followed by a significant decrease i n F2 values into Generation X ( M = 1753.4 Hz, SD = 70.9), who exhibit lower F2 values than their predecessors, t (8.07) = 4.23, p<0.01. In Figure 13 , F2 values remain stable for men. There are no significant differences in F2 values in the male group from Silent ( M = 1900 Hz, SD = 61.3) to Boomer generation ( M = 189.3 Hz, SD = 87), t (9.38) = 0.09, p > 0.05 nor in the transition to Generation X ( M = 1874.2 Hz, SD = 130.6), t (8.66) = 0.27, p > 0.05). 4.4.2 Summary of F2 R esults In sum, I observe that pre - oral /æ/ has moved farther back in the vowel space from the beginning to the end of the 20 th century, with the significant change occurring in the tran sition from Baby Boomer to Generation X speakers, at least for women. In the Silent and Baby Boomer generations, women exhibited fronter realizations than men. With the retraction of pre - oral /æ/ exhibited by women in the middle of the 20 th century, the pattern switched such that men exhibit fronter realizations than women. 4.5 Diphthongal Quality of P re - O ral /æ/ In this section, I examine diphthongal quality of pre - oral /æ/ by measuring the change in F1/F2 from the midpoint of the vowel to its offset. Diphth ongal quality (DQ) is third diagnostic of NCS /æ/. As discussed earlier in Chapter 2 , p re - oral /æ/ is more diphthongal in the Inland North than it is in other regions of North America , which have more monophthongal realizations of pre - oral /æ/ (Boberg & Strassel 2000; Labov, Ash & Boberg 2006) . In what follows, I test whether diphthongal quality of pre - oral /æ/ has changed over ti me for Lansing 104 speakers. The results of a mixed effects regression analysis are in Ta ble 10 ; a visual plot of DQ over apparent time is in Figure 14 . Ta ble 10 Best fit m ixed - e ffects m odel of pre - oral /æ/ Diphthongal Quality (DQ) Predictor Value(s) Estimate Std. Error df t - value Significance (Intercept: boomer female) 253.42 17.83 37.95 14.217 *** GEN GenX - 62.01 25.12 36.76 - 2.469 ** GEN Silent - 12.46 25.17 37.06 - 0.495 [0.624] GENDER male - 63.14 23.9 30.78 - 2.642 ** GEN GENDER GenX male 65.64 34.4 32.58 1.908 [0.065] GEN GENDER Silent male - 22.99 34.37 32.59 - 0.669 [0.508] Significant difference (p < 0.001) is reported as ***; significant difference (p < 0.01) is reported as **; significant difference (p < 0.05) is reported as *, and non - significant appears as []. Figure 14 , below, traces the diphthongal quality of pre - oral /æ/ (in red) across year of birth in the Lansing sample. For comparison, I have included mean DQ values of / / (in blue ) as a reference for a front monophthong. Higher diphthongal quality values on the y - axis indicates more values higher than 300 represent a large amount of transition between the vow its offset, whereas values around 150 represent a difference commensurate with that of a monophthongal realization . 105 Figure 14 Average Diphthongal Quality (Hz) values of pre - oral /æ/, /e /, and / / by speaker y ear of birth in 20th century Lansing. Lansing Speech Corpus (n=36), plus Millennial speakers ( n=21 ) for reference. As displayed in Ta ble 10 and visible in Figure 14 , there is a main effect of generational cohort but only in the t ransition from Baby Boomer to Generation X speakers, where diphthongal quality values fall, i.e. pre - oral /æ/ becomes more monophthongal. Neither social class nor its interaction with the other fixed effects come out as significant and they were thus excl uded from the final model. There is a significant main effect of gender in the model, but this effect disappears in Generation X, and the interaction is not significant in the model. 106 4.5.1 Diphthongal Quality and Gender Figure 15 displays DQ measure over time partitioned by gender. In this figure (and follow - up t - tests confirm it), there is an effect of gender in the first two generations, where women have hi gher DQ values than men. Figure 15 Average Diphthongal Quality (Hz) values of pre - oral /æ/ by speaker year of birth and gender in 20th century Lansing. Lansing Speech Corpus (n=36), plus Millennial speakers ( n=21 ) for reference. In the Silent generation, women exhibit larger DQ values ( M = 234.28 Hz, SD = 81.5) than men ( M = 152.39 Hz, SD = 25.1), t (5.85) = 2.44, p<0.05; and in the Baby Boomer generat ion, women ( M = 253.28 Hz, SD = 57.64) exhibit larger DQ values than men ( M = 183.1 107 Hz, SD = 25.2), t (7.22) = 2.77, p<0.05. This gender effect disappears in Generation X, however, where women (women M = 186.93) and men ( M = 187.44) have similar DQ values, t (9.38) = 0.05,p > 0.05. In other words, men and women /æ/ seems to move in opposite directions over time until converging in Generation X . Within the female group, there is no statistically significant effect of generational cohort on DQ values between silent generation ( M = 234.28 Hz, SD = 81.5) and B aby B oomer generation speakers ( M = 253.28 Hz, SD = 57.64), t ( 8.53 ) = 0.34 , p > 0.05 . There is an effect in the subsequent generational transition whereby Generatio n X women ( M = 186.93, SD = 31.3) have lower DQ values than their predecessors, t ( 8.32 ) = 2.48 , p<0.0 5 . Within the male group there is a statistically significant effect of generational cohort in the first half of the 20 th century. Silent generation men ( M = 152.39 Hz, SD = 25.1) exhibit smaller DQ values than B aby B oomer generation men ( M = 183.1 Hz, SD = 25.2), t (9.96 ) = 2.27 , p<0.0 5 . Subsequently, there is non - significant of generational cohort on DQ values; Generation X men exhibit similar DQ values ( M = 187.44, SD = 41.8 ) as their predecessors t ( 8.21 ) = 0.22 , p > 0.05 . Thus, for women, there was stability and then a decrease in diphthongal quality from the baby boomer to Generation X speakers. This is in contrast to the male group where DQ values rose and then remain stable. The interaction between generational cohort and gender was likely in significant in the regression model because though there was an interaction of gender and generational cohort in the transition from Silent to Baby Boomer generation, the gender effect disappeared in Generation X . 4.5.2 Summary of Diphthongal Quality Results In sum, diphthongal quality of pre - oral /æ/ in Lansing has have shift ed toward monophthongization . Diphthongal quality peaked for women in the 1 920s 1940s and then declined, while that for men peaked in the 1960s and then declined. The transition from a female 108 lead in previous generations to no gender difference in Generation X is reminiscent of the social change we have previously seen for the F1 and F2 of /æ/ in sections 4.3 and 4.4 . Clearly, the post - Boomer period is a critical one for /æ/ in the Lansing speech community. 4.6 Nasal A llophony The last diagnostic of NCS participation to be considered in this dissertation is that of pre - nasal and pre - oral token distribution at the level of the speaker. We know from previous studies that Inland Northern speech is characterized by unconditioned and continuous /æ/ systems, though more recent studies have noted the appearance of the supra - local n asal system (Chapter 2) . In what follows, I examine the effect of generational time on the se systems as measured by Pillai - Bartlett score in twentieth century Lansing (see 2.2.2.6 for details on Pillai - Bartlett) . Table 11 displays the output of an ANOVA analysis on Pillai - Bartlett scores in Lansing . The only significant effect on these scores is of speaker year of birth ; gender and social class ha d non - significant on Pillai - Bartlett scores. Table 11 ANOVA results of Pillai - Bartlett score analysis Df Sum Sq Mean Sq F value Pr(>F) YOB 1 25.13 25.13 21.85 4.53 ** * Residuals 3 4 39.11 1. 15 Figure 16 displays the Pillai - Bartlett scores for each speaker in the sample across year of birth. The higher the Pillai - Bartlett measurement (from 0 t o 1), the greater the separation between pre - oral and pre - nasal token clouds for that speaker. 109 Figure 16 Per speaker Pillai - Bartlett score of pre - nasal and pre - oral /æ/ distribution by speaker year of birth in 20th century Lansing. Lansing Speech Corpus (n=36), plus Millennial speakers ( n=21 ) for reference. In the figure, we see that the proportion of speakers with Pillai - Bartlett scores close to 1 continuously increases from the earliest birth dates onward. This is confir med by the ANOVA analysis ( Table 11 ). Social class, gender, and their interactions with year of birth were not significant in the model. The general picture of the relationship between pre - nasal and pre - oral token cloud overlap in Lansing is one of incre asing separation over the 20 th century. 4.6.1 Summary of Nasal Allophony Results In sum, there has been a change in the relationship between pre - nasal and pre - oral allophones of /æ/ in 20 th century Lansing. The community has increased the distance between /æ/ in these two environments over time. Many speakers born in the 1920s had Pillai - Bartlett scores close to 0 indicating an advanced raised /æ/ system. Speakers born in the 1950s and 1960s have larger 110 Pillai - Bartlett score s indicating nasal conditioning, where their /æ/ systems resembled that of a more common (less advanced) raised /æ/ system. Speakers born in the 1990s have Pillai - Bartlett scores ever closer to zero. The general trend, then, is that t he advanced raised sy stem is gradually giving way to a nasal system. I note that this change is ongoing, however. Though speakers born in the 1990s have higher Pillai - Bartlett scores than their predecessors, none of them have scores of 1, which would indicate a true gap in p honetic space between pre - nasal and pre - oral token clouds. 4.7 Summary: /æ/ in 20 th C entury Lansing The above analysis of the impact of external and internal factors on the realization of /æ/ in 20 th century Lansing has shown linguistic and social reorganizat ion, as summarized in Table 12 and Table 13 . Linguistically, the community has moved from an unconditioned /æ/ system to a nasal system. In opposition to pre - nasal /æ/, the pre - oral allophone has lowered and retracted in the vowel space and has decreased in diphthongal quality. This general trend i s mediated by the fact that there was relative stability between the S ilent and B aby B oomer generations that was followed by a significant d ownward trajectory in each of these measures in the transition to Generation X . I also note that what looks like sta bility between Silent and Baby Boomer generation means is actually the result of gender conditioning such women advanced NCS features during this period while men did not participate in the change. In short , all four of the diagnostic measures of NCS /æ/ collectively demonstrate that the Lansing speech community has shifted away from the NCS toward a higher mean F1 (lower in the vowel space) , lower mean F2 (farther back in the vowel space) , lower mean DQ (more monophthongal) , and higher Pillai - Bartlett sco re s (more separation between pre - nasal and pre - oral allophones) . 111 Table 12 Distribution of /æ/ features across generational time in 20 th century Lansing Silent to Boomer Boomer to Generation X F1 Gender conditioned change Stability F2 Stability Retraction Diphthongal Quality Stability Less diphthongal Nasal conditioning More nasal conditioning More nasal conditioning With regard to the social re - organization of NCS /æ/ characteristics in Lansing, Table 13 summarizes which gender group was most advanced for three of the four acoustic measures explored in this dissertation across each generation; in the table, the NCS version of the feature is more indicative of the specified gender group in that generation . Nasal conditioning , as measured by Pillai - Bartlett score is not included, as the only social dimension relevant for t his measure is generational cohort . I observe that social conditioning with respect to the other three features has changed in the community. Table 13 Summary of social conditions on acoustically salient NCS features in 20 th century Lansing Silent Baby Boomer Generation X NCS Raising (F1) Female Male NCS Fronting (F2) Female Female Male Diphthongization Female Female 112 From the Silent and Baby Boomer generations, women were most advanced in NCS /æ/ raising, fronting and diphthongization. In Generation X, however, as women reversed course and retreated from the NCS, it was men who exhibited more NCS features. In Lansing, women led men in advancing NCS features (unconditioned raising, fronting, diphthongization) in the beginning of the 20 th century . In Generation X, men advanced NCS features while women began to diverge from the range of these measures in prior generation s thus leading the community change towards NCS reversal/rejection. These findings are unsurprising, given the fact that w omen are often the leaders of community change (Labov 1990). The large amount of inter - speaker variation in the Baby Boomer genera tion followed by a switch in social conditioning in Generation X further support the claims in Nesbitt (2018) and Nesbitt, Wagner & Mason ( in press) which argue that NCS features were at their height for speakers born in the middle of the 20 th century and that they began to decline thereafter. In Chapter 5, I explore the development and spread of nasal allophony for /æ/ at the level of the speaker. In Chapter 6 , I discuss the implications of the results of th e phonetic and phonological analys es in this di ssertation with a view to some future potential lines of investigation. 113 PHONOLOGIZATION OF / æ / N ASAL A LLOPHONY IN LANSING 5.1 Introduction I now return to the main research question of this dissertation: What has been the impact of these changes in phonetic targe t and social conditioning of /æ/ on the phonological systems of Lansing speakers? This chapter addresses two questions regarding th e development of a phonological distinction between pre - nasal and pre - oral /æ/ in Lansing, MI. The first concerns the Actuation Problem which asks why the change was initiated. For this, we will ask whether phonological /æ/ - nasal allophony develop ed gradu ally from a phonetic precursor or wa s abruptly posited in the community ? The second concerns the T ransition P roblem : how did the change spread throughout the community? I utilize a combination of methods to answer these questions in this chapter, includi ng impressionistic visualizations of phonetic differentiation between allophones of /æ/ and the analysis of a sub - phonemic judgement task. The methodology for both has been provided in Chapter 3 (section 3.3 ). As a reminder, the data for the trajectory visualizations come from the Lansing Speech Corpus plus the Millennial speakers in Wagner et al (2016) and Nesbitt (2018), while the data for the judgement task come from Lansing Speech Corpus speakers plus others recruited through supplemental methods. As I will show, the results of these analyses suggest that phonologization is gradual in Lansing. The trajectory analysis finds that the pre - nasal allophone is phonologically distinct from the rest of the allophones and that it began to distinguish itself in F1/F2 space among speakers born in the 1930s. Subsequently, the sub - phonemic judgement task shows that no 114 speakers born at that time (or even a decade later) posit a phonological distinction between the allophones. This finding that phonologization was gradual supports the analysis in Chapter 2 of Pillai - Bartlett scores in Lansing over time , which showed the progression of systems from NCS raised to continuous to nasal. The analysis of social factors in this chapter suggests that white - collar (women) respondents are the leaders of this change and that blue - collar respondents are on the trailing end . In what follows, I present the results of the trajectory analysis (sectio n 5.2.1 ). I analyze the responses to the sub - phonemic judgement task in section 5.2.2 . Section 5.3 concludes this chapter with a discussion of the methodological implications of the sub - phonemic judgement task. I consider the implications of these results for our theories of allophonic phonologization specifically and phonological change more generally. As a reminder, this dissertation distinguishes between allophones that are phonologically distinct and those that are distinguished only phonetically. Phonological allophones are distinguished from one another by means of a phonological rule acting on abstract phonological representations and thus have different phonological features. Phonetic allophones have the same phonological features and differ only in phonetic realizat ion. 5.2 Results 5.2.1 Phonetic and P honological A llophony To determine whether any allophone of /æ/ has become phonologically distinguished from the rest of the allophones, I begin with a discussion of trajectories over time for each allophone . To determine whethe r sub - phonemic conditioning of /æ/ in Lansing is phonological, I compare th e analysis of /æ/ with the same for two other short - front vowels; / / and / / . I follow this analysis with a determination of when in the apparent time trajectory divergence is evi dent for /æ/ . For 115 these comparisons, I utilize the production data from the Lansing Speech Corpus plus that from Millennials in the Wagner et al (2016) study. For reference, birth years in the sample are as follows: 1925 1945 = Silent generation, 1946 1964 = Baby Boomer, 1965 1984 = Generation X, 1985 1999 = Millennial. In Figure 17 , I show the trajectory of /æ/ in three following phon ological environments; before fricatives (in red), nasals (in green), and stops (in blue). Each dot on the plot represents a token. I set aside the trajectories of the pre - / l / and pre - affricate allophones because of their low token counts in the sample, N =174 and N=166, respectively. These two allophones are produced on average only 3 times per speaker, which makes any conclusions about conditioning less robust. For transparency, I have included a figure of the trajectories of all five phonological envir onments in appendi x 0 . For now, I focus on a comparison of the trajectories for the pre - nasal (in green), pre - fricative (in red), and pre - stop (in blue) allophones because token counts for these are robust in the sample, N=2235, N=1382, and N=2332, respectively. In Figure 17 , we see a clear distinction between the trajectories of the pre - nasal and pre - oral allophones. The pre - fricative (in red) and pre - stop (in blue) allophones f luctuate in diagonal measurement into Generation X and then decline into the Millennial generation; they move down and backwards in the vowel space over time. This is in contrast to the pre - nasal allophone (in green) which remains relatively stable from t he 1920s onward. 116 Figure 17 Per token diagonal measurement of /æ/ in three following phonological contexts by speaker year of birth in 20 th century Lansing. We can compare the allophonic trajectories of /æ/ with those of / / and / /. In Figure 18 , I show the trajectory of / / (left facet) and / / (right) in the same three following phonological environments; before fricatives (in red), nasals (in green), and stops (in blue). We again ignore the other following environments, as token counts for them are relatively low. 117 Figure 18 Per token diagonal measurement of / / (left facet) and / / (right facet) in three following phonological contexts by speaker year of birth in 20th century Lansing. For both of these vowel classes, we find that the pre - nasal allophone moves in lockstep with the pre - stop and pre - fricative allophones. Therefore, I argue that unlike for /æ/, the pre - nasal allophones of / / and / / are p honetically rather than phonologically distinguished from the rest of the allophones. These vowel classes have a different phonological make - up than /æ/ in terms of nasal conditioning; the latter moving towards a phonological distinction and the former 118 on ly phonetically distinct. I will utilize this distinction (phonetic vs phonological allophony) in the analysis of the judgement data to follow. If the sub - phonemic judgement task is indeed capturing the difference between phonological and phonetic allop hony, respondents should react differently when presented with / / and / / words than when they are presented with /æ/ words. As for a more precise estimation of when in apparent time the pre - nasal and pre - oral allophones of /æ/ beg a n to diverge, we focus on the trend lines in the 1920 to 1940 birth year range in Figure 17 (approximately the Silent generation) . Focusing on the period between 1920 and 1930, the pre - nasal (green line) and pre - fricative allophone (in red) already show distinct trajectories; pre - nas al increasing and pre - fricative slightly decreasing. In this same time window, the pre - nasal (green line) and pre - stop allophone (blue line) are moving in lockstep; both upward. In the 1930s, the pre - stop allophone diverges from that upward trajectory. F rom the 1940s onward, the pre - stop and pre - fricative allophones move in lockstep. In conclusion, visual inspection of the F1/F2 diagonal trajectory of /æ/ in multiple following phonological environments indicates that at the community level, pre - nasal /æ / and pre - oral /æ/ have distinct phonological features, whereas pre - nasal and pre - oral / / and / / are distinguished via phonetic implementation rules. Additionally, we have seen that the nasal allophone diverged from the rest close to 1930 which is in th e Silent generation, (birth year 1925 1945). Therefore, in the analysis of whether phonologization of allophonic splits occur abruptly or gradually in a community, I will focus on Silent generation speakers. 5.2.2 Sub - P honemic J udgement T ask: Actuation and Tran sition of P honologization In the previous section, I showed that pre - nasal /æ/ began to move along a different trajector y from the rest of the allophones in F1/F2 space in the Silent generation (1925 1945 ). As such, the actuation question will be addressed by examining phonologization in this generation. If 119 phonologization was abrupt in Lansing, the difference between pre - nasal and pre - oral /æ/ should be phonological for at least some Silent generation r espondents. If, however, phonologization was gradual in Lansing, the difference between pre - nasal and pre - oral /æ/ should not be phonological for any of the Silent generation respondents but should appear in subsequent generations. Under the gradual hypot hesis , Silent generation respondents will only distinguish the two allophones phonetically. A complete description of the methodology for this task is provided in Chapter 3 (section 3.3 ) , which includes a description of the statistical analyses employed below. I include a brief description of the methods here as a reminder to the reader. D uring the sub - phonemic judgement task, participants were presented with two CVC lexical items and were asked to indicate whether the vowels in the two words sound the same or different to them. Responses from 107 white monolingual English - speaking Lansing natives are analyzed here . In this task, Condition 1 was composed of 5 pairs of CVC words that had the same onset and nucleus /æ/ but differed in whether the coda consonant was nasal or oral, e.g. pat and pan. Condition 2 was composed of 5 pairs of CVC words that had the same onset and nucleus /æ/ but had different oral coda consonants, e.g. pat and pass. The hypothesis for this experiment is that participants who distinguish pre - nasal and pre - oral /æ/ phonologically will be more likely to respond DIFFERENT to Condition 1 pairs than they are to Condition 2 pairs because the latter compare /æ/ tokens that differ only phonetically while the former compare /æ/ tokens that have different phonologica l representations . Those who do not have a phonological distinction between pre - nasal and pre - oral /æ/, however, should respond DIFFERENT to Condition 1 and Condition 2 pairs at the same rate. To determine when the phonological change occurred , I ask in which gener ation do we observe that Condition 1 pairs garner more DIFFERENT responses than Condition 2 pairs . 120 Additionally, the trajectory analysis in section 5.2.1 showed that the two allophones began to differentiate in the ir acoustics in the Silent Generation (date of birth 1920 1945) . T herefore, if this change occurred abruptly, responses between Con dition 1 and Condition 2 must be different in this generation. If difference is not noted in this generation but observed in subsequent generations , allophonic reorganization must have occurred gradually in Lansing. In addition to generation, this analys is includes social class as a factor, as the spread of phonological change appears to be conditioned by social class. Gender was not included as an external factor due to the small number of males across generations in the sample. The binomial mixed - effe cts logistic regression model included Response (same, different) as the dependent variable, and Condition (1,2), Generational Cohort (Silent, Boomer, Gen X, Millennial), and Social Class (Blue - collar, White - collar) as predictor variables, and Participant and Pair as random intercepts. Table 14 Mixed - effects regression model for responses to CæC pairs. Predictor Values Estimate Std. Error df t - value (Intercept: BlueCollar, Cond 1, Boomer) - 2.783 0.37604 - 7.401 *** Cond Cond2 0.232 0.55423 0.419 [0.675] GEN GenX 1.079 0.36434 2.961 ** GEN Millennial 2.075 0.35692 5.812 *** GEN Silent 0.096 0.89294 0.107 [0.914] SocialClass WhiteCollar 1.039 0.19922 5.216 *** Cond GEN Cond2 GenX - 0.862 0.68858 - 1.253 [0.210] Cond GEN Cond2 Millennial - 2.358 0.72244 - 3.264 ** Cond GEN Cond2 Silent NA NA NA NA Significant change (p < 0.001) is reported as ***, significant change (p < 0.01) is reported as **, slightly significant change (p < 0.05) is reported as *, and non - significant appears as []. 121 The best fit model for the data is provided below in Table 14 . Blue - collar, Condition 1, and the Baby Boomer generation are set as the reference level. This model included the significant two - way interaction between Generational Cohort and Condition, and significant main effects of Generational Cohort and Social Class. In what follows, I will report on the interaction results and then the social class effect. F igure 19 displays the distribution of the percentage of responses to lexical pairs over time . In the figure, generational cohort appears on the x - axis and response percentage appears on the y - axis. Condition 1 responses appear on the left and Condition 2 responses are on the right. Different responses are shown in yellow and same responses are in blue. F igure 19 Distribution of responses to pre - nasal and pre - oral pairs (left = condition one) compared to pre - oral and pre - oral pairs (right = condition two) of /æ/ over generational time. 122 The first observation to be made about the figure i s that respondents judge pairs in Condition 2 to be different at least 95% of the time no matter their age. We compare this to the general increase in the amount of different (indicated in blue) over apparent time. The regression analysis did not compare the two conditions in the Silent generation because no Silent generation participants responded different to Condition 1 pairs and the dependent variable (response) required two levels (same, different) . Since different responses for C ondition 1 are at z ero for this group and those for C ondition 2 are at 2%, we would not expect to find a significant effect of condition on responses in this generation this group behaves the same under each condition. In the Baby Boomer generation, though different respon ses to Condition 1 increase to 3%, this is not significant in the model. Although in Generation X, different responses increase to 20.7%, this is also not significant in the model. The only combination of condition and generation that is significantly different from the baseline is found in the Millennial generation. Millennial respondents are significantly more likely to judge pre - nasal and pre - oral /æ/ as different from one another t han pre - oral pairs. Different responses in the Millennial group reach 44% under Condition 1 in this experiment. Thus, while respondents do not judge pre - oral /æ/ pairs to be different, the community is moving towards distinguishing between pre - nasal and pre - oral /æ/. The initiation of this move appears to have occurred in Generation X with the significant increase occurring in the subsequent Millennial generation. To provide a clearer view of the social/external conditions on this change, I turn to an in vestigation into the impact of social class in each generation on responses in this task, which is displayed in Figure 20 . In the figure, as above, gen erational cohort appears on the x - axis and response percentage appears on the y - axis. Condition 1 responses appear on the left and Condition 2 responses are on the right. Different responses are shown in yellow and same 123 responses are in blue. W hite - collar respondents are at the top and blue - collar respondents are on the bottom facet. Figure 20 Distribution of responses to pre - nasal and pre - oral pairs (left = condition one) compared to pre - oral and pre - oral pairs (right = condition two) of /æ/ over generational time and by Social Class (White Collar respondents at the top). In what follows, I will only report on responses to Condition 1 pairs (left facet) , since as we can observe in Figure 20 , different responses to Condition 2 pairs remain below 10% across all generations and social class groups. In the Silent generation, no respondent indicated that Condition 1 pairs were different , regardless of their social class. In the Baby Boomer blue - collar sample, different responses remain at 1%. White - collar Baby Boomer respondents judge 124 Condition 1 pairs to be different 6% of the time. This suggests an increase in perception of a difference , but I note that these respondents also judge Condition 2 pairs to be different 4% of the ti me. In Generation X, blue - collar respondents judge Condition 1 pairs to be different 10.7% of the time while white - collar respondents judge them to be different 30.7% of the time. And finally , in the Millennial generation, blue - collar respondents increase in the amount of different responses to Condition 1 pairs to 25.4%. White - collar Millennials also increase the amount o f different responses to Condition 1 pairs to 60.4%. In sum, I note that while different responses to Condition 2 pairs remain at 1% i n each generation and across social groups, there is a general increase in the amount of different responses to Condition 1 pairs across time in both social class groups. In the Silent and Baby Boomer generations, respondents consider all tokens of /æ/ to be the same, no matter the conditioning environment. There is an increase in perceived difference between pre - nasal and pre - oral tokens in G eneration X . This increase is really only evident in the white - collar sample and the difference between conditions is not statistically significant for them at any rate . There is a larger increase in distinction between the two conditions in the Millennial generation . In this Millennial generation, respondents are significantly more likely to consider pre - nasal and p re - oral pairs to be different than in any other generation. Another c rucial finding is that the aforementioned pattern is more prominent in the white - collar as compared to the blue - collar M illennial sample. Significance in the Millennial generation is only observed in the white - collar group, as opposed to the blue - collar group. Therefore, white - collar respondents a ppear to be leading in the development of a phonological distinction between pre - nasal and pre - oral /æ/ ; they are not only the first to judge /æ/ as different in the two phonetic environments, but younger white - collar Lansingites are significantly more 125 likely th an blue - collar Lansingites to judge the two allophones as different during this task . Th ese findings indicate that while nasal allophony is phonological for some speakers in the Millennial generation, the spread of this change is a more recent development . 5.2.2.1 Phonetic Conditioning One tool for distinguishing phonetic from phonological allophony in this chapter was to compare responses to Condition 1 versus Condition 2 pairs. If a participant judged pre - oral pairs (Condition 2 ) alone to be different in this task, then it must be the case that co - articulatory/acoustic coupling is what is being captured by this experiment rather than any representational difference. The results above indicate that this is not the case respondents do not judge allophones that are not differentiated p honologically to be distinguishable enough to prompt a different response. Another tool utilized here is an analysis of responses of pairs whose conditioning environments are not differentiated due to a phonological rule / / and / / in various environments. For this, I consider responses to lexical pairs in Conditio ns 3 ( pen - pet ) and 4 ( pet - peck ). Figure 21 displays the distribution of responses to the lexical pairs with / / or / / as their relevant vowel across gene rational time. Condition 3 pairs are in the left - hand facet and C ondition 4 pairs are on the right. Different responses are represented in yellow and same responses are represented in blue. The results for these conditions are unsurprising, given the tr ajectory analysis in section 5.2.1 . Respondents in every generation judge / / and / / phonemes before oral and nasal consonants to be the same almost categorically; 2% of respondents in the Millennial generation judge them to be different. 126 Figure 21 Distribu tion of responses to pre - nasal and pre - oral pairs (left = condition one) compared to pre - oral and pre - oral pairs (right = condition two) of / / and / / over generational time. As with the judgements for pre - oral pairs of /æ/, respondents in Lansing do not distinguish between pre - nasal and pre - oral / / phonologically. The same is true of C ondition 4 participants do not consider / / and / / before various oral consonants to be different from one another. Thus, the difference between / / in any of the se environments is merely phonetic rather than part of the phonological make - up of the vowel. The comparison of the results for these vowel classes to those for /æ/ suggests that performance on this task can capture sub - phonemic representation at the leve l of the individual. 127 5.3 Conclusion The results of th ese analyses suggest that /æ/ nasal allophony in Lansing developed gradually. Respondents at the beginning of the separation of pre - oral from pre - nasal /æ/ in phonetic space Silent generation respondent s do not judge this vowel to be different in the two environments. It is one generation later , in the Baby Boomer generation , that we observe some participants in the community judging /æ/ in these two environments to be dif ferent, and three generations later , in the Millennial generation , that these judgements are statistically different from judgements to pre - oral pairs. Thus, the phonological rule was not already present in the community before/as the two allophones began to diverge in acoustic space , in opposition to the prediction of the Big Bang Theory (Janda & Joseph 2003) and Baker, Archangeli & Mielke (2011), and as exemplified by Fruehwald (2016) . The findings in this dissertation suggest that the allophonic rule in Lansing developed only after pre - nasal and pre - oral /æ/ were assuming different phonetic targets, in line with theories of gradual phonologization supported by The Life - Cycle of Phonological Processes (Bermúdez - Otero 2003 , 2013; Bermúdez - Otero & Trousdale 2008; Ramsammy 2015) , Hyman (1975; 2013) , Evolutionary Phonology (Blevins 2004, 2006) , and Exemplar Theory (Pierrehumbert 2001; Garrett & Johnson 2011) . The analysis of social factors provides a picture of how this change spr ead t hroughout the community. The results suggest that the spread of nasal allophony is being led by white - collar community members. While there is a general increase in different responses over apparent time in both the blue - and white - collar community, these judgements first appear in the white - collar G eneration X group, and it is only in the white - collar group in the Millennial generation that different judgements are statistically significant between the two conditions. I n line with the sociolinguisti c literature on supra - local change, phonologization appears to have actuated in the 128 white - collar community first and then spread to t he blue - collar community. Th ough the impact of participant gender on responses was not investigated in this chapter , I will offer some insights about gender based on impressionistic observations in section 6.1 . 129 CONCLUSIONS AND DISC USSION The goals of this disserta tion were to address the Actuation problem of allophonic change in Lansing, i.e. why does change occur at a particular place in a particular time. In order to do this, I also address ed the Embedding and Transition Problems what are the internal and extern al conditions surrounding the chang e and how did the change spread throughout the community. To address these problems, I utilized a combination of analyses. First, using a corpus of natural language data from 36 Lansing natives, I conducted an acoustic a nalysis of /æ/ in F1/F2 space. In particular, I measured changes in F1, F2, diphthongal quality, and relative distributions of pre - nasal and pre - oral /æ/ within each speaker in the sample. I supplemented this with the results of a sub - phonemic judgement t ask administered to 107 Lansing natives via an online survey. Below, I summarize my findings. The results of this dissertation provide some evidence for previous claims in the literature that address these problems but run counter to others. Additionally, these results offer some important new avenues for investigation. Below, I discuss each of these in turn. First, the results of this dissertation confirm that both social/external and linguistic/internal factors are involved in phonological change. Commonplace in the phonological change literature is to assert that phonologization of coarticulation occurs when a mec hanical effect is so strong that it is interpreted by a hearer as a distinct phonetic target. Baker et al. (2011) argue that this is too simple and in fact overpredicts the occurrence of this type of sound change because mis - articulations are so common. They claim that inter - speaker variability in the degree of the phonetic effect is what promotes this change, i.e. the range of the effect must be so great for some speakers that those speakers produce a sound that others 130 perceive as a distinct target. The y also argue that the likelihood of the hearer adopting the novel target in their speech depends on social factors. These two points have allowed them to update coart iculation to lead to sound change depends on the chance alignment of extreme provide clear support for this hypothesis. In Chapter 4 ( Figure 7 ) , I showed that at the turn of the 20 th century, there was minimal inter - speaker variation with respect to conditioning of pre - nasal /æ/ , such that all speakers exhibited very low Pillai - Bartlett scores. In the Baby Boomer generation, however, there is considerable inter - speaker variation so me speakers show no conditioning of the pre - nasal environment while others do, and crucially, there is considerable variation with respect to pre - oral productions which can be accounted for by gender - women exhibit much higher and more forward realization s in this generation than men. Respondents born one generation later then begin to distinguish between the allophones in the sub - phonemic judgement task. It appears that these circumstances are what have driven this phonological change; considerable vari ability in conditioning combined with considerable social conditioning. Thus, the chance alignment of inter - speaker variation and social influence is what appears to have motivated allophonic change in Lansing. Second, through the analysis of Pillai - Bart lett scores, and judgement task results, I have shown that there are t hree types of allophony operating in the Lansing speech community with respect to pre - nasal and pre - oral conditioning environments. The first is phonetically motivated but not controlled , the second is a speaker controlled phonetic effect, and the third is the result of a phonological rule. In the vowel space, speakers who exhibit phonetically implemented but not controlled allophony have overlapping pre - nasal and pre - oral token clouds, and they do not judge 131 /æ/ in these environments to be different. Those who exhibit controlled phonetic allophony exhibit some conditioning of the pre - nasal environment but no clear distinction from the pre - oral cloud. They also do not report that the vowe ls in these two environments sound different from one another . Speakers with a phonological rule, on the contrary, have discretely separated pre - nasal and pre - oral token clouds and are more likely to report that /æ/ is different in these environments. Th is type of allophony is characteristic of speakers in Lansing born in the latter half of the 20 th century , who had a clear distinction between pre - nasal and pre - oral /æ/. I therefore provide evidence of the first three stages of the Life - Cycle of Phonolog ical Processes operating in Lansing: epiphenomenal/absence of conditioning, controlled phonetic implementation, and phonological distinction. Crucially, I show that there is a diachronic progression through these stages in the Lansing community. In th e Silent generation (birth dates 1925 1945) there is a mixture of the first two types of allophony in the community. In the Baby Boomer generation (birth dates 1946 1964), Pillai - Bartlett scores increase , suggesting a shift towards more pre - nasal conditio ning . In Generation X (birth date 1965 1985), the first type decreases even more as Pillai - Bartlett scores rise . By the Millennial generation (birth date 1985 1999), no speaker exhibits the first type, though there is an increase in pre - nasal conditionin g . This generational progression reflects exactly what is predicted in the Life - Cycle of Phonological Processes (Bermúdez - Otero 2007; Bermúdez - Otero & Trousdale 2012) the diachronic progression from phonetically motivated but not controlled allophony (stage 1 of the Life - Cycle), to a speaker controlled phonetic effect (stage 2 of the Life - Cycle), to a phonological rule (stage 3 of the Life - Cycle) but runs counter to the theory of abrupt phonological change (Fruehwald 2013 , 2016; Janda & Joseph 2003) . 132 With regard to the Embedding problem, the results of the acoustic analyses in Chapter 4 show that Lansing participated in the NCS in the earlier half of the 20 th century but began to adopt features characteristic of the supralocal Low Back Merger Shift after the 1960s. During this time, pre - oral /æ/ begins to lower and retract in the vowel space and show less diphthongal movement from the nucleus to the offset of the vowel. I also show that women led both the advancement of the NCS and the adoption of the LBMS. Social class was not significant in these analyses, though it has been in other analyses in the dialect area. I suspect that this is because of the small sample size combined with the change in trajectory in the middle of the 20 th century for all of these measures. The sub - phonemic judgement experiment ( Chapter 5 ) did reveal a social class effect, though as I note below, this sample was imbalanced for gender, such that women ma de up much of the sample. Thus, it may very well be the case that social class conditioning is only relevant for women, as would be expected given that (Labo v 2001: 321) . 6.1 Gender and A llophonic C hange in Lansing It is worth specifying here that it is likely the case that the specific leaders of this change are white - collar women. I was not able to control for gender in th e sub - phonemic judgement task , since the respondent count in the earlier generations is too low for a statistical analysis and male Lansingites are underrepresented in the sample (see Table 7 ) . Gender, however, is an important condition ing factor in the acoustic distribution of /æ/ (see Chapter 4 for details). A visual inspecti on of the distribution of responses during the sub - phonemic judgement task appears to point to this same finding. Figure 22 displays the distribution of responses to Condition 1 pairs only, by generational cohort (x - axis), social class (side facet; white - collar on top), and gender (top facet; female respondents on left). 133 Figure 22 Distribution of responses to pre - nasal and pre - oral pairs of /æ/ over generational time, by social class (White Collar respondents at the top), and gender (women on the left). The crucial thing to note her e is that in Generation X, men of both social class groups and blue - collar wo men pattern the same: although some respond that pre - nasal and pre - oral /æ/ are different, this only occurs 20% of the time for these groups. Generation X white - collar women seem to distinguish between pre - nasal and pre - oral /æ/, with different responses almost at 40%. From this, it appears that white - collar women are driving this change. In the subsequent, Millennial generation, different responses increase for white - collar men and women and for blue - collar women, although the proportion of responses is more substantial for the white - collar respondent s than the blue - collar respondents . Again, these observations are based on a small 134 number of respondents in each group and should be interpreted with caution. It suggests, though, that any study examining this change in this community would find that white - collar women were the first in Lansing to posit this rule, while other community members are on the trailing en d of this change. And, in line with the acoustic analysis in Chapter 4 , blue - collar men are not yet participating. This provides an avenue for futur e studies examining the Actuation , Embedding, and Transition probl e ms of this and perhaps any other allophonic change. 6.2 Internally M otivated C hange f rom A bove Given the results of the sub - phonemic judgement task ( Chapter 5 ), allophonic change in Lansing appears to be a change from above; with white - collar women leading. There are two findings that give me pause in this conclusion, however. First, in Chapter 2 , I showed that in - migration from other dialect/language areas and non - White groups was limited in Lansing ; therefore nasal allophony is likely an inte rnally motivated change, which seems to go against the description of a change from above. This may not be a problem, as some studies have shown that changes from above can be internally motivated (Schilling - Estes & Wolfram 1999) . Another issue for the account that this is a change from above is that the acoustic analyses of spontaneous speech in the Lansing Speech Corpus ( Chapter 4 ) did not reveal an effect of social class this includes the analysis of per - speaker toke n distributions. This latter issue may be accounted for if we consider that the Lansing Speech Corpus was composed of only 36 speakers. Perhaps with more speakers in the sample, social class might prove significant. What is clear from this discrepancy i n results, however, is that contact (or lack thereof) is perhaps more important than social conditioning in differentiat ing abrupt versus gradual changes. 135 6.3 Contact as a C onditioning F actor The departure of the results of the current analysis from those of the few studies supporting a theory of abrupt phonologization bring up what I suspect is a crucial fact about the mechanisms of allophonic phonologization perhaps both abrupt and gradual phonologization are possible (as is the case in the literature on p honemic mergers) and the likelihood of either one occurring in a community is dependent upon some external/social conditioning factor(s). I put forth a conditioning factor here: the presence of inter - dialect contact and the impact of awareness on the chan ge. We know from previous studies on allophonic change that variable realizations lead to reallocation of phonetic environments whereby a phoneme assumes one target in one environment and another target in another environment (Britain 2002; Baker, Archangeli & Mielke 2011; Mielke, Baker & Archangeli 2016; Trudgill & Foxcroft 1978) . The crucial difference between contact - induced phonological changes and those tha t are internally motivated is the difference in magnitude of the difference between realizations , and relative awareness of the change . In changes from above, variable pronunciations are prominent upon contact, and thus phonologization is expected to occu r abruptly, given the right social circumstances. In internally motivated changes, however, variable realizations are less likely to gain social awareness and occur so rapidly. As discussed in Chapter 1 , we know that in the literature on changes towards phonemic mergers, inter - dialect/language contact and awareness condition whether a change is abrupt or gradual. Mergers that are abrupt are character istic of communities undergoing change via contact, while mergers that progress gradually and under the level of awareness occur in communities where inter - dialect/language contact is low (Guy 1990) . There is no reason why allophonic changes would Fruehwald (2013, 2016) finds that allophonic change in Philadelphia was abrupt, and though there is no 136 evidence of in - migration in the statistical analysis he performed on his sample of Philadelphia - born residents, it is still likely that contact can account for abruptness in his data, as Philadelphia is a large city that like other cities in the US have experienced in - migration throughout the 20 th century. The results of the analysis of nasal allophony in this diss ertation finds that the allophonic rule developed gradually, i.e. after the phonetic changes were occurring. Crucially, Lansing is not a contact city and allophony appears to be operating below the level of awareness. Although there was considerable in - mi gration from other Michigan towns into Lansing until the middle of the 20 th century, there is no indication that large numbers of speakers from other dialects or languages have migrated here. If contact and awareness are indeed relevant for allophonic phonologization and I believe they are it is unsurprising then that phono logical change was gradual in Lansing but abrupt in Philadelphia . Berkson et al. (2017) do not address any aspects of contact or social conditioning on /a / phonologization , so I am not sure if this distinction holds . However, given the apparent differenc es between the Philadelphia and Lansing results , my position is that there is a relevant component to our theory of allophonic phonologization that can be tested in the future, and that has been attested in the literature on phonemic mergers in progress (s ee Guy 1990; Herold 1997) namely that there are two mechanisms by which allophony can arise abrupt and gradual and the incidence of each is dependent on when variable realizations and social conditioning are introduced into the community. To be more explicit: I hypothesize that allophonic change is abrupt in communities where there is contact between dialects/languages with different realiza tions of the target phoneme and perhaps some social evaluation but gradual in non - contact communities where the change is relatively unnoticed. This statement can thus account for all phonological changes including changes toward phonemic mergers and al lophonic splits. 137 6.4 Methodological Considerations There are a few aspects of this research that I might consider if I were to conduct this research again. I address some of them below and hope that future studies will take them into account. 6.4.1 Awareness of Variable Phonologies We know from previous studies that sociolinguistic awareness of a phonological change can impact behavior on judgement tasks. An explicit example of the effect of awareness on examination of two mergers in progress in Charleston, South Carolina. Baranowski showed that while the community was moving towards both the cot - caught and pin - pen mergers, participants behaved differently during the judgement tasks that involved these vo wels. While participants were willing to judge cot - type and caught - type words to be the same, they were unwilling to judge pin - type and pen - type words to be the same even though many of the speakers displayed a clear merger of the two sounds. Unlike the cot - caught merger, the pin - pen merger is a marked feature of the South and thus socially undesirable in Charleston. The cot - caught merger, however, is progressing well below the level of awareness in the community. As such, in Charleston, judgements mat ched the production data for the cot - caught merger but not for the pin - pen merger. Participants negatively evaluated pin - pen and were thus less likely to say they were the same sound. What this shows is that the use of a judgement task on phonological c hanges that operate above the level of awareness in a community is not useful. While, as I speculated earlier, raised /æ/ has possibly risen to the status of a marker in the Inland North, in contrast, nasal allophony does not appear to be operating above the level of awareness in Lansing or anywhere else in North America. So, it is assumed here that the issue of awareness is not relevant for this change in progress. Future research on 138 phonological change will do well to ensure that the variable in questio n is operating under the level of awareness when utilizing a same - different judgement task. 6.4.2 Acoustic Salience vs Representation One potential confound of this study is that participants may be judging different productions of pre - oral /æ/, since each g eneration has been exposed to different relevant exemplars across individual lifespans. From the trajectory analysis in the previous chapter, we saw that Silent generation speakers produce pre - nasal and pre - oral /æ/ in the same way, with the tongue higher and more forward in the mouth. Generation X speakers , however, produce pre - oral /æ/ with a slightly lower and much farther back realization than pre - nasal /æ/. So, for Silent generation respondents in the judgment task, the two conditioning environments produce similar acoustics, while for Generation X respondents, the two sounds are acoustically different from one another. Thus, though we find differences between these generations for the judgement task, it is unclear whether the results do indeed refl ect a difference in phonological representation or phonetic realization. What this appears to suggest is that the current analysis falls short of discerning whether for a given respondent, their different judgements are the result of a phonetic difference between pre - nasal and pre - oral /æ/ or if that difference is indeed phonological. I a rgue, however, that the current analysis is indeed indicative of representational knowledge. I f these judgements were about phonetic (rather than phonological) opposit ion between pass and pa ck ), 3 ( pen and peck ), and 4 ( pet and peck ) to be different. We do not see this in the data. Respondents consistently judge the pairs in these con ditions to be the same even though some of these are realized acoustically different from one another. I would argue that this is because the distinction between two separate realizations of the same phoneme is merely the result of 139 phonetic implementation , but for those who report a difference, pre - nasal /æ/ is phonologically distinct from that in these other environments. This suggests to me that though the phonetic distinction between pre - nasal and pre - oral /æ/ is indeed much sharper for Millennial gene ration respondents than for Silent generation respondents, acoustic salience alone within a phoneme cannot account for the difference in responses by the two groups in this task. If indeed this task is only indicative of phonetic opposition, perhaps the o ther diagnostics utilized in this dissertation to tease apart phonological from phonetic allophony (bimodality, and divergent vs lockstep trajectories) are better suited for questions regarding phonologization. 6.4.3 Orthography in American English Another pot ential confound in this study involves the issue of orthography. One of the problems with presenting words in American English to a participant is the grapheme - phoneme mismatch. It is likely that many of the participants in this study would have taken or thography as a cue and made their judgements of lexical pairs accordingly; they would have judged Condition 1 pairs ( pan and pat ) to be the same because they share the same character regardless of whether the corresponding pronunciation of that character in the two words was indeed the same sound to them. This is a common issue with presenting visual stimuli in studies examining phenomena at the phonetic s /phonolog y level. As discussed in Chapter 3 (section 3.3.2.1 ), I attempted to reduce the likelihood that this would happen by including words whose vowel character, , could correspond to a different phoneme in American English (compare ma ke and mack ), and by including pairs of lexical items whose graphemes did not match but whose pronunciations may be similar in the community, e.g. lot and thought. I excluded from the analysis any participant who judged pairs like make and mack to be the same. I cannot definitively say, however, that those who judged pairs with words like make in them to be different, were not relying on 140 orthography while they were judging the target pairs. So, orthography remains a possible confound in this study, the effects of which might be eliminated with the replacement of visual stimuli with auditory stimuli. One might argue that a n auditory perception task might also resolve the issue of Millennial and Silent generation respondents making judgements for differen t pronunciations. For this, participants might be presented with a raised/fronted pre - nasal and a lowered/retracted pre - oral /æ/. Upon asking them to judge whether these sounds are the same or different, a Silent generation respondent who does not posit a n allophonic rule should judge these as the same while a respondent who posits an allophonic rule should judge them as different. The presentation of auditory stimuli would eliminate the two possible confounds in the current study but is likely to introduc e another confound. I purposefully chose not to present participants with pre sentation of auditory stimuli would be a sound solution for the current analysis. I leave this for future research. 6.5 Inland North D ialectology While the results of this dissertation further confirm some previous findings regarding the linguistic patternin g of /æ/ as part of the NCS (raising, fronting, diphthongization, unconditioned system), the generational change and social conditioning results offer some confirmation of some existing claims in the literature, and some new information for our knowledge b ase about the NCS. 141 /æ/ - raising is a rather new phenomenon in Lansing, MI : The results of this dissertation show that NCS /æ/ - raising was adopted in the middle of the 20 th century, which is well after the period identified in the account of Inland North /æ/ - hypothesis is that /æ/ - raising developed in upstate New York in the early 1800s due to leveling across multiple /æ/ systems an d spread westward to other Inland North locales. As Gordon and Strelluf (2017) point out, if this scenario held, raised /æ/ would be evident in western locales at least by the turn of the 20 th gene rations for /æ/ to raise (also see Kerswill & Williams (2000) for a discussion of leveling taking two generations). This dissertation finds that the raising pattern occurred in Lansing at ith recent studies in Chicago, Buffalo, and Grand Rapids (Gordon and Strelluf 2017) where /æ/ - raising is not well attested in speakers born at the turn of the 20 th century. One caveat here is that Lansing is a mid - sized city. If the NCS spread to Lansing via a cascade model of diffusion from larger cities to smaller towns, we would expect Lansing to have adopted NCS features later than larger cities (Callary 1975) - raising, then the larger cities of Chicago, Buffalo, and Grand Rapids should have acquired /æ/ - raising much earlier than is reported in the literature. Thus, I conclude that /æ/ - raising is a more recent phenomenon in the Inland North than su ggested by Labov (2010) . It is still unclear whether Lansing acquired /æ/ - raising from these larger cities. If it did, this was a recent phenomenon, likely happening at the turn of the 20 th century . Where are the NCS speakers? : T he third contribution of this dissertation is a suggestion for future Inland North researchers. For contemporary scholars looking to investigate the NCS, this study and other recent studies show that the available subject pool is waning. The NCS is no 142 lo nger prevalent in Inland Northern cities. The more likely carriers of the NCS in the Inland North are speakers born prior to the 1980s, young blue - collar speakers, and perhaps rural speakers (cf Gordon 2001). Though the latter two groups may also be part icipating in the supra - local changes, no recent studies have ruled these speakers out as possible NCS torchbearers . Indeed, Durian and Cameron (2018) find that the NCS has remained among blue - collar speakers in Chicago. Michiganders are not as peculiar as we once thought : The los s of NCS features in the middle of the 20 th sociolinguistic awareness. The social conditioning observed in this chapter, accompanied by the commentary and style - shifting d iscussed in Chapter 1, suggest that Inland Northerners are somewhat aware of NCS /æ/ and that they assign it indexical social meaning. This can lead us to re - interpret some earlier studies of sociolinguistic attitudes to /æ/ in the Inland North. For exampl e, Niedzielski (1999) found that when asked whether a fellow Detroiter had produced a lower or higher variant of /æ/, Detroit n atives picked out the low variant from a provided range. Although to our knowledge, most white Detroiters at that time exhibit advanced NCS features themselves, including raised /æ/ led them t speech. Therefore, their social evaluation of their dialect as correct/standard overrode their performance on this task. For Niedzielski, this was evidence that the NCS w as well under the level of awareness. Considering more recent studies like the current one and others in the dialect area which show that the NCS version of /æ/ has been retreating for some time, it seems more imply aware at some level that a community lowering of /æ/ had begun. These perception studies were conducted on college students born in the 143 1960s and 1970s, the same generation in which NCS features began to decline in Michigan. This suggests that it m ay not have been entirely the case that social evaluation was overriding performance on the perception task. It appears to be likely that exposure to within - generational social reality more lowered variants available in the community was affecting per formance on the task. From an indicator to a marke r: Changes in the social conditioning over time for /æ/ suggest that NCS realizations have come to be negatively evaluated in Lansing, which points to a possible social motivation for these changes, as di scussed in 4.7 . Given the available evidence in the literature, the NCS was progressing well below the level of awareness in the earlier half of the 20 th century. Women led in the advancement of NCS /æ/ char acteristics. The subsequent decline in NCS /æ/ realization by women (while men continued to advance) in the transition from the Baby Boomer generation to Generation X suggests that NCS /æ/ in Lansing might have gained social salience as a local marker . L abov (1972) defines a marker as a sociolinguistic variable that not only has an interspeaker social distribution (e.g. across social class, neighborhoods or ethnic groups), but is also subject to intraspeaker variation. To confirm that NCS /æ/ has become a marker in Lansing, we would therefore need to conduct an analysis of style - shifting. If the hypothesis is accurate, I would expect speakers born in the middle of the 20 th century to exhibit less NCS - like /æ/ features in their more formal speech, e.g. whe n reading a word list, as compared to in their conversational speech. Thiel and Dinkin (2017) show that while older speakers in Ogdensburg do not style - shift, younger speaker s shift away from NCS features in their more formal speaking style. I suspect th e same pattern to emerge in Lansing. I leave examination of style - shifting in my data for a future project. 144 Another possible avenue for investigating social motivations for these changes is an implicit attitudes test. Labov (2001:229) , citing Sturtevant (1947) , suggests that the social motivation of linguistic change is that the form becomes associated arbitrarily with social traits of the originating group. If this is the case, then we might expect the NCS version of /æ/ to be a marker of older local Lansin g identity. If one were to conduct an attitudinal study, I would expect Michiganders to rate NCS /æ/ in the speech of an older local more favorably than if heard in the speech of a younger (maybe even middle - class) local. 6.6 P honological R epresentations in the Inland North (more questions) Structural account of the LBMS in Lansing : I noted earlier that two components of the LBMS shift are underway in Lansing pre - oral /æ/ lowering and retraction, and the low - back merger. The structural account of the LBMS is one of a pull chain whereby / hich pulls /æ/ and the other short front vowels, / / and / /. Though there is evidence of these components occurring in Lansing ( Nesbitt, Wagner & Mason under rev ision) , it remains to be seen if the LBMS in the Inland North is a pull or push chain. The absence of Generation X speakers made it difficult for us to speculate. The present analysis shows that /æ/ retracted in Generation X. I look forward to examin ing movement of / to better address this question. Tense/Lax distinction : The Inland North version of /æ/ is described as [tense] because it moves upward along the periphery of the vowel space and is described as diphthongal (Labov 1994) . In Lansing, and in other Inland Northern cities, /æ/ is either no longer moving along the periphery of the vowel space and/or is no longer diphthongal. This leads to the hypothesis that there has been a shift in this pho neme from tense to lax. I test this hypothesis in Nesbitt (2018b) by examining whether Inland Northerners treat /æ/ as they do traditionally tense vowels or lax 145 vowels in a syllable parsing task. For this, I asked whether Inland Northerners parse a consonant that followed /æ/ as they would one that followed a tense vowel or a lax vowel. The logic behind this is that intervoc alic consonants in American English are conditioned by the tenseness of the preceding vowel, i.e. when preceded by a long/tense vowel, English speakers will parse them as onsets of the following syllable. The hypothesis was that older Inland Northerners w ould treat /æ/ as they do tense vowels, but younger Inland Northerners would treat it as a lax vowel. The preliminary results of this analysis suggest that indeed, older Inland Northerners treat /æ/ as they do tense vowels, while younger Inland Northerner s treat it as a lax vowel. This analysis is preliminary, though, because results are based on a comparison of just 10 lexical items. In any case, there appears to be a trend towards a change overtime in tense vs lax associations for /æ/ in the dialect ar ea. I therefore look forward to exploring this question in a future project. 6.7 North American Dialectology This dissertation joins the growing literature on the decline of regional speech patterns in North America. One of the most substantial contributions of this analysis to the field is that one of the the unconditioned raising of /æ/ regional /æ/ configuration in place of the nasal system is not confined to the Inland North. We see this trend in various locations across North America, in for example, Kansas City (Strelluf 2014) , Cincinnati (Boberg & Strassel 2000) , New Orleans, L A (Labov 2007) , New York city (Becker & Wong 2009; Becker 2010) , various cities along the east coast between Philadelphia and New York city (Ash 2002) , Philadelphia (Labov, Rosenfelder & Fruehwald 2013; Sneller 2017; Sneller 2019) , and New England (Stanford 2019) , inter alia. Thus, short /æ/ appears to no longe r be a reliable metric for distinguishing regional dialects in North America, at least for 146 speakers born since about 1990. I also noted in Chapter 1 that the second pivot point low - back merger vs distinction appears to be diminishing as more regional vari eties move towards merger. Therefore, dialectologists must either look to other features that distinguish regional varieties of North American English, or entertain the idea that rather than divergence, we are witnessing dialect levelling. The movement t owards the supralocal patterns of low - back merger and the nasal system in various regions suggest the latter. 6.8 Mid - 20 th century and Loss of Regional North American Features The transition from the Baby Boomer generation to Generation X appears to have be en a crucial period for changes to regional varieties in North America. Researchers elsewhere in Michigan have observed the same decline in NCS features over the same period of time (Morgan et al. 2017; Rankinen, Albin & Neuhaus 2019) . Loss of local speech features in the mid - 20 th century is not restricted to the Inland North dialect area. It ha s been reported in various locations across North America, in for example, Ohio (Durian 2012) , Philadelphia (Labov, Rosenfelder & Fruehwald 2013), New England (Stanford 2019), Raleigh, NC (Dodsworth & Kohn 2012) , inter alia. One wonders what similar social conditions were in place in such disparate areas to have motivated the apparent leveling of dialects at about the same time. A few pos sible explanations include (1) a rise in face - to - face contact between speakers of different varieties brought on by increased mobility, (2) the spread of Western regional phonology through increased exposure to television, (3) the shift to an economy incr easingly dominated by service jobs that would require of all three. I leave it for future work to investigate the social motivations behind this broader trend. 147 APPENDICES 148 APPENDIX A Recruitment Flyer for 2018/2019 Sociolinguistic Interviews Attention Lansing natives! Earn $15 for Talking to a Linguist for 1 Hour. My name is Monica Nesbitt. I am a PhD student and researcher in the Michigan State University Sociolinguistics Lab. Our research team is currently interviewing Lansing natives as part of our Lansing Speech project and would like you to participate! The Lansing Speech project aims to better understand Lansing speech and culture by interviewing those who were born and raised here. The recordings from this project will contribute to our growing collection of oral histories of native Lansingites. Each interview lasts about 1 hour, during which we will discuss with you your experiences growing up in Lansing and have you play a few short language games. Though our interviewers will come prepared with topics to discuss, we are also interested in hearing about whatever interests you. Interviews will be scheduled until the end o f August 2018. If you are interested in participating, please make sure you satisfy the below qualifications and then contact me via email nesbit17@msu.edu, a private message on Facebook Messenger, or by cell (909) 997 - 2375 so that we can schedule an inter view. Please help us spread the word and share this with as many native Lansingites that you know. We are especially interested in hearing from multiple generations of the same family! Compensation $15 Qualifications a) +17 years old b) From within 15 m iles of the MI state capitol - Lived there from around age 2 to at least around age 17 - Left the area for no more than 4 years c) Self - identify as white (we will interview non - white residents later in the year!) d) Native speaker of American English e) No history of a hearing or speech disorder 149 APPENDIX B Consent Form for 2018/2019 Sociolinguistic Interviews Thank you for contributing to this research project. Our research team at Michigan State is interested to hear about life and language in Lansing. I can tell you more about the project once the end of this session. and experiences growing up, how your community has changed in your lifetime, how you think it compares to other areas of Lansing specifically and Michigan in general. I have topics and questions fe and the perspectives that you think and then ask you to read aloud a short list of words that have distinctive pronunciations across the U.S. Altogether, the interview will take no more than one hour. Because it will be difficult for me to write down everything you say, I would like to audio - record before we start . Before you begin, I would like to inform you that participation in this research project is voluntary. You have the right to stop participating in this study at any time, and you also have a right to not answer specific questions. You may change your mind at any time and withdraw. There will be no penalty or disadvantage to you under any of these circumstances. You must be 18 or older to participate. Please email me at nesbit17@msu.edu if you have any questions regarding the procedure. If you have con cerns or questions about this study, such as how the data will be used, how to do any part of it, or to report an injury, please contact my supervisor, Dr. Suzanne Evans Wagner, at (517) 355 - 9739, or e - mail wagnersu@msu.edu or regular mail at B - 401 Wells H all, 619 Red Cedar Rd, East Lansing, MI 48824. If you have questions or concerns about your role and rights as a research participant, would like to obtain information or offer input, or would like to register a complaint about this study, you may contact, Protection Program at 517 - 355 - 2180, Fax 517 - 432 - 4503, or e - mail irb@msu.edu or regular mail at 4000 Collins Rd, Suite 136, Lansing, MI 48910. 150 APPENDIX C Interview Questions for 2018 /2019 Sociolinguistic Interviews General What is your pseudonym? What is your birth year? What is your gender identity? What ethnic group (race) would you say you belong to? Have you ever spoken any language besides English? Do you at home? Where were you born and raised? Family Who do you live with at home? Do you have children? How old are they? Where are do they live now? Where did they go to school? Are you married? Where is your spouse from? What do they do for a living? How did you meet your spouse? How about your parents? Did you live with them growing up? Where are they from? What did they do for a living? How did they meet? School and Work What high school did you go to? What middle school did you go to? What elementary school did you go t o? Were you involved in any activities when you were in school? Sports, clubs, etc? What sorts of things did you and your friends do outside of class? What do you think was the racial make - up of your schools? 151 Were inter - racial friendships normal? What abou t social class? Were the students from blue - collar or white - collar families? What about training after high school? Did you attend college or anything? What university did you go to ? What do you do for work? How long have you worked in your current posit ion? How did you get into this line of work that you do now? What other kinds of jobs have you had? Neighborhood Where were you born? What neighborhood did you grow up in? Did you go to your neighborhood schools? Did your friends? Describe your childhood neighborhood. What kinds of families lived there? Were they blue - collar or white - collar? What was the racial make - up of your neighborhood? Was it a close - knit community? What sorts of activities did the neighborhood do? Did all the kids attend th e same schools? Was your neighborhood the same or different from other neighborhoods in the area? Would you describe it as your typical Michigan neighborhood? What neighborhood do you live in now? Why did you move? Where do your friends and family live now? Why did they move? Have you noticed any changes to the neighborhood that you grew up in? How about Lansing? Has it changed at all? Localness What do you think of Lansing as a place to live? Would you recommend it? Which Lansing neighborhoods do you think are the best and which ones are the worst? What sorts of things are there to do in Lansing? Are there any other places in Michigan worth checking out? Are they better than Lansing? Did you or did any of your friends and family work for any of the a uto plants in town? What did they do? How long did they work there? What do they do now? If you could live anywhere, where would it be? Would you rather live in a big city or a small town? Why? Would you describe Lansing as a big city? Do you get out of La nsing much? Ever go to Detroit or Chicago? Anywhere else? 152 Narrative Prompts Do you think the winters in Michigan are getting worse or better? Do you think the roads in M ichigan are worse than anywhere else? Why do you think roads are such an ongoing issue here? Is there anything else you would like to share about your ex periences related to living in Lansing or otherwise? Post - interview questions: When you travel, has any one picked you out as being from Lansing or Michigan by the way you talk? What do they notice? Do you take it a plus or minus from their point of view? Form your own? What do you think makes a Lansing accent? Is that different from a Detroit accent? Is that different from anywhere else in Michigan? 153 APPENDIX D Table 15 Speaker Demo graphics Lansing Speech Corpus plus 21 Millennials from Wagner et al. ( 2016 ) *(Millennials were not included in statistical analyses due to demographic imbalance) Subj # Name Sample Birth Year Generation Gender Social Class 1 MarvinGrinstern Oral Histories Collection 1923 Silent male blue - collar 2 JackDown Oral Histories Collection 1924 Silent male white - collar 3 MarilynShadduck Oral Histories Collection 1924 Silent female blue - collar 4 LeslieMitchell Oral Histories Collection 1925 Silent male blue - collar 5 MabelMcQueen Oral Histories Collection 1925 Silent female blue - collar 6 VernonCook Oral Histories Collection 1927 Silent male blue - collar 7 ShirleySanborn Oral Histories Collection 1930 Silent female white - collar 8 FrancesWest 2010 Sociolinguistic Interview 1932 Silent female white - collar 9 GayleGooslin Oral Histories Collection 1937 Silent female blue - collar 10 JohnDean Oral Histories Collection 1938 Silent male white - collar 11 MaudeRobinson 2010 Sociolinguistic Interview 1939 Silent female white - collar 12 RichardBudd Oral Histories Collection 1941 Silent male white - collar 13 DeanPotter 2018/19 Sociolinguistic Interview 1946 Boomer male white - collar 14 LindaMaxon Oral Histories Collection 1947 Boomer female blue - collar 15 AnnePotter 2018/19 Sociolinguistic Interview 1948 Boomer female white - collar 16 KarelTaborsky Oral Histories Collection 1948 Boomer male blue - collar 17 JosephAbraham Oral Histories Collection 1951 Boomer male blue - collar 18 CherryBomb 2018/19 Sociolinguistic Interview 1952 Boomer female blue - collar 19 GeneralEisenhower 2018/19 Sociolinguistic Interview 1953 Boomer male white - collar 154 20 MikelLienhart Oral Histories Collection 1953 Boomer male blue - collar 21 JuneLowe 2018/19 Sociolinguistic Interview 1956 Boomer female white - collar 22 RandyThayer Oral Histories Collection 1956 Boomer male white - collar 23 ConnieStevens 2018/19 Sociolinguistic Interview 1957 Boomer female white - collar 24 SharonDecker Oral Histories Collection 1961 Boomer female blue - collar 25 AaronPeters 2018/19 Sociolinguistic Interview 1965 Generation X male blue - collar 26 BillMummy 2018/19 Sociolinguistic Interview 1969 Generation X male white - collar 27 ViolaMatt 2018/19 Sociolinguistic Interview 1970 Generation X female white - collar 28 JackPierce 2018/19 Sociolinguistic Interview 1974 Generation X male blue - collar 29 SarahMary 2018/19 Sociolinguistic Interview 1976 Generation X female blue - collar 30 Serendipity Miller 2018/19 Sociolinguistic Interview 1977 Generation X female blue - collar 31 AmandaNelson 2018/19 Sociolinguistic Interview 1982 Generation X female white - collar 32 KingLlama 2018/19 Sociolinguistic Interview 1982 Generation X male white - collar 33 SamHill 2018/19 Sociolinguistic Interview 1982 Generation X male white - collar 34 StaceyMiller 2018/19 Sociolinguistic Interview 1982 Generation X female blue - collar 35 KenWinters 2018/19 Sociolinguistic Interview 1984 Generation X male blue - collar 36 BunnyLlama 2018/19 Sociolinguistic Interview 1984 Generation X female white - collar 37 VickyPine 2014/15 Sociolinguistic Interview 1990 Milllennial female white - collar 155 38 ChelseaWalsh 2014/15 Sociolinguistic Interview 1992 Milllennial female white - collar 39 EllenBordner 2014/15 Sociolinguistic Interview 1992 Milllennial female white - collar 40 EmmaAllen 2014/15 Sociolinguistic Interview 1993 Milllennial female white - collar 41 MelodyWilliams 2014/15 Sociolinguistic Interview 1993 Milllennial female white - collar 42 MichelleRaggle 2014/15 Sociolinguistic Interview 1993 Milllennial female blue - collar 43 SarahStone 2014/15 Sociolinguistic Interview 1993 Milllennial female white - collar 44 BenLangdon 2014/15 Sociolinguistic Interview 1994 Milllennial male white - collar 45 JenniferMason 2014/15 Sociolinguistic Interview 1994 Milllennial female blue - collar 46 JessicaTawnee 2014/15 Sociolinguistic Interview 1994 Milllennial female white - collar 47 KarenPeterson 2014/15 Sociolinguistic Interview 1994 Milllennial female white - collar 48 KarenTimmons 2014/15 Sociolinguistic Interview 1994 Milllennial female blue - collar 49 LucyGarth 2014/15 Sociolinguistic Interview 1994 Milllennial female blue - collar 50 BriDixon 2014/15 Sociolinguistic Interview 1995 Milllennial female white - collar 51 KatherineShort 2014/15 Sociolinguistic Interview 1995 Milllennial female white - collar 52 CarolineTonks 2014/15 Sociolinguistic Interview 1996 Milllennial female white - collar 53 EdBrand 2014/15 Sociolinguistic Interview 1996 Milllennial male white - collar 54 HelenSmith 2014/15 Sociolinguistic Interview 1996 Milllennial female white - collar 156 55 MegPaxton 2014/15 Sociolinguistic Interview 1996 Milllennial female white - collar 56 PeteShaw 2014/15 Sociolinguistic Interview 1996 Milllennial male white - collar 57 StephanieWilson 2014/15 Sociolinguistic Interview 1996 Milllennial female white - collar 157 APPENDIX E Consent Form for Sub - Phonemic Judgement Task Research Participant Information and Consent Form Thank you for contributing to this research project. My name is Monica Nesbitt. I am a Ph.D. d in, but if you contact me at the email provided below, I will be happy to answer any questions that you might have. For this experiment, there are two short tasks. For these tasks, you will answer questions about the words that appear on your computer screen. Before you begin, I would like to inform you that participation in this research project is voluntary. You hav e the right to stop participating in this study at any time, and you also have a right to not answer specific questions. You may change your mind at any time and withdraw. There will be no penalty or disadvantage to you under any of these circumstances. Y ou must be 18 or older to participate. If you are taking this for extra credit, you will receive 1% extra credit towards your final course grade for participating in this experiment. You will receive another percentage point if you have a family member tha t is your grandparent's age complete this survey. If you wish not to participate in this study, your instructor has an equivalent non - research assignment which may be done in place of research participation. Please email me at nesbit17@msu.edu if you have any questions regarding the procedure. If you have concerns or questions about this study, such as how the data will be used, how to do any part of it, or to report an injury, please contact my supervisor, Dr. Suzanne Evans Wagner, at (517) 355 - 9739, or e - mail wagnersu@msu.edu or regular mail at B - 401 Wells Hall, 619 Red Cedar Rd, East Lansing, MI 48824. If you have questions or concerns about your role and rights as a research participant, would like to obtain information or offer input, or would like to register a complaint about this study, you Protection Program at 517 - 355 - 2180, Fax 517 - 432 - 4503, or e - mail irb@msu.edu or regular mail at 4000 Collins Rd, Suite 136, Lansi ng, MI 48910. You indicate your voluntary agreement to participate by navigating to the next page. 158 APPENDIX F Social Media and Email Recruitment Script for Sub - phonemic Judgement Task Do the vowels in BOT and BOUGHT sound the same to you? Whatever the answer, the Sociolinguistics Lab at Michigan State wants to hear from you! from as many places, generations, and ethnic backgrounds . please share with all of your friends and family. 159 APPENDIX G Table 16 Pairs of Lexical Items for Sub - Phonemic Judgement Task Condition Lexical Items Vowel Final Cons Nasality 1 FAN FAT æ nasal - oral 1 TRAM TRACK æ nasal - oral 1 RAN RAG æ nasal - oral 1 PAM PASS æ nasal - oral 1 BAN BAT æ nasal - oral 2 PASS PACK æ oral - oral 2 RASH RAT æ oral - oral 2 PACK PAT æ oral - oral 2 RAT RAG æ oral - oral 2 SAP SACK æ oral - oral 3 RENT REST nasal - oral 3 PEN PET nasal - oral 3 TEN TECH nasal - oral 3 KIM KIT nasal - oral 3 BIN BIT nasal - oral 4 REST LET oral - oral 4 PECK PET oral - oral 4 FETCH FED oral - oral 4 KISS KIT oral - oral 4 BIT BIG oral - oral 160 5 BOSS BOUGHT oral - oral 5 TOT TAUGHT oral - oral 5 STALK STOCK / oral - oral 5 CAUGHT COT / oral - oral 5 DAWN DON / nasal - nasal 6 REST COT / oral - oral 6 PIN PEN / nasal - nasal 6 BOMB BIT nasal - oral 6 SAT SAW æ/ oral - oral 6 MAKE MACK e /æ oral - oral 161 APPENDIX H Table 17 Pillai - Bartlett and Mean F1, F2, and DQ Values for Each Speaker in the Lansing Speech Corpus pl us 21 Millennials from Wagner et al. (2016) *(Millennials were not included in statistical analyses due to demographic imbalance) Name Allophone F1 (Hz) F2 (Hz) DQ (Hz) Pillai - Bartlett MarvinGrinstern pre - nasal 727 1995 236 0.261 MarvinGrinstern pre - oral 809 1850 128 0.261 JackDown pre - nasal 736 1982 141 0.04 JackDown pre - oral 769 1917 182 0.04 MarilynShadduck pre - nasal 712 2204 450 0.152 MarilynShadduck pre - oral 807 2047 303 0.152 LeslieMitchell pre - nasal 742 1996 108 0.075 LeslieMitchell pre - oral 730 1865 119 0.075 MabelMcQueen pre - nasal 722 2009 198 0.038 MabelMcQueen pre - oral 740 2150 201 0.038 VernonCook pre - nasal 697 1972 191 0.079 VernonCook pre - oral 757 1905 156 0.079 ShirleySanborn pre - nasal 646 2059 176 0.27 ShirleySanborn pre - oral 789 2013 177 0.27 FrancesWest pre - nasal 666 2080 326 0.272 FrancesWest pre - oral 812 1875 182 0.272 GayleGooslin pre - nasal 712 1938 320 0.091 GayleGooslin pre - oral 798 1833 178 0.091 JohnDean pre - nasal 678 2111 134 0.371 JohnDean pre - oral 833 2012 155 0.371 MaudeRobinson pre - nasal 605 2179 692 0.351 162 MaudeRobinson pre - oral 728 1907 368 0.351 RichardBudd pre - nasal 609 2050 177 0.428 RichardBudd pre - oral 789 1825 167 0.428 DeanPotter pre - nasal 703 2108 346 0.352 DeanPotter pre - oral 806 1948 223 0.352 LindaMaxon pre - nasal 654 2223 269 0.042 LindaMaxon pre - oral 734 2115 289 0.042 AnnePotter pre - nasal 690 2334 564 0.155 AnnePotter pre - oral 717 2120 335 0.155 KarelTaborsky pre - nasal 729 2062 231 0.318 KarelTaborsky pre - oral 843 1920 193 0.318 JosephAbraham pre - nasal 679 2030 212 0.273 JosephAbraham pre - oral 822 2007 193 0.273 CherryBomb pre - nasal 642 2294 350 0.371 CherryBomb pre - oral 747 1945 233 0.371 GeneralEisenhower pre - nasal 652 2165 320 0.479 GeneralEisenhower pre - oral 788 1783 174 0.479 MikelLienhart pre - nasal 771 1957 205 0.16 MikelLienhart pre - oral 811 1802 158 0.16 JuneLowe pre - nasal 651 2115 318 0.419 JuneLowe pre - oral 790 1831 208 0.419 RandyThayer pre - nasal 651 1986 131 0.186 RandyThayer pre - oral 775 1889 157 0.186 ConnieStevens pre - nasal 646 2194 433 0.244 ConnieStevens pre - oral 715 1895 206 0.244 163 SharonDecker pre - nasal 591 2109 217 0.279 SharonDecker pre - oral 719 2036 221 0.279 AaronPeters pre - nasal 578 2080 261 0.424 AaronPeters pre - oral 779 1960 249 0.424 BillMummy pre - nasal 735 1927 151 0.031 BillMummy pre - oral 719 1825 175 0.031 ViolaMatt pre - nasal 667 2175 453 0.38 ViolaMatt pre - oral 767 1723 186 0.38 JackPierce pre - nasal 649 2199 376 0.363 JackPierce pre - oral 780 2078 222 0.363 SarahMary pre - nasal 628 2150 332 0.541 SarahMary pre - oral 795 1742 211 0.541 Serendipity Miller pre - nasal 607 2094 428 0.589 Serendipity Miller pre - oral 778 1801 142 0.589 AmandaNelson pre - nasal 616 2226 474 0.722 AmandaNelson pre - oral 816 1634 163 0.722 KingLlama pre - nasal 617 2086 210 0.507 KingLlama pre - oral 742 1871 137 0.507 SamHill pre - nasal 679 2001 160 0.347 SamHill pre - oral 783 1811 186 0.347 StaceyMiller pre - nasal 716 2159 363 0.401 StaceyMiller pre - oral 805 1831 183 0.401 BunnyLlama pre - nasal 633 2152 255 0.684 BunnyLlama pre - oral 757 1781 232 0.684 KenWinters pre - nasal 689 1967 153 0.36 164 KenWinters pre - oral 762 1700 155 0.36 VickyPine pre - nasal 599 2144 281 0.448 VickyPine pre - oral 732 1898 150 0.448 ChelseaWalsh pre - nasal 703 2132 325 0.503 ChelseaWalsh pre - oral 817 1808 164 0.503 EllenBordner pre - nasal 648 2114 244 0.615 EllenBordner pre - oral 783 1788 144 0.615 EmmaAllen pre - nasal 709 1941 297 0.614 EmmaAllen pre - oral 866 1627 138 0.614 MelodyWilliams pre - nasal 703 2050 193 0.574 MelodyWilliams pre - oral 845 1717 133 0.574 MichelleRaggle pre - nasal 666 1928 195 0.26 MichelleRaggle pre - oral 778 1690 157 0.26 SarahStone pre - nasal 597 2145 233 0.635 SarahStone pre - oral 821 1754 155 0.635 BenLangdon pre - nasal 627 2170 114 0.552 BenLangdon pre - oral 788 1893 208 0.552 JenniferMason pre - nasal 644 2174 218 0.71 JenniferMason pre - oral 796 1709 150 0.71 JessicaTawnee pre - nasal 634 2024 292 0.562 JessicaTawnee pre - oral 790 1782 139 0.562 KarenPeterson pre - nasal 681 2088 232 0.551 KarenPeterson pre - oral 760 1782 128 0.551 KarenTimmons pre - nasal 661 2101 178 0.499 KarenTimmons pre - oral 796 1729 153 0.499 165 LucyGarth pre - nasal 693 1912 212 0.42 LucyGarth pre - oral 808 1661 130 0.42 BriDixon pre - nasal 585 2165 240 0.722 BriDixon pre - oral 812 1791 151 0.722 KatherineShort pre - nasal 620 2091 358 0.558 KatherineShort pre - oral 802 1796 162 0.558 CarolineTonks pre - nasal 622 2037 172 0.445 CarolineTonks pre - oral 825 1811 180 0.445 EdBrand pre - nasal 746 2069 116 0.297 EdBrand pre - oral 785 1761 145 0.297 HelenSmith pre - nasal 660 2150 186 0.561 HelenSmith pre - oral 799 1719 139 0.561 MegPaxton pre - nasal 653 2145 270 0.623 MegPaxton pre - oral 816 1708 153 0.623 PeteShaw pre - nasal 590 2066 103 0.645 PeteShaw pre - oral 834 1919 155 0.645 StephanieWilson pre - nasal 690 2200 221 0.424 StephanieWilson pre - oral 797 1881 155 0.424 166 APPENDIX I Figure 23 Per Token Diagonal Measurement of / æ / in Five Following Phonological Contexts by Speaker Birth Year in the Lansing Speech Corpus ( p lus 21 Millennial s from Wagner et al. (2016)) 167 REFERENCES 168 REFERENCES Adank, Patti, Roel Smits & Roeland van Hout. 2004. A comparison of vowel normalization procedures for langauge variation research. Journal of the Acoustical Society of America 116. 3009 3107. Ash, Sharon. 2002. The distribution of a phonemic split in the M id - short - a . University of Pennsylvania Working Papers in Linguistics: Selected Papers from NWAV 30 8(3). 1 15. Baker, Adam, Diana Archangeli & Jeff Mielke. 2011. Variability in American English s - retraction suggests a solutio n to the actuation problem. Language Variation and Change 23(3). 347 374. doi:10.1017/S0954394511000135. Baranowski, Maciej. 2013. On the role of social factors in the loss of phonemic distinctions. English Language and Linguistics 17(2). 271 295. doi:10.1 017/S1360674313000038. Baranowski, Maciej & Danielle Turton. 2018. Locating speakers in the socioeconomic hierarchy: towards the optimal indicators of social class. Paper presented at New Ways of Analyzing Variation 47 (NWAV47). New York, New York. Oct. 18 - 21. Bates, Douglas, Martin Maechler, Ben Bolker, Steven Walker, Maechler Martin & Steven - Journal of Statistical Software . doi:10.18637/jss.v067.i01. Becker, Kara (ed.). The Lo w - Back Merger Shift: Uniting the Canadian Vowel Shift, the California Vowel Shift, and Short Front Vowel Rotations Across North America . Durham, NC: Duke UP, for American Dialect Society. Becker, Kara. 2010. Regional dialect features on the Lower East Side of New York City: Sociophonetics, ethnicity, and identity. PhD dissertation. New York University. Becker, Kara & Amy Wing - mei Wong. 2009. The short - a system of New York City English: An update. University of Pennsylvania Working Papers in Linguistics: Selected Papers from NWAV 37 . Berkson, Kelly, Stuart Davis & Alyssa Strickler. 2017. What does incipie nt /ay/ - raising look like?: A response to Josef Fruehwald. Language 93(3). e181 e191. doi:10.1353/lan.2017.0050. Bermúdez - Otero, Ricardo. 2007. Diachronic Phonology. In Paul de Lacy (ed.), The Cambridge handbook of phonology , 497 517. Cambridge: Cambridge University Press. doi:10.1080/07900629948817. Bermúdez - Otero, Ricardo. 2015. Amphichronic explanation and the life cycle of phonological processes. In Patrick Honeybone & Joseph C. Salmons (eds.), The Oxford handbook of 169 historical phonology , 374 399. Oxfor d: Oxford University Press. Bermúdez - Otero, Ricardo & Graeme Trousdale. 2012. Cycles and continua: on unidirectionality and gradualness in langauge change. In Terttu Nevalainen & Elizabeth Closs Traugott (eds.), The Oxford handbook of the history of Englis h , 691 720. New York: Oxford University Press. Bigham, Douglas S. 2008. Dialect contact and accommodation among adults in a university setting. PhD dissertation. University of Texas at Austin. Blevins, Juliette. 2004. Evolutionary phonology: the emergence of sound patterns . New York: Cambridge University Press. Blevins, Juliette. 2006. A theoretical synopsis of Evolutionary Phonology. Theoretical Linguistics 32. 117 166. Boberg, Charles. 2008. Regional Phonetic Differentiation in Standard Canadian English. Journal of English Linguistics . doi:10.1177/0075424208316648. Boberg, Charles & Stephanie M Strassel. 2000. Short - a in Cincinnati. Journal of English Linguistics 28(2). 108 126. Boersma, Paul. 2001. Praat, a system for doing phonetics by computer. Glot Int ernational. Britain, David. 2002. Diffusion, levelling, simplification and reallocation in past tense BE in the English Fens. Journal of Sociolinguistics 6(1). 16 43. doi:10.1111/1467 - 9481.00175. Bureau of Labor Statistics. CPI Inflation Calculator. United States Department of Labor . https://www.bls.gov/data/inflation_calculator.html (10 July, 2019). Callary, Robert E. 1975. Phonological change and the development of an urban dialect in Illinois. Language in Society 4(2). 155 170. Clarke, Sandra, Ford Elms & Amani Youssef. 1995. The third dialect of English: Some Canadian evidence. Language Variation and Change 7. 209 228. Coase, R. H. 2000. The acquisition of fisher body by General Motors. The Journal of Law & Economics 43(1). 15 3 2. - led sound change. PhD dissertation. University of Pennsylvania. and the Northern Cities Shift i n a Chicago neighborhood. Paper presented at New Ways of Analyzing Variation 47 (NWAV47). New York, New York. Oct. 18 - 21. Dinkin, Aaron J. 2009. Dialect Boundaries and Phonological Change in Upstate New York. PhD dissertation. University of Pennsylvania. h ttp://repository.upenn.edu/edissertations. 170 Dinkin, Aaron J. 2011a. Nasal Short - a Systems vs. the Northern Cities Shift. University of Pennsylvania Working Papers in Linguistics: Selected Papers from NWAV 39 17(2). Dinkin, Aaron J. 2011b. Weakening resistan ce: progress toward the low back merger in New York State. Language Variation and Change 23. 315 345. DiPaolo, Marianna. 1988. Pronunciation and Categorization in Sound Change. Lingusitic change & contact: Proceedings of the Sixteenth Annual Conference on New Ways of Analyzing Variation in Language , 84 92. Austin. Dodsworth, Robin & Jon Forrest. 2016. Towards a Sociologically - Grounded View of Occupation in Sociolinguistics. University of Pennsylvania Working Papers in Linguistics: Selected Papers from NWAV 44 22(2). Dodsworth, Robin & Mary Kohn. 2012. Urban rejection of the vernacular: The SVS undone. Language Variation and Change 24(2). 221 245. doi:10.1017/S0954394512000105. sis. Dartmouth College. Driscoll, Anna & Emma Lape. 2015. Reversal of the Northern Cities Shift in Syracuse, New York. U. Penn Working Papers in Linguistics 21(2). Durian, David. 2012. A New Perspective on Vowel Variation Across the 19th and 20th Centuries in Columbus, OH. PhD dissertation. The Ohio State University. Durian, David & Richard Cameron. 2018. Another look at the development of the Northern Cities Shift in Chicago. Presented at NWAV 472 . New York University. Eckert, Penelope. 1988. Adol escent Social Structure and the Spread of Linguistic Change. Language in Society 17. 183 207. Evanini, Keelan. 2009. The permiability of dialect boundaries: A case study of the region surrounding Erie, Pennsylvania. PhD dissertation. University of Pennsylv ania. Evans, Betsy E, Rika Ito, Jamila Jones & Dennis R Preston. 2000. Change on top of Change: Social and Regional Accomodation to the Northern Cities Vowel Shift. In H Bennis, H Ryckeboer & Jan Stroop (eds.), De Toekomst van de Variatielinguistiek (a spe cial issue of Taan en Tongval to honor Dr. Jo Daan on her ninetieth birthday) , 61 86. Fine, Lisa. 2004. The Story of Reo Joe: Work, Kin and Community in Autotown, U.S.A . Philadelphia, PA: Temple University Press. Forrest, Jon & Robin Dodsworth. 2016. Towar ds a Sociologically - Grounded View of Occupation in Sociolinguistics. U. Penn Working Papers in Linguistics 22(2). http://repository.upenn.edu/pwpl/vol22/iss2/5. Fox, Robert Allen & Ewa Jacewicz. 2009. Cross - dialectal variation in formant dynamics of 171 Americ an English vowels. The Journal of the Acoustical Society of America 126(5). 2603 2618. doi:10.1121/1.3212921. Fruehwald, Josef. 2013. The phonological influence on phonetic change. PhD dissertation. University of Pennsylvania. Fruehwald, Josef. 2016. The e arly influence of phonology on a phonetic change. Language 92(2). 376 410. doi:10.1353/lan.2016.0041. Fruehwald, Josef. 2017. Generations, lifespans, and the zeitgeist. Language Variation and Change 29(1). 1 27. doi:10.1017/s0954394517000060. Garrett, Andr ew & Keith Johnson. 2011. Phonetic bias in sound change. UC Berkeley Phonology Lab Annual Report 9. 61. Gold, Elaine & Mireille Tremblay. 2006. Eh? and Hein?: Discourse Particles or National Icons? The Canadian Journal of Linguistics / La revue canadienne de linguistique 51(2). 247 263. Gordon, Matthew J. 2001. Small - Town Values and Big - City Vowels: A Study of the Northern Cities Shift in Michigan . Publication of the American Dialect Society . Duke University Press. Gordon, Matthew J. & Christopher Strelluf. 2017. Working the Early Shift: Older Inland Northern Speech and the Beginnings of the Northern Cities Shift. Journal of Linguistic Geography 4. 31 46. Guy, Gregory R. 1990. The sociolinguistic types of language ch ange. Diachronica 7(1). 47 67. doi:10.1075/dia.7.1.04guy. Hagiwara, Robert. 1997. Dialect variation and formant frequency: The American English vowels revisited. Journal of the Acoustical Society of America 102. 655 658. Hall - Lew, Lauren. 2010. Improved re presentation of variance in measures of vowel merger. Proceedings of Meetings on Acoustics . Acoustical Society of America 9. 1 10. doi:10.1121/1.3460625. Halle, Morris. 1962. Phonology in generative grammar. Word 18(3). 54 72. Hay, Jennifer, Paul Warren & Katie Drager. 2006. Factors influencing speech perception in the context of a merger - in - progress. Journel of Phonetics 34. 458 484. doi:10.1016/j.wocn.2005.10.001. Herold, Ruth. 1997. Solving the actuation problem: Merger and immigration in eastern Pennsyl vania. Language Variation and Change 9(2). 165 189. doi:10.1017/S0954394500001861. Holland, Cory Lin. 2008. Shifting or Shifted? The state of California vowels. PhD dissertation. University of Davis. 172 Holley, Jordan, Robin Dodsworth & Suzanne Evans Wagner. Conceptualizing and coding social class in linguistic corpora. In Christopher Cieri & Malcah Yaeger - Dror (eds.), Dimensions of Linguistic Variation , 1 24. Oxford: Oxford University Press. Hyman, Larry. 1975. Phonology: theory and analysis . New York: Holt, Rinehart and Winston. Hyman, Larry M. 2013. Enlarging the scope of phonologization. In Alan Yu (ed.), Origins of sound change: Approaches to phonologization , 3 28. Oxford University Press. doi:10.1145/66051.66057. Idsardi, William. 2006. Canadian raising, opacity, rephonemicization. The Canadian Journal of Linguistics 51(2). 119 126. Ito, Rika. 2001. Belief, Attitudes, and Linguistic Accommodation: A Case of Urban Sound Change in Rural Michigan. U. Penn Working Papers in Linguistics 7(3). http://repository.upenn.edu/pwpl/vol7/iss3/11. Jacewicz, Ewa, Robert Allen Fox & Joseph Salmons. 2 011. Cross - generational vowel change in American English. Language Variation and Change 23. 45 86. Janda, Richard D. & Brian D. Joseph. 2003. Reconsidering the canons of sound - change: Hist orical Linguistics , 205 219. Amsterdam: John Benjamin Publishing Co. doi:10.1075/cilt.237.14jan. Johnson, Daniel Ezra & Jennifer Nycz. 2015. Partial Mergers and Near - Di stinctions: Stylistic Layering in Dialect Acquisition. University of Pennsylvania Working Papers in Linguistics 21(2). http://repository.upenn.edu/pwpl/vol21/iss2/13%0Ahttp://repository.upenn.edu/pwpl/. Jones, Jamila. 2003. African Americans in Lansing and the Northern cities vowel shift: Language contact and accommodation. PhD dissertation. Michigan State University. Katz, Michael B., Mathew J. Creighton, Daniel Amsterdam & Merlin Chowkwanyun. 2010. Immigration and the new metropolitan geography. Journal o f Urban Affairs 32(5). 523 547. doi:10.1111/j.1467 - 9906.2010.00525.x. Keating, Patricia. 1990. Phonetic representations in generative grammar. Journal of Phonetics 18. 321 334. Kerswill, Paul & Ann Williams. 1999. Mobility versus social class in dialect le velling: evidence from new and old towns in England. Cuadernos de Filología Inglesa 8. 37 57. King, Sharese. 2017. African American Identity and Vowel Systems in Rochester, New York. Paper presented at New Ways of Analyzing Variation 46 (NWAV46). Madison, Wisconsin. Nov. 2 - 5. 173 Kohn, Mary & Carly Stithem. 2015. The Third Vowel Shift in Kansas: A supra - regional shift with regional variation. Paper presented at New Ways of Analyzing Variation 46 (NWAV46). Madison, Wisconsin. Nov. 2 - 5. Labov, William. 1 971. The study of language in its social context. In Joshua A. Fisherman (ed.), Advances in the sociology of language, Vol. 1 , 152 216. The Hague, The Netherlands: Mouton. Labov, William. 1984. Field methods of the project on linguistic change and variatio n. In John Baugh & Joel Sherzer (eds.), Language in Use , 28 53. Englewood Cliffs, NJ: Princeton - Hall. Labov, William. 1991. Three dialects of English. In Penelope Eckert (ed.), New Ways of Analyzing Sound Change , 1 44. New York: Academic Press. Labov, Will iam. 1994. Principles of Linguistic Change, Volume 1: Internal Factors . Oxford: Wiley - Blackwell. doi:10.1002/9781444327496. Labov, William. 2001. Principles of linguistic change, Volume 2: Social Factors . Blackwell. Labov, William. 2007. Transmission and D iffusion. Language 83(2). 344 387. doi:10.1353/lan.2007.0082. Labov, William. 2010. Principles of linguistic change. Vol 3: Cognitive and cultural factors . Language in Society . Labov, William, Sharon Ash & Charles Boberg. 2006. The Atlas of North American English: Phonetics, Phonology and Sound Change . De Gruyter. doi:10.1515/9783110206838. Labov, William & Ingrid Rosenfelder. 2011. The Philadelphia Neighborhood Corpus. University of Pennsylvania. Labov, William, Ingrid Rosenfelder & Josef Fruehwald. 2013. One Hundred Years of Sound Change in Philadelphia: Linear Incrementation, Reversal, and Reanalysis. Language 89(1). 30 65. doi:10.1353/lan.2013.0015. https://doi.org/10.1353/lan.2013.0015. Labov, William, Malcah Yaeger & Richard Steiner. 1972. A Quantitati ve Survey of Sound Change in Progress . Philadelphia, PA: U.S. Regional Survey. Linguistics Data Consortium. 2013. Automatic Alignment and Analysis of Lingusitic Change - Transcription Guidelines . http://projects.ldc.upenn.edu/DASL/SLX/transcription.pdf. Lo banov, B. M. 1971. Classification of Russian vowels spoken by different speakers. Journal of the Acoustical Society of America 49(2). 606 608. Max Planck Institute for Psycholinguistics. 2018. ELAN. https://tla.mpi.nl/tools/tla - tools/elan/. McCarthy, Corri ne. 2011. The Northern Cities Shift in Chicago. Journal of English Linguistics 174 39. 166 187. doi:10.1177/0075424210384226. McClelland, Edward. 2013. . Boolmsbury Publ ishing. Mellesmoen, Gloria. 2016. Discerning the sub - phonemic identity of pre - nasal / æ / in British Columbia English. Strathy Student Working Papers in Canadian English . Michigan State University. 2018. Planning and budgets: Enrollments. https://opb.msu.edu/functions/institution/datadigest/students/students - enrollment.html. Mielke, Jeff, Adam Baker & Diana Archangeli. 2016. Individual - level contact limits phonological complexity: Evidence from bunched and retroflex / /. Language 92(1). 101 140. doi:10.1353/lan.2016.0019. https://muse.jhu.edu/journals/language/v092/92.1.mielke.html%5Cnpapers3://publication/d oi/10.1121/1.417972. Mielke, Jeff, Christopher Carignan & Erik R. Thomas. 2017. The articulatory dynamics of pre - velar and pre - nasal /æ/ - raising in English: An ultrasound study. The Journal of the Acoustical Society of America 142(1). 332 349. doi:10.1121/1.4991348. Milholland, Agatha. 2018. Reversal of the Northern Cities Shift in Buffalo, NY. Paper presented at New Ways of Analyzing Vari ation 47 (NWAV47). New York, New York. Oct 18 - 21. Moreton, Elliott & Erik Thomas. 2007. Origins of Canadian Raising in Voiceless - Coda Effects: A Case Study in Phonologization. Laboratory Phonology 9 37 64. Morgan, Beau - Kevin, Kelsey Deguise, Eric Acton, Daniele Benson & Alla Shvetsova. 2017. Shifts toward the supra - regional in the Northern Cities region: Evidence from Jewish women in Metro Detroit. Paper presented at New Ways of Analyzing Variation 46 (NWAV46). Madison, WI. Nov 2 - 5. Nesbitt, Monica . 2018a. Economic Change and the Decline of Raised TRAP in Lansing , MI. University of Pennsylvania Working Papers in Linguistics: Papers from New Ways of Analyzing Variation 46 (NWAV 46) , vol. 24, 66 76. Nesbitt, Monica. 2018b. Changing accents changes sy llables: The effect of diachronic phonetic change on syllabic representations in the Inland North. Paper presented at New Ways of Analyzing Variation 47 (NWAV47). New York, New York. Oct 18 - 21. Nesbitt, Monica & Alexander Mason. 2016. Dialect Leveling Acro ss the US: Evidence of the third dialect shift in the Inland North. Paper presented at New Ways of Analyzing Variation 45 (NWAV45). Vancouver, British Columbia. Nov. 3 - 6. Nesbitt, Monica, Suzanne Evans Wagner & Alexander Mason. forthcoming. A tale of two s hifts: The Low Back Merger Shift in Lansing, MI. In Kara Becker (ed.), The Low Back Merger shifts across North America . American Dialect Society (PADS). 175 Niedzielsk i, Nancy. 1999. The effect of social information on the perception of sociolinguistic variables. Journal of Language and Social Psychology 18. 62 85. doi:10.1177/0261927X99018001005. Niedzielski, Nancy. 2002. Attitudes Toward Midwestern American English. I n Dennis R. Preston & Daniel Long (eds.), Handbook of Perceptual Dialectology , 323 329. John Benjamin Publishing Co. doi:10.1075/z.hpd2.22nie. Ohala, John J. 1981. The listener as a source of language change. In C Masek, R Hendrick & F Miller (eds.), Paper s from the Parasession on Language and Behavior. , 178 203. Chicago: Chicago Linguistics Society, University of Chicago. Ohala, John J. 1989. Sound change is drawn from a pool of synchronic variation. In L Breivik & Jahr E. H. (eds.), Language Change: Contr ibutions to the study of its causes , 173 198. Berlin: Mouton de Gruyter. Ohala, John J. 1990. The phonetics and phonology aspects of assimilation. In John Kingston & Mary E Beckman (eds.), Papers in laboratory phonology 1: Between the grammar and physics o f speech . Cambridge: Cambridge University Press. Ohala, John J. 1993. The phonetics of sound change. In Charles Jones (ed.), Historical Linguistics: problems and perspectives . London: Longman. Ohala, John J. 2013. The actuation of sound change is a matter of chance. Paper presented at The Phonology Laboratory. University of Chicago. Chicago, IL. Pew Research Center. Definitions. https://www.pewresearch.org/methods/demographic - research/definitions/ (29 May, 2019). Pierrehumbert, Janet B. 2001. Exemplar dynam ics: Word frequency, lenition, and contrast. In Joan Bybee & Paul Hopper (eds.), Frequency and the emergence of linguistic structure , 137 157. John Benjamin Publishing Co. Pierrehumbert, Janet B. 1990. Phonological and phonetic representation. Journal of P honetics 18. 375 394. Preston, Dennis R. 1996. Where the worst English is spoken. In Edgar Schneider (ed.), Focus on the USA , 297 360. Amsterdam: Benjamins, John. Qualtrics. 2005. Provo, Utah. https://www.qualtrics.com (9 May, 2019). R Core Team. 2013. R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing. Ramsammy, Michael. 2015. The life cycle of phonological processes: Accounting for dialectal microtypologies. Linguistics and Language Compass 9(1 ). 33 54. doi:10.1111/lnc3.12102. Rankinen, Wil A, Aaron Albin & TJ Neuhaus. 2019. Decline of the Northern Cities Vowel Shift 176 in western Lower Michigan: Apparent - time evidence of a change in progress. Paper presented at the annual meeting of the American D ialect Society (ADS). New York, New York. Roeder, Rebecca V. 2006. Ethnicity and Sound Change: Mexican American Accomodation to the Northern Cities Shift in Lansing, Michigan. PhD dissertation. Michigan State University. Rosen, Laurence. 2007. Knowledge economy emerges in mid - Michigan . Greater Lansing Business Monthly . Vol. 16. Rosenfelder, Ingrid, Josef Fruehwald, Keelan Evanini, Scott Seyfarth, Kyle Gorman, Hilary Prichard & Jiahong Yuan. 2014. FAVE (Forced Alignment and Vowel Extraction) Program Suite. doi:10.5281/zenodo.22281. Schilling - Estes, Natalie & Walt Wolfram. 1999. Alternative Models of Dialect Death: Dissipation vs. Concentration. Language 75(3). 486 521. Schilling, Mark F., Ann E. Watkins & William Watkins. 20 02. Is human height bimodal? American Statistician 56(3). 223 229. doi:10.1198/00031300265. Seale, Elizabeth & Christine Mallinson. 2018. Rural Voices: Language, Identity, and Social Change Across Place . Lanham, Maryland: Lexington Books. Sneller, Elizabeth. 2017. Mechanisms of phonological change. PhD dissertation. University of Pennsylvania. Sneller, Elizabeth. 2019. Allophonic systems as a variable within individual speakers. In D Lightfoot (ed.), Variable properties in language: Their nature and acquisition . Washington, DC: Georgetown University Press. Stanford, James N. 2019. New England English: Large - Scale Acoustic Sociophonetics and Dialectology . Duke University Press. Strassel, Stephanie M & Charles Boberg. 1996. The Reversal of a Sound Chan ge in Cincinnati. University of Pennsylvania Working Papers in Linguistics (N)WAVES and MEANS: A selection of papers from NWAVE 24 3(1). 19. - study of English in Kansa s City. PhD dissertation. University of Missouri - Columbia. Sturtevant, Edgar H. 1947. An introduction to linguistic science . New Haven: Yale University Press. incrementation, and language change. Language 85(1). 58 108. Thiel, Anja & Aaron J. Dinkin. 2017. Escaping the TRAP: Losing the Northern Cities Shift in 177 Real Time. Paper presented at New Ways of Analyzing Variation 46 (NWAV46). Madison, Wisconsin. Nov. 2 - 5. Thomas, Erik R . 2011. Sociophonetics: An introduction . New York: Palgrave Macmillan. Thomas, Erik R. & Tyler Kendall. 2007. NORM: The vowel normalization and plotting suite. http://ncslaap.lib.ncsu.edu/tools/norm/. Thomas, Erik R., Tyler Kendall, Malcah Yaeger - Dror & Wi lliam Kretzschmar. 2007. Two things sociolinguists should know: software packages for vowel normalization, and accessing linguistic atlas data. Workshop at NWAV 36. University of Pennsylvania, PA. Trudgill, Peter & Tina Foxcroft. 1978. On the sociolinguist ics of vocalic mergers: Transfer and approximation in East Anglia. In Peter Trudgill (ed.), Sociolinguistic patterns in British English , 69 79. London: Arnold. Turton, Danielle. 2014. Variation in English /l/: Synchronic reflections of the life cycle of ph onological processes. PhD dissertation. University of Manchester. Turton, Danielle. 2017. Categorical or gradient? An ultrasound investigation of /l/ - darkening and vocalization in varieties of English. Laboratory Phonology: Journal of the Association for L aboratory Phonology 8(1). 1 31. doi:10.1086/426507. https://doi.org/10.5334/labphon.35. U.S. Department of Commerce. 2017. Household Income: 2017 . Wagner, Suzanne E, Alexander Mason, Monica Nesbitt, Erin Pevan, Matt Savage & Suzanne Evans Wagner. 2016. Rev ersal and Re - Organization of the Northern Cities Shift in Michigan. U. Penn Working Papers in Linguistics 22(2). http://repository.upenn.edu/pwpl/vol22/iss2/19. - a ssociated variable (ay0) in South Philadelphia. Universy of Pennsylvania Working Ppapers in Linguistics 13(1). 393 406. Weinreich, Uriel, William Labov & Marvin Herzog. 1968. Empirical foundations for a theory of language change . Austin: University of Texas Press. Wolfram, Walt & Natalie Schilling - Estes. 1998. American English: Dialects and variation . Malden, MA: Blackwell.