Incomplete Acquisition in the Nominal Domain of Korean by Heritage Language Speakers

by

Eugenia Suh

A thesis submitted in conformity with the requirements for the degree of Doctor of Philosophy Graduate Department of Linguistics University of Toronto

© Copyright by Eugenia Suh 2014

Incomplete Acquisition in the Nominal Domain of Korean by Heritage Language Speakers

Eugenia Suh

Doctor of Philosophy

Graduate Department of Linguistics University of Toronto

2014 Abstract

In this study, I examined heritage language speakers’ acquisition of the nominal domain of

Korean to see how these speakers’ age of onset of exposure to English, their frequency of exposure to Korean, and their usage of Korean could help account for their performance. Fifteen adult simultaneous bilinguals and eighteen adult sequential bilinguals participated in the study.

Their responses were compared to that of eight Korean-dominant bilinguals comprised of a subset of the mothers of the participants (and other similar speakers), who formed the control group, to represent the baseline (i.e., the input that heritage language speakers actually receive).

The participants completed language history questionnaires, which were used to create composite usage and exposure scores for Korean. General measures of the speakers’ ability in

Korean were compiled based on self-rated ability scores and an analysis of their accuracy and productivity on two of the tasks in this study. In order to examine the heritage language speakers’ grammars with respect to the nominal domain of Korean, the participants completed three tasks:

1) a preference task designed to test their sensitivity to various combinations of Case drop and their preferences for structures with or without plural marking and comitative Case markers; 2) an elicited imitation task designed to test their ability to recognize and produce Case markers; and 3) an elicited narration task designed to elicit production of Case markers, the plural marker,

ii and quantificational structures. The results of mixed effects logistic regression models suggest that, with respect to aspects of the Korean nominal domain that are sensitive to differing levels of acceptability, the heritage language speakers showed awareness of these levels overall; different aspects of their language experiences were useful in accounting for differences between speakers. The results from the control group also suggest that the input that heritage language speakers receive is not “faulty”.

iii

Acknowledgments

This thesis would not have been possible without the help of several important people. The biggest thanks go to the members of my committee: Ana Teresa Pérez-Leroux, Ron Smyth, and Yoonjung Kang. I am grateful to Ana for introducing me to the world of language acquisition, and for her insightful comments and suggestions. I still have some of my random scribbles, diagrams, and notes from our earlier meetings, and it’s quite an experience to see what those scribbles have materialized into. I am grateful for her encouragement and advice. I am also grateful to Bill Forrest for all of his help, and for the many and varied conversations we had.

I am indebted to Ron for his keen statistical insight and for his detailed comments on this thesis. I was particularly touched by all of the encouraging emails that he sent my way from Argentina. Each timely email was filled with exactly the right words at exactly the right moments. The wisdom he shared with me and the stories of his experiences really helped put me at ease, especially in the last stages of this thesis. I am also grateful to Ron for graciously arranging his travel schedule around my defense so that he would be available to teleconference.

I am grateful to Yoonjung for every insightful comment and suggestion she made on various aspects of this thesis. I was, and still am, inspired by her dedication, thoroughness, and efficiency in all that she does. Her support and encouragement helped me to keep going, and on more than occasion, she helped nudge me past a mental block. I am also very grateful to her for patiently starting me off on my way to running mixed-effects models in R.

I am thankful for the members of my defense committee: Alana Johns, Jeffrey Steele, and my external examiner, Tania Ionin. Alana has been a part of my world of linguistics from very early on. I have enjoyed learning from her, working with her, and just sitting and chatting with her. She has really brought out the fun side of being a linguist. Jeff generously shared his questionnaire with me years ago; I’m happy to have finally met him. I was amazed at the level of attention and detail he put into annotating my thesis. His comments and insightful questions have given me a lot of food for thought on how to improve and on what else to research. I am also grateful to Tania Ionin for highlighting the important issues and providing such clear comments and suggestions.

iv

A number of people generously shared their resources with me as I worked on this thesis. I am grateful to Ji-Hye Kim and Nina Spada for the materials they shared. I thank the Council of Europe for granting me permission to use their Common European Framework of References for Languages. I thank Marina Sherkina-Lieber, Ulyana Savchenko, and Danielle Thomas for very helpful discussions. I am also grateful to Daphna Heller for allowing me to audit her Quantitative Methods course and learn R, and to Derek Denis for also taking the time to patiently answer my statistics-related questions. All errors are of course my own.

I am thankful for everyone who patiently participated in my pilot tests, provided feedback on my test materials, and provided grammaticality judgments in general at various stages of this thesis. I would also like to thank everyone who participated in my experiments here and in Korea, and I am especially grateful for Ruth Kim’s invaluable help in gathering participants in Korea.

I would like to thank the faculty and staff at all three of our campuses. Mary Hsu, Jill Given- King, Arthus Bihis, Sean Ramrattan, Rosa Ciantar, Belinda Grayburn, and Joanna Szewczyk help take care of all of the details that are crucial in making everything run smoothly; I thank them for all the mini-fires they put out. I would also like to thank Gabriela Alboiu, Peter Avery, and Kasia Wolf. Diane Massam, Keren Rice, Arsalan Kahnemuyipour, Michael Lettieri, and Emmanual Nikiema can always be counted on to stop and ask how you’re doing. This always brought a smile to my face. Many thanks also go to Elizabeth Cowper for guiding me during my MA year and beyond. I appreciate her patience in helping me work out my ideas and thoughts, and for all of the good advice she provided me with. I am grateful for the gentle nudges and reminders; I’ve learned a lot from her. I would also like to thank Michelle Troberg, who I’ve had the pleasure of working with for a number of years now. I appreciate her straightforwardness and honesty, and I love how everything seems so simple and clear at the end of our conversations.

I am grateful for the late-night crew, the members of which have rotated over the years. All of you who worked, and still work, late into the night, thank you for your company. In particular, thank you to Midori Hayashi, Kenji Oda, Manami Hirayama, Christopher Spahr, Radu Craioveanu, Tomohiro Yokoyama, and Ross Godfrey. Thanks also to all of the caretakers at Sidney Smith Hall and to Kevin, our security guy.

I am also thankful for the inhabitants of the dungeon who welcomed me in: Joanna Chociej, Julia Su, and Kate Wu, and the new members who have helped make the dungeon a bit brighter:

v

Emilia Melara, Jessica Mathie, Becky Tollan, and Shayna Gardiner. In particular, I would like to thank Joanna, Julia, and Emilia for inviting me to so many things I wasn’t able to attend. I will attend now!

Thank you to Kyumin Kim, Jaehee Bak, Sarah Clarke, Bridget Jankowski, Richard Compton, Catherine Macdonald, Ulyana Savchenko, Magda Goledzinowska, Michael Barrie, Sandrine Tailleur, LeAnn Brown, Jee-Youn Shin, Maria Kyriakaki, Safieh Moghaddam, Liisa Duncan, Dan Milway, Ailis Cournane, Alex Motut, and Mercedeh Mohaghegh, for all of the laughter we share when we’re together. Thanks also go to Kyumin for allowing me to enjoy so many new experiences!

I am thankful to Haesun Moon and Bo-Yon Koh for their friendship, and for introducing me to SF. I am grateful to Bo-Yon for being my work-out partner and for helping me stay healthy. Kelly Choi completely changed my life when she introduced pivot tables to me; I am ever grateful for that, and for her generosity towards me in all that she does. I am thankful for Shazahra Saith, Stella Park, Jenny Hwang, Miranda and Moses Lee, Hana Yoo, Esther Oh, Rex and Leah Lee, Jenny Lee, Minjung Kim, Esther Park, Jenna Park, and for everyone at Wave and Momentum, who all have been praying hard for me, particularly this past summer.

I owe a lot to my father. I’m so grateful for his quiet encouragement and patience, and for all the ways he shows his support for me in my decisions. I also owe a lot to my mother, who, in some sense, helped prepare me for this journey. I’m also thankful for my brother, and the members of my extended family, who have all been rooting for me. I couldn’t have done it without you.

vi

Table of Contents

Acknowledgments ...... iv

Table of Contents ...... vii

List of Tables ...... xi

List of Figures ...... xxii

List of Abbreviations ...... xxiii

List of Appendices ...... xxiv

Chapter 1 Introduction ...... 1

1 1

1.1 Introduction ...... 1

1.2 The Goals of the Thesis ...... 3

1.3 Roadmap of the Thesis ...... 4

Chapter 2 Heritage Language Speakers ...... 5

2 Heritage Language Speakers ...... 5

2.1 Introduction ...... 5

2.2 Definitions ...... 6

2.3 Heritage Language Speakers’ Grammars ...... 8

2.3.1 HLS’ Grammar as Evidence of Transfer (Influence from the L2) ...... 9

2.3.2 HLS’ Grammar as Evidence of Insufficient Input ...... 11

2.3.3 HLS’ Grammar as Evidence of Incomplete Acquisition ...... 13

2.3.4 HLS’ Grammar as Evidence of Language Loss ...... 14

2.4 Age vs. Input ...... 17

2.4.1 Age ...... 18

2.4.2 Input (and Interactions) ...... 25

2.5 Korean Heritage Language Speakers ...... 30

vii

2.6 Summary ...... 32

Chapter 3 The Nominal Domain of Korean: Areas of Difficulty for HLS ...... 33

3 Aspects of the Korean Nominal Domain ...... 33

3.1 Introduction ...... 33

3.2 Case marking in Korean ...... 35

3.2.1 Case in Korean-Speaking Monolingual Children ...... 42

3.2.2 Previous Literature on HLS’ Comprehension of Case in Korean ...... 45

3.2.3 Case Marking in Korean: A Pilot Study ...... 49

3.3 -Tul ‘PL’ and Classifiers: Introduction ...... 55

3.3.1 Semantic Requirements and Syntactic restrictions on -Tul ...... 59

3.3.2 Previous Literature on the Acquisition of Number and Classifiers ...... 65

3.3.3 Experiment (Suh 2006) ...... 67

3.4 Conjunctions and Comitative Case Markers in Korean ...... 73

3.4.1 Comitative Case Markers ...... 75

3.4.2 Clausal Conjunctions ...... 79

3.4.3 Conjunctions: Pilot Study ...... 81

3.5 Summary ...... 91

Chapter 4 The Study: Questions, Hypotheses, and Methods ...... 95

4 The Study ...... 95

4.1 Introduction ...... 95

4.2 Questions and Hypotheses ...... 95

4.2.1 Case Marking and Case Drop in Korean ...... 95

4.2.2 -Tul, Classifiers, and Quantificational Structures ...... 97

4.2.3 Comitative Case Markers ...... 99

4.3 Participants and Methods ...... 100

4.3.1 Elicited Narration Tasks ...... 101

viii

4.3.2 Preference Task ...... 102

4.3.3 Elicited Imitation Task ...... 107

Chapter 5 Description of the Participants ...... 110

5 Participants ...... 110

5.1 Age of Onset of Exposure to English and Time Spent in Korea ...... 110

5.2 Usage and Exposure ...... 113

5.3 The Heritage Language Speakers’ Ability in Korean ...... 118

5.3.1 Self-Rated Ability in Korean ...... 119

5.3.2 Imitation Scores ...... 129

5.3.3 Subordination Index, Mean Length of Utterance, and Number of Different Words ...... 133

5.4 Summary ...... 146

Chapter 6 Results ...... 149

6 Results ...... 149

6.1 Case ...... 149

6.1.1 Case Drop: Heritage Language Speakers’ Preferences ...... 149

6.1.2 Elicited Imitation of Nominative and Accusative Case Markers ...... 160

6.1.3 Production of Case Markers in Elicited Narratives ...... 166

6.1.4 Case Marking: Analysis of Individuals ...... 176

6.1.5 Summary ...... 185

6.2 Plural Marking and Classifiers ...... 187

6.2.1 Preferences regarding Use of -Tul ...... 188

6.2.2 Production of -Tul and Quantificational Structures in Elicited Narratives ...... 195

6.2.3 Plural Marking and Classifiers: Analysis of Individuals ...... 199

6.2.4 Summary ...... 201

6.3 Comitative Case ...... 202

ix

6.3.1 Results of Preference Task ...... 203

6.3.2 Production of Comitative Case Markers in Elicited Narratives ...... 207

6.3.3 Comitative Case: Analysis of Individuals ...... 213

6.4 Summary ...... 214

Chapter 7 Discussion and Conclusions ...... 216

7 Summary ...... 216

7.1 Findings and Discussion ...... 216

7.2 Limitations of the Study and Areas for Future Research ...... 220

7.3 Contributions to the Field ...... 223

References ...... 225

Appendices ...... 239

x

List of Tables

Table 1. Case Drop in English and Korean ...... 42

Table 2. Complexity of Structures Produced: Mean Clauses Per Utterance ...... 51

Table 3. Accuracy of Structures Produced: Mean Error-Free Clauses (EFC) Per Utterance ...... 52

Table 4. (In)Accuracy of Structures Produced (Mean Errors) ...... 52

Table 5. Mean Proportion of Case-related Errors ...... 53

Table 6. Mean Proportion of Case Drop ...... 54

Table 7. Summary of Plural Marking in English ...... 65

Table 8. Summary of Variables Conditioning Plural Marking in Korean ...... 65

Table 9. Predicted Results for Production and Acceptance of - and -tul ...... 72

Table 10. Nominal Conjunctions: English vs. Korean ...... 81

Table 11. Clausal Conjunctions: English vs. Korean ...... 81

Table 12. Mean Proportion of Nominal Conjunctions Produced ...... 84

Table 13. Mean Proportion of Clausal Conjunctions Produced ...... 85

Table 14. Mean Proportion of Conjunctions Attached to Second Conjunct of Subject Nouns

([N1-NOM N2-CONJ]) ...... 90

Table 15. Case Marking and Case Drop: English vs. Korean ...... 92

Table 16. Plural Marking: English vs. Korean ...... 92

Table 17. Conjunctions: English vs. Korean ...... 93

Table 18. HLS’ Age-Related Information ...... 111

xi

Table 19. Distribution of HLS by Type of Bilingual and Number of Years Spent in Korea ..... 111

Table 20. Distribution of HLS by AoO and Number of Years Spent in Korea ...... 112

Table 21. HLS’ Median Responses to Questions Used to Create a Composite Score for Use of Korean ...... 114

Table 22. HLS’ Median Responses to Questions Used to Create a Composite Score for Exposure to Korean ...... 116

Table 23. Distribution of Simultaneous Bilinguals by Exposure to and Use of Korean ...... 117

Table 24. Distribution of Sequential Bilinguals by Exposure to and Use of Korean ...... 117

Table 25. Distribution of HLS by Exposure to and Use of Korean ...... 117

Table 26. Summary of HLS’ Language Experience Information ...... 118

Table 27. Summary of HLS’ Self-Ratings on their Ability in Korean based on a Four-point Likert Scale ...... 119

Table 28. Summary of HLS’ Self-Rating on their Ability in English based on a Four-point Likert Scale ...... 120

Table 29. Summary of the Difference between the HLS’ Self-Ratings on their Ability in Korean and English based on a Four-point Likert Scale ...... 121

Table 30. Summary of HLS’ Self-Ratings on their Ability in Korean based on Specific Criteria ...... 122

Table 31. Model Comparisons of Single Predictor Models to the Null Model of the HLS’ Self- rated Ability Scores (Model 1) ...... 123

Table 32. Model Comparisons of Two-Predictor Models to a Model with the HLS’ Usage Scores as a Predictor of the HLS’ Self-rated Ability Scores (Model 4) ...... 124

xii

Table 33. Model Comparisons of a Three-Predictor Model to a Model with the HLS’ Usage Scores and the Number of Years Spent in Korea as Predictors of the HLS’ Self-rated Ability Scores (Model 7) ...... 124

Table 34. Model 7: Generalized Linear Model Predicting HLS’ Self-rated Ability Scores in Korean Using the Number of Years in Korea and Usage Scores as Predictors ...... 125

Table 35. Summary of HLS’ Self-Rated Ability according to their Levels of Exposure to Korean and the Number of Years They Spent in Korea ...... 126

Table 36. Summary of HLS’ Self-Rated Ability according to their Usage and Number of Years Spent in Korea ...... 127

Table 37. Summary of HLS’ Median Self-Rated Ability according to their Levels of Exposure to Korean and their AoOs ...... 128

Table 38. Summary of HLS’ Median Self-Rated Ability according to their Usage of Korean and their AoOs ...... 128

Table 39. Summary of HLS’ Imitation Scores as a Measure of Ability in Korean ...... 130

Table 40. Model Comparisons of Single Predictor Models to a Null Model of the HLS’ Imitation Scores (Model 10) ...... 131

Table 41. Model Comparisons of Two-Predictor Models to a Model with Only the HLS’ Usage Scores as a Predictor of the HLS’ Imitation Scores (Model 13) ...... 131

Table 42. Model 13: Generalized Linear Model Predicting HLS’ Imitation Scores Using the HLS’ Usage Scores as a Predictor ...... 132

Table 43. Summary of HLS’ Proportion of Morphemes Repeated Correctly in the Elicited Imitation Task according to HLS’ Usage of Korean and their AoOs ...... 133

Table 44. Summary of the HLS’ Structures in their Elicited Narratives ...... 135

Table 45. Correlation Table of HLS’ Language History Variables and Measures of their Ability in Korean ...... 136

xiii

Table 46. Model Comparisons of Single Predictor Models to a Null Model of the HLS’ SIs (Model 18) ...... 137

Table 47. Model Comparisons of Two-Predictor Models to a Model with the HLS’ AoOs as a Predictor of the HLS’ SIs (Model 19) ...... 137

Table 48. Model Comparisons of Two-Predictor Models to a Model with the HLS’ Number of Years in Korea as a Predictor of the HLS’ SIs (Model 20) ...... 138

Table 49. Model 23: Generalized Linear Model Predicting HLS’ SIs Using the HLS’ AoOs and Usage Scores as Predictors ...... 138

Table 50. Model 25: Summary of a Generalized Linear Model Predicting HLS’ SIs Using the HLS’ Number of Years in Korea and Usage Scores as Predictors ...... 138

Table 51. Summary of HLS’ SIs according to the HLS’ Usage of Korean and their AoOs ...... 139

Table 52. Model Comparisons of Single Predictor Models to a Null Model of the HLS’ MLUs (Model 27) ...... 140

Table 53. Model Comparisons of Two-Predictor Models to a Model with the HLS’ AoOs as a Predictor of the HLS’ MLUs (Model 28) ...... 140

Table 54. Results of Likelihood Ratio Tests Comparing Various Two Predictor Models to a Model with the HLS’ Number of Years in Korea as a Predictor of the HLS’ MLUs (Model 29) ...... 140

Table 55. Model 28: Generalized Linear Model Predicting HLS’ MLUs Using the HLS’ AoOs as a Predictor ...... 141

Table 56. Model 29: Generalized Linear Model Predicting HLS’ MLUs Using the HLS’ Number of Years in Korea as a Predictor ...... 141

Table 57. Summary of HLS’ MLUs according to the HLS’ Usage of Korean and their AoOs . 142

Table 58. Summary of HLS’ MLUs according to the HLS’ Usage of Korean and their Number of Years Spent in Korea ...... 143

xiv

Table 59. Model Comparisons of Single Predictor Models to a Null Model of the HLS’ NDWs (Model 36) ...... 144

Table 60. Model Comparisons of Two-Predictor Models to a Model with the HLS’ Exposure Scores as a Predictor of the HLS’ NDWs (Model 40) ...... 144

Table 61. Model 40: Generalized Linear Model Predicting HLS’ NDWs Using the HLS’ Exposure Scores as a Predictor ...... 145

Table 62. Summary of HLS’ NDWs according to the HLS’ Exposure to Korean and their Number of Years Spent in Korea ...... 145

Table 63. Summary of HLS’ NDWs according to the HLS’ Exposure to Korean and their AoOs ...... 146

Table 64. Summary of Significance of Various Aspects of the HLS’ Language Backgrounds in Predicting their Measures of Ability in Korean ...... 147

Table 65. Correlation Table of Measures of HLS’ Ability in Korean ...... 147

Table 66. Median and Mean Proportion of Grammatical Distractors Correctly Selected in the Preference Task ...... 149

Table 67. Median and Mean Proportion of Correct Responses Showing Preferences of No Drop > Acc Drop > Nom Drop > Both Dropped in the Preference Task ...... 150

Table 68. Mean Proportion of Correct Responses Showing the Expected Preferences according to the Case Drop Hierarchy in the Preference Task ...... 151

Table 69. Likelihood Ratio Tests Comparing Single Predictor Models to a Null Model Predicting the HLS’ Correct Selections in the Preference Task (Model 44) ...... 153

Table 70. Likelihood Ratio Tests Comparing Two Predictor Models Predicting the HLS’ Correct Selections in the Preference Task with a Model with the Case Combination Preferred as a Predictor (Model 46) ...... 154

xv

Table 71. Model 46: Mixed-effects Logistic Regression Model Predicting the HLS’ Correct Selections in the Preference Task Using the Case Combination Expected to be Preferred as a Fixed Effect Predictor ...... 155

Table 72. Model 46: Mixed Effects Logistic Regression Model Predicting the HLS’ Correct Selections in the Preference Task Using the Case Combination Expected to be Preferred as a Fixed Effect Predictor ...... 156

Table 73. Likelihood Ratio Tests Comparing Single Predictor Models to a Null Model Predicting the HLS’ Correct Selections in the Preference Task (Model 49) ...... 157

Table 74. Likelihood Ratio Tests Comparing Two-Predictor Models to a Model with the Case Combination Preferred as a Fixed Effect Predictor of the HLS’ Correct Selections in the Preference Task (Model 50) ...... 158

Table 75. Likelihood Ratio Test Comparing a Model with an Interaction to a Two-Predictor Model Predicting the HLS’ Correct Selections in the Preference Task (Model 58) ...... 158

Table 76. Model 59: Mixed Effects Logistic Regression Model Predicting the HLS’ Correct Selections in the Preference Task Using an Interaction between the Case Combination Expected to be Preferred and the HLS’ Exposure Scores ...... 159

Table 77. Number of Tokens Removed from the Data from the Elicited Imitations ...... 160

Table 78. Mean Proportion of Nominative and Accusative Case Markers Correctly Repeated in the Elicited Imitation Task ...... 161

Table 79. Mean Proportion of Case Markers Correctly Repeated in the Elicited Imitation Task ...... 162

Table 80. Likelihood Ratio Tests Comparing Single Predictor Models to a Null Model of the HLS’ Correct Imitations of Case Markers in the Elicited Imitation Task (Model 60) ...... 163

Table 81. Likelihood Ratio Tests Comparing Two-Predictor Models to a Model of the HLS’ Correct Imitations of Case Markers in the Elicited Imitation Task with the Type of Case Marker Expected as a Fixed Effect Predictor (Model 61) ...... 164

xvi

Table 82. Likelihood Ratio Tests Comparing a Model with an Interaction to a Two-Predictor Model Predicting the HLS’ Correct Imitations of Case Markers in the Elicited Imitation Task (Model 66) ...... 165

Table 83. Model 66: Mixed Effects Logistic Regression Model Predicting HLS’ Correct Imitations of Case Markers in the Elicited Imitation Using the Type of Case Required and the HLS’ AoOs as Fixed Effect Predictors ...... 166

Table 84. Median and Mean Proportion of Nominative and Accusative Case Markers Correctly Produced in Elicited Narratives ...... 167

Table 85. Median and Mean Proportion of Case Markers Produced in their Elicited Narratives by Type of Case Required ...... 168

Table 86. Likelihood Ratio Tests Comparing Single Predictor Models to a Null Model Predicting the HLS’ Correct Production of Case Markers in their Elicited Narratives (Model 71) ...... 170

Table 87. Likelihood Ratio Test Comparing a Two-Predictor Model to a Model with Case as a Predictor of the HLS’ Correct Production of Case Markers in their Elicited Narratives (Model 73) ...... 171

Table 88. Likelihood Ratio Test of a Model with an Interaction to a Two-Predictor Model Predicting the HLS’ Correct Production of Case Markers in their Elicited Narratives (Model 74) ...... 171

Table 89. Model 75: Summary of a Mixed Effects Logistic Regression Model Predicting the Participants’ Proportion of Case Markers Correctly Produced in their Elicited Narratives by an Interaction for the Type of Case Marker Required and the Grouping of the Participants ...... 172

Table 90. Likelihood Ratio Tests Comparing Two-Predictor Models with a Model with Case as a Predictor of the HLS’ Correct Production of Case Markers in their Elicited Narratives ...... 173

Table 91. Likelihood Ratio Test Comparing a Model with an Interaction to a Two-Predictor Model of the HLS’ Correct Production of Case Markers in their Elicited Narratives ...... 174

xvii

Table 92. Model 79: Mixed Effects Logistic Regression Model Predicting the HLS’ Correct Production of Case Markers in their Elicited Narratives Using their Usage Scores and the Type of Case Required as Fixed Effect Predictors ...... 174

Table 93. Raw counts of Overproduction of Case Markers in the Elicited Narratives...... 175

Table 94. Summary of HLS’ Proportion of Sentences Correctly Preferred according to their Levels of Exposure to Korean and their AoOs ...... 177

Table 95. Summary of HLS’ Proportion of Sentences Correctly Preferred according to their Usage of Korean and their AoOs ...... 178

Table 96. Summary of HLS’ Proportion of Case Markers Repeated Correctly according to their Levels of Exposure to Korean and their AoOs ...... 179

Table 97. Summary of HLS’ Proportion of Case Markers Repeated Correctly according to their Usage of Korean and their AoOs ...... 179

Table 98. Summary of HLS’ Production of Nominative Case Markers in their Elicited Narratives according to their Levels of Exposure to and Use of Korean ...... 180

Table 99. Summary of HLS’ Production of Accusative Case Markers in their Elicited Narratives according to their Levels of Exposure to and Use of Korean ...... 181

Table 100. Summary of HLS’ Production of Nominative Case Markers in their Elicited Narratives according to their AoO and and their Usage of Korean ...... 182

Table 101. Summary of HLS’ Production of Accusative Case Markers in their Elicited Narratives according to their AoOs and and their Usage of Korean ...... 183

Table 102. Summary of HLS’ Production of Nominative Case Markers in their Elicited Narratives according to their AoOs and and their Exposure to Korean ...... 184

Table 103. Summary of HLS’ Production of Accusative Case Markers in their Elicited Narratives according to their AoOs and their Exposure to Korean ...... 185

Table 104. Summary of HLS’ Performance on Case-Related Tasks ...... 187

xviii

Table 105. Summary of the Influence of the HLS’ Background Variables on their Performance on Case-Related Tasks ...... 187

Table 106. Median and Mean Proportion of Sentences with -tul on Specific Nouns Correctly Selected in the Preference Task ...... 188

Table 107. Median and Mean Proportion of Sentences without -tul on Classifiers Correctly Selected in the Preference Task ...... 189

Table 108. Likelihood Ratio Tests Comparing Single Predictor Models to a Null Model Predicting the HLS’ Correct Selections in Various Plural Contexts in the Preference Task (Model 82) ...... 190

Table 109. Likelihood Ratio Test Comparing a Two-Predictor Model to a Model with the Plural Context as a Fixed Effect Predictor of the HLS’ Correct Selections in the Preference Task (Model 84) ...... 190

Table 110. Model 84: Mixed Effects Logistic Regression Model Predicting the HLS’ Correct Selections in the Preference Task Using Context as a Fixed Effect Predictor ...... 191

Table 111. Model 84: Mixed Effects Logistic Regression Model Predicting the HLS’ Correct Selections in the Preference Task Using Context as a Fixed Effect Predictor ...... 192

Table 112. Likelihood Ratio Tests Comparing Single Predictor Models to a Null Model Predicting the HLS’ Correct Selections in Plural Contexts in the Preference Task (Model 86) 193

Table 113. Likelihood Ratio Tests Comparing Two-Predictor Models to a Model with the Context as a Predictor of the HLS’ Correct Selections in Plural Contexts in the Preference Task (Model 87) ...... 194

Table 114. Likelihood Ratio Test Comparing a Model with an Interaction to a Model with the HLS’ Usage Scores and the Context as Fixed Effect Predictors of the HLS’ Correct Selections in Plural Contexts in the Preference Task (Model 86) ...... 194

xix

Table 115. Model 94: Mixed Effects Logistic Regression Model Predicting the HLS’ Correct Selections in the Preference Task Using the HLS’ Usage Scores and Context as Fixed Effect Predictors ...... 195

Table 116. Proportion of Tokens of Plural -tul Produced in the Elicited Narratives in Plural Contexts ...... 196

Table 117. Distribution of Quantificational Structures Produced in Elicited Narratives ...... 197

Table 118. Production and Position of Classifiers Produced according to Group ...... 198

Table 119. Summary of HLS’ Performance with Respect to Plural Marking and Quantified Structures ...... 202

Table 120. Summary of the Influence of HLS’ Background Variables on their Performance with Plural Marking and Quantified Structures ...... 202

Table 121. Median and mean Proportion of Sentences with -ul kaciko ‘ACC take-and’ with Inanimate Correctly Preferred over Sentences with Comitative Case Markers . 203

Table 122. Likelihood Ratio Tests Comparing Single Predictor Models to a Null Model

Predicting the HLS’ Selection of Sentences with -ul kaci-ko ‘ACC take-and’ in the Preference Task (Model 96) ...... 204

Table 123. Likelihood Ratio Test Comparing Two-Predictor Model to a Model Using Context as a Predictor of the HLS’ Correct Selections of Sentences with -ul kaci-ko ‘ACC take-and’ in the Preference Task (Model 98) ...... 204

Table 124. Model 98: Mixed Effects Logistic Regression Model Predicting the HLS’ Correct

Selections of Sentences with -ul kaci-ko ‘ACC take-and’ in the Preference Task ...... 205

Table 125. Likelihood Ratio Test Comparing Single Predictor Models to a Null Model Predicting the HLS’ Correct Selections of Sentences with -ul kaci-ko ‘ACC take-and’ in the Preference Task (Model 100) ...... 206

xx

Table 126. . Likelihood Ratio Test Comparing Two-Predictor Models to a Model Using Context to Predict the HLS’ Correct Selections of Sentences with -ul kaci-ko ‘ACC take-and’ in the Preference Task (Model 101) ...... 206

Table 127. Model 101: Mixed Effects Logistic Regression Model Predicting the HLS’ Correct

Selections of -ul kaci-ko ‘ACC take-and’ in the Preference Task with Context as a Fixed Effect Predictor ...... 207

Table 128. Distribution of Comitative Case Markers Produced in the Elicited Narratives ...... 208

Table 129. Distribution of Levels of Speech Used in Elicited Narratives by Participants ...... 210

Table 130. Distribution of Comitative Case Markers Incorrectly Produced in Elicited Narratives ...... 211

Table 131. Number of Speakers who Produced Errors with Comitative Case Markers ...... 212

Table 132. Proportion of Comitative Case Markers Provided in Elicited Narratives ...... 212

Table 133. Proportion of Accusative Case markers Correctly Produced (with kaci-ko ‘take-and’) on Inanimate ...... 213

Table 134. Summary of HLS’ Performance with Respect to Comitative Case Markers ...... 215

Table 135. Summary of the Influence of the HLS’ Background Variables on their Performance on Comitative Case-Related Tasks ...... 215

xxi

List of Figures

Figure 1. Mean Proportion of Nominative and Accusative Case Markers Correctly Produced in Elicited Narratives ...... 169

xxii

List of Abbreviations

I – level I register II – level 2 register III – level 3 register IV – level 4 register Acc – accusative Case Cond – conditional Dat – dative Case Decl – declarative Gen – genitive Case Hon – honorific Hyp – hypocoristic Imp – imperative Nmlz - nominalizer Nom – nominative Case Pres – present tense Pst – past tense Q – question (interrogative) Rel - relativizer

In this thesis, I use the Yale Romanization system (Martin 1992) for transcriptions of Korean.

xxiii

List of Appendices

Appendix Table A1. Summary of HLS’ Attitudes towards Korean and Speaking Korean ...... 239

Appendix Table A2. Summary of HLS’ Language Use in Childhood ...... 237

Appendix Table A3. HLS’ Language Use with Family and Friends ...... 238

Appendix Table A4. HLS’ Use of Korean in Various Situations ...... 239

Appendix Table A5. HLS’ Passive Exposure to Korean ...... 240

Appendix Table A6. HLS’ Written Interaction in and Exposure to Korean ...... 240

Figure A1. Summary of HLS’ Responses on Language Questionnaire ...... 241

Appendix Table A7. Correlation Table for Measures Used in Usage Scores for HLS ...... 241

Appendix Table A8. Correlation Table for Measures Used in Exposure Scores for HLS ...... 242

Appendix Table A9. Summary of the HLS’ Language Histories ...... 243

Appendix Table A10. Summary of HLS’ Language Histories and Measures of Ability (as High or Low) ...... 244

Appendix Table A11. Summary of Individual Simultaneous Bilinguals’ Performance with Case Marking on the Preference Task, Elicited Imitation Task, and the Elicited Narration Task ...... 245

Appendix Table A12. Summary of Individual Sequential Bilinguals’ Performance with Case Marking on the Preference Task, Elicited Imitation Task, and the Elicited Narration Task . 246

Appendix Table A13. Summary of Individual Simultaneous Bilinguals’ Performance (as Pass or Fail, based on 0.75 Threshold) with Case Marking on the Preference Task, Elicited Imitation Task, and the Elicited Narration Task ...... 247

Appendix Table A14. Summary of Individual Sequential Bilinguals’ Performance (as Pass or Fail, based on 0.75 Threshold) with Case Marking on the Preference Task, Elicited Imitation Task, and the Elicited Narration Task ...... 248

Appendix Table A15. Proportion of Types of Quantified Structures Produced in Elicited Narratives ...... 249

xxiv 1

Chapter 1 Introduction 1 1.1 Introduction

Definitions of heritage language speakers (HLS) vary (e.g., Merino 1983, Grosjean 1998, Fishman 2001, Lynch 2003, Van Deusen-Scholl 2003, Valdés 2005, Polinsky & Kagan 2007, Polinsky 2008, and Benmamoun, Montrul, & Polinsky 2010), as heritage language acquisition is a relatively new field. The group of speakers that I am interested in is those who have received input in a heritage language (in particular, Korean) from a young age and who have at least basic aural and oral proficiency in the language. HLS are often assumed to have grammars which are incomplete in some respect. Although they are exposed to their heritage language from birth, their resulting grammars often differ from that of monolingual speakers living in a country where the heritage language (HL) is the language of the majority (e.g., in the country where the HLS may have immigrated from). HLS are often classified into two groups: simultaneous or sequential bilinguals, depending on their age of acquisition of the language of the majority (e.g., English). For example, speakers who grew up with two languages from birth or at least before the age of three are considered to be simultaneous bilinguals, whereas sequential bilinguals are speakers who were exposed to a second language after first acquiring their heritage language (i.e., speakers who exclusively used and heard the HL until at least the age of four). However, this classification is not always the most informative. Since the language experiences of HLS are often quite different, a study of HLS’ language experiences can help us investigate which factors affect the successful, or unsuccessful, acquisition of the HL. Where the language experiences of HLS are unequal, I predict that more frequent production of the HL will be more important than receiving frequent input in the language. In this respect, it is also important to test the HLS’ actual input (i.e., the language of their parents) and use this as the baseline, in order to test the assumption that the input that HLS receive is quantitatively and/or qualitatively different from that which monolingual children are generally exposed to in the “home country” (i.e., the country which the HLS’ parents immigrated from) (Polinsky & Kagan 2007, Polinsky 2008).

2

In order to investigate these issues, I examined the grammar of HLS of Korean as compared to their mothers (the HLS’ main source of input at the initial point of acquisition). In particular, I focus on aspects of the nominal domain which differ between English and Korean, two typologically different languages. For example, Korean is a language which has a fairly flexible word order. The meaning is recoverable even in non-canonical sentences, because nouns are generally marked with Case markers which identify their role in the sentence. In the canonical

sentence in (1), -ka ‘NOM’ indicates that Mincwu is the subject, and -ul ‘ACC’ indicates that pap ‘rice’ is the object.

(1) Mincwu-ka pap-ul mek-nun-ta. Minjoo-NOM rice-ACC eat-PRES-DECL ‘Minjoo is eating.’

In colloquial language, nominative, accusative, and genitive Case markers can often be omitted (i.e., Case drop can occur). Even without the overt case markers, (1) is still grammatical, as shown in (2) below.

(2) Mincwu-Ø pap-Ø mek-nun-ta. Minjoo-NOM rice-ACC eat-PRES-DECL ‘Minjoo is eating.’

Case markers are not dropped equally across the board. A Case drop hierarchy has been observed in the production of Korean, where genitive Case markers are dropped the most often, followed by accusative Case markers and nominative Case markers: Genitive > Accusative > Nominative (K.-h. Kim & J.-i. Kwon 2004, H. Lee 2006). Since (1) and (2) are both grammatical, monolingual Korean-speaking children often do not attach the appropriate Case markers. Thus, overuse of Case drop (or not producing overt Case markers regularly) by adults can sometimes be perceived as “baby talk” (S.-H. Lee et al. 2009), and sentences such as (2) are frequently produced by HLS.

An elicited narration task, an elicited imitation task, and a preference task were used to investigate the HLS’ use of Case marking and other aspects of the Korean nominal domain. The sentences produced by the HLS in the elicited narration task were analyzed in terms of the number of nominative and accusative Case markers produced, the types of quantificational structures produced, and the number of comitative Case markers produced. The sentences were also analyzed in terms of the level of embedding present (i.e., the mean number of clauses

3 produced per utterance), the mean number of words produced per utterance (MLU), and the number of different words (NDW) produced as measures of the HLS’ ability in Korean. The number of morphemes correctly repeated back in the elicited imitation task was used as another measure of ability, in addition to self-assessments of the HLS’ ability in Korean. The data from the elicited imitation task were also analyzed for the number of nominative and accusative Case markers imitated correctly. The preference task was used to test the HLS’ sensitivity to various combinations of Case drop, and also to test the HLS’ preferences for structures with or without plural marking and comitative Case markers. The HLS completed questionnaires on their language experiences and history; this information was used to compile composite usage and exposure scores, and these were used as predictors in mixed effects logistic regression models, along with the HLS’ ages of onset of extended exposure to English. The results indicate that different aspects of the HLS’ language experiences were useful in predicting their performance.

1.2 The Goals of the Thesis

The main goal of this study is to look into the grammars of HLS of varying levels of ability in the Korean language, their HL, and to investigate how age of onset of exposure to English, the amount of input received in the HL, and the frequency of usage of the HL can help predict the HLS’ performance in the production of various markers and structures in the nominal domain. Often studies are conducted on students who are already enrolled in Korean language classes at a university, who have been grouped according to the level of the course, and who are presumably able to write in the heritage language. Rather than testing speakers of a particular proficiency to see how they perform, I investigate how aspects of the HLS’ language experiences can help account for the HLS’ performance, irrespective of their overall ability in the language. Since HLS are known to have stronger listening and speaking skills, compared to their writing and reading skills, I tested speakers in an oral modality, without the prerequisite of being able to write in Korean, to see how “incomplete” these speakers’ grammars were. Another goal of this study was to compare Korean HLS’ performance to that of their actual baseline (i.e., the language of their parents), which has not been done in very many studies on Korean, with the exception of S. Park (2010) and Kang & Nagy (2013), to also determine whether the HLS are receiving “faulty input”. I examine aspects of the nominal domain which are sensitive to factors which involve various domains of grammar: (nominative and accusative) Case, plural marking

4 and quantificational structures, and comitative Case marking to investigate how these aspects are represented in HLS’ grammars.

1.3 Roadmap of the Thesis

The rest of the thesis is structured as follows. In the following chapter, I present the definitions of heritage language speakers that will be used in this study, and provide an overview of some of the literature on heritage language speakers and their grammars, and the issue of whether age of arrival or exposure and input in Korean is better at accounting for the nature of HLS’ grammars. Chapter 3 discusses the details of the aspects of the nominal domain of Korean to be tested in the HLS, including the marking of (nominative and accusative) Case, number, and comitative Case marking. For each phenomenon, the syntactic structures assumed, and the relevant literature on studies which have investigated these structures amongst different populations (e.g., monolingual Korean-speaking children, Korean-English bilinguals, and second language learners of Korean) is discussed. Chapter 4 presents the questions to be investigated in the thesis, the hypotheses, and the methodology of the experiment used to explore the questions posed. A detailed summary of the participants and their language experiences is given in Chapter 5. Section 5.3 presents a description of the HLS’ ability in Korean. The HLS’ self-ratings on two questionnaires, as well as their performance on an elicited imitation task are provided. The HLS’ elicited narratives are also analyzed to obtain various measures of proficiency, including a subordination index (i.e., the mean number of clauses per utterance), the mean length of their utterances (MLU), and an estimate of the size of their vocabulary by a measure of the number of different words produced (NDW). In Chapter 6, Section 6.1 presents the results of the HLS’ performance with respect to nominative and accusative Case marking, Section 6.2 presents the results of the HLS’ performance with respect to the use of the plural marker in Korean and their use of various structures to show quantification, and Section 6.3 presents the results of the HLS’ use of comitative Case markers. Chapter 7 presents a discussion of the results and concludes the thesis.

5

Chapter 2 Heritage Language Speakers 2 Heritage Language Speakers 2.1 Introduction

A traditional distinction among subfields of language acquisition has been the age of the speakers under investigation and whether these speakers are learning a single language or more than one. Child language acquisition is the study of the acquisition process of monolingual children, whereas second language acquisition (SLA) generally investigates the acquisition process of adults who are learning a second language (L2) after having already acquired their mother tongue. Excluding cases of language impairment, under normal circumstances, a monolingual child will successfully acquire language, whereas adult L2 learners often fall short of targetlike norms in their L2. When a native speaker of English in Canada learns French in elementary school, the result is often additive bilingualism (e.g., Lambert 1973); however, when a native speaker of an immigrant language (i.e., a heritage language (HL)) begins learning L2 English in Canada at a young age, many have observed subtractive bilingualism, with the loss of the HLS’ “L1” (e.g., Wong Fillmore 1991). If monolingual children are expected to successfully acquire their L1, it is odd that bilingual children would “lose” their L1 (their HL) due to the acquisition of an L2. Many explanations for this loss or attrition have been put forth, such as a lack of (sufficient) input in the HL, heritage language speakers (HLS) being exposed to an L2 at too early an age (accompanied with insufficient input in their HL) (Montrul 2008), HLS’ recognition of the L2 as a language of higher status or value in the community, leading to a desire to align with the culture of the L2 (Jeon 2008), or HLS’ exposure to input which may differ from that which monolingual children are exposed to (Paradis & Navarro 2003, Sanchez 2004, Rothman 2007, among others). Heritage language acquisition has thus been researched from several different perspectives. Foreign language teachers are primarily interested in which aspects of HLS’ grammars are non-target-like and which need to be explicitly corrected as HL speakers (HLS) “re-learn” their HL (especially when compared to the needs of non-heritage L2 learners). Sociolinguists are interested in the role of cultural influences, the group with which speakers identify themselves, and other background factors that play a role in determining HLS’ motivation to learn and maintain an HL or not, as well as the processes of intergenerational

6 language loss and diachronic language change. L2 or bilingual language researchers are interested in discovering the causes of variation or divergence in HLS’ grammars, as well as the differences between HLS’ grammars and monolingual children’s grammars. Each area contributes an essential part to our understanding of HLS and HLA. Since the language experiences of HLS are often quite different, a study of the language histories of HLS can help us determine which factors affect the successful, or unsuccessful, acquisition of an L1 from input which may be quantitatively and/or qualitatively different from that which monolingual children are generally exposed to in the “home country” (i.e., the country from which the HLS’ parents immigrated). In addition, we can learn in what ways HLS’ grammars are different from those of monolingual speakers of the HL and to what extent they are evidence of language change via attrition in the HLS’ parents’ language or diachronic change in (colloquial) varieties of the HL as it is spoken in the home country. Since HLA is a field that is quite broad in scope and spans many different research interests, the terminology used to describe HLS and their language situations can vary according to the specific area of HLA that researchers are interested in. In the following section, I will outline the terms that will be used to describe HLS in this thesis, followed by characteristics of HLS grammars and the different explanations for these characteristics. Section 2.4 will present the arguments for and against using the age of first exposure to the language of the majority as a measure of how proficient HLS will be in their HL as adults, and section 2.5 will present details of the language experiences of Koreans in Toronto.

2.2 Definitions

HLS are an understudied population that has been gaining more attention in the past couple of decades (Polinsky 1996, 2008; Montrul 2002, 2004; Lynch 2003, J.-H. Kim 2004a, 2004b; J.-H. Kim, Montrul, & Yoon 2005, Valdés 2005, Polinsky & Kagan 2007, J.-H. Kim, Montrul & Yoon 2009, among others), and definitions of HLS vary. Some linguists (Fishman 2001, Van Deusen- Scholl 2003) assert the importance of distinguishing between heritage languages based on the socio-political context of language contact (e.g., immigrant languages, colonial languages, and indigenous languages). More commonly, HLS are simply defined as speakers of a particular ethnic background, where the language of the majority is not the same as the HL (Fishman 2001), or speakers of a “home” language (Valdés 2005). Merino (1983), Grosjean (1998), Lynch (2003), Valdés (2005), Polinsky & Kagan (2007), Polinsky (2008), and Benmamoun et al. (2010) define HLS as speakers who learn their HL (e.g., Russian) as their L1 but then subsequently switch

7 their language of dominance to the language of the majority (e.g., English), often once they begin attending school at age 5.1 The typical HLS is an unbalanced bilingual who is more proficient in the language of the majority, with little or no formal education in their HL, and the bulk of research on HLS has focused on this type of speaker.

Although unbalanced bilinguals comprise the majority of HLS, it is still possible for HLS to become balanced bilinguals. HLS can be distinguished according to their language (usage) preferences (e.g., whether they use their HL as their primary language or their secondary language) and in the order of acquisition of their HL relative to the language of the majority (i.e., whether their HL is their first or second language). HLS also differ in the age at which they were first exposed to the language of the majority. For example, speakers who were exposed to two languages before the age of three are considered to be simultaneous bilinguals2, whereas speakers who were exposed to a second language after having acquired their L1 are considered to be sequential bilinguals. Thus, it is important to characterize HLS appropriately, given the range in their levels of proficiency and types of language experiences. I am interested in HLS who have at least basic aural and oral proficiency in their “home language”, an “immigrant language” (e.g., Korean) which differs from the “language of the majority” (e.g., English), regardless of their language of preference (i.e., whether they are English-dominant or Korean- dominant). This includes simultaneous bilinguals and sequential bilinguals who immigrated (to Canada) by the age of 10. I will refer to the language of the majority (or the societal or ambient language) as the “L2” for ease of exposition. Although there is no question that this language is the L2 for sequential bilinguals, strictly speaking, it is not the “second” language that simultaneous bilinguals are exposed to. In the latter case, it is simply the “other” (first) language that the bilinguals are exposed to, and “L2” is not intended to imply that there is any particular difference in the status of the two languages or in the timing of the learning of the HL and “L2” for these speakers. Similarly, “L1” and “HL” will be used interchangeably, and “baseline” will

1 Benmamoun et al. (2010, p. 15) describe HLS as those who were born in a country where their dominant language is spoken, which presumes that HLS are more dominant in the language of the majority – or they arrived in the US early enough to learn the language (i.e., English) as a “bilingual” – i.e., by the age of 4-5 (Schwartz 2004, Unsworth 2004, G. Cho, F. Shin, & Krashen 2004). Thus, their definition of a “bilingual” appears to only apply to simultaneous or very early sequential bilinguals.

2 There is some debate on who should be classified as a simultaneous bilingual. Traditionally the cut-off age has been considered to be at 3 years old (Grosjean 1987, among others), although a recent phonetic study used 5-6 years of age as the cut-off (Lee & Iverson 2011).

8 be used to refer to the input that HLS receive in their HL (typically the language of their immigrant parents).

2.3 Heritage Language Speakers’ Grammars

Research on HLS has highlighted differences between the baseline (assumed to be the language as it is spoken in a country where monolinguals exist, or as the HL as it is spoken by immigrant adults) and HLS’ grammars. For example, HLS have been found to have non-native-like accents depending on their language experiences (Au, Knightly, Jun, & Oh 2002, Oh, Jun, Knightly, & Au 2003, Knightly, Jun, Oh, & Au 2003) and limited lexicons, especially with respect to academic areas (hence the use of the term “kitchen language” to refer to HLS’ HL), resulting in the use of frequent code-switching (Kaufman & Aronoff 1991, Silva-Corvalán 1994, Polinsky & Kagan 2007). HLS’ morphosyntax also often differs from the baseline. Studies on tense and aspect (Kaufman & Aronoff 1991, Polinsky 2006, Polinsky 2008, Laleko 2010) and the nominal domain (T. Lee 2011), as well as HLS’ interpretations of anaphors (J.-H. Kim 2007) have revealed that HLS often have reduced systems, resulting in fewer contrasts in the language, often accompanied by a reanalysis of existing boundaries. In addition, due to their often limited use of the HL outside the home, HLS often overuse informal language and have difficulty with honorifics (Byon 2003, J. S. Lee 2008).

Some linguists attribute these differences to incomplete acquisition and/or subsequent reanalysis and transfer from the L2 (Kim & Montrul 2004, Kim, Montrul & Yoon 2004, Montrul 2002, Montrul 2004, Montrul 2006), and others view HLS’ grammars as developing on par with other languages that develop under minimal input, such as creoles (Polinsky 2008). If it is the case that HLS receive insufficient input, it is assumed that differences between their HL grammars and the baseline are due to incomplete acquisition. For example, if HLS’ grammars are similar to an earlier developmental stage in monolingual children, this would be evidence of incomplete acquisition. It is assumed that the HLS were unable to receive sufficient input to progress beyond that particular development stage, resulting in arrested development (Montrul 2008) (akin to “fossilization” in L2 learners). HLS’ grammars could show variation that matches the input they receive (i.e., their baseline), but if HLS’ grammars have properties which are more similar to those of their L2 rather than their baseline, this could be attributed to influence from the L2, as well as incomplete acquisition. However, if the HL grammar shows evidence of simplification in

9 line with universal principles of markedness in a way that is different from both the L2 and the L1 grammar of monolingual speakers of the HL, it could be likely that the HLS may have acquired the HL grammar fully as a child and then subsequently started simplifying the grammar. We would assume some degree of attrition had occurred. Some of the relevant studies on HLS and the claims that have been made will be reviewed below.

2.3.1 HLS’ Grammar as Evidence of Transfer (Influence from the L2)

Since HLS are exposed to more than one language, one obvious possible source of variation in their HL grammar is the influence of the other language (i.e., the L2). If the HL and the L2 differ in a particular property, and we see that the HL grammar is more like the L2 than the baseline, we can assume that the L2 has had an influence on the HL. If the grammar is neither like the baseline nor like the L2 grammar with respect to particular properties, the grammar may simply be a more unmarked option cross-linguistically.

Silva-Corvalán (2008) describes transfer in the lexical-pragmatic domain3 of Spanish-English bilinguals4 born in the United States. She found that transfer was evidenced in structures where parallels existed in Spanish and English but with different constraints on their usage. For example, English complementizers are optional in relative and complement clauses, whereas Spanish complementizers are obligatory in (spoken) General Spanish5, but optional in complement clauses in written registers. Bilinguals, however, allowed complementizers to be optional even in colloquial language for verbs such as creer ‘believe’, pensar ‘think’, and saber ‘know’, 6 presumably due to negative transfer from English. English input also appeared to influence lexical choices; high frequency expressions in English were (inappropriately) directly translated into Spanish (e.g., #tener un buen tiempo 'lit: have a good time', rather than pasar un

3 Rather than a Minimalist view (Chomsky 1995), where vocabulary items from the lexicon are merged into structures, Silva-Corvalán adheres to a view of syntax where the formation of syntactic structures occurs on a separate level from the insertion of lexical items. Thus, she analyzes the data from a perspective where structure is built independently of pragmatics, which in turn determines whether a particular lexical item can be inserted into the structure or not.

4 Silva-Corvalán refers to these speakers as bilinguals, but they are akin to the group I call HLS.

5 This is the term that Silva-Corvalán uses; I assume it refers to the formal variety of Spanish that is taught to L2 learners in language classes.

6 Silva-Corvalán notes that this is also allowed in colloquial language in certain dialects, which in turn may also have been influenced by contact with indigenous languages with optional complementizers (p. 217).

10 buen rato 'lit: pass a good moment'). Another example of transfer was attested in the bilinguals’ use of null subjects in Spanish. In the (Spanish) baseline, null subjects are used pragmatically to represent continuing topics, but in English, subjects are generally overtly expressed, irrespective of their pragmatic role. 7 Thus, direct influence from English could have led to a decrease in the HLS’ use of null subjects, yet the opposite occurred. Spanish-English bilinguals used even more null subjects than what would have been pragmatically acceptable in the baseline, despite the unacceptability of null subjects in English. Silva-Corvalán (2008) argues that this is an example of indirect transfer. Even though the bilinguals did not converge on a grammar that was more “English-like” by making the overt expression of subjects obligatory, their grammars were L2- like in that the pragmatic distinction in the null or overt realization of subjects in the bilinguals’ grammars was lost due to the lack of this distinction in the L2 (see also Montrul 2004). Another area of transfer was in the position of overt subjects. Discourse-new subjects need to be in post- verbal position, but the bilinguals had higher rates of discourse-new subjects in preverbal position. Since English subjects are always in preverbal position, this could be viewed as the direct transfer of syntax, but Silva-Corvalán argues that this is also an example of indirect transfer, where the pragmatic constraint of new information being placed postverbally is affected, due to the lack of this constraint in English.

However, as has been noted in several studies (Montrul 2008, among others), it may be premature to conclude that the L2 is the primary source of influence, since most studies on HL acquisition have only compared various HLs with English. English is a relatively conservative language with respect to many properties (e.g., only genitive Case is obligatory and marked regularly; otherwise Case marking is only visible on pronouns), thus it can sometimes be difficult to ascertain whether non-target-like properties of the HL grammar are due to influence from the L2 or due to a preference for unmarked options within the parameters of UG. It is also possible for the baseline (i.e., HLS’ input) to influence HLS’ grammars in a way that superficially appears to look like incomplete acquisition of the HL due to influence from the L2.

7 The only exception is with diary-style writing, where the subject (i.e., “I”) is frequently dropped.

11

2.3.2 HLS’ Grammar as Evidence of Insufficient Input

Polinsky and Kagan (2007) and Polinsky (2008) note the importance of comparing the grammar of HLS to its proper baseline – the actual input language of HLS, dialectal or otherwise, and not from the (often far-removed) language spoken as the standard in the country where the HL is spoken by monolingual speakers (i.e., the country the speaker and/or the speakers’ parents immigrated from; what Polinsky refers to as “the metropoly”). For example, it has been noted that the input that Spanish HLS receive is often a regional dialect which is different from the standard (i.e., that which is taught in schools in the metropoly). Thus, in the absence of formal instruction in the language, another source of variation in HLS’ grammar could be their input, which can consist of more variation than the standard due to dialectal differences (Sanchez 2004, Rothman 2007). There are two ways that variation can affect the input: the quantity of input provided to HLS may be different from that provided to monolinguals (Barrena, Ezeizabarrena, & Garcia 2008), as well as the quality of input provided (Rothman 2007).

Paradis and Navarro (2003) predicted that the Spanish-English bilingual child Manuela8 in the CHILDES database (www.childes..cmu.edu) would overuse overt subjects in Spanish when compared to two age-matched monolingual Spanish children, since subjects are generally overt in English, whereas in Spanish, a null-subject language, the use of overt and null subjects is pragmatically constrained, as mentioned above. This was maintained in their data, although an analysis of Manuela’s parental input revealed that her parents produced a higher percentage of overt subjects when compared to the parents of the other children. Thus, Manuela may have simply been matching the input that she was receiving from her parents (i.e., the baseline language). Paradis and Navarro suggested that this may not have necessarily been due to the influence of English, but perhaps also due to the Panamanian Spanish dialect (with Cuban influences) that was spoken by her parents, since it appears that, unlike standard Spanish, null subjects are disappearing from Caribbean dialects. Thus, in this case, the results of influence from Panamaian Spanish would be the same as the results of influence from English. However, what perhaps should also have been emphasized by Paradis and Navarro is that although the distribution of pronouns and lexical subjects amongst Manuela’s father’s overt subjects did not differ from that of the Spanish monolingual children’ parents (i.e., more lexical subjects

8 Manuela lived in the UK.

12 appeared overtly than pronouns), Manuela’s mother’s data differed from that of all other parents by having equal proportions of overt lexical subjects and pronouns. This is significant because Manuela’s mother was actually a native speaker of English and an L2 learner of Spanish; Manuela’s mother’s data appears to have been influenced by her L1 English, since subjects need to be overtly expressed in English, regardless of whether they are lexical subjects or pronouns. Thus, Manuela would have received input in English, English-influenced Spanish, and Cuban Spanish, which means that her HL grammar was likely influenced by transfer from both her L2 and by her baseline (part of which was also influenced by the same L2 and part of which consisted of data from a dialect which differs from the standard spoken in Spain), which would all have influenced Manuela to produce a higher percentage of overt subjects in her HL (Spanish) than monolingual speakers of Spanish. Paradis and Navarro note that the percentage of null subjects used in Caribbean dialects of Spanish seem to be an example of language change in progress, it is impossible to tell what played the larger role in influencing Manuela’s HL grammar, but it should be clear that Manuela did not receive the same quality of input that Spanish-speaking monolingual children would have received.

S. Park (2010) conducted an interesting study on the influence of parental input on Korean- English simultaneous bilinguals’ ability to produce the three-way voicing contrast in Korean stops. In Korean, voice onset time (VOT) and the pitch (F0) of the following vowel are the phonetic cues used to distinguish between fortis, lenis, and aspirated stops. For example, VOT differentiates the stops as follows: aspirated > lenis > fortis, where aspirated stops have the longest VOT and fortis stops the shortest. The F0 of the following vowel also differentiates the stops, but in the following order: aspirated > fortis > lenis, where aspirated stops cause the highest F0 and lenis stops the lowest. In contrast, in English, there is only a 2-way voicing contrast between stops and VOT is the main cue used to distinguish between voiceless and voiced stops.9 S. Park tested five children who were ten years old, four of whom had been born in Canada and one who had immigrated to Canada when she was three years old, and compared their production of Korean stops to that of their parents. She found that all of the children were able to use appropriate phonetic cues to differentiate each of the stops, and that many of the

9 There appears to be a drop in F0 with voiced stops even in English (see Hombert, Ohala, & Ewan (1979) for details).

13 differences between the preferences in the children’s use of various cues often matched their parents’ preferences for the use of various cues.

2.3.3 HLS’ Grammar as Evidence of Incomplete Acquisition

“Incomplete acquisition” is often included in descriptions of HLS, given that most HLS switch their language of dominance at a young age, resulting in less use of and exposure to the HL (e.g., Montrul 2008, Benmamoun et al. 2010, among others). Incomplete acquisition can refer to the process of arrested development, where development ceases in a particular language, or it can refer to the end result – a grammar with non-target-like characteristics (Montrul 2008). Montrul (2008) notes that incomplete acquisition and language loss are not mutually exclusive, since it is possible for a single grammar to exhibit evidence of incompletely acquiried properties as well as attrited properties.

Traditionally, especially from the pedagogical perspective, studies on HLS have compared the language of HLS to that of monolingual speakers of the language in order to ascertain deficiencies in the HLS’ grammar which need to be fixed through proper language education, as per the Error Analysis Approach (Corder 1967)10. Research on students enrolling in language courses in university has shown that the needs of HLS differ from those of L2 learners. With respect to phonology, Oh et al. (2003) found that early exposure to and production of Korean made a significant difference in determining whether speakers were able to produce and perceive phonemes particular to Korean (e.g., the three-way—unaspirated/plain, tense, and aspirated— voicing distinction in stops) when they were re-learning their HL in university. Childhood hearers and childhood speakers of Korean were able to distinguish phonemes in Korean comparably to native speakers11, whereas adult L2 non-heritage learners of Korean had significantly greater difficulty. In terms of production, only childhood speakers of Korean reliably produced the three-way contrast. Since children are generally able to perceive phonetic differences before they are able to produce them, we can conclude that the childhood hearers exhibited incomplete acquisition. This could also be interpreted as transfer, since English only

10 This is also described as the “target deviation perspective” by Klein (1998: 535).

11 This is Oh et al.'s term, but these speakers were selected from the same university as the other participants and thus presumably would be better classified as Korean-dominant bilinguals.

14 has a two-way voicing distinction (i.e., aspirated and unaspirated) in stops. Thus, we can assume that influence of the L2 had led to incomplete acquisition in the HLS.

2.3.4 HLS’ Grammar as Evidence of Language Loss

Language loss, or attrition, has traditionally been described in terms of the difficulties that speakers have with their L1 upon immigrating to a country where the dominant/societal language is different from the speakers’ L1. Benmamoun et al. (2010) argue that attrition is not an appropriate description of what happens with HLS, since we can only talk about the loss of features which were fully acquired at a certain point in time, although they suggest that only properties which have been found to be fully acquired by 4-5 year olds should be considered to have undergone attrition if it is found that HLS have difficulty with those particular properties. Due to the lack of extensive longitudinal studies on HLS, and given that HLS generally follow a delayed timeline of acquisition (e.g., Gathercole (2007) and references therein) , we often cannot say for certain whether a feature had been completely acquired by a particular HLS or not.

Kaufman and Aronoff’s (1991) roughly two-year longitudinal study showed four stages of attrition in a HLS of Hebrew who had immigrated to the US at 2;6 (i.e., who was a simultaneous bilingual). The child initially (2;9–3;1) only used code switching, substituting L2 nouns and verbs while using her L1, and then entered a stage (3;1–3;2) where she was completely bilingual and fluent in both languages in terms of the age-appropriate levels attested in monolinguals. From the third stage (3;2–3;5), the child began inserting non-target-like L1 verbal forms into L2 structures – behaviour which is not evidenced in the language of monolingual children. At the fourth stage (3;5–4;6), the child overgeneralized a single verbal template of Hebrew to all contexts, showing evidence of regularization and loss. This particular child attended nursery school in English from 2;7 and was given intensive exposure to English at school from 2;8 (while continuing to use Hebrew at home), but began to avoid speaking in Hebrew during the third stage of her language attrition. Thus, within six months of exposure to English, she already showed evidence of attrition in the verbal domain of Hebrew.

In lieu of a longitudinal study, Polinsky (2008) compared utterances produced spontaneously by a male child (9;7) and a male adult (23 years old) HLS of Russian who had similar language backgrounds with respect to their exposure to and use of Russian. Both HLS showed evidence of a simplified Case and aspectual system in Russian. In oral narratives, the HLS tended to overuse

15 a default Case (i.e., nominative) and reassigned the function of the accusative Case to mark indirect objects, which are normally marked with dative Case. Aspectual marking appeared to be lexically determined in the HLS' grammar, and rather than using the appropriate aspectual affixes, analytic means (e.g., adding a light verb) were used to express aspect. An important finding was that the child's language was significantly more target-like than that of the adult, despite both of them being HLS of Russian in the US who had used Russian until they entered kindergarten, after which they had switched to using English as their primary language. The adult's speech was slower and also contained more instances of code switching, which shows that the adult had more difficulty overall with speaking Russian. This would support a view where language attrition had subsequently occurred in the grammar of the adult due to his extended exposure to English and lack of input in Russian. Polinsky suggests the difference in the two grammars, with further research, could prove to be on par with the restructuring that occurs in the construction of creole grammars (see also Polinsky 1996), with the possible additional influence of transfer from English, since English has a much simpler Case system than Russian. The fact that both the child and the adult showed evidence of simplification in their grammars shows that, at least for the two speakers involved in the study, Case and aspect were most likely never quite fully acquired. Thus, these two HLS would be considered as examples of L2 influence leading to incomplete acquisition and attrition. However, although the HLS had similar ages of arrival, other information on their language backgrounds is needed to see whether these two speakers were actually comparable in terms of use of and exposure to the HL.

Benmamoun et al. (2010) discuss the Regression Hypothesis, which is the most controversial theory for HL acquisition. It was proposed by Jakobson (1941) to account for language loss in the L1 of adults. The theory proposes that mastery of features, principles, and constructions is “lost” in the reverse order of which the features were acquired, such that constructions acquired last will be the first to go. Keijzer (2007) found partial support for the Regression Hypothesis in his study on attrition in L1 Dutch-speaking adults who had immigrated to Canada after the age of 15 and who had lived in Canada for at least 20 years. His speakers were 40-66 years old and were age-matched with individuals in the Netherlands who had similar language backgrounds. The former group constituted the “attriters”, the latter group was the control group, and he included a group of “acquirers” who were 13-14 year olds. He found that the attriters performed like the acquirers on several of the features tested, and both groups generally performed

16 differently from the control group. This is strange, given that the features are described in the literature as being fully acquired by monolingual children by the age of five. In Keijzer’s (2007) study, the results correlated with the participants’ level of education, such that participants with postsecondary education were more likely to perform better than those with only elementary or secondary education. Keijzer (2007) and Montrul (2008) admit that it is possible that full acquisition only occurs upon schooling and with the help of literacy. Thus, it is important to recognize that language “acquisition” does not in fact end in childhood, but that there are several constructions where only formal instruction would lead to “complete” acquisition.

Some linguists focus on the development of the HL grammar as a situation where two languages are in contact (i.e., the HL and the L2).12 Silva-Corvalán (1991) describes language loss as the mirror image of creolization (p. 164). Polinsky and Kagan (2007) adopt the terminology used for creoles (e.g., basilect, mesolect, and acrolect) to classify HLS according to their level of proficiency and the degree of similarity between their language and that of their input (i.e., the target language; e.g., Korean). Creoles are developed by speakers with minimal input, and similarly, the language of HLS is also developed under circumstances where speakers have a relatively lower degree of exposure to their HL (than to their L2, when compared to monolinguals who receive all of their input in the baseline). Just as “basilect” is used to describe the form of a creole differing the most from the lexifying language13, “basilect” is used by Polinsky and Kagan to identify the variety of the HL produced by very low proficiency speakers whose grammars differ greatly from that of the target language. Acrolects are creole varieties which are the most similar to the lexifying language (and thus correspondingly can refer to the language of high proficiency HLS) and mesolects are an intermediate variety (between basilects and acrolects) and can be used to describe the language of intermediate HLS. However, the use of the terms “acrolect”, “mesolect”, and “basilect” suggests that the superstrate language should be the baseline and the substrate language should be the L2, which conflicts with the actual status of the two languages, since the L2 is actually the “language of prestige” since it is the language of the majority. Creolization is a creative process whereby different pidgins develop a

12 Technically, this could depend on whether HLS have access to a community of speakers or not, but immigrant parents commonly speak to their children in the HL, regardless of whether the children respond in the HL or not, so it is likely that input in the HL and input in the L2 is present.

13 This is the language considered to contribute the most to the creole’s lexicon and is also sometimes referred to as the superstrate language.

17 more regularized and complex grammar with the second generation of speakers, despite minimal input. With HLS, it is assumed that insufficient input leads to a breakdown in the HL grammar. However, it is not the case that attrition in HL grammars leads to pidgin-like grammars. Thus, it is theoretically incongruous to borrow terms from creolization to describe HL grammars.

2.4 Age vs. Input

The above studies show that clear differences exist between HL and monolingual grammars, resulting in what looks like incomplete acquisition and/or attrition. Studies on L2 acquisition often suggest that the difference between L2 interlanguage grammars and L1 grammars is related to the age of acquisition of the L2 learner. This is an extension of the Critical Period Hypothesis, which was originally proposed to account for the different language outcomes in (child) L1 acquisition—normally developing children are able to become fluent native speakers of a language, but children who do not receive linguistic input before puberty may never become fully competent in a language14. The assumption is that the neural plasticity of the brain decreases, around the age of puberty as specialization occurs in the brain. Thus, without enough exposure to linguistic input before the “critical period”, children are expected to not be able to master a language. Researchers (Bley-Vroman 1989, 1990; Scovel 1988) have proposed a critical period for L2 acquisition as well, given that not all L2 learners are successful at acquiring target- like grammars. This view has been labeled the Fundamental Difference Hypothesis (Bley- Vroman 1989, 1990), and attributes the difference between successful first language acquisition and the often “unsuccessful” adult second language acquisition to the loss of brain plasticity that occurs after the critical period once the brain becomes “specialized” in a particular language. However, this is an insufficient explanation for HL acquisition, since HLS (often) learn their HL before the age of five, which should be well before the plasticity of the brain is affected, yet aspects of grammar that are fully acquired by monolinguals before the age of 5 are not always target-like in HLS’ grammar. Montrul (2008) argues that age of acquisition is still relevant for HL acquisition, but that it is the age of exposure to the L2 (or societal language) that determines the likelihood of attrition or incomplete acquisition to occur in HL grammars, such that the

14 Fromkin, Krashen, Curtiss, Rigler, & Rigler (1974) provide an update on the case of Genie, a girl who was abused and deprived of language until the age of thirteen. Although they note significant improvement in her language ability, Genie’s language comprehension was at about the same level as that of a two- or two-and-a-half- year-old child.

18 earlier the age of exposure the greater the likelihood of attrition or incomplete acquisition. This is in contrast with others who assume that the degree of exposure to the baseline and the degree of use of the HL plays the biggest role in determining whether attrition or incomplete acquisition will occur. Both views will be presented below.

2.4.1 Age

Montrul (2008) draws on studies on simultaneous bilingualism, early and late child bilingualism, adult second language acquisition, L1 attrition, and L2 fossilization to argue that the research all points to age effects being the cause of different types of outcomes in bilingualism. Specifically, with respect to HL acquisition, she proposes that if L1 (HL) attrition or incomplete acquisition occurs, it will be more severe in simultaneous bilinguals than sequential bilinguals. Although she reiterates several times that amount and type of exposure, along with other factors, are also important, she assumes that the age of onset of bilingualism is the main determinant of the outcome of bilingualism. Based on studies of unbalanced bilinguals (i.e., individuals who are more dominant in one language over the other), she argues that the earlier the age of onset of exposure to an L2, the greater the HLS’ proficiency in the L2, but also the greater the degree of incomplete acquisition in the HL. Montrul suggests that HLS will only have relatively high proficiency in their HL if they have (high) exposure to the baseline input15 at least until the end of the critical period. Age is the most important variable, because early exposure to the L2 causes the eventual breakdown and gaps in the core competence of simultaneous HLS, whereas attrition in the HL (L1) of late (adult) bilinguals only affects speakers’ performance, rather than their competence. This suggests that low performance on particular types of experimental tasks can sometimes merely be evidence of performance errors whereas other times it can be interpreted as severe errors in competence. However, we can never test competence directly; we can only ever make conclusions about competence via speakers’ performance.

Yeni-Komshian (2009) found that age of arrival was the best predictor in determining how native-like HLS’ Korean pronunciation was. She found that the older the participants were when they immigrated to the States, the more native-like they sounded in their HL, and the less native- like they sounded in their L2. However, HLS who emigrated between the ages of 12 and 13 were

15 Mere exposure to the baseline input seems to be insufficient. For Montrul, the assumption appears to be high exposure by virtue of living in the metropoly.

19

fairly balanced in their pronunciation of both languages, even though they did not sound completely native-like in either language.

A recent study by T. Lee (2011) is relevant to the present study, since it examines Korean HLS’ acquisition of the unergative-unaccusative distinction in verbs by using classifier constructions. Due to the lack of classifiers in English, classifiers are typically difficult for Korean HLS to acquire. 16 T. Lee examined how the possibility, or impossibility, of adverb modification interacts with the possibility of quantifier float in Korean classifier constructions depending on the type of verb. The regular order for classifier constructions in Korean is N Num17 Cl (as in (3), with the numeral and classifier following the head noun, but with certain types of arguments, quantifier float is possible such that the Num and Cl are not adjacent to the head N. For example, in (4) (T. Lee’s 6b), the object noun chayk ‘book’ is separated from the Num and Cl sey kwen

‘three Cl (book)’ by the intervening adverbial phrase tosekwan-eyse ‘library-at’, resulting in the linear order of N Adv Num Cl.

(3) Haksayng-i tosekwan-eyse chayk sey-kwen-ul ilk-ess-ta. student-NOM library-at book three-CL (book)-ACC read-PST-DECL ‘A student read three books at the library.’

(4) Haksayng-i chayk-ul tosekwan-eyse sey-kwen ilk-ess-ta. student-NOM book-ACC library-at three-CL (book) read-PST-DECL ‘A student read three books at the library.’

For transitive verbs such as ilk- ‘read’, this type of quantifier float with the direct object (i.e., an internal argument) is grammatical. However, when a Num and Cl are associated with the subject (i.e., the external argument) of a transitive verb (5), quantifier float is unacceptable (6) (T. Lee’s 6a).

(5) Haksayng-i sey-myeng tosekwan-eyse chayk-ul ilk-ess-ta. student-NOM three-CL (human) library-at book-ACC read-PST-DECL ‘Three students read books at the library.’

16 Different classifiers are used for different types of nouns. For example, the classifier myeng is used with human nouns, the generic classifier kay is used with inanimate nouns, the classifier kulwu is used for counting trees, and tay is used when counting machines such as cars or computers. There are over 50 classifiers in Korean.

17 Num stands for Numeral here.

20

(6) ?/*Haksayng-i tosekwan-eyse sey-myeng chayk-ul ilk-ess-ta. student-NOM library-at three-CL (human) book-ACC read-PST-DECL ‘Three students read books at the library.’

Intransitive verbs only have one argument, but they are often divided into two groups: unaccusative and unergative verbs. Unaccusative verbs such as tochakha- ‘arrive’ are assumed to have subjects that are base-generated as internal arguments of the verb (i.e., as complements of V), whereas unergative verbs such as wul- ‘cry’ are assumed to have subjects that are external arguments base-generated outside of the VP. Interestingly, quantifier float is allowed with the

internal argument (e.g., sonnim-i ‘guest-NOM’ in (7), T. Lee’s 7b) of unaccusative verbs such as tochak-ha- ‘arrive’, whereas it is not allowed with the external argument (e.g., yeca-ka ‘woman-

NOM’ in (8), T. Lee’s 7a) of unergative verbs such as wul- cry’. Thus, the availability of quantifier float appears to be related to whether the Num and Cl are associated with an internal or external argument.

(7) Sonnim-i ce hoteyl-ey ney-myeng tochak-hay-ss-ta. guest-NOM that hotel-at four-CL (human) arrive-do-PST-DECL ‘Four guests arrived at that hotel.’

(8) ?/*Yeca-ka ce pang-eyse sey-myeng wul-ess-ta. woman-NOM that room-at three-CL (human) cry-PST-DECL ‘Three women cried in that room.’

Using a written acceptability judgment task, where sentences with quantifier float were presented without context and participants were asked to judge sentences based on a scale of one (“fully unacceptable”) to four (“fully acceptable”),18 T. Lee tested whether HLS were sensitive to differences in the acceptability of quantifier float with various types of arguments. There were 16 distractor sentences, which were all grammatical simple transitive sentences (presumably without classifiers). Thirteen US-born HLS and fourteen Korean-born HLS who had immigrated to the US between the ages of four and fourteen and were taking intermediate or advanced university-level Korean courses were tested, and twelve native speakers of Korean, between their mid-twenties and mid-forties, who had completed college in Korea, formed the control group.

18 The control group was asked to rate the sentences on a five-point scale (-2 to 2, with -2 being “fully unacceptable” and 2 being “fully acceptable”), since their results were actually obtained via a larger study. The scores from the control group were converted to a four-point scale for T. Lee’s (2011) study, but the algorithm for converting these scores is not given.

21

Only participants who accepted at least 80% of the distractors as being grammatical and who were 80% consistent in accepting sentences with quantifier float involving transitive objects and 80% consistent in rejecting sentences with quantifier float with transitive subjects were considered to be “consistent” in their judgments. Thus, although all participants rated all of the sentences, only those who were (at least 80%) consistent in their judgments on sentences which did not involve intransitive verbs were included in the analysis for sentences with quantifier float involving either unaccusative or unergative verbs. T. Lee found that HLS who were born in the US were not sensitive to the distinction between unaccusative and unergative verbs; amongst the speakers who were able to successfully judge quantifier float as being acceptable with transitive objects and unacceptable with transitive subjects, the US-born HLS rated sentences with quantifier float with both types of intransitive verbs as being equally ungrammatical (mean score: 2.1 (0.89) for unaccusatives and 1.79 (0.84) for unergatives). The Korean-born HLS were able to notice a difference in the acceptability of quantifier float with the internal argument of unaccusatives (mean score: 1.67 (0.82)) and the external argument of unergatives (mean score: 1.11 (0.32)). Although the difference was statistically significant in a two-way ANOVA, these speakers found quantifier float to be only slightly more acceptable with the internal argument of unaccusatives than with unergatives, but not nearly as acceptable as with transitive objects. (mean score: 3.30 (0.55)). T. Lee interprets these results as an indication that HLS have incomplete knowledge of quantifier float and suggests that this is tied to the age of onset of exposure to English.

However, there are some reasons to question this conclusion. One reason is the complexity of the structures that were tested. HLS are known to have difficulty with classifiers in general, even without quantifier float and even selecting the appropriate classifier (see fn. 15) can be a challenge. For example, the classifier myeng is only appropriate with nouns referring to humans, but HLS may use it with nouns referring to animals as well, or they may also overgeneralize the use of the generic classifier kay, which is only appropriate with inanimate nouns, to animate nouns, such as animals, as well. Some lower proficiency HLS may not even use classifiers at all when counting nouns (Shin & Milroy 1999, see also Section 3.3.2 for discussion). HLS are also known to have difficulty with scrambling, where elements of a sentence are moved, resulting in an order other than the canonical SOV word order. Both scrambling and quantifier float are fairly

22 uncommon in colloquial language. Thus, it would not be surprising for HLS to have difficulty judging the acceptability of sentences with quantifier float.

T. Lee argued that since the HLS were able to distinguish the difference in acceptability of quantifier float with transitive subjects and transitive objects, it would be possible to test the sensitivity of HLS to the difference in acceptability of quantifier float with the arguments of unaccusative and unergative verbs in order to see whether they were sensitive to unaccusative and unergative verbs in general. However, low ratings were given to the majority of sentences involving quantifier float with unaccusative verbs, and there was significant variation depending on the particular unaccusative verb tested. For example, four of the nine Korean-born HLS judged sentences with quantifier float with the verb thayena- ‘be born’ to be acceptable, yet only one HLS was consistent in judging quantifier float with the rest of the unaccusative verbs to also be acceptable. It appears that these sentences are more complicated than expected. T. Lee’s (6b) and (6a) are repeated here as (9) and (10). In (9), the classifier kwen ‘CL (book)’ can only be associated with the inanimate transitive object chayk ‘book’; the classifier cannot be misinterpreted as being associated with the human subject haksayng ‘student’. However, in (10), the human classifier myeng can only be associated with the human subject haksayng ‘student’, yet the sentence is ungrammatical with quantifier float. The reason may be that, in Korean, topic drop is quite common in colloquial language, and sentences are generally arranged to have (or drop) given information before introducing new information, unless (sentence-initial) focus is desired. Information structure appears to also be relevant in terms of which types of Case marking are allowed or disallowed in a sentence. Recall that HLS had less difficulty in accepting quantifier float with sentences involving the unaccusative verb thayena- ‘be born’. There is a limited set of things that can ‘be born’; if the test sentence involved ‘babies’, speakers would be more likely to accept ‘babies’ to be “given” information, with the Num and Cl as being focused or “new” information. In (10), the nominative Case marking on the subject noun haksayng ‘student’ seems to suggest that the noun is focused; its sentence-initial position also seems to suggest that it is being focused. Since sey-myeng ‘three-CL (human)’ is introduced after the intervening adverbial phrase tosekwan-eyse ‘library-at’, it seems as though the classifier is given (old) information, which leads to confusion in processing the sentence, since there is no reading where it would be appropriate to introduce a noun associated with a classifier as “new” or “focused” with the actual classifier being “old” or “given” information. The sentence can,

23

however, be felicitous if it is clear that haksayng ‘student’ is the topic, perhaps by adding -tul

cwung-eyse ‘-PL amongst-at’ to attain the interpretation ‘amongst the students’ (with the assumption that there is a particular group of students of which both the speaker and hearer are

aware) and it is made clear that sey-myeng ‘three-CL (human)’ is new information by adding the

nominative Case marker -i ‘NOM’, as in (11).

(9) Haksayng-i chayk-ul tosekwan-eyse sey-kwen ilk-ess-ta. student-NOM book-ACC library-at three-CL (book) read-PST-DECL ‘A student read three books at the library.’

(10) ?/* Haksayng-i tosekwan-eyse sey-myeng chayk-ul ilk-ess-ta. student-NOM library-at three-CL (human) book-ACC read-PST-DECL ‘Three students read books at the library.’

(11) Haksayng -tul cwung-eyse tosekwan-eyse sey-myeng-i chayk-ul ilk-ess-ta. student-PL amongst-at library-at three-CL (human)-NOM book-ACC read-pst-DECL ‘Three students read books at the library.’

Similarly, T. Lee’s (7b) involving an unaccusative verb improves in acceptability if the following changes are made, as in (12), and T. Lee’s (7a) involving the unergative verb can become acceptable, as in (13) below.

(12) Sonnim-tul cwung-eyse ce hoteyl-ey ney-myeng-i tochak-hay-ss-ta. guest-PL amongst-at that hotel-at four-CL (human)-NOM arrive-do-PST-DECL ‘Four guests arrived at that hotel.’

(13) Yeca -tul cwung-eyse ce pang-eyse sey-myeng-i wul-ess-ta. woman-PL amongst-at that room-at three-CL (human)-NOM cry-PST-DECL ‘Three women cried in that room.’

Thus, the acceptability of quantifier float is improved or worsened by the presence or absence of (different types of) Case marking and is also dependent on how well the sentence follows the expected information structure, rather than being solely dependent on the type of argument (internal vs. external) or the type of verb (e.g., unaccusative or unergative). Given that two members of the control group had to be excluded from the analysis based on their failure to meet the threshold of consistency in their judgments on quantifier float with transitive verbs, it could be the case that at the very least, more context needs to be provided in order to determine the acceptability of a sentence involving quantifier float.

24

In terms of the test design, it could be possible that the distractor sentences (grammatical transitive sentences) were so simple and straightforwardly acceptable that all the test items seemed unacceptable in comparison. It would be helpful to know how the participants would have judged the acceptability of simple sentences with a Num and Cl, but no quantifier float. However, having these types of sentences as distractors would likely have drawn participants’ attention straight to the distinction between the placement of the Num and Cl with respect to the noun they were associated with. A more useful distractor could have been to have simple sentences with OSV word order. Another possibility for the control group’s inconsistent behaviour is that quantifier float may not be as acceptable as it is expected to be according to the literature. This is particularly possible given that the modality of the experiment was written, rather than oral. As with instances of scrambling, specific intonation may be necessary in order to facilitate appropriate interpretations of the sentence.

Some of the US-born HLS had to be excluded from the test group based on failure to accept 80% of the distractor sentences. Since the test sentences all involved quantifier float, simple transitive sentences without any quantifier float should have been fully acceptable in comparison. This suggests that the US-born HLS may not have been as “fluent” as they were described to be by T. Lee. Although the specific ages of all of the HLS were not provided, T. Lee notes that none of the Korean-born speakers who had immigrated to the US before the age of eight were consistent in their judgments, and the only HLS who consistently rejected sentences with quantifier float with unergatives and accepted sentences with quantifier float with unaccusatives had moved to the US at the age of 14. It appears to be significant that the only heritage language speaker who was able to successfully (consistently) reject quantifier float with unergatives and accept quantifier float with unaccusatives was someone who had immigrated to the US as a high school student. If the HLS were registered in university-level courses at the time of testing, they were likely between the ages of 17 and 21, which means that this particular HLS would have spent only a few years in the US. It may be the case that this sensitivity to the restrictions on quantifier float does not arise until high school and may in fact be an ability that is taught explicitly through formal instruction, as T. Lee appears to allude to in fn. 12, where she discusses Rothman’s (2007) observation that inflected infinitives are only learned through formal instruction in Brazilian Portuguese. Thus, it would be useful to conduct a similar study with a more robust number of monolingual native speakers of Korean in Korea. However, it is interesting that two of the US-

25 born HLS were actually more successful than some of the Korean-born HLS in identifying the difference in acceptability of quantifier float with unaccusatives vs. unergatives. Since the US- born HLS were presumably exposed to English from birth, their performance seems to suggest that age of onset of exposure to English cannot account for this. All HLS were able to read, write, and speak Korean before enrolling in their university-level Korean course. In general, speakers had to have immigrated to the US at the age of eight or older in order to distinguish between the acceptability of quantifier float with transitive objects as compared to transitive subjects, except one speaker who had moved at the age of four was able to make this distinction. T. Lee notes that the difference between this speaker and the other speakers who had immigrated before the age of eight was that she reported speaking Korean “all the time” (p. 165). This is a strong indication that language use (i.e., output) and input are important in determining proficiency in the HL; thus, it is debateable as to whether this study actually supports an analysis of age of onset of exposure to English as being responsible for HLS’ ability in their HL.

2.4.2 Input (and Interactions)

Montrul restricted her definition of HLS to those who are unbalanced bilinguals and significantly weaker in their HL than in their L2, and who have “limited” or “reduced” input. She argued that due to early exposure to the L2, simultaneous bilinguals would be more vulnerable to incomplete acquisition than sequential bilinguals who develop their L1 fully before being exposed to another language. However, it has been found that it is likely that different critical/sensitive periods exist for different domains of grammar (e.g., age six for phonology and age fifteen for morphology and syntax (see Long 1990 and references therein)), so it is unclear whether this is expected to be true of all sequential bilinguals or only those who are beyond the oldest age for any critical/sensitive period. Given that individuals exist who are not significantly weaker in their HL, or who are at least highly proficient in the HL, this suggests that age is not the ultimate determinant of proficiency. Polinsky (2006), for example, found no statistical correlation between HLS’ age of acquisition of the L2 and their proficiency in the HL, and Montrul (2008) also acknowledged that certain aspects of grammar are sometimes only learned through formal instruction.

Thus, rather than age of arrival or age of first exposure to the L2, it appears that other aspects of HLS’ language background may play a bigger role in influencing the outcome of HL acquisition.

26

Some of these include: the language that HLS use in the home, whether the HLS have received any formal instruction on the HL, the birth order of the HLS, how extensive their exposure to the HL is, whether they are literate in the HL (Benmamoun et al. 2010 and references therein), and whether they have access to a community of speakers. Generally speaking, the HLS who become more proficient in their HL than the average HLS are those who have had one or more of the following experiences: they use the HL with their parents at home, have taken university-level language courses in the HL, 19 are the firstborn child, 20 seek out various forms of media in the HL and engage in leisure reading (McQuillan 1996), and have close friends who speak the HL and are interested in the culture (Yi 2008).

Jia’s (1998) cross-sectional study found that age of arrival (AoA) was able to account for a large percentage of the variance in the performance of L1 Chinese-speaking children on oral grammaticality judgment tasks in their L1 Mandarin Chinese and L2 English after at least five years of exposure to English in the United States (US).21 In addition, HLS’ self-rated proficiency in their HL was significantly negatively correlated with their self-rated proficiency in the L2. Thus, speakers who arrived in the US at an earlier age tended to have higher proficiency in their L2 and speakers with later ages of arrival tended to have lower proficiency in their L2, and HLS who rated themselves as having higher proficiency in their HL tended to also rate themselves as having lower proficiency in their L2, and vice-versa. In a longitudinal study, Jia found that younger children (i.e., those under the age of 9) immersed themselves in the English language and American culture and quickly became more comfortable with their L2, which subsequently led to attrition in their HL. In contrast, older children (i.e., those over the age of 9) preferred to immerse themselves in Chinese culture and were more likely to preserve their HL, often at the expense of development in their L2 English. The 9-year-old children in Jia’s study were equally

19 J. S. Lee (2002, 2008), among others, has shown that HL schools (i.e, Saturday or Sunday schools) are typically ineffective in helping HLS improve to native-like proficiency in their HL. One possible reason why university-level language courses are more effective could also be related to the level of motivation that HLS who enroll in these classes have to learn the HL. HLS in Saturday language schools, for example, generally have low motivation to learn the HL and many resent having to attend classes on the weekend. Students who enroll in university language courses, on the other hand, pay tuition and have specifically chosen to attend that course (over many other courses available to them).

20 Polinsky (2006), for example, found that HLS generally only spoke English with their siblings; thus, HLS who had siblings generally showed greater language erosion.

21 Jia (1998) described her participants as L2 learners; I will refer to them as HLS and refer to their L1 as their HL for ease of exposition.

27 comfortable with either language. Thus, children who arrived in the US at an earlier age tended to switch their language of dominance from their HL to their L2, whereas the older children preferred to maintain their HL as their dominant language. Jia labeled this the Dominant Language Switch and Maintenance Hypothesis. Thus, rather than age of exposure, or age of onset of bilingualism, the significant factor appeared to be what types of language choices the speakers were making and the degree to which they created opportunities to use either language. It is thus possible that speakers who make conscious decisions to use and maintain both languages could end up being balanced bilinguals. Jia’s study also provides evidence of the effect of formal instruction. Older children (aged 15 and 16 years old) who had received 3-4 years of formal instruction in English in China and Taiwan initially performed better on the oral English grammaticality judgment task and the oral Mandarin-to-English translation task than the younger children. However, they were outperformed by HLS who had arrived in the US at an earlier age after a year of exposure to English in the US. Jia notes that part of the reason is that younger children are still developing their lexicon and some of the concepts and ideas they learn as they grow are learned only in their L2 (i.e., English) and not in their HL. Jia concluded that the greater initial HL proficiency interfered with the older children’s acquisition of English, since the speakers not only had to learn the L2, but also had to maintain their HL. Jia’s study, however, did not include simultaneous bilinguals. The situation is slightly different for simultaneous bilinguals, since they are exposed to both languages before the age of three and are generally not afforded a choice in their language environments.22

Jia’s study highlights the importance of exploring the different ways that HLS may actively create their language environments, inasmuch as they have control over them. This is important, since children, in particular, are often viewed as passive learners who are exposed to particular types of input or not. This is sometimes unintentionally supported by findings that the role of the mother is important in the development of HLS’ HL. For example, Kondo (1998) and Kondo- Brown (2001) found that mothers played an important role in creating opportunities for HLS to use and receive exposure to and instruction in the HL. For example, the mothers of HLS

22 The exact nature of the language input that very young children receive depends on the language attitudes and proficiencies of their parents. Immigrant parents who value assimilation into the environment could increase the child's exposure to English, whereas those who value preservation of cultural heritage could increase the child's exposure to the HL (Jeon 2008). In other types of bilingualism, some parents employ the One Parent One Language method. However, Montrul (2008) argues that children are aware of the “unequal” status of the two languages they are being exposed to.

28 encouraged visits to their homeland, pushed their children to attend weekend language schools, and also tried to speak to their children in the HL as much as possible. H.-S. Kim (2005) found that the language that HLS used with their parents from birth till age five was significant in determining how target-like their HL was. Jia (1998) also found that the number of L2 speakers (of English) in the home had an effect on HLS’ HL (i.e., Chinese) proficiency. The HLS’ mothers’ proficiency in their L2 had an effect on the HLS’ L2, but not on their HL. However, despite the efforts of HLS’ parents to encourage the use and learning of the HL, it appears that this is insufficient in allowing HLS to improve in proficiency in their HL, as evidenced by the low proficiency speakers prevalent amongst HLS, especially by the time these speakers reach high school and university. Even with formal instruction, HLS may not improve significantly in their proficiency in the HL. For example, HLS are known to have difficulty with honorifics and more formal registers (H.-S. Kim 2001, Byon 2003). Byon (2003) found that although Korean language teachers used the marker -yo when addressing the whole class to demonstrate how -yo is used to mark politeness, students in HL classrooms rarely responded by using -yo when speaking to the teacher,23 showing persistent difficulty with this more formal register. J. S. Lee & S. Shin (2008) also note that HL education (for elementary and high school students) mainly occurs at weekend schools which are generally run by Korean churches and volunteers, and that poor teaching techniques, not entirely relevant material, and lack of motivation to learn and attend classes on weekends contribute to little improvement in the HL via these classes.

Thus, although mothers may encourage their children to receive formal instruction in the HL, this is not necessarily highly effective in helping HLS to maintain their HL. Kondo (1998) discusses a case study of a university student, Alan, who had low intermediate proficiency in Japanese despite having attended Japanese language school for 11 years and having lived in Japan until age 3, only speaking Japanese. He was introduced to English at age 3, and despite continued exposure to Japanese, did not use it very much. He spoke English to his parents and was reported to speak in English to classmates in Japanese school as well. As has often been observed, many HLS tend to respond to their parents in English even when addressed in the HL (H.-S. Kim 2005). In an extreme case, Jia (1998) describes an incident where a 5-year old child from China/Taiwan was asked to interpret what her teacher was saying to her mother in Chinese,

23 This would be quite unacceptable in Korea, where students are required to show respect to their teachers (and anyone older than them) by using -yo. See also Section 3.4.1 for further discussion on register in Korean.

29 yet the child simply reiterated what the teacher had said in English and refused to speak in Chinese or did not have the vocabulary to explain what had happened in Chinese. As Jia (1998) notes, children are developing their vocabulary and learning new concepts as they grow, thus it is possible that children who arrive in an English-speaking country learn particular concepts only in their L2, and perhaps never in their HL.

However, these types of HLS sometimes develop a renewed interest in improving proficiency in their HL. Affective factors have been studied in HLS who choose to enroll in university language courses to learn, or “relearn”, their HL. G. Cho, K.-S. Cho, & Tse (1997) found that Korean HLS enrolled in language classes in order to increase their level of communication with their parents and to help them overcome any “language barriers”, as well as to learn more about their heritage and identity and to expand their job opportunities. However, many HLS failed to finish the course due to a lack of confidence in speaking the language and also due to a lack of support from their language instructors, who may have been ill-equipped to effectively teach both L2 learners and HLS in the same classroom.

McQuillan (1996) found that courses that focused on improving literacy and promoting leisure reading resulted in an improvement in Spanish HLS’ proficiency in their HL. He conducted 10-week experimental university courses for HLS to demonstrate the effectiveness of Free Voluntary Reading, where students were mainly simply given the opportunity to select books to read and to discuss them with peers. Some of the benefits that McQuillan notes are that students received exposure to other dialects (including the dialect used by their parents), as well as the standard, and were also introduced to aspects of their culture. Effectiveness was measured in terms of the HLS’ attitudes towards reading in the HL and also in terms of the number of words the HLS recognized in a vocabulary test before the course and after the completion of the course. Although the sample sizes of his courses were small, the results suggest that reading in the HL helped improve the HLS’ confidence in reading, which also resulted in the HLS continuing to read on their own even seven months after the course they had enrolled in had ended. We can interpret these results as a voluntary decision by the HLS to increase their exposure to the HL, which we would also expect to have a positive effect on their proficiency in the HL as well. Kondo (1998) also discusses a case study of a university student who had improved in his attitude toward his HL (Japanese) and reading proficiency only after becoming interested in manga (Japanese comic books). Despite previous attempts by the student’s mother to encourage

30 him to learn Japanese, it was only after she had sent him to Japan to attend an international school that the student developed an interest in Japanese. Thus, his proficiency rapidly improved once he had personal motivation to learn the language, and with the appropriate motivation he was able to attain literacy in his HL, which helped him to improve in proficiency in his HL.

Yi (2008) also found that literacy was considered important to HLS who had advanced proficiency in their HL. Yi studied the use of Korean by two 1.5 generation (i.e., HLS who had immigrated to the US from Korea after first learning Korean in Korea) immigrants in high school and found that they used written language to socialize with their peers via notes passed at school or online blogs and websites. They used the Internet to maintain connections with friends and family members in Korea via email and instant messages. Yi thus emphasizes the importance of having Korean peer networks where HLS can interact in the HL in order to help them maintain the HL. It is thus more likely that one will find that HLS who are more proficient (than the average HLS) are precisely those who seek out more opportunities to use the language and who are exposed to more input. A similar conclusion was reached by Tse (2001) in her study of the commonalities of HLS who were literate in their HL. She found that rather than formal instruction in the language or use of the HL with family members, having peers who used the HL and who were literate in the HL made the biggest difference in determining whether a HLS would become literate in the HL or not. For example, HL-literate peers often encouraged activities such as going to karaoke bars to sing songs in the HL or reading comic books in the HL, which often motivated the HLS to become literate in the HL as well.

From the above studies, we can see that along with assessing the proficiency of HLS in their HL, it is important to also gather information on the extensiveness of their peer networks in the HL, how immersed they are in HL media and culture, whether they have received formal instruction in the HL, and whether they are literate in the HL. I will now examine Korean HLS in further detail to see how they can be characterized, especially in Toronto, Canada.

2.5 Korean Heritage Language Speakers

Of the 2,320,165 immigrants amongst the 5,072,075 people living in the Toronto Census Metropolitan area (CMA), according to the 2006 census, 55,265 were Koreans (Statistics Canada 2007); Koreans were the ninth biggest visible minority population. The number of Koreans has more than doubled compared to the 24,440 Koreans in Toronto in 1996 (Statistics Canada 1997);

31 there were 36,505 Koreans in 2001 (Statistics Canada 2002). The number of Koreans in the Toronto CMA is growing and reflects an influx of immigrants. Census data on intergenerational language transmission has shown an increase in the number of Korean-speaking children born in Canada. Amongst the children born in Canada between 1981 and 2006 whose mother’s L1 was Korean, the percentage of children whose L1 was also Korean rose from 32% in 1981 to 54% in 2006 (Statistics Canada 2007) (i.e., more mothers whose L1 was Korean were raising their children to be native speakers of Korean in 2006 compared to in 1981). This could simply be due to new immigrant mothers arriving with older children who were more comfortable with Korean, which could influence them to speak in Korean even to their children born in Canada, or it could be an indication that the Korean language may actually be preserved better now than in the past. For example, HLS of Korean are often described as having low proficiency in the language, despite having relatively good listening comprehension (H.-S. Kim 2005). Korean HLS often do not speak in Korean with their siblings or peers (J. S. Lee 2008), or with their parents – even if they are addressed in Korean (H.-S. Kim 2005, Polinsky & Kagan 2007), as was evidenced by Jia for child HLS. Thus, they often have limited vocabularies and are only familiar with informal registers, which suggests that Korean could follow the pattern of other immigrant languages in the US, whose speakers have switched to English within three generations (Wiley 2001). However, the Korean community is still relatively young and many families have not been in Canada for more than three generations. In recent years in Toronto, peers who are fluent native speakers of Korean abound, due to an influx of international students, new immigrants, and travellers. Exposure to Korean media is also readily available, given that Korean movies, dramas, and music are receiving attention all over Asia, increasing the international appeal of the Korean language and Korean media amongst other immigrant groups in Toronto as well; Korean movies are played in theatres in Toronto, and Korean music can be heard on the radio and in schools. Opportunities to teach English in Korea also abound, and for the past several years, the tourist industry has broadcast commercials promoting travel to Korea. Thus, although Benmamoun et al. (2010) observe that many HLS have low proficiency in their HL due to low use of the HL and prolonged lack of exposure (to the baseline), it is possible that this situation may change in Toronto, given that there are many more opportunities to obtain exposure to Korean.

S.-Y. Shin (2010) found that 5–8-year-old children and their teachers at a Sunday school at a Korean church in Los Angeles appeared to be strategically using code-switching in order to

32 adhere to Korean cultural norms which include respecting social hierarchies. The teachers and some of the students spoke in Korean when giving instructions in order to assert their authority in terms of their role (e.g., as teacher) or age (e.g., if they were older than the person they were talking to) or in order to show respect by using Korean terms of address. Even children who were English-dominant used Korean terms of address with adults to show respect and to avoid having to address them by name or to say you, which would be considered quite rude in Korean culture. The children also used Korean to avoid uncomfortable topics or words (e.g., the use of Korean ttong ‘feces’ for poo). S.-Y. Shin’s observations are based on a very small sample size, but if these findings are generalizable, they suggest that it is possible that this use of code-switching could be a regular part of HLS’ language situation at home as well (i.e., parents may be using Korean to assert authority and to enforce the Korean culture of respecting elders). Given the above literature, one goal of this study is to explore how various types of input (e.g., media) and interactions (with family members, friends, etc.) may contribute to HLS’ proficiency in their HL.

2.6 Summary

Given the unique nature of the language exposure that HLS receive in the languages they acquire, it is most likely that all of the above factors play a role in the development of their HL. HLS begin with fairly rich input24, but this is provided when they are young children and still developing cognitively and have relatively less complex communicative needs. A switch to the language of the majority (i.e., English) often occurs when they are still young (around five years of age) and thus development of their English increases exponentially, whereas the development of their HL often stagnates, leading to incomplete acquisition and/or attrition. This thesis focuses on aspects of the nominal domain which differ in English and Korean as potential areas of incomplete acquisition in the language of HLS of Korean. In particular, I examine the use of Case marking, plural marking and classifiers, and conjunctions and comitative Case markers in Korean and investigate which language background factors are relevant in determining the proficiency of HLS in their HL.

24 Given that the actual quantity and quality of a child’s input is determined by factors such as the child's birth order and the language attitudes of the parents towards the HL and English, the specific amount of input for any particular child would be impossible to guess, but we can assume that the majority of HLS receive the greatest input in their HL as children.

33

Chapter 3 The Nominal Domain of Korean: Areas of Difficulty for HLS 3 Aspects of the Korean Nominal Domain 3.1 Introduction

English and Korean are typologically different languages. English is a head-initial language with a fairly rigid SVO word order and subjects are obligatory. Korean is a head-final SOV language with a fairly flexible word order and topic drop is common in Korean; yet in the nominal domain the languages appear to have some similarities. Both English and Korean are nominative- accusative languages and they have the same linear order of morphemes in their noun phrases, which has led some scholars to propose that Korean is actually head-initial, at least in the nominal domain. However, there are still important differences. The types of Case markers available are different, along with how regularly Case is marked overtly. In addition, although the linear order of morphemes is similar, their syntax is different. There are no articles in Korean, which suggests that DPs in Korean may not exist. A reconsideration of the structure/syntax according to the morphemes and types of phrases available in Korean presents an analysis where the head-final structure is preserved in the nominal domain. Thus, the languages are still different in the nominal domain despite generally having the same linear word (and morphemic) order in this domain. This makes this domain an area ripe for transfer or cross-linguistic influence. In this thesis I examine the following phenomena in the nominal domain of Korean HLS’ grammars: (i) Case, (ii) Classifiers and plural marking (including quantificational structures), and (iii) comitative Case markers.

With each of the above phenomena, English presents more straightforward evidence for the presence or absence of particular markers, but Korean shows a lot of optionality in the grammar, albeit optionality which is sensitive to various variables/preferences. Case marking is realized differently in English and Korean. Case is not visible in English, other than with genitive ’s and in the pronouns, where Case is realized as suppletion. Since English has a fairly strict SVO word order, word order reveals the Case of the noun phrase. In Korean, although Case marking is obligatory in the written register, it is optional in colloquial language. This optionality is not

34 random, but rather is sensitive to various variables, such as whether the sentence is in the canonical word order or not,25 as well as the person, animacy, and definiteness of the noun. Thus, learning Korean involves not only learning the form of the morphological Case markers, but also the varying levels of acceptability of using Case drop or not. Thus, a study on Case marking and Case drop in Korean HLS’ grammars would help shed light on how HLS deal with optionality in the grammar.

A similar type of optionality is available with plural marking in Korean. Whereas English regularly marks the plural on count nouns, plural marking in Korean is sensitive to the animacy and specificity of the noun, as well as the particular word order of elements in the noun phrase. In addition, Korean count nouns generally require classifiers, but the acceptability of -tul also interacts with the presence or absence of classifiers, so -tul and classifiers will be discussed together in further detail later on in this chapter.

Another difference between English and Korean lies in their conjunctions. Although conjunctions may not be an area that is typically relevant to the nominal domain, there are specific properties of conjunctions that are unique to Korean and are not found in English, which make them an interesting area of investigation. Conjunctions in Korean differ in morphology, syntax, and selection. For example, some of the conjunctions in Korean are phonologically conditioned, unlike English conjunctions, and structurally, conjunctions in English attach to the final conjunct, whereas Korean conjunctions attach to the first conjunct.

Given the differences between English and Korean, these aspects of the nominal domain – Case, classifiers, plural marking, and conjunctions – pose a potential area of difficulty for HLS of Korean, since influence from English would lead them to make the wrong conclusions about Korean. In addition, given that learning of the HL by HLS usually results in arrested development when HLS begin schooling in English, HLS could have grammars which reflect an earlier level of development in monolingual children. It may also be the case that although the HLS mastered these aspects of the nominal domain at an earlier age, with the increased exposure to English, attrition of the HL may have occurred. Investigating the nominal domain of Korean is

25 It is sometimes suggested that the option to move things around in a sentence in a language is only available when overt Case marking allows the grammatical role of the noun to be recoverable.

35

an interesting area of study, since it will help shed light on the nature of the grammars of HLS. In the sections that follow, I will examine each aspect of the nominal domain in Korean in turn.

3.2 Case marking in Korean

Although English and Korean are both nominative-accusative languages, the extent of overt Case marking used in the two languages is very different. In English, nominative and accusative Case are not marked overtly on full noun phrases; word order reveals what kind of Case a noun phrase has. In (14), since Gina precedes the dog, even without overt Case marking we know that Gina has (null) nominative Case marking and the dog has accusative Case marking. Similarly, since the dog precedes Gina in (15), we expect the dog to have nominative Case and Gina to have accusative Case, which makes the sentence semantically infelicitous.

(14) Gina walked the dog.

(15) #The dog walked Gina.

Genitive ’s appears on full noun phrases (e.g., Mary’s dog); otherwise, Case marking only appears on personal pronouns and a different form exists depending on the Case (e.g., his vs. him, etc.), except for you and it, which are the same regardless of Case.26 Thus, even with the personal pronouns, Case is not always distinctively marked in English.

In contrast, Case marking is much more pervasive in Korean. The nominative and accusative Case markers and the Contrastive marker (Suh 2004) -nun are phonologically conditioned; -ka

‘NOM’, -lul ‘ACC’, and -nun ‘CONTR’ occur when attached to bases ending in vowels, as in Mina,

swukcey ‘homework’, and ecey ‘yesterday’ in (16), whereas -i ‘NOM’, -ul ‘ACC’, and -un ‘CONTR’ occur when attached to bases ending in consonants, as in Thaypin, swuyeng ‘swimming’, and onul ‘today’ in (17).

(16) Mina -ka ecey-nun swukcey-lul hay-ss-ta. Mina-NOM yesterday-CONTR homework-ACC do-PST-DECL ‘Mina did (her) homework yesterday (but not on any other days).’

26 In addition, you is the same form that is used for the singular and the plural and the object pronouns are the same as dative pronouns, as in He thanked him and He gave him a raise.

36

(17) Thaypin-i onul-un swuyeng-ul hay-ss-ta. Taebin-NOM today-CONTR swim-ACC do-PST-DECL ‘Taebin swam today (but not on any other days).’

Case marking allows for freer word order variations in Korean. Since the grammatical role of the noun is identified with a Case marker, moving elements around in a sentence does not change the meaning of the sentence; thus, any element of the sentence other than the verb can be moved.27 In colloquial language, and often in child-directed speech, Case drop of some or all of the Case markers is common, as in (18).

(18) I-ke(-l) enni(-hanthey) cwu-Ø-lkka? This-thing(-ACC) older.sister(-DAT) give-PRES-Q ‘Should (we) give this to (your) older sister?’

However, Case drop is not completely optional, as noted by S.-H. Lee, et al. (2009) – overusing Case drop can be unnatural and result in what sounds like “baby talk”. Case drop restricts the number of word order variations, since recoverability becomes difficult with non-canonical word orders. In other words, the acceptability of Case drop partially depends on whether the noun is a subject or an object.28 Thus, although it is possible to drop the Case markers on both the subject and the object in the canonical word order (i.e., SOV), as illustrated in (20)–(22), it is only acceptable to drop the object Case (24), and not the subject Case, when the object has been fronted to sentence-initial position, as illustrated by the awkwardness of (25) and (26).

(19) Swuci -(ka) Cayho-(lul) coha-hay-ss-e. Suji-NOM Jaeho-ACC like-do-PST-DECL ‘Suji liked Jaeho.’ (SOV; (NOM)-(ACC))

(20) Swuci -ka Cayho-Ø coha-hay-ss-e. Suji-NOM Jaeho-ACC like-do-PST-DECL ‘Suji liked Jaeho.’ (SOV; NOM-Ø)

(21) Swuci -Ø Cayho-lul coha-hay-ss-e. Suji-NOM Jaeho-ACC like-do-PST-DECL ‘Suji liked Jaeho.’ (SOV; Ø-ACC)

27 It appears that it is acceptable to move even the verb (e.g., to sentence-initial position) for at least some speakers of contemporary Korean (Shinjung Park, p.c.).

28 With Japanese, it has been noted (Fry 2001) that nouns that are closer to the verb tend to undergo Case drop more frequently.

37

(22) Swuci -Ø Cayho-Ø coha-hay-ss-e. Suji-NOM Jaeho-ACC like-do-PST-DECL ‘Suji liked Jaeho.’ (SOV; Ø-Ø)

(23) Cayho -lul Swuci-ka coha-hay-ss-e. Jaeho-ACC Suji-NOM like-do-PST-DECL ‘Suji liked Jaeho.’ (OSV; ACC-NOM)

(24) Cayho -Ø Swuci-ka coha-hay-ss-e. Jaeho-ACC Suji-NOM like-do-PST-DECL ‘Suji liked Jaeho.’ (OSV; Ø-NOM)

(25) #Cayho -lul Swuci-Ø coha-hay-ss-e. Jaeho-ACC Suji-NOM like-do-PST-DECL ‘Suji liked Jaeho.’ (OSV; #ACC-Ø)

(26) Cayho -Ø Swuci-Ø coha-hay-ss-e. Jaeho-ACC Suji-NOM like-do-PST-DECL #‘Suji liked Jaeho.’/‘Jaeho liked Suji.’ (OSV; #Ø-Ø)

Similarly, when the contrastive Case marker is used instead of a grammatical Case marker, it is possible to drop the accusative Case marker in the canonical (SOV) word order (28) and in the non-canonical (OSV) word order (32), but it is only possible to drop the nominative Case marker in the canonical word order (30), but not the non-canonical one (34). (34) is only felicitous when the null Case marker is interpreted as a dropped accusative Case marker. It should be noted that even though Case drop of the nominative Case marker in SOV word order (30) and Case drop of the accusative Case marker in OSV word order (32) is allowed when the other noun in their respective sentences is marked with contrastive Case, the sentences become identical in linear word order, thus they are ambiguous between readings where the sentence-initial non-Case-marked noun is interpreted as a subject or as an object.

(27) Swuci-nun Cayho-lul coha-hay-ss-e. Suji-CONTR Jaeho-ACC like-do-PST-DECL ‘Suji liked Jaeho.’ (SOV; CONTR-ACC)

(28) Swuci-nun Cayho-Ø coha-hay-ss-e. Suji-CONTR Jaeho-ACC like-do-PST-DECL ‘Suji liked Jaeho.’ (SOV; CONTR-Ø)

(29) Swuci-ka Cayho-nun coha-hay-ss-e. Suji-NOM Jaeho-CONTR like-do-PST-DECL ‘Suji liked Jaeho.’ (SOV; NOM-CONTR)

38

(30) Swuci-Ø Cayho-nun coha-hay-ss-e. Suji-NOM Jaeho-CONTR like-do-PST-DECL ‘Suji liked Jaeho.’/ ‘Jaeho liked Suji.’ (SOV; Ø-CONTR)

(31) Cayho-lul Swuci-nun coha-hay-ss-e. Jaeho-ACC Suji-CONTR like-do-PST-DECL ‘Suji liked Jaeho.’ (OSV; ACC-CONTR)

(32) Cayho-Ø Swuci-nun coha-hay-ss-e. Jaeho-ACC Suji-CONTR like-do-PST-DECL ‘Suji liked Jaeho.’/‘Jaeho liked Suji.’ (OSV; Ø-CONTR)

(33) Cayho-nun Swuci-ka coha-hay-ss-e. Jaeho-CONTR Suji-NOM like-do-PST-DECL ‘Suji liked Jaeho.’ (OSV; CONTR-NOM)

(34) Cayho-nun Swuci-Ø coha-hay-ss-e. Jaeho-CONTR Suji-NOM like-do-PST-DECL #‘Suji liked Jaeho.’/‘Jaeho liked Suji.’ (OSV; #CONTR-Ø)

Accusative Case can also be dropped on the verbal nouns in light verb constructions involving the light verb ha- ‘do’ (Han & Rambow 2000, S.-H. Lee & Park 2008, among others). In these cases, the object noun appears to have been incorporated into the verb phonologically,29 as in kongpu-hay- ‘study’ in (36), where the noun kongpu ‘study’ combines with the light verb ha- ‘do’. In contrast, genitive Case can be dropped regularly, generally irrespective of word order, as illustrated in (37) and (38). This leads to the following hierarchy of acceptability in terms of felicity in dropping Case: Genitive > Accusative > Nominative.30

29 An alternate view is that what occurs in these cases is not Case drop but Case insertion, as suggested by S.-H. Lee & Park (2008); barring the introduction of more substantial historical evidence, I will assume that these are still instances of Case drop.

30 There is some debate as to whether accusative Case is dropped more often than nominative Case. Kim & Kwon (2004) and H. Lee (2006) argue that this is the case, whereas Kwon (1989), Hong et al. (1998), and S.-H. Lee & Park (2008) argue that nominative Case is dropped more often; Kwon (1989) proposed that Case markers which are less lexical are dropped more often, as in the following hierarchy: Nominative > Accusative > Genitive > Adverbial (temporal) > Adverbial (instrumental) > Adverbial (comparative). However, as S.-H. Lee & Park (2008) note, Kwon (1989) did not perform a statistical analysis, and even though Hong et al. (1998) performed a statistical analysis, their corpus is relatively small. S.-H. Lee & Park (2008) analyzed various written (public speeches and newspaper editorials and articles) and spoken (informal personal conversations and formal TV debates) registers in the Sejong corpus for Case drop. Overall, they found that nominative Case was dropped more often (15%) than accusative Case (10%). However, in conversation, accusative Case was dropped more often (49%) than nominative Case (31%) – even when instances of accusative Case drop on the verbal nouns in light verb constructions with ha- ‘do’ were omitted, and Case drop occurred much more frequently overall in conversation than in any of the other registers. No statistical analysis of the data was conducted, but if the differences are significant, S.-H. Lee & Park’s results appear to contradict their conclusions. Since Case drop is generally identified as a phenomenon occurring in colloquial

39

(35) Ku-ka kongpu-lul hay-ss-ta. he-NOM study-ACC do-PST-DECL ‘He studied.’

(36) Ku-ka kongpu-hay-ss-ta. he-NOM study-PST-DECL ‘He studied.’

(37) Hila-(ka) Cihay-(uy) moca-(lul) pillye-ss-ta. Hera-NOM Jihae-GEN hat-ACC borrow-PST-DECL ‘Hera borrowed Jihae’s hat.’ (SOV)

(38) Cihay-(uy) moca-(lul) Hila-ka pillye-ss-ta. Jihae-GEN hat-ACC Hera-NOM borrow-PST-DECL ‘Hera borrowed Jihae’s hat.’ (OSV)

H. Lee (2006) found that accusative Case was dropped more often than nominative Case (46.8% vs. 28.1%, respectively) in the Call Friend Korean (CFK) corpus (1996), even when instances of Case drop with light verbs were excluded from the data. The grammatical function of the noun (i.e., subject vs. object) also correlated with several other factors determining Case drop. These factors included grammatical person, animacy, and definiteness.31 Case was dropped on local (i.e., first or second person) subjects more often than on third person subjects, but Person did not affect the rates of Case drop on objects. Case was dropped more often on human and other animate subjects than inanimate ones, whereas Case was dropped more often on inanimate objects than on human and other animate objects. With respect to definiteness, H. Lee (2006) analyzed pronouns, names, and “other”—the latter consisting of nouns which would require careful analysis in order to determine whether they were definite or not. Assuming that pronouns and names are “strongly definite” nouns, H. Lee found an interaction for Case drop between definiteness and the grammatical function of the noun—Case was dropped more often on strongly definite subjects than on other types of subjects, whereas Case was dropped more often

language, it is odd that S.-H. Lee & Park chose to conflate the results to form their conclusions. Their results point to formality being a factor in determining the frequency of Case drop, but they do not necessarily provide evidence for nominative Case being dropped more frequently than accusative Case. See Aissen (2003) for analyses and further discussion of different Case marking languages.

31 Another factor noted by H. Lee (2006), but not pursued in her analysis of the CFK corpus, is the style or register of the discourse. Case is more likely to be dropped in less formal discourses and when participants in the discourse are more familiar with one another (S.-H. Lee & Park 2007).

40

on other types of objects (vs. strongly definite objects).32 The hierarchies for susceptibility to Case drop on subjects are as follows (H. Lee 2006, p. 74): Person: First person > Second person > Third person, Animacy: Human > Animate > Inanimate, and Definiteness: Personal pronoun > Name > Other; the reverse is true of objects. This likely relates to the prototypicality of the argument and the recoverability of Case—animate (human or not) nouns are more likely to be interpreted as , and thus the default interpretation would be to assume that nominative Case has been dropped, whereas inanimate nouns are generally , and thus it would be assumed that accusative Case on the noun has been dropped. If these assumptions do not hold, Case marking is more likely to be retained. Lee notes that this is in line with split ergative-accusative languages, where Case is dropped on canonical types of arguments and is more likely to be retained on non-canonical ones.

Even though Korean is a head-final language and English is a head-initial language, the linear order of morphemes in the nominal phrases is the same for both languages. However, structurally, English and Korean differ, since DPs are obligatory in English and are headed by articles, such as the and a, and the genitive Case marker. The (simplified) 33 structure of an English sentence is given in (39).

(39) IP qp DP I’ 2 3 D NP I VP | 4 ei The boy V DP | ep borrowed DP D’ 3 3 D NP D NP | 4 | 6 the man ’s umbrella

32 The same results were found by S.-H. Lee & Park (2008).

33 The structure is simplified here. I assume a VP-internal hypothesis, where the subject of the sentence is base-generated in spec-vP and raises to spec-IP. I also assume that the object is base-generated as the complement of the VP and then raises and “tucks in” to spec-vP.

41

I assume the lack of overt definite or indefinite articles in Korean to be an indication that there are no DPs in Korean; rather, Case phrases (KPs) form the outer functional layer of nominal phrases, allowing them to appear as arguments in a sentence. I assume that KPs are present in Korean even when Case drop occurs. The (also simplified) structure of the sentence in (37) above would be as follows:

(40) CP qp IP C qp | KP I’ -ta 2 to ‘DECL’ NP K VP I 5 | ro | Hila -ka KP V -ss ‘Hera’ ‘NOM’ ru | ‘PST’ KP K’ pillye- 1 2 ‘borrow’ NP K NP K 5 | 5 | Cihay -uy moca -lul ‘Jihae’ ‘GEN’ ‘hat’ ‘ACC’ To summarize, Case marking is different in English and Korean. In English, Case marking only occurs regularly for genitive Case; other Cases are expressed only in suppletive forms of personal pronouns. The genitive Case markers and pronouns head DPs in English. The lack of regular Case marking makes English word orders very restrictive. Case marking is much more prevalent in Korean than in English. All nouns in Korean are marked for nominative, accusative, and genitive Case. The nominative and accusative Case markers and the contrastive marker have

two forms which are phonologically conditioned. -Ka ‘NOM’, -lul ‘ACC’, and -nun ‘CONTR’ attach

to bases ending in vowels; -i ‘NOM’, -ul ‘ACC’, and -un ‘CONTR’ attach to bases ending in consonants. These Case markers head KPs in Korean. Regular Case marking allows for more word order variations in Korean. Case marking is optional in colloquial language, but when it occurs, word orders are more restricted. Case is susceptible to drop in the following hierarchies: Type of Case: Genitive > Accusative > Nominative, Person: First person > Second person > Third person, Animacy: Human > Animate > Inanimate, and Definiteness: Personal pronoun > Name > Other. This is summarized in Table 1.

42

Table 1. Case Drop in English and Korean

Case English Korean

Gen Case is always overt on N Most likely Case to be dropped: Gen > Acc > Nom

Acc Case is never marked on N; Word order: SOV > OSV only ever visible on PN Person: (3p > 2p, 1p)34

Animacy: Inanimate > Human, Animate

Definiteness: Other > Personal PN, Name

Nom Case is never marked on N; Word order: SOV, *OSV only ever visible on PN Person: 1p, 2p> 3p

Animacy: Human, Animate > Inanimate

Definiteness: Personal PN, Name > Other Note. N=noun, PN=pronoun.

Since most HLS switch to English as their dominant language at the onset of schooling, it is possible that HLS experience arrested development, resulting in an interlanguage grammar that reflects an earlier developmental stage of Case acquisition in monolingual children. Although the input to HLS is different, since most HLS are simultaneous bilinguals (i.e., they receive input in Korean and English and must sort out their separate grammars), it is useful to examine the acquisition of Case by monolingual Korean-speaking children.

3.2.1 Case in Korean-Speaking Monolingual Children

The hypotheses described above are evidenced in monolingual Korean-speaking children. Y.-j. Kim (1997) examined Case in the L1 acquisition of Seoul (standard) Korean by gathering longitudinal data by audio-recording 5 children between 1;7 and 3;5 in interactions with their primary caregivers. She found that between 1;8 and 2;0 children went through a stage where -ka was the only form used productively for the nominative Case marker. It took the children seven to eight months longer to begin using the allomorph -i correctly. Until then, they entered a period of variation, where -ka or -ika was used in place of -i, as in (41) (Y.-j. Kim’s (28)) below.

34 This hierarchy was not found to be statistically significant by Lee (2006).

43

(41) Halmeni, samchon-*ka thokki manci-e thokki, samchon-*i-ka grandmother uncle-NOM rabbit touch-DECL rabbit uncle-NOM ‘Grandma, uncle is touching the rabbit.’ (P 2;1)

Y.-j. Kim attributes the delay in children’s acquisition of nominative -i to a number of factors.

Children could be having problems with the distribution of -i ‘NOM’ due to its homophony with hypocoristic -i35. In speech, it is quite common to add hypocoristic -i to names ending in consonants. Attaching hypocoristic -i to Thaypin in (42) creates a base that ends in a vowel, thus the -ka form of the nominative Case marker is used. Thaypin-i ‘Taebin’ in (43) is ambiguous. The -i could be interpreted as the correct phonological form of the nominative Case marker (since Thaypin ends in a consonant), or since hypocoristic -i is frequently added to names ending in consonants, Thaypin-i could be interpreted as an instance of (nominative) Case drop. Both are grammatically possible interpretations; thus, only the speaker would know for certain which interpretation was intended.

(42) Thaypin-i-ka swuyeng-ul hay-ss-ta. Taebin-HYP-NOM swimming-ACC do-PST-DECL ‘Taebin swam.’

(43) Thaypin-i swuyeng-ul hay-ss-ta. Taebin-NOM/HYP? swimming-ACC do-PST-DECL ‘Taebin swam.’

Children use -i before the topic marker -nun, which would be acceptable if -i were a hypocoristic marker (44), but unacceptable if it were a nominative Case marker (45). Alternatively, (45) could be interpreted as an infelicitous use of the hypocoristic marker instead of an infelicitous use of the nominative Case or topic markers.

(44) Minhyek -i-nun hakkyo-ey ka. Minhyek-HYP-TOP school-to go ‘Minhyuk is going to school.’

(45) *Haks ayng-i-nun hakkyo-ey ka. student-NOM-TOP school-to go ‘The student is going to school.’

35 This is referred to as a hypocoristic suffix by Martin (1992) and an endearment marker by Sohn (1999).

44

Similarly, children also use -i incorrectly with words with the delimiter -to (see (46), Y.-j. Kim’s (31)), even though the nominative Case marker is supposed to be in complementary distribution with this morpheme and the use of -i as a hypocoristic marker would be equally infelicitous, since taycang ‘captain’ is a job title being used as a term of address. Not only is the hypocoristic marker disallowed with job titles, it is a form of familiarity (and informality) for people who are the same age or younger than the speaker. Thus, its use is especially inappropriate with honorific -nim – a marker which is supposed to show respect towards the listener.36 Children also overgeneralize the use of the hypocoristic marker to non-human/non-animate nouns, as shown in (47), Y.-j. Kim’s (32).

(46) taycangnim-*i-to na-o-sey-yo. 37 captain-also move.out-come-HON:IMP-POL ‘Captain, (you) too come out.’ (C 2;6)

(47) pholkhein-*i-ka(=i) ilehkey manh-a. forklift truck-NOM so be.many-DECL ‘(There) are so many forklift trucks.’ (C 2;2)

In the above cases, it is impossible to tell whether -i is being used as a nominative Case marker or the hypocoristic marker; however, we can assume that homophony of the two morphemes is causing the delay in acquisition.

If -i in the above example is a misuse of the hypocoristic marker, children could be assuming that the nominative Case marker is -ka and the accusative Case marker is -lul to satisfy their cross-linguistically attested preference for a 1:1 mapping of form to function. The use of -ka where -i is required is an (albeit less frequent) error produced by monolingual children, but it is supported given that i) children may be misanalyzing noun stems, ii) -ka is more salient than -i, and iii) -ka occurs more frequently than -i in children’s input. Y.-j. Kim suggests that these

children may be misanalyzing noun stems and assuming -i ‘NOM’ to be part of the noun stem, as they tend to do when the copula i ‘be’ is present, as shown below (part of Y.-j. Kim’s (40)), where the child incorrectly attaches the copula i to the noun os ‘clothes’ when only the noun os ‘clothes’ would be appropriate.

36 This use of the hypocoristic marker in conjunction with an honorific term is sometimes used sarcastically as an insult or in jokes in adult speech. 37 Y.-j. Kim uses the following abbreviations: HON: honorific, IMP: imperative, POL: polite level of speech.

45

(48) Adult: os-i-ya. clothes-COP-DECL ‘(They) are clothes.’

Child: osi? ‘Clothes?’ (P 1;9)

Y.-j. Kim also notes that -ka is more salient than -i. Since -i attaches to stems ending in consonants, the final consonant of the stem becomes the onset of the syllable containing -i. In addition, -ka appears more frequently in children’s input. In Y.-j. Kim’s study, the adult caregivers of the children produced -ka over -i in ratios of 2:1 to 8:1. Since children seem to know early on that -ka is not used with bases ending in consonants, Y.-j. Kim suggests -ika may be used by children as the nominative Case marker to satisfy their preference for a 1:1 mapping of form to function. Thus, all of the above factors may play a role in children’s errors with nominative Case marking at different points of their acquisition. Y.-j. Kim also notes a difference in the acquisition of the nominative and accusative Case markers—monolingual children acquire the accusative Case marker about a year after the nominative Case marker (based on S. Lee & S. Pae’s 1989 study). In addition, the accusative Case marker is dropped more often than the nominative Case marker in children’s input and their production (based on S.W. Cho’s 1981 study). The performance of monolingual children can be compared to that of HLS, a couple studies of which will be discussed in the following section.

3.2.2 Previous Literature on HLS’ Comprehension of Case in Korean

M. Song, O’ Grady, S. Cho, and M. Lee’s (1997) study found that even 8-year-old HLS were unable to use Case marking to help them distinguish between the subject and the object of OSV sentences, despite the fact that monolingual children performed above chance by age 4. O’ Grady, O.-S. Lee, and J.-H. Lee (2011) discuss the results in terms of the lack of one-to-one mapping of form to meaning in Korean. They point to the frequency of Case drop in speech, the relative infrequency of OSV word orders in general and in cases where Case would not be redundant (i.e., where pragmatics would not help listeners determine the and ), and the fact that nominative and accusative Case are used differently depending on whether the noun is animate or inanimate, and definite or indefinite. Song et al.’s (1997) study focused solely on comprehension and specifically on cases where Case marking is the only clue that can be used to

46

determine who is the and who is the , thus we need to investigate HLS’ actual production of Case markers in general.

A characteristic of the HLS enrolled in Ko’s (2005) low-intermediate level Korean course is that they were insensitive to the discourse properties of the contrastive marker -nun and the nominative Case marker -ka (see C.-M. Lee 1989 for a discussion of the different uses of each marker), using them interchangeably or only using -ka exclusively.38

S.-H. Lee, Jang, & Seo (2009) conducted a corpus study on various errors related to Case and other particles39 in Korean. They examined written data from beginner and intermediate heritage and non-heritage (L2) speakers taking language courses in Korea and the United States. Heritage learners were identified as those whose parents spoke Korean at home and who could understand Korean but did not necessarily speak or write it. Lee et al. divided errors into the following six types: omission, addition, replacement, malformation, paraphrasing, and spacing. Omission is not always an error, since Case drop is common in Korea, but its overuse can result in awkward and unnatural sentences, as discussed in Section 3.1, thus it was important to keep track of the proportions of Case drop in general. Errors of addition included the insertion of extra or redundant particles, as in the “fossilized idiomatic expression" below (Lee et al. 2009: p. 534; morphemic glosses added):

(49) Ku salam-Ø/*-i iss-Ø-canha That person-NOM exist-PRES-DECL ‘As for that person,’

(49) is used as a topicalizing expression to contextualize the statements to follow ‘concerning the person mentioned’ and the nominative Case marker must be dropped. When the nominative Case marker is present, the sentence loses its idiomatic meaning and is interpreted literally as ‘But that person is there.’ Replacement errors included the substitution of particles with incorrect particles (e.g., using -eyse ‘from’ instead of -ey ‘to’), whereas malformation errors were those with

38 Only a single speaker was able to use -ka and -nun appropriately; testing revealed that the speaker had greater proficiency overall in the language than their classmates and was transferred to a higher level course.

39 Lee et al. (2009) distinguish between lexical Case and discourse/modal particles. Lexical Case is comprised of structural Case markers (e.g., nominative -ka, accusative -lul and genitive -uy) and inherent Case markers (e.g., dative -eykey, goal -ulo and -ey, locative -eyse, instrument -ulo, etc.), whereas discourse/modal particles consist of the topic marker (their terminology) -nun and delimiters such as -to ‘also’ and -man ‘only’.

47

incorrect forms for particles which are phonologically conditioned (e.g., using -lul instead of -ul for a noun that ends in a consonant). Paraphrasing errors were those where longer morphemes or phrases were used in lieu of a single particle or vice-versa, as in (50) below (Lee et al.’s (7)), where the “topic marker” -un should have been used instead of the copula -ilako and (51), where (pat)-nun kes-i should be replaced with (pat)-key.

(50) Hayngbok -ilako salam-mata taluta Happiness-copula person-each different ‘As for happiness, every person has different ideas.’

(51) Sang -ul pat-nun kes-i (pat-key) twoymyen kipun-i cohci. 40 award-ACC receive-REL thing-NOM (receive-E) become feeling-NOM good ‘If I come to receive an award, I feel good.’

Spacing errors were orthographical errors where particles, which are bound morphemes, were not attached to the word they modify and were written as separate words. This group of errors is not particularly relevant to the present study, since I elicited oral data.

Lee et al. (2009) found the following hierarchy of error types in the data (p. 539): Omission > Replacement > Spacing > Malformation > Addition > Paraphrasing. The majority of errors were omission or replacement errors for all groups. These errors occurred in roughly the same proportions for all groups, except non-heritage beginner learners produced a larger proportion of replacement errors overall (~10% higher than all groups). Intermediate heritage learners differed from the other groups in producing fewer addition and paraphrasing errors and the beginner and intermediate heritage learners produced a greater number of malformations than the non-heritage learners. Lee et al. attribute the increased proportions of the latter error type to HLS’ (often sole) exposure to colloquial vs. written language. However, as Lee et al. concede, these remain at the level of observations, since they failed to count the total number of instances where particles should have been produced by each learner and also because of the small corpus size; thus it is unknown whether the differences are statistically significant or not. In addition, because all of the particles (structural Case, inherent Case, and discourse/modal particles) have been grouped together, the results are rather skewed, especially for the count of “omission” errors, given that, whereas the omission of Case markers is often licit in colloquial language, the omission of other

40 REL refers to Relativizer; however, it is unclear what E refers to in the morphemic gloss. -Key is a verbal particle often appearing with twoy-myen ‘become-COND’ to mean ‘if (I) X (this)’.

48 particles is generally not acceptable. Analyzing each particle individually would help present a clearer picture of the specific nature of difficulties that the various types of learners have. Another difficulty in interpreting the data lies in the fact that data was collapsed across learners. The results are presented as overall proportions of total errors produced by all members of the group, rather than the means of the proportions of errors produced by each learner in the group, even though there was likely a significant amount of interspeaker variation. Thus, these individual differences are masked by the presentation of the overall group proportions.

Lee et al. also examined the structural complexity of the learners’ writing by comparing the number of paragraphs, words, and sentences produced by the speakers. They note that heritage beginner learners produced more words and sentences than non-heritage learners and they conclude that the intermediate heritage learners produced more structurally complex sentences than the intermediate non-heritage learners based on the greater number of paragraphs produced. They note that the difference between the beginner and intermediate heritage learners is larger than between their non-heritage counterparts in terms of the number of sentences (476:299 vs. 381:327, respectively). However, here again, the numbers presented reflect overall totals for each group of speakers rather than the mean proportions of the individuals in the group. Thus, it is possible that the lack of a significant difference between the beginner and intermediate non-heritage (and the intermediate heritage) learners could be due to one or two individuals producing a significant proportion of the total number of sentences for the group. Also, it is unclear how the total number of words, sentences, and paragraphs can be used as an overall measure of structural complexity. It is misleading to use the number of sentences as a gauge of complexity given that a sentence in Korean can consist of several clauses or just a single clause. Thus, we might assume that the greater number of sentences produced by the beginner heritage learners indicates that they produced simpler sentences than the intermediate heritage learners who might have produced fewer sentences, but coupled with the observation that they produced a greater number of paragraphs, are likely to have produced more complex (i.e., multi-clausal) sentences. However, this might not actually be the case. Using the number of words produced as a measure of structural complexity is also misleading since individual speakers could simply be producing paraphrases (resulting in a greater number of words). Moreover, it is unclear what was counted as a word. As noted above, beginner learners have great difficulty with spacing when writing, which results in a greater number of “words”.

49

Thus, although the study presents interesting trends for the various learner groups and produces a checklist of errors which should be analyzed with respect to particles, the results must be treated with extreme caution, given that all individual differences were masked and no statistical tests were run on the data. In addition, precisely because HLS often tend to write words the way they are pronounced (resulting in a high number of orthographical errors), we need to examine whether Case drop is actually occurring in contexts which are acceptable for colloquial language or whether the learners’ use of Case drop is inappropriate even for colloquial language.

3.2.3 Case Marking in Korean: A Pilot Study

Given the stages of Case acquisition in monolingual Korean-speaking children, if HLS have interlanguage grammars which reflect earlier stages of Case acquisition, we would predict a disparity in their mastery of the nominative and accusative Case markers and in their respective phonological forms. Low proficiency HLS would be expected to have a one-to-one form to function mapping for the nominative and accusative Case markers (e.g., -ika ‘NOM’ and -lul

‘ACC’)—most likely also accompanied by difficulty with the distribution of hypocoristic -i, and it is possible that HLS may have more difficulty with accusative Case markers than nominative ones. Dropping the accusative Case marker more frequently than the nominative Case marker would actually mimic adult monolingual Koreans’ speech; thus, lower proficiency HLS may appear more “native-like” in this manner, yet we would also expect them to drop accusative (and nominative) Case markers without being sensitive to pragmatic considerations (i.e., in instances where Case drop would actually be inappropriate). To examine how HLS actually use Case markers, I conducted a study with fourteen HLS of Korean (seven males and seven females) aged 19 to 29 years old (mean age: 24 years old). They were divided into a low and high proficiency group (nine speakers and five speakers, respectively), based on their scores on a Korean cloze test (Kim & Montrul 2004a, 2004b). There were two control groups. The first group consisted of six Korean-dominant bilingual speakers of Korean in Canada (four males and two females), aged 19 to 26 years old (mean age: 22 years old). The length of stay in Canada for the Korean-dominant bilinguals (henceforth, KS-Can) was two months to seven months (mean: five months). Given that the KS-Can may have been susceptible to transfer from English and may have already begun to undergo attrition, another group of eight monolingual/Korean-

50

dominant bilingual41 speakers in Korea (henceforth, KS-Kor) (2 males and 6 females), aged 18 to 34 years old (mean age: 24 years old) were also tested. The latter group had relatively little exposure to English and had had fewer opportunities or situations requiring the use of English than the former.

The task was an oral elicited narration task, where participants were shown pictures in a wordless picture book42 and asked to describe what was happening in the pictures in Korean. The task was audio recorded, transcribed, coded, and analyzed for the productivity of morphemes, complexity of structures, and accuracy in morphemes and structures produced, similar to the methodology used by Kanno, Hasegawa, Ikeda, and Ito (2005) for Japanese. Structural complexity was measured by counting the number of clauses per utterance. Clauses modifying nouns (i.e., relative clauses) were not counted as a separate clause, since adjectives appear in relative clause structures in Korean (52) and very few morphologically simple adjectives exist in Korean (see M.-J. Kim 2007 for further details); thus speakers who were more liberal in their use of “adjectives” would have had inflated numbers for the number of clauses produced otherwise.

(52) [Khi khu-Ø-n] ai Height big-PRES-REL child ‘Tall child’

Accuracy was calculated based on corrections made by an adult Korean-dominant bilingual.43 “Errors” consisted of grammatical errors and discourse phenomena, such as afterthoughts and null pronouns which were not licensed (i.e., illicit topic drop). For example, in (53) the PP mun-ey ‘at the door’ is added as an afterthought and follows the verb, even though the verb is generally supposed to be in sentence-final position. (54) was produced as a description of a picture where a woman outside a bakery has a plate of cookies and a dog is begging for food in front of her. A group of children are copying the dog and also begging; in the next picture they each have a cookie in their mouths and are walking somewhere together in a line. (54) is

41 Due to Korea’s push in international affairs, many Koreans currently in the workforce today are required to have at least basic knowledge of English and English is now taught from pre-school on. Koreans’ comprehension is generally better than their production of English, and most of the participants in the current study described their overall competence in English to be at a beginner level.

42 The same picture book was shown to the HLS and the KS-Can, but a different picture book was shown to the KS-Kor, due to the lack of availability of the first book.

43 Some speakers were also rated by a second judge, but the judge was not able to complete scores for all speakers.

51

pragmatically odd for several reasons. With the complementizer -se ‘because’ in the subordinate clause, the sentence is interpreted as ‘Since the lady gave us (some) bread, (we’re) following Carl somewhere.’ The subject of the subordinate clause is acwumma ‘lady’, but the subject of the following clause is different (i.e., wuli ‘we’ are the ones who left and not the woman); thus, this would be a case of inappropriate topic drop. The sentence is only pragmatically appropriate if the subject wuli ‘we’ is overt.

(53) W uli-nun Carl-ilang cangnan-ul ha-ko iss-Ø-ta, mun-ey. We-TOP Carl-COMIT trick-ACC do-KO be-PRES-DECL door-at ‘We are playing around with Carl—at the door. (E10)

(54) [[Acwumma -ka wuli-hanthey ppang-ul cwe-se ][ Ø Carl ttala lady-NOM we-to bread-ACC give-because (we) Carl follow

eti ka-n-ta.] somewhere go-PRES-DECL ‘(The) woman gave us bread, so (we) are following Carl somewhere.’ (E10)

Some other errors included choosing the wrong lexical item (e.g., ccoch- ‘chase away’ instead of ccocha-ka ‘chase after (lit: chase go)’ and using the plural marker twice in the same word, as in

ay-tul-tul-i ‘child-PL-PL-NOM’. The results are given in Table 2–Table 4.

Table 2. Complexity of Structures Produced: Mean Clauses Per Utterance

Clauses Utterances Clauses/Utterance

Group N M SD M SD M SD

Low HLS 7 33.0 13.9 22.3 14.9 1.61 0.43

High HLS 5 25.8 5.7 16.6 0.9 1.56 0.39

KS-Can 6 28.7 9.3 16.8 1.8 1.68 0.39

KS-Kor 8 37.4 20.0 15.8 3.5 2.31 1.04

52

Table 3. Accuracy of Structures Produced: Mean Error-Free Clauses (EFC) Per Utterance

EFC Utterances EFC/Utterance

Group N M SD M SD M SD

Low HLS 7 13.7 10.8 22.3 14.9 0.61 0.30

High HLS 5 10.2 5.6 16.6 0.9 0.62 0.36

KS-Can 6 22.8 5.9 16.8 1.8 1.35 0.23

KS-Kor 8 30.5 15.0 15.8 3.5 1.88 0.72

Table 4. (In)Accuracy of Structures Produced (Mean Errors)

Errors/Clause Errors/Utterance

Group N M SD M SD

Low HLS 7 1.09 0.53 1.77 0.90

High HLS 5 0.92 0.42 1.50 0.95

KS-Can 6 0.25 0.15 0.45 0.31

KS-Kor 8 0.22 0.23 0.55 0.57

The HLS and the KS-Can performed similarly in terms of the complexity of their structures; the KS-Kor produced more complex structures than all other groups. However, the KS-Can and KS-Kor did not differ in terms of accuracy – both groups were noticeably more accurate than the HLS. There was high individual variation in the structural complexity of the elicited data, thus it would be more accurate to examine the errors performed per clause, rather than per utterance.

Similar to the Japanese HLS in Kanno et al.’s (2005) study, there was less code-switching than might have been expected. Only 6% of HLS’ “errors” and 2% of the KS-Can’s “errors” involved code-switching (e.g., push-up in (55)), whereas the KS-Kor did not code-switch at all.44 Both groups of HLS had much higher rates of Case drop than the KS-Can and KS-Kor. Nouns where

44 Although code-switching can sometimes be evidence of advanced proficiency in both languages, this was not the case for the HLS’ use of code-switching in the present study. Code-switching appeared to be used to fill gaps in the HLS’ vocabulary (e.g., park for kongwen ‘park’), whereas the KS-Can used code-switching to fill lexical gaps in Korean (e.g., khwukhi ca ‘cookie jar’—cookie jars are not used in Korea) or with words which were already (partial) borrowings from English (e.g., chokholeyt wuyu ‘chocolate milk’ instead of chokho wuyu ‘chocolate milk’).

53

Case markers were replaced with Contrastive Case (i.e., -nun) or other types of particles (e.g., -hanthey ‘to’) which cannot co-occur with nominative or accusative Case markers were not counted as instances of Case drop.

(55) Ay -tul-i Carl-hanthey cap-hye-kaciko push-up-ul hay-ss-ta. Child-PL-NOM Carl-to catch-PASS-because push-up-ACC do-PST-DECL ‘The children were caught by Carl, so they did push-ups.’ (E1)

If a particle or Case marker was used incorrectly, this was classified as a “wrong Case marker”, and if no particles or Case markers were produced on the noun, this was counted as an instance of a “missing Case marker”. “Wrong forms” were instances where the incorrect phonological form of a Case marker was used. The types of Case errors produced are summarized in Table 5.

Table 5. Mean Proportion of Case-related Errors

Missing Case Wrong Case Wrong form

Group N M SD M SD M SD

Low HLS 7 0.39 0.27 0.04 0.09 0.00 0.00

High HLS 5 0.19 0.16 0.01 0.02 0.01 0.02

KS-Can 6 0.02 0.03 0.00 0.01 0.00 0.00

KS-Kor 8 0.04 0.03 0.01 0.02 0.00 0.00

From the table, we can see that although the HLS had higher rates of Case drop, they generally picked the correct Case marker and only had very slight difficulty in selecting the correct phonological form for the Case markers which had more than one form. It might be somewhat unexpected that the HLS had little difficulty with the form of the Case markers, but since the story was centred on the activities of a dog and a group of children, most of the sentences started

with kangaci ‘puppy’ or ay-tul ‘child-PL’, so there was not a lot of variation in the nouns used. In spontaneous speech, the number of these errors could rise. Low HLS appear to have difficulty with the syntax of structures with Case marking in Korean (by dropping Case 39% of the time); the high HLS dropped Case less often, but still much more often than the KS. This suggests that the HLS could have the incorrect impression that nouns are selected from the lexicon with or without Case markers overtly attached. If this is the case, we would expect them to have the structure in (16) (repeated here as (56)).

54

(56) CP wo IP C qp | NP I’ -ta 5 2 ‘DECL’ Mina-ka VP I ‘Mina-NOM’ 2 | NP V -ss 5 | ‘PST’ swukcey-lul hay- ‘homework-ACC’ ‘do’ ‘Mina did (her) homework.’

This may be unexpected given the literature that HLS have less difficulty with syntax than other areas (Montrul 2002, 2004, 2006, among others).

Although the HLS dropped Case markers more often than the KS, they followed the pattern of the KS in terms of the types of Case markers that were dropped, as shown in Table 6. Whereas nominative Case was marked fairly regularly, accusative Case was optional for the low HLS and genitive Case was rarely marked, even though genitive Case is the only Case that is marked regularly in English. The HLS’ Case drop followed the hierarchy of vulnerability to Case drop evidenced in the monolingual children in Y.-j. Kim’s (1997) study: Genitive > Accusative > Nominative. Partial transfer from English resulted in a higher frequency of Case drop across the board, rather than a preservation of the genitive Case alone.

Table 6. Mean Proportion of Case Drop

Nominative Accusative Genitive

Group N M SD M SD M SD

Low HLS 7 0.29 0.32 0.47 0.37 0.67 0.00

High HLS 5 0.09 0.14 0.27 0.13 0.67 0.58

KS-Can 6 0.01 0.02 0.03 0.04 0.13 0.25

KS-Kor 8 0.02 0.04 0.01 0.03 0.22 0.25

55

To summarize, the results show that Case was almost completely optional for the low HLS, which suggests that they may have a non-target-like structure associated with Case markers, even though Case markers appear in the correct place (i.e., following the noun). The high HLS did not have as much difficulty as the low HLS in producing Case markers. All HLS did not struggle with choosing the correct phonological form of the Case markers and they performed similarly to the KS in dropping Case markers according to the following hierarchy: Genitive > Accusative > Nominative, which shows that if there was any influence from English, it resulted in the lower proportions of Case marking overall; there was no boost in the number of genitive Case markers marked overtly. Another area of the nominal domain which could cause potential difficulty for HLS is the use of the plural marker -tul and classifiers, which will be discussed below.

3.3 -Tul ‘PL’ and Classifiers: Introduction

English and Korean use different strategies for marking the plural. In English, a noun can only refer to a plural number of entities if -s 'plural' is present, as in (57) below (the structure of the nominal phrase is given in (58)). In Korean, a bare noun45 is ambiguous between a singular and a plural reading, as illustrated in (59) below.

(57) Helen bought magazine*(s).

(58) NumP 3 Num NP | 5 -s magazine

(59) Hyen a-ka capci-lul sa-ss-ta. Hyun-Ah-NOM magazine-ACC buy-PST-DECL ‘Hyun-Ah bought a magazine/magazines.’

Although English -s is obligatory in plural contexts, Korean has both a plural marker (-tul) and a classifier system for counting nouns. Korean -tul can occur on non-nominal elements as well as nominal ones and is obligatory, optional, or disallowed, depending on the context. For example, in (60) and (61) -tul attaches to verbs and adverbs respectively to create a distributive reading.

45 I will refer to nouns with or without Case marking as “bare” nouns.

56

(60) Chinkwu -(tul)-i nolle-tul ka-ss-ta. friend-(PL)-NOM play- PL? go-PST-DECL ‘(My/your/his/her/their) friends went out (to play/on vacation) (separately).’

(61) Haksayng -(tul)-i il-ul ppali-tul hay-ss-ta. student-(PL)-NOM work- ACC fast-PL? do-PST-DECL ‘The students (each) worked quickly.’

These properties of -tul call into question whether -tul is a true plural marker; if -tul is used with non-nominal elements and is not always required in plural contexts, it could be marking something other than plurality. However, since “non-nominal” and nominal -tul are so different in terms of meaning and distribution, most linguists consider them to be separate morphemes (Sohn 1999)46. Thus, for the purposes of the present study, I will focus on nominal -tul alone and leave non-nominal -tul aside for the purposes of the present study and simply note that non-nominal -tul may cause extra confusion for HLS as they learn the distribution of nominal - tul, which is also sensitive to animacy and specificity, as will be discussed in further detail below.

The use of -tul is not always obligatory because Korean also uses classifiers to count nouns. It has been argued that nouns in Korean are all mass nouns and that classifiers are needed to individuate properties into countable units.

(62) Ku-nun chinkwu-ka manh-ass-ta. He-CONTR friend-NOM be.many-PST-DECL ‘He had many friends.’

(63) Ku-nun chinkwu twu myeng-i iss-ess-ta. He-CONTR friend two CL-NOM exist-PST-DECL ‘He had two friends.’

Chierchia 1998, Borer 2005, among others, have suggested that plural markers and classifiers occur in complementary distribution, such that a language that has a plural marker will not have classifiers, and vice-versa. However, languages which have both plural marking and classifiers exist (e.g., Armenian, Japanese, Chinese, and Korean). For Armenian (Borer 2005, data from M. Siegler), there is evidence to suggest that even though plural marking and classifiers both exist, they do not co-occur in the same noun phrase, thus salvaging the generalization that plural

46 Sohn (1999) refers to non-nominal -tul as “particle -tul”, and nominal -tul as “suffix -tul".

57 marking and classifiers are in complementary distribution cross-linguistically. However, since - tati in Japanese and -men in Chinese co-occur with classifiers, it has been argued that -tati (Nakanishi & Tomioka 2004) and -men (Cheng & Sybesma 1999) are collective markers rather than markers of plurality.47 However, Korean -tul appears to be more like a plural marker than any of the morphemes discussed in the other languages.

When a classifier is present, the most common word order is for the head noun to precede the numeral and classifier (Martin 1992) (i.e., N > Card > Cl). In structures with classifiers, there is a syntactic restriction against having -tul attach to the classifier, but -tul is optional on the noun in colloquial speech (64). The structure is given in (65)48. I assume that cardinal numbers head their own phrases (CardPs) and classifiers also head their own phrases (ClPs). Plural marking would occur in a number phrase (NumP).

(64) Minwu-ka kay-(tul) tases mali-(*tul)-lul khiwe-ss-ta. Min-Woo-NOM dog-(PL) five CL-(*PL)-ACC raise-PST-DECL ‘Min-Woo raised five dogs.’

47 In general, it is difficult to make cross-linguistic generalizations on the nature of classifiers and any interaction with plural marking given that classifiers and plural marking work differently in each of the classifier languages. Differences exist in the types of nouns that can be modified by classifiers and the necessity of any genitive marking, as well as the possible interpretations which arise from classifiers and /or plural markers, and the word orders that are possible. See Carson 2000 and Nemoto 2005, and references cited therein, for further discussion on the differences in the properties of classifiers across classifier languages.

48 This structure differs from those proposed for Chinese and Japanese. Chinese classifiers generally precede the noun and Chinese is an SVO (head-initial) language. Japanese is an SOV (head-final) language and classifiers follow the noun, similar to Korean. However, classifiers in both Chinese and Japanese have different properties from those in Korean; thus, it is very difficult to develop a unified syntactic analysis for all three languages. Some linguists have proposed head-initial structures for the noun phrases of these languages. Some linguists assume that cardinal numbers and classifiers head separate phrases, whereas others, based on the observation that numerals and classifiers generally co-occur, prefer to analyze the numeral of classifier constructions as a clitic-like element as part of a numeral+classifier complex that occurs in the specifier of NPs or as the head of ClPs or Num(eral)Ps (Kawashima 1998).

58

(65) KP 3 ClP K 3 | CardP Cl -lul 3 # ‘ACC’ NP Card mali 5 # ‘CL’ kay tases ‘dog’ ‘five’ ‘five dogs’

Thus, it appears that Korean does not conform to the generalization that classifiers and plural markers do not coexist in the same structure. However, there is also evidence to suggest that the head noun is in a separate phrase from the classifier phrase. For example, it is also possible to attach another (accusative) Case marker to the head noun, as in (66) below.

(66) Minwu -ka kay-(tul)-(ul) tases mali-(*tul)-lul khiwe-ss-ta. Min-Woo-NOM dog-(PL)-(ACC) five CL-(*PL)-ACC raise-PST-DECL ‘Min-Woo raised five dogs.’

Especially in these cases, there has to be a break in intonation between the head noun and the classifier; thus, it appears that the head noun kay-tul-ul ‘dogs’ is being topicalized in a VP-external Topic phrase. If this is the case, then we can preserve the generalization that plural markers and classifiers do not co-occur in the same structure cross-linguistically. However, there is an asymmetry between subject nouns and object nouns. For example, it is not as natural to add an additional nominative Case marker to subject nouns, as in (67) below.

(67) Kay -(tul)-(i) tases mali-(*tul)-ka ttwie tanye-ss-ta. Dog-(PL)-(NOM) five CL-(*PL)-NOM run go.around-PST-DECL ‘Five dogs ran around.’

Although English does not have a generalized classifier system, there are measuring units which superficially pattern with classifiers. They are used for grouping entities or counting mass nouns. Regular plural marking applies in these cases: plural marking is obligatory on the measure words,

59

and is also obligatory on the head noun if it is a count noun such as student in (68), but is disallowed if the head noun is a mass noun such as tea in (69).49

(68) We met five group*(s) of student*(s).

(69) I drank five cup*(s) of tea(*s).

As we can see, plural markers and classifiers (and measure words) pattern differently in English and Korean. Plural marking is always obligatory with count nouns in English, but can be optional or even disallowed in certain contexts in Korean. English has measure words, but does not have classifiers, but Korean regularly makes use of classifiers for counting quantities of nouns. The factors affecting the distribution of nominal -tul – animacy, specificity, and the presence of classifiers – will be discussed in the following section.

3.3.1 Semantic Requirements and Syntactic restrictions on -Tul

Plural marking appears to be optional in Korean; however, -tul is sensitive to the animacy and specificity of the noun, and also to the presence of classifiers. Overt plural marking is much more likely to occur when the noun is animate (vs. inanimate), and it is most likely to occur with human nouns (Song 1975, Nemoto 2005; Biermann 1982, Downing 1996, for Japanese). Thus, we can think of a hierarchy of preference for -tul as follows: human nouns > other animate nouns > inanimate nouns.50 It sounds much more natural to have the bare noun, even if the inanimate noun tol ‘rock’ is intended to refer to more than one entity, as in (70) (structure of the nominal

phrase is given in (71))—it is slightly awkward to have -tul present. However, the sentence sounds natural if -tul is attached to the animate noun kangaci ‘puppy’ in (72) (structure of the nominal phrase is given in (73)). The sentence in (74) sounds best if -tul is present on the human noun chinkwu 'friend'.

49 Here, we are ignoring the “Universal Sorter” meaning where teas refers to different types of teas.

50 The hierarchy discussed here and the obligatoriness of -tul with nouns that are specific (as discussed below) is similar to the factors constraining Case drop discussed above. Croft (1990; see also Dixon 1979, Silverstein 1981, Corbett 2000, Croft 2003) proposes the following Animacy Hierarchy to account for differential Case-marking and number marking cross-linguistically: 1st person pronoun, 2nd person pronoun > 3rd person pronoun > proper names > human common N > non-human animate common N > inanimate common N.

60

(70) Hyenswu-ka tol-(#tul)-ul poa-ss-ta. Hyun-Soo-NOM rock-(#PL)-ACC see-PST-DECL ‘Hyun-Soo saw rocks.’

(71) KP 3 NP K 5 | tol -ul ‘rock’ ‘ACC’ ‘rock/rocks’

(72) Hyenswu-ka kangaci-(tul)-ul poa-ss-ta. Hyun-Soo-NOM puppy-(PL)-ACC see-PST-DECL ‘Hyun-Soo saw puppies.’

(73) KP 3 NumP K 3 -ul NP Num ‘ACC’ 5 # kangaci -tul ‘dog’ ‘PL’ (74) Hyenswu-ka chinkwu-tul-ul manna-ss-ta. Hyun-Soo-NOM friend-PL-ACC meet-PST-DECL ‘Hyun-Soo met friends.’

Although animacy can influence whether -tul occurs or not, the specificity of the noun also plays a role.

Plural nouns that are specific require -tul. A noun is most clearly specific when there is a demonstrative present, since demonstratives are deictic; thus, in these cases -tul is obligatory (75) (Chang 1996) or “generally required” (Sohn 1999).

(75) Hyeymi -ka ku ai-*(tul)-ul towa-cwu-ess-ta. Haemi-NOM that child-*(PL)-ACC help-give-PST-DECL ‘Haemi helped the/those children.’

Since demonstratives in Korean do not inflect for number, -tul only attaches to the noun (structure given below in (76)).

61

(76) KP 3 NumP K 3 -ul NP Num ‘ACC’ 3 # DemP N -tul 4 5 ‘PL’ ku ai ‘that’ ‘child’ ‘the/those children’

This contrasts with English, where plural marking occurs on both the demonstrative and the noun, as illustrated in (77). The structure of the nominal phrase is given in (78).

(77) Helen helped those teacher*(s).

(78) DP 3 D NumP | 3 those Num NP | 5 -s teacher

Since animacy only creates a preference (or dispreference) for -tul, we could expect the specificity of the noun to override any dispreference for -tul (e.g., in the case of inanimate nouns—we would expect specific inanimate nouns to still be marked with -tul). However, another factor interacts with animacy and specificity to determine whether -tul is realized or not when referring to different quantities of nouns—whether or not there is a classifier present.

There are two ways in which quantified nouns affect the occurrence of -tul – as a co-ocurrence restriction (or dispreference) and as syntactic restrictions on -tul. For example, the presence of a quantificational predicate/adjective such as manh- 'a lot' or a cardinal number such as twu 'two' triggers a dispreference for -tul (79) (structure with a cardinal number given in (80)). In these cases, -tul appears to be redundant, but is still optional (Kim 1987, Song 1988, Chang 1996, Sohn 1999). This differs from English, where quantificational adjectives and cardinal numbers trigger a requirement for plural marking on the noun (81) (structure given in (82)).

62

(79) Swuhi -ka manhun/twu chinkwu-(tul)-ul sakwi-ess-ta. Soo-Hee-NOM many/two friend-(PL)-ACC befriend-PST-DECL ‘Soo-Hee made a lot of/two friends.’

(80) KP 3 NumP K 3 # 1 NP Num -ul 3 # ‘ACC’ 1 CardP N i -tul 2 # ‘PL’ N/NP2 Card namca | | ‘man’ ei sey ‘three’ (81) Susan made a lot of/two friend*(s).

(82) CardP 3 Card NumP | 3 two Num NP | 5 -s friend

Although syntactically the structures are very different in the Korean and English, the linear word order is the same. In Korean, an alternative word order is also possible, where the cardinal number follows the noun (Sohn 1999), as in (83) (structure of nominal phrase in (82)). In these cases, -tul may be optional to some speakers on the noun, but it is always disallowed on the cardinal number.51 This structure is never grammatical in English (see (85)).

(83) Cenghwa-ka namca seys-(*tul)-ul ttala tanye-ss-ta. Jung-Hwa-NOM man three-(*PL)-ACC follow go.around-PST-DECL ‘Jung-Hwa followed three men around.’

51 Dropping the classifier can sound stylistically awkward to some speakers in certain contexts and is usually avoided, especially for quantities greater than nine and for non-humans (Sohn 1999).

63

(84) KP 3 CardP K 3 # NP Card -ul 5 # ‘ACC’ namca seys ‘man’ ‘three’ ‘three men’

(85) *Mark saw men three.

However, in Korean, when a cardinal number is present, classifiers are generally used as well,52 but there is a syntactic restriction on the usage of -tul when classifiers are present.

As mentioned above, -tul cannot attach to classifiers, but is optional on the head noun. (64) and (65) are repeated here as (86) and (87).

(86) Minwu -ka kay-(tul) tases mali-(*tul)-lul khiwe-ss-ta. Min-Woo-NOM dog-PL five CL-(*PL)-ACC raise-PST-DECL ‘Min-Woo raised five dogs.’

(87) KP 3 ClP K 3 | CardP Cl -lul 3 # ‘ACC’ NP Card mali 5 # ‘CL’ kay tases ‘dog’ ‘five’ ‘five dogs’

This is also true in other word orders, such as when the head noun follows the cardinal number 53 and the classifier (which has genitive Case marking) (i.e., Card Cl-GEN N), as in (88).

52 In general, only human nouns can appear with cardinals alone; otherwise the classifier is obligatory. In other words, the classifier can only be dropped for human nouns.

53 -Tul appears to sound more natural with higher numbers. Thus, although (86) may sound stylistically awkward to some speakers, many more speakers will agree that (i) below sounds perfectly fine.

64

(88) Minwu-ka tases mali-(*tul)-uy kay-(tul)-lul khiwe-ss-ta. Min-Woo-NOM five CL-(*PL)-GEN dog-(PL)-ACC raise-PST-DECL ‘Min-Woo raised five dogs.’

In these cases, I assume the head noun kay ‘dog’ is co-indexed with an empty noun in its KP complement (see (89)). This type of structure is roughly analogous to English measure phrases, except that plural marking is different in the two languages.

(89) KP1 3 NumP K1 3 | NP1 Num -ul 3 | ‘ACC’ 2 1 KP N i -tul 3 # ‘PL’ ClP K2 kay 3 # ‘dog’ CardP Cl -uy 2 # ‘GEN’ N/NP2 Card mali | | ‘CL’ ei tases ‘five’ As can be seen from the above discussion, several conditions interact in determining whether the

use of -tul ‘PL’ is felicitous or not. -Tul is preferred with animate nouns vs. inanimate nouns, and it is obligatory with nouns that are specific. When a classifier is present, -tul is optional on the head noun. The classifier is also optional with human nouns. Thus, although plural -s is obligatory in English, HLS must learn that the distribution of -tul is much more restricted (i.e., it only occurs in a subset of the instances where English -s occurs). Since there are numerous contexts where -tul is disallowed or dispreferred in Korean, it would be difficult to learn the restrictions constraining the usage of -tul (e.g., as related to animacy, specificity, whether quantifiers or classifiers are present, etc.) without negative evidence. In addition, HLS of Korean would also have to learn how to produce appropriate structures (with or without classifiers) when

(i) Minwu-ka payk mali-(*tul)-uy kay-tul-lul khiwe-ss-ta. Min-Woo-NOM one hundred CL-(*PL)-GEN dog-(PL)-ACC raise-PST-DECL ‘Min-Woo raised one hundred dogs.’

65 counting in Korean. Formal instruction on Korean would not help in the former task, since textbooks and instructors often simplify their explanations of -tul. A Korean language professor at a major Canadian university mentioned that he only introduces -tul as a regular plural marker, equivalent to -s in English, because it would be “impossible” to teach a non-native speaker or an HLS the intricacies of the distribution of -tul. Thus, even if HLS were to receive formal instruction, it would not help them in distinguishing the particular properties of -tul that differ from those of English -s. English plural marking is summarized in Table 7, and the factors constraining the usage of -tul in Korean are summarized in Table 8.

Table 7. Summary of Plural Marking in English

Type of noun Plural marking

All count nouns Obligatory PL marking

Noun in Measure phrase Numeral Measure-PL of N-PL

Table 8. Summary of Variables Conditioning Plural Marking in Korean

Condition Acceptability of or preference for plural marking

Animacy Human N, Animate N > Inanimate N

Specificity Specific N*(-tul); Non-specific N(-tul)

Quantificational Structure N(-tul) Num Cl (*-tul)-Case Num Cl(*-tul)-GEN N(-tul)-Case

3.3.2 Previous Literature on the Acquisition of Number and Classifiers

Parodi et al. (1997, 2004) investigated the L2 acquisition of L1 functional categories by L1 speakers of typologically different languages – they tested the L2 German of L1 speakers of Korean, Turkish, and Romance languages. Unlike Wakabayashi, Parodi et al. assume that definiteness and plurality head different phrases (D and Num, respectively). They found that the L2 production of plural inflection in informal oral interviews in longitudinal and cross-sectional studies differed according to the regularity of plural inflection found in the speakers' L1s. Native speakers of Turkish and Romance languages, whose L1s mark plural regularly, marked plural inflection in L2 German more regularly than native speakers of Korean speakers, who produced plural inflection the least often because the Korean plural marker (i.e., -tul) is optional, but

66 dispreferred in the presence of a numeral or quantifier (Kim 1987:894). However, Parodi et al. note that lexical expressions of plurality (e.g., numerals and quantifiers) were used more often than plural inflection by all L2 learners. This makes it difficult to determine whether the L1 Korean speakers produced plural inflection less often because plural marking is optional in Korean, as Parodi et al. claim, or if their initial strategy was to use numerals and quantifiers in plural contexts and then their L1 influenced them to avoid also marking plural overtly on the noun, or if they were simply following the general trend to use numerals and quantifiers to indicate plurality (vs. overt plural inflection). However, since the use of lexical expressions of plurality was a strategy used by all L2 learners, we can assume that even if the L1 Korean speakers were simply adopting this general strategy of substituting lexical markers of plurality for nominal inflection, their L1 tendency to avoid plural inflection when lexical markers of plurality are present most likely caused more of them to persist in this stage than the other L2 learners (if this were truly a “stage” for them as well). Thus, these speakers appear to be obeying the grammar of their L1 rather than their L2.

Shin and Milroy (1999) studied the acquisition of plural marking in Korean and English (as well as other English grammatical morphemes, which will not be discussed here) by simultaneous and sequential bilinguals in New York City who had acquired Korean at home in Korea or the US or Argentina (one speaker), but learned English in school. These six and seven year olds did not mark plural regularly on English nouns on an elicited production task, which Shin and Milroy attribute to the optionality and “infrequency” of -tul on Korean nouns. The children produced the following responses:

(90) swupak han kay N > Card > Cl watermelon one CL ‘one watermelon’

(91) swupak hana N > Card watermelon one ‘one watermelon’

(92) *han kay swupak Card > Cl > N one CL watermelon ‘one watermelon’

(93) ? han swupak Card > N one watermelon ‘one watermelon’

67

They consider the word order in (93) to be a “borrowing” from English, since monolingual Korean-speaking children have not been found to produce this word order, despite its limited acceptability with certain nouns (i.e., it is more acceptable with human nouns than other types of nouns; see Sohn 1999). The children did not produce classifiers very often, and when they did, they only produced the general classifier kay (which should be restricted to inanimate nouns), regardless of the animacy of the noun, which is an error that monolingual Korean-speaking children produce at much earlier stages of acquisition. Since the bilingual children made the same types of semantic errors associated with classifiers that monolingual Korean-speaking children make, but they also produced word orders which are more common in English than in Korean, unlike monolingual Korean-speaking children, Shin and Milroy suggest that the difficulty the children had with quantified structures may be a more general difficulty with Korean’s SOV word order, compared to the SVO word order of English, which could be a difficulty with word order alone, rather than with the classifier. Nevertheless, their results indicate a difficulty in the acquisition of both semantic (e.g., infelicity of using kay ‘CL’ with human nouns) and syntactic restrictions on classifiers (e.g., a classifier cannot precede the noun that it modifies unless it has a genitive Case marker) by the simultaneous and sequential bilingual children. The bilinguals also evidenced difficulty in acquiring English inflection, but whereas these errors will likely disappear, Shin and Milroy speculate that the bilinguals will most likely weaken in their Korean language ability progressively as English becomes their dominant language. We can thus assume that, at least for this particular group, Korean-English bilinguals who switch their language of dominance to English may never fully acquire Korean plural marking and structures involving classifiers.

3.3.3 Experiment (Suh 2006)

Given that other studies have suggested that HLS do not have great difficulty with the syntax of their HL (Montrul 2002, 2004, 2006, among others), in order to determine whether there was a difference in HLS’ understanding of the semantic requirements and syntactic restrictions on the usage of -tul ‘plural’ in Korean, in Suh (2006) I tested 14 adult HLS of Korean in Canada (divided into low and high proficiency groups) and compared them to a control group of 9 Korean-dominant bilingual speakers in Canada (henceforth, KS-Can).

68

I expected HLS with low input in Korean and/or low ages of onset of exposure to English to not use classifiers and to overuse -tul—i.e., without being sensitive to animacy or specificity. If a classifier is used, I expected these speakers to incorrectly allow -tul to occur on the classifier, because since Korean is a head-final language, these speakers may assume that the rightmost element is the head of the phrase even though it is not the head noun. This is possible because classifiers sometimes appear to function as head nouns (i.e., they occur alone, without the head noun) (see Koo 2008 for the grammaticalization of classifiers in Korean). For example, at a restaurant it is common for the server to ask the following question (94), where the honorific human classifier pun is used without a head noun (e.g., sonnim ‘guest’).

(94) Twu pun-i-sey-yo? Two HON.CL-be-HON-DECL ‘For two? (lit: Are there two of you?)’

Exposure to these types of sentences could lead HLS to produce sentences with just classifiers and no head nouns, such as in (95).

(95) *Twu pun-tul-ul manna-ss-eyo. Two HON.CL-PL-ACC meet-PST-DECL ‘(I) met two people.’/‘I met them.’

(96) * Kay twu mali-tul-i cina-ka-ss-e. Dog two CL-PL-NOM pass.by-go-PST-DECL ‘Two dogs passed by.’

If -tul were omitted, (95) would be grammatical if two people (higher in social status or age, etc.) had been mentioned in the previous discourse. In this case, the sentence is interpreted as ‘I met those two people.’ However, in a discussion without previous mention of anyone, (95) is ungrammatical not only because -tul is present, but also because the head noun is missing. (96) is ungrammatical because -tul cannot attach to classifiers. HLS who use Korean more often may use classifiers correctly if syntax is indeed intact, but transfer from English could lead them to still overgeneralize the use of -tul to inanimate nouns and classifiers. In addition, given the different choices of quantificational structures available in Korean, transfer from English could lead the HLS who do not use Korean often to produce more structures without classifiers or if they do produce classifiers, they would likely use classifiers prenominally (Num Cl N). The structures with prenominal classifiers are ungrammatical if the genitive Case marker is not attached to the classifier; HLS with less input in Korean would likely not be aware of this

69 requirement and would produce these ungrammatical structures. This would be transfer due to the similarity to the word orders of quantificational structures in English; it would be evidence for no transfer with respect to Case marking, given that the genitive Case marker is the only Case marker that is marked regularly on nouns.

The instructions for the experiment were provided in English to ensure that the HLS would understand how to complete the tasks. The experiment itself consisted of two paper-and-pencil tasks and an oral task. The first task consisted of a fill-in-the-blank task designed to elicit bare nouns, nouns with demonstratives, and nouns with classifiers (eight of each type of noun, using four animal nouns and four human nouns) in order to determine participants’ usage of -tul. The first section was designed to elicit neutral nouns (e.g., bare nouns). Participants were asked questions such as “Who did Yumi meet?”, followed by uninflected vocabulary items provided in brackets. Pictures depicting people or animals were provided below each question to show the subject that more than one entity was involved in the action. Below each picture was a sentence missing its verb phrase. Participants were required to fill in the blank. The text of the test item is given below in (97).

(97) Ywumi -ka nwukwu-lul manna-ss-ni? (haksayng) (mannata) Yumi-NOM who-ACC meet-PST-Q (student) (to meet) ‘Who did Yumi meet?’

Ywumi-ka ______. Yumi-NOM ‘Yumi….’

Vocabulary items were provided in brackets in order to control the particular object nouns that were produced (and to reduce the effect of inherent preferences for the use or omission of -tul with particular nouns), and also to ensure that even participants who had lower levels of proficiency in Korean would be able to complete the task. An example of the expected response is given in (98), where the object noun haksayng ‘student’ either has the plural marker -tul attached or not.

70

(98) Ywumi -ka haksayng(-tul)-ul manna-ss-e. Yumi-NOM student(-PL)-ACC meet-PST-DECL.IV ‘Yumi met (some) students.’

The second section elicited specific nouns (e.g., nouns with demonstratives). Participants were presented with an introduction explaining that they were supposed to tell Yumi what to do. Thus, when prompted with the question “What should I do here?”, participants were expected to answer the question with a command, using the vocabulary items provided. The accompanying picture (see (99) below) depicted two groups of almost identical-looking animals or people, in order to elicit the use of a demonstrative (e.g., i ‘this’ or ce ‘that over there’) to differentiate between the two groups. An example involving a singular entity was provided to ensure that participants knew how to answer the questions. An example of the expected answer to (99) is given in (100), where the demonstrative ce ‘that over there’ is used with the noun provided (kaykwuli ‘frog’), and the plural marker is either attached to the noun or not.

(99) Yeki -se mues-ul hal-kka? (kaykwuli) (capta) here-at what-ACC do-Q (frog) (to catch) ‘What should (I) do here?’

______.

(100) Ce kaykwuli(-tul)-lul capa-la. That.over.there frog(-PL)-ACC catch-IMP.III ‘Catch those frogs (over there).’

The third section elicited quantified nouns (e.g., nouns with classifier phrases). In order to avoid presenting prompts with the test nouns with (or without) plural marking, only a general instruction was provided at the beginning of the section asking participants to answer the question of “How many…”. A blank line was provided below a picture depicting plural entities;

71

the text of the test item is given in (101). As with the other sections, vocabulary items were provided. An example of the expected answer is given in (102), where the classifier used with animals (i.e., mali) is used, either with the plural marker -tul or not.

(101) (pyengali) (khiwuta) chick raise

Ywumi-ka ______. Yumi-NOM

(102) Ywumi -ka pyengali sey mali(*-tul)-lul khiwe-ss-e. Yumi-NOM chick three CL(-PL)-ACC raise-PST-DECL.IV ‘Yumi raised three chicks.’

In order to prevent questions from other parts of the test (presented in Korean) from affecting the results of the elicited production task, this task was always administered first. The purpose of the task was to test the sensitivity of the participants to the various types of nominals available in Korean. Given no prompts, speakers transferring English syntactic structure to Korean could overuse -tul as a plural marker (e.g., with classifiers) or they could apply English syntactic structure (e.g., the cardinal number, noun construction) to the Korean nominal phrases. There are several options for indicating quantities of nouns, as discussed in Section 3.3.

The second task consisted of an acceptability judgment task. Participants were presented with 32 short paragraphs. Each paragraph was followed by a test sentence in Korean. The sentences included nouns with demonstratives and nouns with classifier phrases. The object noun in the test sentence would either have or not have -tul attached to it. Participants were then asked to judge whether the test sentence was acceptable or not, based on the given context, based on a 5- point Likert scale, ranging from -2 to 2. Participants could rate the sentence as either: “unacceptable”, “sounds a little awkward—I wouldn’t say it like this”, “not sure”, “sounds okay in this context—I would say it like this”, and “perfect”. An example is provided in (103) below.

(103) Yesterday my son Dong-Yub spent the whole afternoon running around our backyard with a net. After five hours, he came in and proudly said to me,

“Nay-ka napi yeses mali-lul capa-ss-e!” ______I-NOM butterfly six CL-ACC catch-PST-DECL ‘I caught six butterflies!’

72

There were two versions of the test, and sentences were randomized (same order for all tests) and counterbalanced for -tul. Quantifiers and classifiers were avoided as much as possible in the stories in order to reduce the risk of priming the acceptability of the use or omission of -tul in the classifier constructions on other questions of the test. -Tul was never used. The acceptability judgment task always followed the elicited production task in order to prevent the priming of particular syntactic structures as well as the use or omission of -tul. The acceptability judgment task tests whether HLS are sensitive to the effect of specificity on the usage of -tul and the syntactic restrictions on -tul (i.e., its ability to attach to classifiers). Those analyzing -tul as a plural marker should use the plural marker regularly in all contexts, whereas the KS should show sensitivity to the semantic and syntactic restrictions on the usage of -tul, as illustrated in Table 9 below.

Table 9. Predicted Results for Production and Acceptance of - and -tul

Condition

Specific nouns Quantified nouns

Group - -tul - -tul

HLS *  * 

KS *   *

The oral task was an elicited narration task, where participants were shown pictures in a wordless picture book54 and asked to describe what was happening in the pictures in Korean. The task was audio recorded, transcribed, coded, and analyzed following the methodology of Kanno et al. (2005), as it related to the Korean data. The data was then analyzed for Case- and postposition-related errors, and other errors in general, in order to determine the HLS’ usage of the appropriate morphemes.

The results on paper-and-pencil elicited production and acceptability judgment tasks differed according to the task. All groups, including the control group, had unexpectedly low production of -tul in all contexts of the elicited production task. The only distinction was in the counting

54 The same picture book was shown to the HLS and the Korean-dominant bilingual speakers in Canada, but a different picture book was shown to the Korean-dominant bilingual speakers in Korea.

73 contexts, where all groups correctly avoided the use of -tul almost completely (only one low HLS produced -tul in this context). However, on the acceptability judgment task, there was a clear difference between the HLS and the control group. The low HLS had a strong preference for -tul in specific contexts – they rejected sentences without -tul. However, in the counting contexts, both groups of HLS failed to discriminate between sentences with or without -tul. Thus, there appeared to be at least partial transfer from English, since classifier-like measure words in English are pluralized, as in three cups of coffee; full transfer would have predicted plural marking to also appear on the head noun, as in phrases such as five groups of boys. The variation in the NS’ responses also shows the importance of further research, particularly on the language input that HLS receive and are exposed to, given that the data from the KS-Can did not always follow what was expected by the literature.

3.4 Conjunctions and Comitative Case Markers in Korean

English and Korean also differ in the properties of their conjunctions. In English, the same conjunctions (e.g., and) can be used to conjoin nouns and clauses, but in Korean different conjunctions are used. Conjunctions in Korean are an interesting area of study because they are markers which attach directly to the noun (or to the verb, for clausal conjunctions). When any other markers, such as plural marking and postpositions are present, these markers occur at the end (i.e., the right edge of the nominal phrase) (104). They occur in complementary distribution with all Case markers other than genitive Case, as illustrated in (105) and (106), respectively. Thus, they can also be called comitative Case markers (Lee & Ramsay 2000), which is the term I will adopt.

(104) Nala -ka ai-tul(-eykey)55-hako elun-tul-eykey phyenci-lul sse-ss-ta. Nara-NOM child-PL(-to)-COMIT adult-PL-to letter-ACC write-PST-DECL ‘Nara wrote a letter/letters to the children and the adults.’

(105) Nay -ka [kongchayk-kwa-(*lul) yenphil-ul] sa-ss-ta. I-NOM notebook-COMIT-(*ACC) pencil-ACC buy-PST-DECL ‘I bought a notebook and a pencil.’

55 The postposition -eykey can generally only precede the comitative Case marker (on the same noun) in colloquial language. In conjoined structures, the postposition is usually omitted and only present on the second/final conjunct.

74

(106) Nay -ka [kunye-wa-*(uy) ipyel-ey tayhayse] sse-ss-ta. I-NOM her-COMIT-*(GEN) separation-at about write-PST-DECL ‘I wrote about (my) break-up with her.’

In (105), it is ungrammatical to have the accusative Case marker and the comitative Case marker -kwa occur on the same noun. However, in (106), the genitive Case marker and the comitative Case marker -wa both attach to the same noun. Without the comitative Case marker, the sentence is interpreted as ‘I wrote about her break-up’ and without the genitive Case marker, the sentence is interpreted as ‘I wrote with her about break-ups’; thus, both markers are necessary.

Korean conjunctions differ in type (i.e., nominal vs. clausal), morphological form, and their syntactic environments. There are three main comitative Case markers: -wa, -hako, and -lang. Unlike the invariant free morpheme and in English, nominal conjunctions (Case markers) in Korean are suffixes. Two of the Case markers (-wa/kwa and -lang/ilang) are phonologically conditioned, whereas the other is not (-hako), and the appropriateness of the particular type of comitative Case marker used is dependent on the particular level of speech used. In English, and attaches to the second56 conjunct, whereas in Korean the comitative Case markers can attach to the first conjunct and any other non-final conjuncts present. Alternatively, it is possible to attach the comitative suffix to the second conjunct to create the meaning of ‘with’; 57 however, the suffix cannot attach to inanimate nouns to mean ‘with’ (i.e., the suffix is not appropriate for ). Korean clausal conjunctions consist of the complementizer suffix -ko ‘and’, which attaches to the first conjunct, and the free morpheme kuliko ‘also’, which attaches to the second conjunct. We will examine the differences between the conjunctions in more detail below.

56 I use “first” and “second” to refer to “first and any non-initial conjuncts” and “final conjunct”, respectively (i.e., “first” and “second” are appropriate when only two conjuncts are being conjoined, but the same generalizations are true when there are more than two conjuncts.

57 Sohn (1999) labels the morpheme differently according to its syntactic position: “connectives” refer to the former case, where the conjunction attaches to the first noun, and “comitatives” refer to the latter case, where the conjunction attaches to the second noun, and he notes that the morphemes can have a reciprocal reading with certain types of verbs, as shown below in (i). I will refer to both types of morphemes as comitative markers for ease of exposition. (i) Ywuna-ka Hyencwung-kwa kyelhon-hay-ss-ta. Yuna- NOM Hyunjoong- COMIT marriage-do-PST-DECL ‘Yuna and Hyunjoong got married.’

75

3.4.1 Comitative Case Markers

The first difference between English and and the Korean comitative Case markers is in their morphological form. Whereas English and is a free morpheme, the Korean comitative Case markers are suffixes. These Case markers occur in complementary distribution with other Case markers, other than the genitive. Two of the comitative Case markers (i.e., -wa/kwa and -lang/ilang) are phonologically conditioned, whereas the other (i.e., -hako) is not. -Wa and -lang occur with bases ending in vowels (e.g., Miya in (107)) and -kwa and -ilang occur with bases ending in consonants (e.g., Minceng in (108)). For ease of exposition, I will refer to the comitative Case markers by the forms that occur with bases ending in vowels (i.e., -wa and -lang). Comitative Case markers are generally accompanied by the adverb hamkkeh ‘together’.

(107) Miya -wa hyeymi-ka hamkkeh nol-ass-ta. Miya-COMIT Haemi-NOM together play-PST-DECL ‘Miya and Haemi played together.’

(108) Minceng -kwa swumi-ka hamkkeh nol-ass-ta. Minjung-COMIT Sumi-NOM together play-PST-DECL ‘Minjung and Sumi played together.’

The main distinction amongst the comitative Case markers lies in the level of speech that they are appropriate for. There are four main levels of speech in Korean (i.e., sets of grammaticalized markers of formality and social distance) (Sohn 1999, Choo & Kwak 2008, among others), which are distinguished by the complementizers58 they take. Their corresponding complementizers are: -(su)pnita (deferential – “level I”), -a/eyo (polite – “level II”), -ta (plain – “level III”), and -a/e (familiar59 – “level IV”). Levels I and II are the most formal registers and are generally used to show respect to people older or higher in status than the speaker. The levels are generally interchangeable, but level I is most commonly used by men and level II is most commonly used by women (Choo & Kwak 2008, among others). Levels III and IV are used towards people who are younger or very close (in social distance) to the speaker. Level III is

58 Certain lexical items are also deferential (e.g., cinci ‘food/meal: deferential’ vs. pap ‘food/meal: neutral’), but these can be used irrespective of the level of speech.

59 I follow the terminology used by Sohn (1999), but I rename -a/e as ‘familiar’, rather than ‘intimate’. In Sohn (1999), ‘familiar’ refers to the complementizer -ney, which is a form generally restricted to an older generation of speakers.

76 most commonly used in writing, as is level I, and levels II and IV are the most common levels of speech used in discourse. -Wa is the most formal marker and is generally not used with level IV. -Hako is not as formal; it has the widest distribution and can be used with all levels of speech. -Lang is the most casual marker and is generally restricted to informal speech (i.e., levels III and IV), but can generally be used with all levels of speech other than level I (level of speech 60 shown after DECL).

(109) Miya -lang hyeymi-ka kathi nol-ass-ta. Miya-COMIT Haemi-NOM together play-PST-DECL.III ‘Miya and Haemi played together.’

(110) Minceng -kwa swumi-ka kathi nol-ass-supnita. Minjung-COMIT Sumi-NOM together play-PST-DECL.I ‘Minjung and Sumi played together.’

(111) Mina -hako swumi-ka kathi nol-ko iss-e-yo Mina-COMIT Sumi- NOM together play-KO be-PRES-DECL.II ‘Mina is playing with Sumi.’

Syntactically, I follow Cowper and Hall (2000) in analyzing conjunction phrases as monadic conjunction phrases (&Ps) which are headed by conjunctions (or comitative Case markers) that can take a conjunct as their complement. The &P attaches as an adjunct of the first conjunct. Whereas English and heads an &P with the second conjunct as its complement (111) (modified from Cowper & Hall 2000’s (15a)), in Korean, the comitative Case markers head an &P with the first conjunct as its complement (113).

(112) DP wo DP &P 6 ei sticky keys & DP and 6 missing strings

60 Sohn (1999) describes -kwa as a formal conjunction and -hako and -lang as informal conjunctions. Choo and Kwak (2008) describe -kwa as being appropriate for written/formal language, -hako as being appropriate for colloquial language, and they describe -lang as having a “more spoken flavour” (Choo & Kwak 2008, p. 258). Choo and Kwak also mention the infrequent conjunction mich, which is generally restricted to quite formal written language, and which is incompatible with human nouns. Due to the infrequency of this conjunction and its incompatibility with the other conjunctions, I leave a study on this conjunction for future research.

77

(113) KP wo &P KP ru ru NP61 & NP K 5 | 5 | Miya -wa hyeymi -ka ‘Miya’ ‘and’ ‘Haemi’ ‘NOM’ ‘Miya and Haemi’

It is possible for nominal conjunctions to be attached to the second conjunct for nouns in subject position, but in these cases the conjunctions are interpreted as ‘with’, as shown by the English gloss in (114) below.

(114) Hyeymi -ka miya-wa hamkkeh nol-ass-ta. Haemi-NOM Miya-with together play-PST-DECL ‘Haemi played with Miya.’

(115) CP qp IP C qp | KP I’ -ta ei ri ‘DECL’ KP &P VP I ru 2 3 | NP K NP & AdvP VP -ass 5 | 5 | 5 5 ‘PST’ Hyeymi -ka miya -wa hamkkeh nol- ‘Haemi’ ‘NOM’ ‘Miya’ ‘COMIT’ ‘together’ ‘play’ ‘Haemi played with Miya.’

61 As mentioned above, nominal conjunctions and Case markers are in complementary distribution in Korean. Although the “topic” marker -nun is generally also in complementary distribution with Case markers, unlike other Case markers, it can co-occur with postpositions and conjunctions. (i) Nakato tway. Taman miya-wa-nun an tway. go.out become however Miya-with-TOP not become ‘(You) can go out. But not with Miya.’

78

More than one noun can be marked with -wa in a sentence, but the first (116) or the final noun (117), depending on the intended meaning of the sentence, must be Case-marked with the nominative Case.

(116) Hyeymi-ka miya-wa swumi-wa minceng-kwa kathi nol-ass-ta. Haemi-NOM Miya-COMIT Sumi-COMIT Minjung-COMIT together play-PST-DECL ‘Haemi played with Miya, Sumi, and Minjung.’

(117) Hyeymi-wa miya-wa swumi-wa mincengi-ka62 kathi nol-ass-ta. Haemi-COMIT Miya-COMIT Sumi-COMIT Minjung-NOM together play-PST-DECL ‘Haemi, Miya, Sumi, and Minjung played together.’

All of the above comitative Case markers can be used with animate and inanimate nouns (e.g., chayk ‘book’ in (118)) when they mean ‘and’ (i.e., when they attach to the first conjunct), but there is a semantic restriction against the use of the comitative Case with an inanimate noun to mean ‘with’ (i.e., when the comitative Case marker attaches to the second conjunct). For example, in (119) below, when Yuna plays with the person Semy, -hako attaches to the animate noun saymi ‘Semy’ to mean ‘with’, but when Yuna plays with a ball, -hako cannot attach to the inanimate noun kong ‘ball’ (120) or to Yuna (121). Rather, the accusative Case marker -ul needs to attach to the inanimate noun kong ‘ball’, and the verb kaci-ko (lit: ‘get-and’) is added to convey the meaning that the ball is what Yuna played with.63

(118) Yuna-ka [chayk-hako yenphil-ul] sa-ss-ta. Yuna-NOM book-COMIT pencil-ACC buy-PST-DECL ‘Yuna bought a book and a pencil.’

62 The name ‘Minjung’ is modified here with the hypocoristic marker -i in order to make the sentence sound more natural.

63 It is possible for two inanimate nouns to be conjoined. In this case, the comitative Case marker can only attach to the first conjunct. In (i) and (ii) below, the subject noun has been dropped, but the verb philyo ‘need’ is one of the verbs in Korean which requires multiple (nominative) Case marking. (i) Kongchayk-hako yenphil-i philyo-hay-ss-ta. notebook-COMIT pencil-NOM need-do-PST-DECL ‘(I) needed (a) notebook(/s) and (a) pencil(/s).’ (ii) *Kongchayk-i yenphil-hako philyo-hay-ss-ta. notebook-NOM pencil-COMIT need-do-PST-DECL ‘(I) needed (a) notebook(/s) and (a) pencil(/s).’

79

(119) Yuna-ka Saymi-hako nol-ass-ta. Yuna-NOM Semy-COMIT play-PST-DECL ‘Yuna played with Semy.’

(120) #Yuna-ka kong-hako nol-ass-ta. Yuna-NOM ball-COMIT play-PST-DECL ‘Yuna played with a ball.’

(121) #Yuna-hako kong-i nola-ss-ta. Yuna-COMIT ball-NOM play-PST-DECL ‘Yuna played with a ball.’

(122) Yuna-ka kong-ul kaci-ko nol-ass-ta. Yuna-NOM ball-ACC take-and play-PST-DECL ‘Yuna played with a ball.’

The incompatibility between the inanimate noun and hako ‘with’ in (122) appears to lie in the fact that Yuna is an , but kong ‘ball’ is similar to an . According to the Coordination Constraint, unlike elements (e.g., an and an ) cannot be conjoined. For nouns with an -like role, the postposition -ulo/lo should be attached, as in (123), where pyektol ‘brick’ refers to the material that was used to build the house.

(123) Yuna -ka pyektol-lo cip-ul ci-ess-ta. Yuna-NOM brick-INSTR house-ACC build-PST-DECL ‘Yuna built a house with bricks.’

Thus, although Korean comitative Case markers can mean ‘and’ or ‘with’ depending on the position of the conjunct they attach to, the Coordination Constraint still applies and unlike constituents cannot be conjoined. In these instances, the inanimate noun receives accusative Case marking and the verb kaci-ko ‘take-and’ is added or the postposition -lo ‘with’ is attached to the noun. A summary of the properties of comitative Case markers will be presented in the following section once we have examined the properties of clausal conjunctions in Korean.

3.4.2 Clausal Conjunctions

In English, and can be used to conjoin clauses. In Korean, the clausal conjunction kuliko or the complementizer -ko is used to conjoin clauses. Kuliko is a free morpheme meaning ‘also’ that occurs at the left periphery of the second clause (124), whereas the complementizer -ko ‘and’ attaches to the verb of the first clause (125). Kuliko and -ko do not just mean 'and'; they can also mean 'after'. A more neutral way to express that more than one activity occurred (i.e., without

80 reference to any particular temporal ordering), is by attaching -ki-to ‘nominalizer-also’ to the verbal root (126). There are several other clausal conjunctions corresponding to and in English which will not be discussed here.

(124) A y-tul-i kongpu-lul hay-ss-ta. Kuliko nol-ass-ta. Child-PL-NOM study-ACC do-PST-DECL also play-PST-DECL ‘The children studied. They also/then played.’

(125) A y-tul-i kongpu-lul ha-ko nola-ss-ta. Child-PL-NOM study-ACC do-and play-PST-DECL ‘The children studied and (then) played.’

(126) A y-tul-i kongpu-to ha-ko nol-ki-to hay-ss-ta. Child-PL-NOM study-also do-and play-NMLZ-also do-PST-DECL ‘The children studied and (also) played.’

In summary, there are three ways in which Korean conjunctions differ from English ones: i) type, ii) morphological form, and iii) syntax. Unlike English, Korean conjunctions which attach to nouns are different from those that attach to clauses and their morphological forms also differ. The nominal conjunctions are comitative Case markers which are suffixal; -lang and -wa are phonologically conditioned, -hako is not. The Korean conjunctions differ in terms of the levels of speech with which they are appropriate for. The syntax of Korean conjunctions differ from those in English. English conjunctions attach to the second conjunct to form an &P, but in Korean conjunctions form an &P with the first conjunct; when they attach to the second conjunct, they mean ‘with’, but they cannot mean ‘with’ for nouns – kaci-ko ‘take-and’ must be used instead. The clausal conjunctions differ morphologically – one is a free morpheme and the other is a (bound) complementizer; the former attaches to the second conjunct, the latter attaches to the first conjunct. These aspects of Korean conjunctions make them susceptible to transfer from English, since all of these conjunctions would be replaced by and or with (or also) in English, which attaches to the second conjunct. Conjunctions in English and Korean are summarized in Table 10–Table 11. As noted above, conjunctions in English are the same for nouns and clauses, but the data is repeated in Table 11 for ease of exposition.

81

Table 10. Nominal Conjunctions: English vs. Korean

Condition English Korean

Morphological Free morpheme Suffix (e.g., -hako) status (e.g., and)

Phonologically No Yes for -lang and -wa; conditioned? No for -hako

Sensitive to No -wa: ok for spoken I, II, written I, II, III; register? -lang ok for spoken II, III, IV; -hako ok for all levels–spoken and written

Syntactic position N1 [CONJ N2] [N1-COMIT] N2-NOM ‘and’ or N1-NOM [N2-COMIT] ‘with’, but N2 cannot be ; [N1-COMIT] N2-ACC ‘and’

Table 11. Clausal Conjunctions: English vs. Korean

Condition English Korean

Morphological Free morpheme Free morpheme (e.g., kuliko ‘also’) status (e.g., and) or Suffix (e.g., -ko ‘and’)

Phonologically No No conditioned?

Sensitive to No No register?

Syntactic position Clause1 [CONJ Clause2] Clause1 [kuliko Clause2] or [Clause1-ko] Clause2

The data leads us to several questions, which will be discussed in the following section.

3.4.3 Conjunctions: Pilot Study

Given the restrictions on the contexts within which each of the various nominal and clausal conjunctions are used, we would expect conjunctions to be an area of difficulty for heritage language speakers to master. The general problems of incomplete acquisition and transfer lead us to the following questions:

82

1) Do HLS have difficulty in using comitative Case markers and clausal conjunctions appropriately?

2) Do HLS have the correct syntax for constructions involving conjunctions and comitative Case markers?

3) Do HLS use the correct form for comitative Case markers that are phonologically conditioned?

4) Do HLS understand the semantic restrictions and requirements on the usage of comitative Case markers?

Since and is used to conjoin nouns and clauses in English, we would expect HLS with strong transfer to overgeneralize the use of a particular conjunction for both nominal and clausal coordination. Since -hako has the widest distribution, we would expect the conjunction that is used to conjoin both nouns and clauses to be -hako (127). In addition, since the dummy ‘do’ verb is ha- in Korean, HLS could misanalyze -ha-ko ‘do-and’ to be the (monomorphemic) comitative Case marker -hako (128).

(127) *Kongpu -hay-ss-ta. Hako nol-ass-ta. study-do-PST-DECL with play-PST-DECL ‘(We) studied and played.’

(128) Kongpu -ha-ko nol-ass-ta. study-do-KO play-PST-DECL ‘(We) studied and (then) played.’

Errors such as using hako ‘and’ as a free morpheme would be expected of low proficiency speakers who do not recognize that separate clausal conjunctions exist and who are applying English syntactic structure to Korean by placing the second conjunct, rather than the first, in the complement position of the &P.

A less common error could be using the clausal conjunction kuliko ‘also/and’ as a nominal conjunction. Since children/L2 learners have a cross-linguistic preference for a one-to-one mapping of form to function (Slobin 1973), we would expect lower proficiency learners to perform these kinds of errors (i.e., using an incorrect free morpheme instead of the correct bound one) more frequently than higher proficiency learners. In (129), the use of the clausal

83

conjunction kuliko instead of one of the comitative Case markers, such as -hako or -lang is awkward.64

(129) #S aca kuliko wenswungi-lul pwa-ss-ta. lion also monkey-ACC see-PST-DECL ‘(I) saw a lion and a monkey.

Either strategy—using the comitative Case marker -hako ‘and’ inappropriately (i.e., in contexts

where another construction, such as -lul kaci-ko ‘-ACC take-and’, would be more appropriate) or using the clausal conjunction kuliko ‘also/and’ to conjoin both nouns and clauses—would also be expected if HLS’ grammars are undergoing simplification, since neither form is phonologically conditioned and a grammar where a single conjunction conjoins both nouns and clauses would be simpler than a grammar that uses separate conjunctions.

If HLS use -wa at all, we would expect lower proficiency HLS to either overgeneralize -wa to all conjuncts, including the final noun (130), which would normally take Case marking ((117) repeated here as (131)).

(130) *Hyeymi-wa miya-wa swumi-wa mincengi-wa kathi nol-ass-ta. Haemi-COMIT Miya-COMIT Sumi-COMIT Minjung-COMIT together play-PST-DECL ‘Haemi, Miya, Sumi, and Minjung played together.’

(131) Hyeymi-wa miya-wa swumi-wa mincengi-ka kathi nol-ass-ta. Haemi-COMIT Miya-COMIT Sumi-COMIT Minjung-NOM together play-PST-DECL ‘Haemi, Miya, Sumi, and Minjung played together.’

Since -hako is not phonologically conditioned, we would expect it to be the easiest to learn and use. I expect HLS to overgeneralize the use of -hako with , since with in English is not sensitive to animacy.

64 Choo and Kwak (2008) note that the clausal conjunction kuliko can be used to conjoin nouns in the following example (transcription, glosses, and emphasis my own): (i) Ccacang hana, ccamppong twul, kuliko thangswuyuk hana cwu-seyyo. Black.bean.noodles one spicy.seafood.soup two also sweet.and.sour.pork one give-IMP ‘One black bean noodles, two spicy seafood soups, and one sweet-and-sour pork, please.’ However, it should be noted that the example they provide is from what someone would say when ordering in a restaurant – a situation which often produces violations of regular grammatical rules (e.g., the acceptability of two waters (vs. the usual two glasses of water) when ordering), thus it is unclear how widespread this usage is. Informal interviews with native speakers of Korean revealed that kuliko is more acceptable in colloquial language when there is more than one object and someone is pointing and talking about each one in turn. Generally, minimally there must be more than one object in question.

84

The written register of English is slightly more formal than colloquial language, but the same conjunctions are used for both registers; thus, we would expect errors to abound in choosing appropriate nominal conjunctions for the level of speech being used (e.g., using -lang with the most formal level of speech, etc.). In addition, since many HLS only have one register (i.e., an informal register), they might not have even been exposed to the more formal conjunctions such as -wa. Anecdotally, Korean adults tend to use -hako more often than -lang, so even though -lang is the form used with more informal registers, HLS with little exposure to Korean would be expected to only use -hako. To the best of my knowledge, there has been no previous research on the acquisition of conjunctions in Korean by monolinguals or HLS or second language learners. I will discuss the results of a pilot study below.

The speech of the same speakers discussed above in Section 3.2.3 was transcribed and the speakers’ use of conjunctions was analyzed. Only their use of conjunctions meaning ‘and’ or ‘with’ were counted.65 Since the same conjunctions are used to conjoin nouns and clauses in English, Korean HLS with strong transfer from English would be expected to overgeneralize a particular comitative Case marker (e.g., -hako) to contexts where clausal conjunctions were required or a particular clausal conjunction (e.g., kuliko ‘also’) where nominal conjunctions were required. Table 12 and Table 13 show the means of the nominal and clausal conjunctions that were produced by the speakers, respectively.

Table 12. Mean Proportion of Nominal Conjunctions Produced

-wa ‘COMIT -hako ‘COMIT -lang ‘COMIT *kuliko Group N (more formal)’ (neutral)’ (casual)’ ‘also’

Low HLS 7 0.00 0.60 0.33 0.06

High HLS 5 0.11 0.35 0.54 0.00

KS-Can 6 0.98 0.00 0.02 0.00

KS-Kor 8 0.75 0.00 0.25 0.00

65 Other conjunctions such as ku taum-ey ‘afterwards (lit. ‘that following-at’)’ were used by some of the speakers in Korea, but these were not counted in the overall count, since they were not used by the HLS. Once again, it should be noted that the group in Korea were describing pictures from a different book than that shown to the HLS and the KS-Can.

85

Table 13. Mean Proportion of Clausal Conjunctions Produced

Group N -ko ‘and’ *-hako ‘COMIT’ kuliko ‘also’ Noun-to ‘also’66

Low HLS 7 0.66 0.02 0.30 0.02

High HLS 5 0.88 0.00 0.00 0.13

KS-Can 6 0.95 0.00 0.02 0.03

KS-Kor 8 0.73 0.00 0.08 0.20

From Table 12 and Table 13 we can see that most HLS67 recognized that different conjunctions are used to conjoin nouns and clauses. One low HLS overgeneralized the clausal conjunction kuliko ‘also’ to contexts where a nominal conjunction was required, as in (132) below. (132) Ay -tul kuliko kangaci-lang cangnankam kakey aph-ey po-n-ta Child-PL also puppy-COMIT toy store front-at see-PRES-DECL ‘The children and the puppy are looking inside the toy store. (literally: (The) children also with (the) puppy see from (the) front of (the) toy store.)’ (E7)

In (132), the first conjunct is not Case-marked and two conjunctions are used—the clausal conjunction kuliko, which attaches to the second conjunct, and the comitative Case marker -lang, which also attaches to the second conjunct. If kuliko was intended to conjoin the nouns ay-tul

‘child-PL’ and kangaci ‘puppy’, the second conjunct kangaci should not have the comitative Case marker -lang. However, if -lang was intended to conjoin the nouns, the first conjunct ay-tul should have Nominative Case marking. It thus appears that the speaker is stuck at a stage where kuliko is being overgeneralized as a comitative Case marker—most likely due to transfer from English, but where there has been exposure to, and perhaps tentative uses of, the comitative Case

66 -To ‘also’ is a conjunction that can be used to conjoin clauses, but it attaches to nouns, as illustrated below, where -to attaches to nolay ‘song’ in (i) and chayk ‘book’ and yenphil ‘pencil’ in (ii). The sentence in (ii) is an instance of verb ellipsis. Chayk-to ‘book-also’ is part of a clause whose verb sa-ko ‘buy-and’ has been elided. (i) Swuyeng-ul hayss-ta. Nolay-to pwul-ess-ta. Swim-ACC do-PST-DECL song-also sing-PST-DECL ‘(I) swam. (I) also sang songs.’ (ii) Chayk-to yenphil-to sa-ss-ta. Book-also pencil-also buy-PST-DECL ‘I (not only) bought a book and(/but also) a pencil.’

67 One low HLS and one high HLS did not produce any clausal conjunctions at all; thus it is unclear whether they are aware that the nominal conjunctions are not used to conjoin clauses.

86 marker -lang. The use of the clausal conjunction kuliko ‘also’ to conjoin nouns is also consistent with the speaker’s grammar undergoing simplification or attrition, since this could be motivated by a preference for the use of a free morpheme over a bound one.

Although the preferred strategy for all groups of speakers was to conjoin clauses using the clausal conjunction -ko, one low HLS overgeneralized -hako to contexts where a clausal conjunction was required.

(133) Incey -nun cip-ey wa-se ay-tul-i phikon-hay-se ttang-eyse ca-ko Now-CONTR home-to come-and child-PL-NOM tired-do-because ground-at sleep-and

hako kay-hako cwuin-un chimtay-se ca-ko emma-ka pang-ey tulewa-se and dog-with owner-CONTR bed-at sleep-and mom-NOM room-to enter-and

ay-tul-i ca-nun ke kunyang po-ko iss-eyo child-PL-NOM sleep-PRES.REL thing just see-KO be-PRES-DECL.II ‘Now after coming home, the children are tired so they are sleeping on the ground and also the dog and his owner are sleeping on the bed and the mom has entered the room and is just looking at the children.’ (E9)

As can be seen in (133), this speaker produced quite complex sentences using a string of conjunctions. Despite using the clausal conjunction -ko, the speaker also uses the nominal conjunction -hako to conjoin clauses. Given that this is immediately followed by the use of -hako as a nominal conjunction, it is likely that the overgeneralization of -hako ‘with’ as a clausal conjunction is due to transfer from English, since the same morpheme and would have been used to conjoin the clauses and the nouns.

With respect to morphological accuracy, although the HLS struggled with other Case markers, the majority of HLS produced the correct form for the conjunctions they used. Of the HLS who used -wa or -lang, only one low HLS used the incorrect phonological form. In (134), since the base ay-tul ‘child-PL’ ends in a consonant, -ilang is required. This is an indication of incomplete acquisition of the comitative Case marker -lang.

(134) kay -ka incey cip aph-ey nawa-se ay-tul-lang nol-ko iss-e-yo dog-NOM now house front-at come.out-and child-PL-with play-KO be-PRES-DECL.II ‘The dog has now come out to the front of the house and is playing with the children.’ (E9)

One of the low HLS undergenerated -hako; the speaker produced the Accusative Case marker -lul instead, as shown in (135) below.

87

(135) #Acwumma -hako acessi-lul kathi nol-ass-supnita woman-COMIT man-ACC together play-PST-DECL.I ‘(We) played with the woman and man.’ (E5)

In (135), the topic of the sentence (i.e., ‘the children and the dog’) has been dropped,68 and acumma ‘woman’ and acessi ‘man’ are incorrectly presented as the objects of ‘playing’. Since both are active participants in ‘playing’, and since the nominative/contrastive Case marker should have appeared on the dropped topic, acumma ‘woman’ and acessi ‘man’ both need to be marked with the comitative Case marker -hako (or another appropriate comitative Case marker). This also indicates that the comitative Case markers were not completely acquired by this speaker.

The groups of speakers differed in the types of comitative Case markers they produced. The low HLS preferred using -hako, the high HLS preferred -lang, and the KS-Can and KS-Kor both preferred using -wa to conjoin nouns. None of the low HLS produced the more formal comitative Case marker -wa, and all HLS preferred the two comitative Case markers that are used in colloquial speech (i.e., -hako and -lang), which is not unexpected, since HLS are often more proficient in oral vs. written production. There was some interspeaker variation. Three of the low HLS produced -hako exclusively, one speaker produced -lang exclusively, one low HLS did not produce comitative Case markers at all, and two of the low HLS used more than one type of comitative Case marker to conjoin nouns. It is interesting to note that when a single type of comitative Case marker was produced by the low HLS, it was usually the comitative Case marker that is not phonologically conditioned (i.e., the simplest form, -hako), as would be predicted if the HLS’ grammars had undergone simplification. All of the low HLS who produced -lang had difficulty. One speaker often used the clausal conjunction kuliko together with a -lang-marked conjunct (see (132)), and the other two speakers had difficulty choosing the correct phonological form of the comitative Case marker and had other errors, such as with the placement of the comitative Case marker (e.g., kathi-lang ‘together-COMIT’, where -lang is attached to the adverb ‘together’). The high HLS preferred producing the very colloquial, but phonologically conditioned, and thus more complex, comitative Case marker -lang, and there was only a single instance of -lang being produced incorrectly (out of 14 tokens produced by all

68 This is a common phenomenon in Korean, where D(iscourse)-linked topics are dropped.

88 high HLS). This suggests that (at least for HLS) speakers go through the following stages in their acquisition of conjunctions: i) overgeneralization of clausal conjunction kuliko to nominal contexts, ii) exclusive use of -hako, iii) introduction of -lang, and iv) introduction of -wa, with intermediate stages in between. It is not incorrect to use more than one type of comitative Case marker, given that -hako and -lang are fairly interchangeable,69 but whether it is appropriate to use both comitative Case markers depends on the level of speech that is used.

The HLS generally used level III. Although potentially all comitative Case markers are acceptable with this level of speech, level III is more commonly used in written language rather than speech70. One high HLS used -wa and -lang when using the level III register, which creates a mismatch between the level of speech and the choice of comitative Case marker: if level III is being used as a written register, -wa is acceptable, but if it is being used in a colloquial register, -lang would be more appropriate and the more formal -wa would be dispreferred. Thus, the speaker appears to not have completely acquired the register restrictions on the comitative Case markers. Other HLS who used more than one type of comitative Case marker used -hako and -lang interchangeably, which are both acceptable for level III. One low HLS incorrectly mixed the use of levels I and III. The use of these levels indicate a preference for a written register for this task, but the comitative Case marker used by this speaker was -lang, which, as mentioned above, is more appropriate for colloquial registers. Thus, this speaker appears to have incomplete acquisition of register-appropriate levels of speech. None of the low HLS used the more formal comitative Case marker -wa, even though it would have been acceptable in the use of level III as a written register. As mentioned above, the absence of -wa in the low HLS’ data is not unexpected, since low HLS generally do not have a lot of exposure to written Korean and are more proficient in (casual) speech than writing.

69 As noted above, Choo and Kwak (2008) distinguish between the two conjunctions as being “colloquial” and “having a spoken flavour” (p. 258).

70 Level III is not used very often in speech. If the speaker and hearer have a “close” enough relationship that level III or IV is acceptable, level IV is the level that is appropriate. Choo and Kwak (2008) note that level III (her level 4) can be used for sarcasm or to make general commentaries, as in (i) below (p.7; transcription and glosses my own), but that otherwise level IV (their level 3) “sounds more gentle and intimate” (p.6). (i) Nemu an tway-ss-ta. too.much not become-PST-DECL.III ‘That’s so sad.’ (Choo & Kwak’s translation: ‘Too bad. I’m sorry to hear that.’)

89

The KS-Can all used the deferential level of speech (i.e., level I), and all appropriately used the most formal comitative Case marker, -wa. Only one KS-Can mixed the use of the more colloquial -lang and the formal -wa, which was unexpected, but the speaker still had a clear preference for using -wa with level I speech. Whether -lang becomes acceptable if a story is being read aloud or retold is unclear. The KS-Kor used a variety of speech levels (average: level II). Except for two speakers who used level III exclusively, the KS-Kor used one of the more formal levels (i.e., level I or II). For these speakers, when they used level III, it was either to quote commands being given to people depicted in the picture or to add an afterthought, as in the second sentence of (136) below: Ta honca iss-Ø-ko ‘and (they’re) all alone’. The preceding sentence ends in -yo, the complementizer that is used to indicate level II speech. In order to maintain level II speech, -yo should also be attached to the following sentence. The absence of -yo makes this level III or level IV speech.71 However, this sentence appears to be an afterthought, and if the clause had been conjoined with (and preceding) the predicate of the first sentence weylowun kes kath-ay-yo ‘(they) seem lonely’, -yo would not have been required to maintain level II speech, since complementizers indicating the level of speech only occur on the matrix clause. Thus, these types of examples do not appear to be actual switches in formality or levels of speech.

(136) Ku taum-ey-nun ay-tul-i woylowun kes kath-ay-yo. That following-at-CONTR child-PL-NOM lonely-PRES.REL thing seem-PRES-DECL.II ‘(On the next page), it looks as though the children are lonely.’

Ta honca iss-Ø-ko. All alone exist-PRES-and ‘And they’re all alone.’ (K17)

Most of the KS-Kor used the formal comitative Case marker -wa. One speaker used -wa and -lang interchangeably with level II, which is acceptable, since -wa is odd with level IV and -lang is odd with level I. One speaker produced only one comitative Case marker and this was -lang, despite using level I, which was unexpected.

With respect to the syntax of comitative Case markers, given that English conjunctions form constituents with the second conjunct, low HLS, or HLS strongly influenced by English, were expected to attach the comitative Case marker to the second conjunct even though it is generally

71 In order to be level I speech, the clause kureh-Ø-supnita ‘be.so-PRES-DECL.I’ needs to be attached.

90 preferable to attach the comitative Case marker to the first conjunct in Korean. Thus, I analyzed the proportion of comitative Case markers which were attached to the first conjunct or the second conjunct. In cases where the subject was dropped, it was counted as an instance of a comitative Case marker attaching to the second conjunct, since Korean follows a topic-comment structure and old information precedes new information. A noun can only be dropped if it is old information and would thus appear on the left periphery if it were overt.

Table 14 shows that whereas the KS-Can and KS-Kor preferred to attach the comitative Case marker to the first conjunct, both groups of HLS preferred to attach the comitative Case marker to the second conjunct. Unexpectedly, the effect of transfer from English was stronger in the high HLS, who attached the comitative Case marker to the second conjunct 96% of the time, than the low HLS, who attached the comitative Case marker to the second conjunct 71% of the time.

Table 14. Mean Proportion of Conjunctions Attached to Second Conjunct of Subject Nouns

([N1-NOM N2-CONJ])

Group N M SD

Low HLS 7 0.75 0.39

High HLS 5 0.97 0.06

KS-Can 6 0.46 0.29

KS-Kor 8 0.16 0.22

This indicates that strong transfer from English influenced their preference in attaching the comitative Case marker to the second conjunct of subject nouns (i.e., where the position of the comitative Case marker was optional).

If the comitative Case marker is attached to the second conjunct, the comitative Case marker is interpreted as ‘with’. However, the comitative Case marker cannot attach to inanimate nouns with an role. One low HLS and one high HLS produced this error, as illustrated below. In (137), ay-tul ‘child-PL’, pumo-nim ‘parents-HON’, and kangaci ‘puppy’ are the active participants and kong ‘ball’ is the inanimate object () they are playing with. Thus, it is ungrammatical to attach -hako to kong ‘ball’; instead of a comitative Case marker, -ul kaci-ko

91

‘ACC take-and’ is required. The use of -hako with an is most likely evidence of transfer from English,72 given that when -hako attaches to the “second” conjunct, as it does in (137), it is interpreted as ‘with’ and English with would be appropriate for a ball (as well as their parents and the dog), as illustrated by the English translation provided below.

(137) *A y-tul-i pumo-nim-tul-hako kangaci-hako kong-hako nol-ko iss-e-yo Child-PL-NOM parents-HON-PL-COMIT puppy- COMIT ball- COMIT play-KO be-PRES-DECL.II ‘The children are playing with a ball with their parents and the dog.’ (E4)

To summarize, the majority of HLS appeared to have distinct conjunctions for conjoining nouns and clauses, but some had evidence of incomplete acquisition – they appear to be stuck at a stage where these two types of conjunctions are still being sorted out and used in the correct contexts by either overgeneralizing kuliko to contexts where comitative Case markers should be used or overgeneralizing -hako to contexts where clausal conjunctions should be used, which is also evidence of transfer from English. The HLS had a strong preference for -hako, the comitative Case markerthat is not phonologically conditioned and which does not have any restrictions pertaining to the level of speech used, which is evidence of simplification. The low HLS preferred using -hako, whereas the high HLS preferred using -lang—both are colloquial comitative Case markers. The HLS showed evidence of incomplete acquisition of the various levels of speech and the appropriateness of various comitative Case markers with these levels of speech. There was also strong transfer from English in the HLS’ structure for sentences containing comitative Case markers. The HLS preferred to attach the comitative Case marker to the second conjunct, as occurs in English, whereas the KS-Can and KS-Kor preferred attaching the comitative Case marker to the first conjunct. When the comitative Case marker attaches to the second conjunct, there is a semantic restriction on it attaching to ; HLS’ use of comitative Case markers (vs. the expected -ul kaci-ko ‘-ACC take-and’) in these contexts is also strong evidence of transfer, since English with has no such restrictions.

3.5 Summary

The results of the pilot study indicate the importance in obtaining HLS’ (and KS’) current usage of Korean, as it relates to their proficiency. It is also important to consider the HLS’ input, since

72 The absence of literature on the acquisition of comitative Case markers by monolingual Korean-speaking children does not allow us to rule out the possibility of incomplete acquisition by the HLS.

92 not all errors can be attributed to transfer from English and may be influenced by their input, especially if it consists of a language under attrition. Aspects of the HLS’ nominal domain were non-target-like in certain respects. For reference, the differences between English and Korean are summarized in Table 15–Table 17 below. Table 15. Case Marking and Case Drop: English vs. Korean Condition English Korean: Acceptability of Case drop

Types of Case GEN marked regularly on Case marked on all N (and PN), but marking all N (and PN) can be dropped in following hierarchy: GEN > ACC > NOM NOM and ACC only marked on PN (and not N); Case drop not allowed

Word order N/A NOM: SOV, *OSV ACC: SOV > OSV

Person NOM and ACC marked on NOM: 1st person, 2nd person > 3rd person 1st and 3rd person, but (ACC: 3rd person > 2nd person, 1st person) not on 2nd person

Animacy N/A NOM: Human, Animate > Inanimate ACC: Inanimate > Human, Animate

Definiteness N/A NOM: Personal PN, Name > Other ACC: Other > Personal PN, Name

Table 16. Plural Marking: English vs. Korean Condition English Korean

Animacy PL obligatory on all N Plural optionally marked, often according to following hierarchy: Human, Animate > Inanimate

Specificity PL obligatory on all N Specific N*(-tul); Non-specific N(-tul)

Quantificational Numeral Measure-PL of N-PL N(-tul) Num CL(*-tul)-Case Structure Num CL(*-tul)-GEN N(-tul)-Case

93

Table 17. Conjunctions: English vs. Korean Condition English Korean

Morphological Free morpheme (e.g., and); N conjunctions: suffixes (e.g., -wa, -hako, status same for N and clauses -lang);

Clausal conjunctions: suffix (e.g., -ko) or free morpheme (e.g., kuliko)

Phonologically No Yes for -wa and -lang; conditioned? No for -hako; No for clausal conjunctions

Sensitive to No -wa: ok for spoken I, II, written I, II, III, register? -lang ok for spoken II, III, IV -hako ok for all levels–spoken or written; No for clausal conjunctions

Syntactic N1 [CONJ N2] [N1-CONJ] N2-NOM ‘and’ or position with N1-NOM [N2-CONJ] ‘COMIT’, but not if nouns N2 is ; [N1-CONJ] N2-ACC ‘and’

Syntactic Clause1 [CONJ Clause2] Clause1 [kuliko Clause2] or position with [Clause1-ko] Clause2 clauses

The HLS’ usage of Case marking appears to have been affected by transfer from English (since English does not mark Nominative and Accusative Case on nouns other than pronouns), but also perhaps by the degree of Case drop that is present in their input. They dropped Case markers much more often than the KS and the accusative and genitive Case markers were not used regularly. The HLS’ overall proportions of Case drop with respect to the type of Case followed the pattern of the KS (i.e., Genitive > Accusative > Nominative). Since the accusative and genitive Case markers are not marked regularly by the low HLS, we might expect attrition to eventually eliminate the use of Case markers at all.

The HLS used -tul differently, depending on the context (e.g., specific contexts vs. counting contexts), but on the acceptability judgment task it was revealed that English had a greater influence on what they determined to be correct. The low HLS had an even stronger preference for -tul in the specific contexts (since this would be reinforced by English) and both low and high HLS allowed -tul to attach to classifiers (as plural marking attaches to measure words in English).

94

The data from the HLS’ use of comitative Case markers shows that low proficiency speakers have some difficulty in selecting appropriate conjunctions to use in nominal vs. clausal contexts. In terms of syntax, although the HLS appeared to have the correct structure, transfer of English appears to have influenced the HLS’ (especially for the high HLS) preference for structures where the comitative Case marker was attached to the second conjunct (as is the case with English conjunctions) even though it is possible for the same comitative Case marker to be attached to the first conjunct in Korean. Thus, although we might expect the English linear order of N1-conjunction-N2 to influence the HLS’ preference for structures which maintain this order (i.e., where the comitative Case marker attaches to the first conjunct, since comitative Case markers attach as suffixes to the noun in Korean), the HLS actually preferred structures which maintained the structure of English (i.e., where the comitative Case marker attaches to the second conjunct). Although there were not very many tokens of inanimate nouns being used with an role, the results suggest that even high HLS may have difficulty in ascertaining that comitative Case markers can only mean ‘with’ when attached to animate nouns (unlike English with, which attaches freely to both animate and inanimate nouns). The choice of the HLS’ level of speech (i.e., level III) made it difficult to ascertain whether the speakers were aware of the differing appropriateness of the various levels of speech, since potentially all of the comitative Case markers are acceptable with this level (depending on whether level III is being used in a written or oral context), but it should be noted that level III is generally more appropriate in written registers and is more common in the speech of young children. Thus, there was still a mismatch between the level of speech chosen and the comitative Case markers that were used (e.g., the more colloquial -lang was used with the “written” register of level III).

The results suggest that rather than causing across the board English-like errors, transfer from English may influence the proportions of errors that are made – many errors that HLS make are also evidenced (in much lower proportions) in the language of KS. Thus, it is important to consider whether HLS have had sustained exposure to Korean or whether they have had interrupted exposure to Korean, and whether their length and type of input results in quantitative differences in their preferences for the use of particular morphemes and structures. The details of the study conducted for this thesis are discussed in the following chapter.

95

Chapter 4 The Study: Questions, Hypotheses, and Methods 4 The Study 4.1 Introduction

The main question of my study is whether age of exposure, to an L2 or the language of the majority (i.e., English), or the degree of input and use of the HL are better predictors of HLS’ performance on tasks testing the acquisition of various morphemes in the nominal domain in Korean. HLS have been noted to have listening and speaking skills that are superior to their reading and writing skills (H.-S. Kim 2005, among others), thus it is important to test their oral ability, rather than their ability in written Korean. Testing the HLS orally allows us to reach speakers of various levels of proficiency, including those who are able to speak Korean, but not necessarily write in Korean. In the following section, I will discuss the language-specific questions I will be investigating for each of the various aspects of the nominal domain in Korean, along with my hypotheses.

4.2 Questions and Hypotheses

4.2.1 Case Marking and Case Drop in Korean

With respect to Case marking in Korean, I am interested in the following questions:

1) Do HLS recognize that Case marking occurs regularly in Korean?

2) Are HLS sensitive to differences in the acceptability of dropping the accusative Case marker vs. the nominative Case marker?

3) Are HLS able to select the correct phonological form of the nominative and accusative Case markers?

We could predict slightly different outcomes based on HLS’ language experiences. With respect to the first two questions, Case marking is an area where the influence of English could lead HLS to mark Case overtly in a way that is different from their baseline. If HLS were allowing their L1 or L2 English grammars to influence the grammar of their HL, HLS could fail to produce any

96 nominative or accusative Case markers, given that nouns in English are not marked overtly with nominative or accusative Case markers in English (Case marking is only visible in the pronominal system).

Case marking is optional in Korean, but the acceptability of Case drop is conditioned by several factors. In order to be sensitive to the statistical variability in Case marking, sufficient input in Korean would be required. For speakers who do not receive very much input in Korean, I would expect a gradient level of acceptability to not be evidenced in their use of Case drop. For these speakers, Case marking is predicted to be completely optional (i.e., marked overtly only half of the time). This would only be expected of speakers with very low usage of and exposure to Korean. Speakers who receive more input in Korean would be expected to be sensitive to the relative rates of Case drop, and are expected to accept Case drop with accusative Case markers more frequently than Case drop with nominative Case markers. I predict that speakers who not only are exposed to Korean frequently, but who also use Korean frequently, will drop accusative Case markers more frequently than nominative Case markers in their own productions as well. Speakers who do not use Korean as frequently, but who still may receive more input in Korean than other speakers, may be sensitive to the greater likelihood (and acceptability) of dropping the accusative Case, but they may not be as consistent in producing Case markers, compared to their baseline. For example, although they may produce nominative Case markers more frequently than accusative Case markers, they may not produce Case markers overall as frequently as their baseline.

It should be noted that the factors conditioning the acceptability of Case drop are in fact in line with cross-linguistic universals that are present in languages which employ differential Case marking. For example, in languages with differential subject marking, human nouns in subject position are often not Case-marked, given that they are prototypical subjects (Croft 2003, Aissen 2003). Similarly, in languages with differential object marking, the prototypical objects (i.e., inanimate nouns) are generally not Case-marked (Croft 2003, Aissen 2003). Rather, it is the non- canonical subjects (e.g., inanimate subjects) and objects (e.g., human objects) that are Case- marked. Thus, if HLS are resorting to unmarked principles in UG, they are not expected to have difficulty with effects such as animacy. Thus, I would not expect HLS to have difficulty with the semantic requirements on Case marking.

97

With respect to the third question, it is important to note that the Case markers in Korean are phonologically conditioned, whereas the English Case marker that is marked regularly, the genitive Case marker, is not phonologically conditioned. The English genitive Case marker attaches to the right edge of the nominal phrase, and the same form is used regardless of the base. If HLS’ heritage language grammar is influenced by their English grammar, we could expect HLS to produce only one form for nominative Case (e.g., -ka) and one form for accusative Case (e.g., -lul). This could also be true of HLS who do not receive a lot of input in Korean, given that monolingual Korean-speaking children first learn Case marking by going through a stage where they use the nominative Case marker -ka exclusively. Speakers who receive a lot of input but who do not use Korean very often could use both forms of the nominative and accusative Case markers, but perhaps not correctly all of the time. In particular, given the homophony of the form of the nominative Case marker which attaches to bases ending in consonants and the hypocoristic marker -i, these speakers could produce forms such as -ika ‘NOM’, if they do not recognize that when -i appears with -ka it must be the hypocoristic marker. Alternatively, another possibility is that the overall high frequency of the hypocoristic marker in Korean (especially in child-directed speech) means that HLS are likely to recognize -i as a separate morpheme from -ka in forms where both markers attach to a noun.

4.2.2 -Tul, Classifiers, and Quantificational Structures

With respect to the plural marker -tul and classifiers, the following questions arise:

1) Are HLS aware of the semantic requirements and syntactic restrictions on the use of -tul?

2) Do HLS learn the use of classifiers at all (i.e., do the types of quantified structures that HLS produce include classifiers)?

For HLS, the heritage language is acquired as an L1. With respect to the semantic requirement on the use of plural marking in Korean, HLS are expected to recognize the semantic difference between having the plural marker expressed overtly on the noun or not. Thus, I would expect HLS to not have difficulty in consistently marking the plural on specific nouns (i.e., nouns which are preceded by demonstratives). In English, the requirement for plural marking is not tied to the specificity of the noun—it is always required on plural nouns, but any influence from English on

98 the HLS’ heritage language grammars would be expected to result in positive transfer, since English nouns that are specific also require overt plural marking. The unmarked setting of UG is also to have overt plural marking on nouns that are specific, thus any resort to the unmarked would results in HLS producing overt plural marking on specific nouns. HLS who receive less input in Korean could interpret plural marking to be optional on specific nouns, since it is also optional on bare nouns, and produce it inconsistently (i.e., half of the time) on specific nouns. But it is more likely that they would still be able to use plural marking correctly, given that UG and transfer would both lead them to mark the plural on nouns that are specific. HLS who do not use Korean very often are also likely to attach the plural marker consistently on specific nouns.

With respect to the syntactic restriction on the use of -tul, speakers who do not receive very much input in Korean or who do not use Korean very often, influence from English could lead them to incorrectly attach plural marking to the classifier, given that measure words (e.g., bottle) require overt plural marking in English (e.g., five bottles of beer). However, cross-linguistically, Borer (2005) notes that plural marking and classifiers tend to occur in complementary distribution, such that languages which have classifiers often do not have plural markers. If HLS were resorting to the unmarked setting from UG, we could thus expect HLS to not make the error of attaching -tul to classifiers. Speakers who do not receive a lot of input in Korean could fail to produce classifiers at all, making it impossible to determine whether they have acquired the syntactic restriction against plural marking on classiifiers in Korean. Speakers who do not use Korean very often could be expected to incorrectly produce plural marking even on classifiers, perhaps due to transfer from English.

The use of quantificational structures not involving classifiers (or the avoidance of classifiers) could be more common with speakers who do not receive a lot of input in Korean. Reduced input in Korean could mean that HLS would not be exposed to the range of different quantificational structures in Korean, and they may resort to producing structures which are common in English. For example, structures such as Num N or where a classifier occurs prenominally (e.g., Num Cl N) could be used more frequently than structures with postnominal classifiers, as was found with the bilingual children in Shin and Milroy’s (1999) study. HLS who do not use Korean very often could also produce a subset of the quantificational structures available to them. For example, they could simply produce postnominal classifiers.

99

4.2.3 Comitative Case Markers

The questions that will be investigated regarding comitative Case markers in Korean are the following:

1) Do HLS understand the semantic restrictions and requirements on the usage of comitative Case markers?

2) Do HLS use the correct form for comitative Case markers that are phonologically conditioned?

With respect to the semantic restriction against the use of the comitative Case marker in Korean to inanimate , I would expect transfer from English to allow this overgeneralization of the comitative Case, given that the same morpheme with is used in play with Nora and play with a ball. HLS who do not receive very much input in Korean may never have heard -lul kaci-ko ‘ACC take-and’, and could also be expected to incorrectly allow comitative Case markers in Korean to be used with . Speakers who do not use Korean very often may show acquisition of this restriction if they have received sufficient input in Korean.

With respect to the preference, or dispreference, of the various comitative Case markers with particular levels of speech, transfer from English could lead HLS to not be sensitive to the different levels of acceptability, and these speakers could use a single Case marker across the board (.e.g, -hako). HLS who do not receive a lot of input in Korean may not even be fully aware of the different types of comitative Case markers that exist in Korean. They may only have been exposed to -hako as a child, for example. These speakers may not have been exposed to different levels of speech either; for example, they may only have heard level IV speech directed towards them. Without sufficient input in Korean, HLS would not be expected to show sensitivity to the statistical differences in how often a particular type of comitative Case marker is used with a particular level of speech. I would expect HLS who did not use Korean very often to have difficulty with the discourse-constraints on the acceptability of the different comitative Case markers, given that these speakers may only ever produce one or two different levels of speech. These speakers are expected to allow any of the comitative Case markers with the different levels of speech, regardless of the particular level.

100

4.3 Participants and Methods

Participants were recruited from a local Canadian university, a Korean-Canadian church, and through personal contacts. Thirty-three HLS of Korean (14 males and 19 females) participated in the study; their ages ranged from 18 to 35. Twenty-two of the participants were born in Korea; seventeen were from Seoul and five participants were from outside of Seoul. A selection of the mothers of the bilinguals, as well as other mothers,73 (total: eight speakers; three from Seoul, five from outside Seoul) served as the control group (i.e., the baseline), since the majority of HLS use Korean primarily at home and their main source of Korean input is often from their parents. Most of the testing occurred at a Korean-Canadian church, and the sessions took about one hour.

The experiment consisted of three oral tasks: a) an elicited narration task, b) a preference task, and c) an elicited imitation task, which were administered in that order. All instructions and experiment questions were digitally recorded in Korean by a Korean-dominant bilingual74. The questions were recorded at a moderate pace in order to simulate the pace of a regular conversation. The questions were then edited using Audacity (http://audacity.sourceforge.net/) to keep the volume and tempo consistent. Participants were asked to listen to the instructions and to respond to various prompts that were played from an audio file on a laptop. In order to maintain interaction in Korean as much as possible, any requests or clarification questions asked by participants during the course of the experiment were responded to in Korean by the experimenter, an English-dominant bilingual with advanced proficiency in Korean, regardless of whether the participants asked questions in English or Korean. Most of the participants tried to ask questions in Korean.

In addition to the main tasks, the participants completed two language history questionnaires and two self-assessments of their ability in Korean, which were completed after the preference task and before the elicited imitation task. These materials were in English. The first language history questionnaire (Steele 2002) elicited information on the participants’ first language and language

73 Due to the lack of availability of many of the parents of the participants, other adults who had children who were around the same age of the participants were selected to participate.

74 This speaker was a 35-year-old female who was raised in Seoul, South Korea and who had immigrated to Canada at the age of 16 years old.

101 use with others in the home during childhood, their education and current use of Korean, as well as a self-assessment of their ability in Korean and English. The second language history questionnaire elicited more details on the nature of the participants’ input and usage of Korean, including who they communicated with in Korean, estimates of the language proficiency of the speakers who spoke to them in Korean, the types of Korean media they were exposed to, and their attitudes towards their use of Korean. The second language self-assessment (Common European Framework of Reference for Languages: Learning, Teaching, Assessment (Council of Europe 2001)) provided specific descriptions of what speakers should be able to accomplish in the language, divided according to six different levels of ability. The information from the language history questionnaires and the self-assessments were used to compile composite usage and exposure scores for the HLS, and to provide an external measure of their general ability in Korean. The details of the main tasks of the study, as well as the motivations for using these tasks will be discussed below.

4.3.1 Elicited Narration Tasks

The first task was an elicited narration task, designed to elicit the use of plural morphology, Case marking, and conjunctions. The participants were shown pictures in a wordless picture book and asked to describe in Korean what was happening in the pictures. The elicited narration task was audio recorded, transcribed, coded, and analyzed following a methodology similar to that used by Kanno et al. (2005) for Japanese elicited speech. The number of clauses, as indicated by the use of subordinate or coordinate conjunctions, per utterance was used to determine the subordination index (SI) of the speakers’ utterances, and the mean length of the HLS’ utterances (MLU) was also calculated as a measure of the speakers’ fluency. The number of different words (NDW) used by the speakers was also analyzed as an estimate of the speakers’ vocabularies. The number of plural-, Case- and postposition-related errors was also calculated. Since the vocabulary required to describe the pictures was fairly simple and accessible, the task provided the added benefit of being able to see the complexity and level of detail that was chosen by the speakers to describe the various scenes depicted, as well as the types of structures that were used. Having the participants describe the pictures rather than produce their own spontaneous narratives on a particular topic makes it possible to also evaluate the accuracy of the statements they made (e.g., whether the context was plural or not).

102

4.3.2 Preference Task

The second task was a preference task. For this task, participants were presented with 56 short paragraphs. Each paragraph ended in a question similar to ‘But do you know what happened?’ and was followed by two test sentences. Participants were asked to write “1” or “2” on a piece of paper, according to whether the first or second sentence sounded more appropriate given the preceding context. Participants were not allowed to listen to questions more than once. The preference task tested HLS’ allowance of different combinations of Case drop, whether HLS are sensitive to the effect of specificity on the usage of -tul and the syntactic restrictions on -tul (i.e., its ability to attach to classifiers), as well as the HLS’ allowance of structures which have -hako

‘COMIT’ attached to inanimate . Of the 56 test items, 24 of the test items compared sentences with or without Case drop. For these sentences, eight different verbs were used (one intransitive and seven transitive), and there were four possible variations of Case drop – what I will refer to as “No Drop” (both the nominative and accusative markers were attached to the appropriate nouns), “Acc Drop” (only the accusative Case marker was dropped; the nominative Case marker was overt), “Nom Drop” (only the nominative Case marker was dropped; the accusative Case marker was overt), and “Both Dropped” (both the nominative and accusative Case markers were dropped). Each of the four cases was presented 12 times, and each possible combination (of pairs of cases) was presented four times each, as in No Drop vs. Nom Drop in (138). Eight of the test items compared sentences with nouns with demonstratives with or without -tul, as in ku kay-lul ‘that dog-ACC’ vs. ku kay-tul-ul ‘that dog-PL-ACC’ (139), and eight of the items compared sentences with classifier phrases with or without -tul attached to the classifier, as in napi yeses mali-lul ‘butterfly six CL-ACC’ vs. napi yeses mali-tul-ul ‘butterfly six

CL-PL-ACC’ in (140). Eight test items contained nouns with -hako ‘COMIT’ or -lul kaciko ‘-ACC take-and’ with inanimate , as in kelley-hako ‘rag-COMIT’ vs. kelley-lul kaci-ko

‘rag-ACC take-and’ in (141). The subject noun was always a Korean name and any plural or comitative Case markers were attached to the object noun.

(138) Hyenci -ka kalpi-lul mek-ko siphe-se il-cwuil-tongan namca Hyunji-NOM beef.rib-ACC eat-KO want-because one-week-during male

chinkwu-hanthey kalpi sa tal-la-ko hay-ss-e. haciman friend-DAT beef.rib buy ask.for-IMP-REL do-PST-DECL.IV but

103

siktang-ey tule-ka-se mwe-l sikhi-n cwul al-a? restaurant-at enter-go-and what-ACC order-PST.REL thing know-PRES-DECL.IV ‘Hyunji wanted to eat beef ribs, so for one week (she) asked (her) boyfriend to buy (her) beef ribs. But when she went into the restaurant do you know what (she) order?’

Hyenci-ka pokkum-pap-ul sikhye-ss-e. ______Hyunji-NOM fried-rice-ACC order-PST-DECL.IV ‘Hyunji ordered fried rice.’

Hyenci pokkum-pap-ul sikhye-ss-e. ______Hyunji fried-rice-ACC order-PST-DECL.IV ‘Hyunji ordered fried rice.’

(139) Sungcwun -i-nun yel sal-i-Ø-ntey, kay-lul nemu coha-hay-se ney Seungjoon-HYP-NOM ten year-be-pres-but, dog-ACC too.much like-do-because four

sal ttay-puthe sayng-il senmul-i hangsang kay han mali-ye-ss-e. year time-from birth-day present-NOM always dog one CL-be-PST-DECL.IV

isa-ka-l ttay kay-lul han mali-to mos teyli-ko ka-se move-go-FUT time dog-ACC one CL-even cannot take.along-KO go-because

wul-myense ilehkey mal hay-ss-e. cry-while this.way language do-PST-DECL.IV ‘Seungjoon is ten years old, but because (he) loved dogs so much, ever since (he) was four (his) birthday present was always a dog. But when he moved, because he couldn’t take even one dog (with him), (he) said this, crying:’

“Nay-ka ku kay-lul salang-hay-ss-e.” ______I-NOM that dog-ACC love-do-PST-DECL.IV ‘I loved that dog.’

“Nay-ka ku kay-tul-ul salang-hay-ss-e.” ______I-NOM that dog-PL-ACC love-do-PST-DECL.IV ‘I loved those dogs.’

(140) Tongyep -i-nun nala tani-nun pelley-lul nemu coha-hay-Ø-Ø Dongyub-HYP-TOP fly go.around-PROG bug-ACC very like-do-PRES-DECL.IV

Enu nal kongwen-ey ka-ss-ta wa-se tongyep-i-nun ilekhey Some day park-to go-PST-AFTER come-AFTER Dongyub-HYP-TOP this.way

mal-hay-ss-e talk-do-PST-DECL.IV ‘Dongyub loves flying insects. One day (he) went to the park and when (he) came back, Dongyub said this:’

104

“Nay-ka napi yeses mali-lul capa-ss-e!” ______I-NOM butterfly six CL-ACC catch-PST-DECL.IV ‘I caught six butterflies!’

“Nay-ka napi yeses mali-tul-ul capa-ss-e!” ______I-NOM butterfly six CL-PL-ACC catch-PST-DECL.IV ‘I caught six butterflies!’

(141) Minswu -ka chinkwu-ney cip-ey nol-le ka-ss-e. pap mek-Ø-ko Minsoo-NOM friend-place house-at play-in.order.to go-PST-DECL.IV rice eat-PRES-and

selkeci-lul hay cwu-ki-lo hay-ss-nuntey, minswu-ka cepsi-lul washing.dishes-ACC do give-NMLZ-intend.to do-PST-but, Minsoo-NOM dishes-ACC

takk-ul ttay molu-Ø-ko mwe-l ha-n cwul al-a? wipe-FUT time not.know-PRES-and what-ACC do-PST.REL thing?? know-PRES-DECL.IV ‘Minsoo went over to a friend’s house. After dinner (he) had said (he) would do the dishes, but when Minsoo wiped the dishes do you know what (he) did?’

Minswu-ka kelley-hako cepsi-lul takka-ss-e. ______Minsoo-NOM rag-COMIT dish-ACC wipe-PST-DECL.IV ‘Minsoo wiped the dishes with a rag.’

Minswu-ka kelley-lul kaci-ko cepsi-lul takka-ss-e. ______Minsoo-NOM rag-ACC take-and dish-ACC wipe-PST-DECL.IV ‘Minsoo wiped the dishes with a rag.’

There were eight distractors - four items comparing sentences which differed in grammaticality or acceptability due to the lexical items used (142), and four items which compared sentences differing in grammaticality due to the nominal or verbal particle used (143).

(142) Unmi -ka miswul swukcey-ka hangsang manh-a-Ø. Wenlay-nun Eunmi-NOM art homework-NOM always a.lot-PRES-DECL.IV normally-CONTR

swukcey-lul ala-se cal ha-Ø-nuntey, ecey-nun sikan-i homework-ACC know-and well do-pres-but, yesterday-CONTR time-NOM

mocala-se musun il-i iss-ess-nun cwul al-a? not.enough-because what work-NOM exist-PST-REL thing know- PRES-DECL.IV ‘Eunmi always has a lot of art homework. Usually (she) finishes it on (her) own, but yesterday because (she) didn’t have enough time, do you know what happened?’

Unmi-ka swukcey-lul an hay-ss-e. ______Eunmi-NOM homework-ACC not do-PST-DECL.IV ‘Eunmi didn’t do (her) homework.’

105

Unmi-ka swukcey-lul ceyil cal hay-ss-e. ______Eunmi-NOM homework-ACC most well do-PST-DECL.IV ‘Eunmi’s homework was the best.’ (lit.: ‘Eunmi did (her) homework the best.’)

(143) Unse -ka kunyang cip-ey iss-Ø-ko siphe-ss-nuntey, kapcaki sa-ko Eunsuh-NOM just home-at exist-PRES-and want-PST-but, suddenly buy-KO

sipe-ss-ten os-i sayngkak-i na-se mwe-l ha-n want-PST-REL clothes-NOM thought-NOM appear-because what-ACC do-PST.REL

cwul al-a? thing know-PRES-DECL.IV ‘Eunsuh just wanted to stay at home, but suddenly (she) remembered the (piece of) clothing (she) had wanted to buy, so do you know what (she) did?’

Unse-ka kakey-hanthey ka-ss-e. ______Eunsuh-NOM store-DAT go-PST-DECL.IV ‘Eunsuh went to(dative) the store.’

Unse-ka kakey-ey ka-ss-e. ______Eunsuh-NOM store-to go-PST-DECL.IV ‘Eunsuh went to the store.’

There were two versions of the test, and the test items were semi-randomized (in the same order for both versions) and counterbalanced for order of presentation of each pair of test sentences. Quantifiers and classifiers were avoided as much as possible in the stories in order to reduce the risk of priming the acceptability of the use or omission of -tul ‘plural’ in the classifier constructions on other questions of the test. -Tul was never used. Training items were provided before the task and participants were allowed to listen to the training items as many times as needed in order to familiarize themselves with the task. However, once the actual task began, test questions were only presented once to elicit speakers’ first impressions and to minimize the participants’ reliance on prescriptive grammar. The preference task always followed the elicited narration task in order to prevent the priming of particular syntactic structures as well as the use or omission of -tul.

Preference tasks are intended to tap into speakers’ grammars indirectly by asking speakers to choose between two test sentences, given a particular context. The sentence a participant prefers is assumed to be part of their grammar. Preference tasks are useful in eliciting judgments on sentences and structures that may not appear very frequently in spontaneous production or which are sensitive to subtle factors. With acceptability judgment tasks, it can be difficult eliciting clear

106

numerical differences between sentences which are perfectly grammatical and sentences which are slightly anomalous, for example, if a particular Case marking is missing, since there are many contexts where Case drop is optional. Thus, preference tasks are useful in testing whether distinctions/assertions made about grammar in the literature are maintained when speakers are forced to choose between two options. If the phenomenon is truly variable (and perhaps unconstrained), we would expect no difference in the percentage of times speakers selected one option over the other. Otherwise, when forced to make a choice, we would expect speakers to have a clearer preference for a particular option. This is also useful if there are other possibilities that come to mind for the speaker, which would otherwise affect their response on an acceptability judgment task. For example, in order to test HLS’ acceptance of conjunctions, I

presented sentences either with -hako ‘COMIT’ or -ul kaci-ko ‘-ACC take-and’ attached to inanimate . A pilot test with late sequential bilinguals (i.e., speakers who had arrived in Canada at the age of 15 or older) revealed that these bilinguals were not distinguishing

between sentences with -hako ‘COMIT’ and sentences with -ul kaci-ko ‘-ACC take-and’ by having low acceptability of both structures. Interviews with the participants following the task revealed that they sometimes thought that the postposition -ulo ‘using’ would sound more natural. On an acceptability judgment task, if participants felt that -ulo ‘using’ would sound more appropriate in the given context, since -ulo is often used with , as in (144) below, their data could show low acceptability, or even rejection, of both types of sentences (i.e., with -hako (145) and -ul kaciko (146). In interviews following the task, participants stated that this was the reason for their low acceptability of both structures.75

(144) Hyenswu-ka cwumek-ulo pyek-ul kkay-ss-ta. Hyun-Soo-NOM fist-INSTR wall- ACC break-PST-DECL.III ‘Hyun-Soo broke a wall with (his) fist.’

(145) Hyenswu-ka cwumek-hako pyek-ul kkay-ss-ta. Hyun-Soo-NOM fist-COMIT wall- ACC break-PST-DECL.III ‘Hyun-Soo broke a wall with (his) fist.’

75 Even though participants assigned low positive values to sentences using the comitative Case marker -hako with instruments, they admitted that the sentences sounded strange and that they would not use -hako in the contexts given. This version of the acceptability judgment task had been recorded by an English-dominant Korean bilingual with advanced proficiency in Korean and thus it is possible that the participants had higher tolerance for strange- sounding structures in their interactions with English-dominant Korean bilinguals. This could account for their reluctance to assign negative values to the structures, even though negative values were supposed to be assigned to sentences which they would not produce themselves.

107

(146) Hyenswu-ka cwumek-ul kaci-ko pyek-ul kkay-ss-ta. Hyun-Soo-NOM fist-ACC take-and wall- ACC break-PST-DECL.III ‘Hyun-Soo broke a wall with (his) fist.’

By forcing a comparison only between the comitative Case marker -hako (which should always be disallowed with ) and -ul kaciko (which can often be used interchangeably with -ulo), we would be able to see more clearly whether -hako was the preferred option for HLS despite only having -ul kaciko (and not -ulo) as a competitor. This could elicit useful information which would otherwise be masked in an acceptability judgment task. The contextualizing stories provided also helped control the way in which the test sentences could be interpreted since the discourse surrounding the test sentence was the same for each participant. This is useful for phenomena that are sensitive to several different factors (e.g., the acceptability of Case drop). In order to quantify the results, the participants’ responses were coded according to whether they picked the correct sentence from the pair of sentences presented to them.

4.3.3 Elicited Imitation Task

In order to determine the speakers’ proficiency and as another measure of the speakers’ use of various nominal morphemes, an elicited imitation task was used. The participants were asked to listen to an audio recorded sentence and were then give three seconds to judge whether it was true or not by writing either “one” for ‘true’, “two” for ‘false’, or “three” for ‘not sure’ on a piece of paper. This was followed by a one second beep, after which the participants were given ten seconds to repeat back the sentence they had heard, similar to Erlam’s (2006) methodology. Participants were only allowed to listen to each sentence once. The task for the present study consisted of twenty-four test items and eight distractors. There were eight sentences with classifiers (four with -tul and four without), eight sentences with (four with -hako

‘COMIT’, four with -ul kaciko ‘-ACC take-KO’), and eight sentences with dative Case markers (four with -hanthey ‘to (colloquial)’ and four with -eykey ‘to (more formal)’), and the eight distractors consisted of sentences with different tenses (four with present tense and four with future tense). An example is given below in (147), where the comitative Case marker -hako is attached incorrectly to the kawi ‘scissors’. Participants were expected to write “1” to indicate that the sentence was true and then either repeat back the sentence with -hako still

attached or to correct the sentence by using -ul kaci-ko ‘-ACC bringing’ instead.

108

(147) Elin-i-tul-un kawi-hako nol-myen wihem-ha-Ø-ta. Young-person-PL-CONTR scissors-COMIT play-COND dangerous-do-PRES-DECL.III ‘It’s dangerous for children to play with scissors.’

There was a mixture of sentences with different truth values. The sentences were semi-randomized and counterbalanced across two versions. The participants were given four examples as training items. For one of the training items, it was pointed out that the sentence was actually ungrammatical. The participants were then shown how the sentence should be corrected, but were not explicitly instructed to correct any ungrammatical sentences they encountered. This was to encourage spontaneous corrections inspired by an incompatibility with their grammars and to avoid having the participants deliberately change the sentence to avoid having to mention the target structure.

The main assumption behind the use of elicited imitation tasks is that speakers can only produce what is possible or allowed in their grammars, even if mere repetition of a particular sentence or structure is required (Munnich et al. 1994). Having a time delay and taking the focus off the actual repetition of the sentence by having the participants first determine whether the sentence is true or false helps restrict the participants’ access to prescriptive grammar and keeps the focus on the content of the sentence, rather than its form. Under the pressure of time, participants are expected to internalize the content of the sentence and then repeat back their interpretation of the sentence using different structures or words, if necessary, depending on what is available in their grammars (Erlam 2006). Spontaneous corrections in particular help give insight to the nature of the participants’ internal grammars. Elicited imitation tasks have been used successfully in various studies76, including those on bilingual language (HL) acquisition, such as Erlam (2006) and Pérez-Leroux et al. (2011), as both an independent measure of proficiency and as a test for speakers’ acquisition of various phenomena. Thus, for the current study, the elicited imitation task was also intended to serve the dual purpose of providing a measure of the participants’ proficiency as well as giving an indication of whether HLS were truly familiar enough with the correct usage of conjunctions, classifiers, Case markers, tense, and -tul to repeat back the morphemes correctly and make any appropriate changes to ungrammatical sentences. In order to calculate the participants’ scores on this task, the sentences the participants produced were

76 This method was also used by Sherkina-Lieber (2011) as an effective test of production for receptive bilinguals who were HLS of Inuttitut.

109 transcribed and the transcriptions were checked and confirmed by a Korean-dominant bilingual. The nouns were coded for plural marking and Case marking, and the verbs were coded for tense.

In the following chapter, I present the language histories of the participants who participated in this study. Given that HLS comprise a heterogeneous group of varying levels of proficiency, I also discuss different measures of the participants’ ability in Korean, in order to assess their relative ability in Korean on aspects not related to the nominal domain.

110

Chapter 5 Description of the Participants 5 Participants 5.1 Age of Onset of Exposure to English and Time Spent in Korea

In order to test whether age of onset of exposure to English was a good predictor of the HLS’ performance, the participants were initially grouped according to their age of arrival (AoA) (with the assumption that this corresponded to their age of onset of sustained exposure to English). The first group consisted of 15 simultaneous bilinguals (five males and ten females) who were born in Canada or who had emigrated from Korea at the age of three or younger. Their ages at the time of testing ranged from 19 to 30 years old (median age: 25 years old, IQR: 4). The second group consisted of 18 sequential bilinguals (nine males and nine females) who had emigrated between the ages of four and ten77, with the exception of one speaker who was born in Canada, but had moved to Korea when he was eight months old. According to the criteria that simultaneous bilinguals are those who are exposed to more than one language by the age of three, he was technically a simultaneous bilingual, but from birth till the age of eight months he likely would have only had very limited exposure to English before moving to Korea and he had returned to Canada when he was much older, at age eight. His responses on the language history questionnaires confirmed that his language experiences were quite different from the rest of the simultaneous bilinguals, thus this speaker was grouped with the sequential bilinguals, rather than the simultaneous bilinguals. The sequential bilinguals’ ages ranged from 18 to 35 years old (median age: 20 years old, IQR: 4).

Some of the other HLS had also moved back and forth between Canada (or another English- speaking country, such as the US or UK) and Korea at various points in their life. For these HLS, instead of their age of arrival in Canada, the age at which they first moved to and lived in an English-speaking country (min. one year) was used as their age of onset of sustained exposure to English (AoO), since the age of first (sustained) exposure to another language is what is considered to affect the development of a heritage language. The median age of onset for the

77 Montrul (2002) found that the L1 of speakers who immigrated to another (L2-speaking) country after age 9 were less likely to undergo attrition.

111 simultaneous bilinguals was 0.0 years old (IQR: 0.0) and the median age of onset for the sequential bilinguals was 8.0 years old (IQR: 2.8). However, the age of onset of sustained exposure to English does not necessarily correspond to the age at which sustained exposure to Korean ended, especially for the HLS who moved around during their childhood. The data is summarized in Table 18 below.

Table 18. HLS’ Age-Related Information

Simultaneous bilingualsa Sequential bilingualsb

Characteristic Mdn (IQR) M (SD) Range Mdn (IQR) M (SD) Range

Age 25.0 (4.0) 24.7 (3.2) 19–30 20.0 (3.8) 22.6 (5.1) 18–35

AoO 0.0 (0.0) 0.6 (1.2) 0–3 8.0 (2.8) 7.4 (1.9) 0–10

# of yrs. in Korea 0.4 (2.8) 1.6 (2.2) 0.0–7.0 8.1 (2.0) 7.9 (1.9) 4.1–10.0 Note. The number of years spent in Korea includes the length of time the HLS lived in Korea or stayed for vacation. a N=15. b N=18.

Contingency tables showing the distribution of the HLS according to their status as simultaneous or sequential bilinguals compared to the number of years they had spent in Korea (low vs. high), and also comparing their ages of onset of exposure to English (low vs. high) compared to the number of years they had spent in Korea (low vs. high) are given below as Table 19 and Table 20, respectively.

Table 19. Distribution of HLS by Type of Bilingual and Number of Years Spent in Korea

# of yrs. in Korea Simultaneous bilinguals Sequential bilinguals

Low 14 2

High 1 16

112

Table 20. Distribution of HLS by AoO and Number of Years Spent in Korea

Age of onset of English

# of yrs. in Korea Low High

Low 16 0

High 1 16

The results of Table 19 show that the grouping of the HLS as simultaneous vs. sequential bilinguals may not be completely accurate for at least three of the HLS. For example, one speaker was born in Canada, but moved back and forth between Korea and English-speaking countries, and thus spent longer in Korea than the other simultaneous bilinguals. For the sequential bilinguals, compared to the rest of the group, the HLS who had arrived at the age of four, and who had begun to receive exposure to English at this age, would actually be grouped with the simultaneous bilinguals in terms of the length of time they had actually spent in Korea, if the speakers were divided into two even groups, based on the number of years they had spent in Korea. When the HLS with AoO of four years of age were included in the “low” group (see Table 20), only the speaker who had moved back and forth had atypical experience (i.e., had spent longer in Korea) compared to the rest of the group.

To measure how long the HLS received sustained exposure to Korean, the number of years that the HLS had spent in Korea, including any vacations, was independently totaled based on their responses on the language background questionnaires. In general, this number was close to their age of onset of exposure to English (Kendall’s correlation was large and highly significant (τ=0.87, z=6.6, p<.001), but for the HLS who had moved back and forth to and from Korea, the number of years spent in Korea provided a more accurate picture of the HLS’ exposure to Korean. The median number of years that the simultaneous bilinguals had spent in Korea was 0.4 years (about five months) (IQR: 2.8 years) and the median number of years spent in Korea by the sequential bilinguals was 8.1 years (IQR: 2.0 years). After dividing the participants according to their age of arrival in an English-speaking country, there was roughly an eight year difference between the simultaneous and sequential bilinguals in their ages of onset of sustained exposure to English and the number of years they had spent in Korea. However, there are opportunities for simultaneous bilinguals to receive input in, and use, Korean even in Canada, so the HLS were

113 asked to provide detailed information on the various avenues through which they are exposed to Korean, as well as how often they speak Korean.

5.2 Usage and Exposure

The HLS’ ages of onset of exposure to English and the total number of years the HLS spent in Korea provide us with an estimate of how much input they received in Korean and how much they used Korean. However, in order to get a better picture of what kinds of language experiences they actually had, the HLS were asked to provide information on different aspects of their language usage and exposure. Their responses showed that the sequential bilinguals in general had more exposure to Korean, used Korean more, and were more confident in using Korean than the simultaneous bilinguals (see Appendix Tables A1–A6 and Figure A1 for summaries of responses to language experience questionnaires). In order to condense and compare the HLS’ usage of and exposure to Korean, composite scores were calculated based on the HLS’ responses to ten of the questions (five for each measure) on the language history questionnaires. The questions were selected by the experimenter to be representative of different sources of input (i.e., exposure) and output (i.e., usage). Since much of the input that HLS receive is from their family members or friends, the HLS’ responses to the number of friends they spoke to in Korean, how often they did not code-switch to English while they were speaking Korean (i.e., the less English they used, the higher their score), how often they used Korean (vs. English) at home and in social situations, and the percentage of text/instant messages they sent in Korean (vs. English) were compiled and converted to numerical scores, if necessary. For example, the HLS were asked to list the number of friends who they spoke to in Korean “always”, “frequently”, “occasionally”, and “never”. Each level of frequency was assigned a score from zero to three, where three represented “always” and zero represented “never”, and this score was multiplied by the number of friends the HLS listed as speaking to with that frequency. The responses were totaled and then divided by the number of friends, and this value was used as a measure of how often the HLS spoke to their friends in Korean. For how often the HLS code-switched, the participants were asked to indicate on a four-point Likert scale (“always”, “frequently”, “occasionally”, or “never”) how often they mixed in English while speaking in Korean. Their responses were then converted to numerical scores, such that higher numbers indicated less code-switching (i.e., a “three” represented “never” and a “zero” represented “always”). No conversions were needed for the other categories, since the results

114 represented percentages. The HLS were grouped together and ranked in terms of increasing usage, such that “one” corresponded to the participant who used Korean the least, and “thirty- three” corresponded to the participant who used Korean the most amongst the HLS tested. The rankings from each category were then totaled for each participant; the results are summarized in Table 21 (see Appendix A7 for an unadjusted correlation table for the HLS’ responses on these different categories comprising their usage score for Korean).

Table 21. HLS’ Median Responses to Questions Used to Create a Composite Score for Use of Korean

Simultaneous bilinguals (N=15) Sequential bilinguals (N=18)

Mdn score Mdn rank Range of Mdn score Mdn rank Range of (IQR) (IQR) scores (IQR) (IQR) scores

Spoke Korean to 0.8 (0.7) 13.0 (11.5) 0.1 – 1.9 1.0 (1.3) 14.0 (15.5) 0.0 – 2.2 friends

Avoided code- 1.0 (1.5) 9.0 (14.5) 0.0 – 2.0 1.3 (1.0) 15.0 (13.0) 0.0 – 2.0 switching

% Korean used at 30.0 (42.5) 12.0 (11.0) 0.0 – 95.0 55.0 (51.3) 18.0 (14.8) 0.0–100.0 home

% Korean used in 20.0 (25.0) 19.0 (15.0) 1.0 – 50.0 10.0 (46.3) 15.0 (24.0) 0.0 – 80.0 social situations

% of texts sent in 1.0 (7.5) 15.0 (21.5) 0.0 – 50.0 1.0 (31.3) 15.0 (25.0) 0.0 – 95.0 Korean

Sum of usage 57.5 (50.2) 60.0 (24.5) 36.0–127.0 75.0 (99.1) 74.5 (45.0) 17.0–142.0 totals

Overall, the sequential bilinguals used Korean more often than the simultaneous bilinguals tested (median sum of rankings: 74.5 (IQR: 45.0) for the former and 60.0 (IQR: 24.5) for the latter). The median scores of the two groups were fairly similar for how often they spoke to their friends in Korean, how often they avoided code-switching, and how often they sent text messages in Korean, but the sequential bilinguals reported higher (median) usage of Korean at home (55% vs. 30% for the simultaneous bilinguals), as expected, and unexpectedly, they reported lower use of Korean in social situations (10% vs. the simultaneous bilinguals’ median reported usage of 20%), which was unexpected. However, there was more variation in the responses from the sequential bilinguals and the range was greater; both groups had speakers who did not really

115 speak Korean at all in social situations, but the highest percentage of Korean used by the simultaneous bilinguals was 50%, whereas the highest percentage of Korean used by the sequential bilinguals was 80%.

To measure the HLS’ exposure to Korean, the HLS’ responses on the number of family members that spoke to them in Korean, the number of hours of TV they watched in Korean per week, the number of Korean movies they watched per year, the percentage of music they listened to in Korean, and the number of hours they spent online browsing in Korean per week were compiled. The HLS were asked how often various family members spoke to them in Korean on a four-point Likert scale (“always”, “frequently”, “occasionally”, or “never”), and similar to the way the responses were converted for the number of friends that the HLS spoke to in Korean, these responses were scored as a “three” if a family member “always” spoke to the participant in Korean and a “zero” if the family member “never” spoke in Korean. These responses were totaled and divided by the total number of family members that the HLS provided responses for, and these values were used to represent how much input the HLS received in Korean from their families. The other categories used to measure the HLS’ exposure involved numerical data and did not need to be converted. The participants were grouped together and ranked according to their responses in each category and these rankings were then summed together for each participant to create a composite score representing their input in Korean. The results are summarized in Table 22 below (see Appendix A8 for an unadjusted correlation table for the HLS’ responses on these different categories comprising their exposure score for Korean).

116

Table 22. HLS’ Median Responses to Questions Used to Create a Composite Score for Exposure to Korean

Simultaneous bilinguals (N=15) Sequential bilinguals (N=18)

Mdn score Mdn rank Range of Mdn score Mdn rank Range of Category (IQR) (IQR) scores (IQR) (IQR) scores

How often family 2.0 (0.6) 10.0 (8.0) 0.8 – 3.0 2.5 (0.3) 17.0 (7.0) 1.5 – 3.0 members speak in Korean

# of hrs. of 2.0 (3.8) 15.0 (12.0) 0.0 – 7.0 2.5 (6.8) 14.0 (24.8) 0.0– 28.0 Korean TV/wk.

# of Korean 2.0 (3.0) 11.0 (13.0) 0.0–10.0 2.0 (7.4) 11.0 (21.5) 0.0– 60.0 movies/yr.

% of music in 10.0 (36.0) 10.0 (14.5) 0.0–70.0 20.0 (40.0) 18.0 (13.0) 0.0– 95.0 Korean

# of hrs. spent 0.0 (2.0) 1.0 (19.5) 0.0 – 6.0 2.0 (4.8) 20.0 (26.8) 0.0– 25.0 browsing in Korean/wk.

Sum of exposure 16.0 (37.6) 59.0 (43.5) 11.0–99.0 28.9 (60.2) 86.5 (61.8) 8.0–151.0 scores

Similar to their responses to the usage-related questions, the sequential bilinguals reported higher exposure to Korean than the simultaneous bilinguals overall (median sum of rankings for sequential bilinguals: 86.5 (IQR: 61.8), compared to 59.0 (IQR: 43.5) for the simultaneous bilinguals). The sequential bilinguals’ reported exposure to Korean was higher in all categories, except for the median number of Korean movies they watched per year, which was the same for both groups of bilinguals, but the sequential bilinguals had greater variance in their responses (IQR: 21.5), compared to the simultaneous bilinguals (IQR: 13.0). Contingency tables showing the distribution of the HLS according to their levels of input and usage of Korean are given below.

117

Table 23. Distribution of Simultaneous Bilinguals by Exposure to and Use of Korean

Exposure

Usage Low High

Low 6 3

High 2 4

Table 24. Distribution of Sequential Bilinguals by Exposure to and Use of Korean

Exposure

Usage Low High

Low 6 1

High 2 9

Table 25. Distribution of HLS by Exposure to and Use of Korean

Exposure

Usage Low High

Low 12 4

High 4 13

Table 23 shows that the simultaneous bilinguals were a diverse group with varying levels of exposure to and usage of Korean. The largest group amongst the simultaneous bilinguals consisted of speakers who received relatively less input in Korean and who used Korean relatively less than the other speakers. Table 24 shows that an equal number of sequential bilinguals had similarly low levels of exposure to and usage of Korean, but the largest group amongst the sequential bilinguals consisted of speakers who received high levels of input in Korean and who used Korean relatively more often than the other HLS. Only a few speakers had unequal levels of exposure and usage of Korean. When the HLS were combined together, the results show (Table 25) that the HLS could be split into roughly equal groups of speakers who either had both less input and lower usage of Korean or who had both more input and higher usage of Korean. The remaining speakers could be split into equal groups of speakers who used Korean less, compared to the high levels of input they received, or who used Korean quite often,

118 but who received relatively little input in Korean. The language experience information to be used in the analysis of the HLS’ performance is summarized in Table 26 below (see Appendix A9 for summaries for each participant).

Table 26. Summary of HLS’ Language Experience Information

Simultaneous bilinguals (N=15) Sequential bilinguals (N=18)

Category Mdn (IQR) M (SD) Range Mdn (IQR) M (SD) Range

Age of onset of 0.0 (0.0) 0.6 (1.2) 0.0 – 3.0 8.0 (2.8) 7.4 (1.9) 4.0 – 10.0 English

# of yrs. in Korea 0.4 (2.8) 1.6 (2.2) 0.0 – 7.0 8.1 (2.0) 7.9 (1.9) 4.1 – 10.0

Sum of usage 60.0 (24.5) 66.0 (26.7) 36.0–127.0 74.5 (45.0) 78.7 (37.6) 17.0–142.0 rankings

Sum of exposure 59.0 (43.5) 62.5 (27.7) 11.0 – 99.0 86.5 (61.8) 80.4 (42.7) 8.0–151.0 rankings

Table 26 shows that the simultaneous and sequential bilinguals differed in language experiences, with the biggest difference being in their exposure (i.e., input) to Korean. The participants were asked to perform different experimental tasks in order to analyze various aspects of their Korean. The methods for these tasks will be discussed in the following section.

5.3 The Heritage Language Speakers’ Ability in Korean

Several measures were used to determine the HLS’ general ability in Korean. The HLS were asked for self-ratings of their ability, and their elicited narratives and elicited imitations were analyzed for additional measures. The subordination index (SI), mean length of utterance (MLU), and the number of different words used (NDW) by the HLS were calculated from the HLS’ elicited narratives to measure the level of embedding of the HLS’ structures, the average length of their sentences, and to measure their vocabulary. From the elicited imitation task, the number of verbal morphemes correctly repeated by the HLS was calculated as well, as a passive measure of their familiarity with the grammar. Each measure of the HLS’ ability in Korean will be discussed in detail below.

119

5.3.1 Self-Rated Ability in Korean

The HLS were given two questionnaires to rate their ability in Korean. One was intended to elicit a direct comparison of the HLS’ ability in Korean and English, and the other was intended to determine the types of tasks that HLS were able to accomplish in Korean. On the first questionnaire, the HLS were given a four-point Likert scale (beginner, intermediate, advanced, or near-native) and asked which level they considered themselves to be at in reading, writing, speaking, listening, and overall in Korean. Their responses were converted to numerical scores as follows: beginner: 1, intermediate: 2, advanced: 3, and near-native: 4. A summary of the HLS’ responses is shown below in Table 27; information on their median scores are provided, given that the numbers correspond to ordinal data.

Table 27. Summary of HLS’ Self-Ratings on their Ability in Korean based on a Four-point Likert Scale

Simultaneous bilinguals (N=15) Sequential bilinguals (N=18)

Category Mdn (IQR) M (SD) Range Mdn (IQR) M (SD) Range

“Reading” 2.0 (1.0) 2.1 (0.9) 1 – 4 2.0 (1.0) 2.3 (0.8) 1 – 4

“Writing” 2.0 (1.0) 1.9 (0.7) 1 – 3 2.0 (1.0) 1.7 (0.7) 1 – 3

“Speaking” 2.0 (0.5) 2.1 (0.8) 1 – 4 3.0 (1.0) 2.5 (0.9) 1 – 4

“Listening” 2.0 (1.0) 2.5 (1.0) 1 – 4 3.0 (0.8) 2.8 (1.0) 1 – 4

“Overall” 2.0 (1.5) 2.1 (0.8) 1 – 3 2.5 (1.0) 2.4 (0.8) 1 – 4

Overall median 2.0 (1.5) 2.0 (1.1) 1 – 3 2.5 (1.0) 2.5 (0.6) 1 – 4

Overall sum 10.0 (5.5) 10.7 (3.8) 5–16 12.5 (3.0) 11.7 (3.8) 5–19

The median scores of the simultaneous bilinguals were the same across the board (2.0, corresponding to an intermediate level; median sum of scores: 10.0, IQR: 5.5); the majority of the simultaneous bilinguals rated themselves as having beginner-to-intermediate level ability in Korean overall. In contrast, the sequential bilinguals assigned themselves higher scores on speaking and listening, and overall the majority of the sequential bilinguals rated themselves as having intermediate-to-advanced proficiency (overall median: 2.5, IQR: 1.0; median sum of scores: 12.5, IQR: 3.0). A Mann-Whitney U-test for independent samples revealed that the

120 difference between the medians was not significant (U=107, p=.29), which suggests that despite their different ages of onset of exposure to English, the simultaneous bilinguals did not appear to be judging their ability in Korean significantly lower than the sequential bilinguals judged themselves to be.

The HLS were also asked to rate their ability in English using the same scale78 in order to compare these self-ratings with their self-ratings in Korean (see Table 28 for results). As above, the median scores are presented, given that the numbers represent ordinal data. All of the simultaneous bilinguals rated themselves at the highest level of proficiency in all categories (median: 4.0, IQR: 0.0; median sum of scores: 20.0, IQR: 0.0). The sequential bilinguals’ median scores in each category were also at the highest level of proficiency (4.0), but there was more variation in their responses; the sequential bilinguals’ scores generally ranged between advanced and near-native-like ability (overall median: 4.0, IQR: 0.0; median sum of scores: 19.0, IQR: 1.8). A Mann-Whitney U-test for independent samples indicated that the difference in the overall median scores between the two groups was not significant (W=150, p=.21).

Table 28. Summary of HLS’ Self-Rating on their Ability in English based on a Four-point Likert Scale

Simultaneous bilinguals (N=15) Sequential bilinguals (N=18)

Category Mdn (IQR) M (SD) Range Mdn (IQR) M (SD) Range

“Reading” 4.0 (0.0) 4.0 (0.0) 4–4 4.0 (0.0) 3.8 (0.5) 2–4

“Writing” 4.0 (0.0) 4.0 (0.0) 4–4 4.0 (1.0) 3.6 (0.6) 2–4

“Speaking” 4.0 (0.0) 4.0 (0.0) 4–4 4.0 (0.8) 3.7 (0.6) 2–4

“Listening” 4.0 (0.0) 4.0 (0.0) 4–4 4.0 (0.0) 3.8 (0.5) 2–4

“Overall” 4.0 (0.0) 4.0 (0.0) 4–4 4.0 (0.0) 3.7 (0.6) 2–4

Overall median 4.0 (0.0) 4.0 (0.0) 4–4 4.0 (0.0) 3.8 (0.4) 2–4

Overall sum 20.0 (0.0) 20.0 (0.0) 20–20 19.0 (1.8) 18.6 (2.5) 10–20

78 This questionnaire was based on a questionnaire for an L2 study, thus the highest rating that participants could give themselves was “near-native-like”, although some, especially the simultaneous bilinguals, mentioned that they had “native” proficiency in English.

121

The difference between the HLS’ self-ratings on their ability in Korean and their ability in English was calculated (see Table 29) to determine whether the participants were Korean- dominant or English-dominant. The simultaneous bilinguals were clearly English-dominant, given that their self-ratings for English were all at the highest level (4.0), whereas their self- ratings for Korean were generally at the intermediate level (2.0; overall median difference: -2.0 (IQR: 1.5)). For the sequential bilinguals, the difference between their self-ratings on the two languages was similarly two levels lower for reading and writing in Korean, compared to English, but only one level lower in the other categories (overall median difference: -1.0 (IQR: 1.0)). A Mann-Whitney U-test revealed that the difference in the two groups’ overall median difference scores was marginally significant (U=85.5, p=.06). The results suggest that the sequential bilinguals were overall closer to being balanced in ability in the two languages, at least in speaking and listening, even though overall, like the simultaneous bilinguals, they were English-dominant.

Table 29. Summary of the Difference between the HLS’ Self-Ratings on their Ability in Korean and English based on a Four-point Likert Scale

Simultaneous bilinguals (N=15) Sequential bilinguals (N=18)

Category Mdn (IQR) M (SD) Range Mdn (IQR) M (SD) Range

“Reading” -2.0 (1.0) -1.9 (0.9) -3– 0 -2.0 (1.0) -1.4 (1.0) -3– 0

“Writing” -2.0 (1.0) -2.1 (0.7) -3– -1 -2.0 (2.0) -1.9 (1.1) -3– 0

“Speaking” -2.0 (0.5) -1.9 (0.8) -3 – 0 -1.0 (2.0) -1.2 (1.3) -3– 1

“Listening” -2.0 (1.0) -1.5 (1.0) -3 – 0 -1.0 (1.0) -1.0 (1.1) -3– 1

“Overall” -2.0 (1.5) -1.9 (0.8) -3– -1 -1.0 (1.0) -1.3 (1.1) -3– 1

Overall median -2.0 (1.5) -2.0 (1.1) -3– -1 -1.0 (1.0) -1.4 (1.1) -3– 0

Overall sum -10.0 (5.5) -9.3 (3.8) -15– -4 -7.0 (5.3) -6.9 (4.9) -15–2

To get a clearer picture of what the HLS considered themselves to be capable of in Korean, the HLS were also asked to rate themselves using a questionnaire which listed various skills and activities that the speakers would be able to accomplish at six different specified levels of ability in the categories of spoken interaction, spoken production, listening, reading, and writing (from

122 the Common European Framework of Reference for Languages: Learning, Teaching, Assessment (Council of Europe 2001)). The results are summarized in Table 30.

Table 30. Summary of HLS’ Self-Ratings on their Ability in Korean based on Specific Criteria

Simultaneous bilinguals (N=15) Sequential bilinguals (N=18)

Category Mdn (IQR) M (SD) Range Mdn (IQR) M (SD) Range

Spoken interaction 4.0 (1.0) 3.6 (1.1) 1 – 5 4.0 (1.8) 4.1 (1.1) 2 – 6

Spoken production 3.0 (0.5) 3.2 (0.9) 2 – 5 3.0 (1.0) 3.4 (1.3) 1 – 6

Writing 2.0 (1.5) 2.2 (1.0) 1 – 4 3.0 (0.8) 2.7 (1.1) 0 – 4

Listening 4.0 (2.0) 4.0 (1.4) 1 – 6 5.0 (1.0) 4.4 (1.1) 2 – 6

Reading 3.0 (0.5) 2.9 (0.9) 1 – 5 3.0 (1.0) 3.4 (1.2) 1 – 6

Overall median 3.0 (0.5) 3.2 (0.9) 1 – 5 4.0 (1.0) 3.6 (1.0) 1 – 5

Overall sum 16.0 (6.0) 15.9 (4.6) 9–25 19.0 (3.0) 17.9 (5.0) 6–25

Compared to their self-ratings on the first questionnaire, there was more variation in the HLS’ self-ratings on the different categories, not only because there were more levels, but also likely due to the re-evaluating the HLS had to do in order to see if they met the specifications on what they should be capable of at a particular level of ability. On the first questionnaire, the simultaneous bilinguals rated themselves consistently at the intermediate level (2.0), whereas on the second questionnaire their median self-ratings ranged from 2.0 to 4.0 (out of 6) across the various categories. With the more fine-grained distinctions between the levels, these bilinguals tended to rate themselves the highest on their spoken interaction79 (median score: 4.0, IQR: 1.0) and listening skills (median score: 4.0, IQR: 2.0), and the worst on writing (median score: 2.0, IQR: 1.5). Similarly, the sequential bilinguals rated themselves the highest on spoken interaction (Mdn: 4.0, IQR: 1.8) and listening (Mdn: 5.0, IQR: 1.0). The sequential bilinguals’ lowest self- ratings were on their ability in writing, spoken production, and reading (Mdn: 3.0), but the range of their self-ratings on writing (0–4) differed from that of the other categories (1–6 or 2–6). This suggests that, like the simultaneous bilinguals, the sequential bilinguals were the least confident

79 Spoken interaction referred to the ability to carry on a conversation with someone; spoken production referred to the ability to present one’s ideas clearly, for example, in a presentation.

123 in their writing skills. The HLS’ self-ratings were in line with the generalizations about HLS having better speaking and listening skills than reading and writing skills. The difference between the median scores of the simultaneous bilinguals (3.0 (IQR: 0.5)) and the sequential bilinguals (4.0 (IQR: 1.0)) was not significant in a Mann-Whitney U-test (U=97.5, p=.16).

In order to determine whether there was a relationship between the HLS’ self-ratings and their language backgrounds discussed in Section 5.1 and Section 5.2 (i.e., their AoO, the number of years they had spent in Korea, and their usage of and exposure to Korean), various generalized linear models with the various aspects of the HLS’ language backgrounds included as predictors were fit to the HLS’ overall median self-ratings on their ability in Korean using R. Since the HLS’ ages of arrival were significantly correlated with the number of years the HLS had spent in Korea (τ=0.87, z=6.6, p<.001), only one of these measures was entered into any particular model. In addition, the HLS’ composite usage and exposure scores were converted to z-scores before they were entered into the models.

The HLS’ language background variables were added into models as predictors, and the models were evaluated additively, starting with single predictor models (Models 2–5). The models were compared to the null model (i.e., a model with just an intercept; Model 1) using likelihood ratio tests to determine whether the more complex models were better at accounting for the variation in the data. The likelihood ratio tests were performed by using one-way ANOVAs in R (see Table 31 for the results).

Table 31. Model Comparisons of Single Predictor Models to the Null Model of the HLS’ Self- rated Ability Scores (Model 1)

Model Predictor RSS G p

1 (Null model) 40.0

2 AoO 34.4 4.9* .03

3 # of yrs. in Korea 31.1 8.8** <.01

4 Usage 24.6 19.2*** <.001

5 Exposure 34.3 5.6 .03 Note. RSS = residual sum of squares.

124

All models were significantly better than the null model at accounting for the variation in the data. The HLS’ usage scores accounted for more of the variation than the other predictors (residual sum of squares: 24.6 vs. residual sum of squares of the null model: 40.0; highly significant at G=19.2, p<.001), thus two predictor models were compared using likelihood ratio tests to the model with just the HLS’ usage scores as a predictor (Model 4). The results are given in Table 32 below.

Table 32. Model Comparisons of Two-Predictor Models to a Model with the HLS’ Usage Scores as a Predictor of the HLS’ Self-rated Ability Scores (Model 4)

Model Predictors RSS G p

4 Usage 24.6

6 Usage + AoO 23.5 1.4 .23

7 Usage + # of yrs. In Korea 21.7 4.0* .05

8 Usage + Exposure 24.6 0.1 .78 Note. RSS = residual sum of squares.

Table 32 shows that only a model with the number of years the HLS had spent in Korea as an additional predictor with the HLS’ usage scores (Model 7) was significantly better (G=4.0, p=.05) at accounting for the variation in the data than the model with just the HLS’ usage scores as a predictor (Model 4). In order to determine whether adding the HLS’ exposures scores into the model improved in accounting for the data, the full model (Model 9) was compared to Model 7 in a likelihood ratio test (see Table 33).

Table 33. Model Comparisons of a Three-Predictor Model to a Model with the HLS’ Usage Scores and the Number of Years Spent in Korea as Predictors of the HLS’ Self-rated Ability Scores (Model 7)

Model Predictors RSS G p

7 Usage + # of yrs. in Korea 21.7

9 Usage + # of yrs. in Korea + Exposure 21.0 0.9 .34 Note. RSS = residual sum of squares.

125

Adding the HLS’ exposures scores into the model (Model 9) only resulted in an improvement of 0.7 in the residual sum of squares; this model was not significantly better (G=0.9, p=.34) at accounting for the data; thus, this model was rejected in favour of Model 7. A summary of Model 7 is given in Table 34.

Table 34. Model 7: Generalized Linear Model Predicting HLS’ Self-rated Ability Scores in Korean Using the Number of Years in Korea and Usage Scores as Predictors

Predictor B (SE) t p

Intercept 3.0 (0.3) 11.3*** <.001

# of years in Korea 0.1 (0.0) 2.0* .05

Usage 0.6 (0.2) 3.6** <.01

Every increase in the HLS’ number of years spent in Korea was associated with about a 0.1 point increase in the HLS’ self-rated ability score, and every increase in one standard deviation in the HLS’ usage scores was associated with about a 0.6 point increase in their self-rated ability score as well. The r2 of Model 7 was 0.46, and the adjusted r2 was 0.42, which indicates that this model was able to account for 42% of the variation in the HLS’ self-rated ability scores in Korean. The residual standard error was 0.9 with 30 degrees of freedom, which is quite low, and the model was significant at F(2,30)=12.6, p<.001. Rather than how much input the HLS received in Korean, it appears that how often they actually spoke in Korean and how long they had spent in Korea may have factored in the most in how the HLS rated themselves in ability in Korean.

126

Table 35. Summary of HLS’ Self-Rated Ability according to their Levels of Exposure to Korean and the Number of Years They Spent in Korea

Exposure

# of yrs. in Korea Low High

Low

Mdn (IQR) 3.0 (1.5) 3.0 (0.0)

M (SD) 3.0 (1.4) 3.0 (0.6)

Number of speakers 10 6

High

Mdn (IQR) 3.5 (1.8) 4.0 (1.0)

M (SD) 3.3 (1.2) 4.0 (0.8)

Number of speakers 6 11

Table 35 shows that when the speakers were grouped according to whether they had spent relatively little time in Korea or relatively more time in Korea, and according to whether they had less or more exposure to Korean, the speakers who had spent more time in Korea rated themselves as being more proficient.

127

Table 36. Summary of HLS’ Self-Rated Ability according to their Usage and Number of Years Spent in Korea

Usage

# of yrs. in Korea Low High

Low

Mdn (IQR) 3.0 (1.8) 3.0 (0.0)

M (SD) 2.8 (1.3) 3.3 (0.8)

Number of speakers 10 6

High

Mdn (IQR) 3.0 (1.5) 4.0 (1.0)

M (SD) 3.0 (0.9) 4.2 (0.8)

Number of speakers 6 11

Table 36 shows that when the speakers were grouped according to whether they had spent relatively little time in Korea or more time in Korea, and according to how often they used Korean, the speakers with the higher levels of usage of Korean gave themselves higher ratings on their ability in Korean.

The results were similar when the HLS were divided according to their AoOs and their usage and exposure scores. Table 37 shows that, rather than their exposure scores, the HLS’ ages of onset of extended exposure to English appear to have played a bigger role in the self-ratings they gave themselves. When we compare this to when their usage scores were taken into account, as in Table 38, their usage scores were better predictors than their AoOs in predicting how high the HLS would rate themselves to be in Korean.

128

Table 37. Summary of HLS’ Median Self-Rated Ability according to their Levels of Exposure to Korean and their AoOs

Exposure

AoO Low High

Low

Mdn (IQR) 3.0 (1.5) 3.0 (0.5)

M (SD) 3.0 (1.4) 3.3 (1.0)

Number of speakers with ability>3 3/10 2/7

High

Mdn (IQR) 3.5 (1.8) 4.0 (0.8)

M (SD) 3.3 (1.2) 3.9 (0.7)

Number of speakers with ability>3 3/6 7/10

Table 38. Summary of HLS’ Median Self-Rated Ability according to their Usage of Korean and their AoOs

Usage

AoO Low High

Low

Mdn (IQR) 3.0 (1.8) 3.0 (1.0)

M (SD) 2.8 (1.3) 3.6 (1.0)

Number of speakers with ability>3 3/10 2/7

High

Mdn (IQR) 3.0 (1.5) 4.0 (0.8)

M (SD) 3.0 (0.9) 4.1 (0.7)

Number of speakers with ability>3 2/6 8/10

As another measure of the HLS’ ability, the HLS’ responses in the elicited imitation task were analyzed, as will be discussed in the following section.

129

5.3.2 Imitation Scores

There were two versions of the elicited imitation task. From each version, 17 grammatical sentences were selected for analysis.80 In version 1, the sentences ranged from 9–-20 syllables (mean: 17 syllables) and consisted of 8–18 morphemes (mean: 13 morphemes). In version 2, the sentences ranged from 11-21 syllables (mean: 17 syllables) and consisted of 10–-18 morphemes (mean: 14 morphemes). The data was analyzed in two different ways. Initially, a morpheme-for- morpheme count analysis was conducted to see how accurate the participants were in repeating back the words they had heard. Morphemes, rather than words, were counted since Korean is an agglutinative language. Lexical additions (as well as the plural morpheme) were not penalized, unless they resulted in ungrammaticality. Paraphrases were also generally accepted (e.g., teyly-e ka-n-ta ‘take-e go-PRES-DECL.III’ instead of teyli-ko ka-n-ta ‘take-KO go-PRES-DECL.III’), unless they reduced the overall number of morphemes (i.e., they eliminated an obligatory context for a morpheme, as in twul-ul ‘two-ACC’ for twu myeng-ul ‘two CL-ACC’, where the classifier is missing) or caused a significant change in meaning. The proportion of verbal morphemes the HLS repeated correctly was counted as a separate measure of ability. The verbal tense markers and complementizers were counted together, since the appropriateness of a particular tense marker depends on the complementizer used. A summary of the results is given in Table 39.

80 Four native speakers of Korean were asked to determine whether the lists of sentences were grammatical or not. Only sentences which were accepted by all four speakers were used.

130

Table 39. Summary of HLS’ Imitation Scores as a Measure of Ability in Korean

Simultaneous bilinguals (N=15) Sequential bilinguals (N=18)

Measure Mdn (IQR) M (SD) Range Mdn (IQR) M (SD) Range

Lexical verbs correctly repeated 38.0 (5.5) 35.0 (6.3) 20 – 41 39.0 (3.5) 36.4 (6.0) 17 – 41

Lexical verbs required 41.0 (2.0) 40.2 (1.1) 39 – 42 40.5 (2.0) 40.1 (1.1) 39 – 42

Tense/comp correctly repeated 36.0 (9.5) 32.9 (7.9) 14 – 40 39.0 (2.5) 36.2 (6.7) 15 – 41

Tense/comp required 41.0 (2.0) 40.2 (1.1) 39 – 42 40.5 (2.0) 40.1 (1.1) 39 – 42

Morphemes correctly repeated 74.0 (15.0) 67.9 (14.0) 37 – 81 78.0 (6.0) 72.7 (12.6) 32 – 82

Morphemes required 82.0 (4.0) 80.4 (2.2) 78 – 84 81.0 (4.0) 80.2 (2.2) 78 – 84

Proportion of morphemes correctly repeated 0.93 (0.18) 0.85 (0.17) 0.5 – 1.0 0.98 (0.07) 0.91 (0.16) 0.4 –1.0

Similar to the method used to analyze the HLS’ self-rated ability scores above, various models were also fit to the data in order to see whether a relationship existed between the HLS’ proportion of verbal morphemes correctly repeated and the HLS’ language backgrounds. The results of the likelihood ratio tests comparing models with a single predictor to a null model are given in Table 40.

131

Table 40. Model Comparisons of Single Predictor Models to a Null Model of the HLS’ Imitation Scores (Model 10)

Model Predictor RSS G p

10 (Null model) 0.9

11 AoO 0.8 3.5· .07

12 # of yrs. in Korea 0.8 4.6* .04

13 Usage 0.7 9.9** <.01

14 Exposure 0.7 5.9* .02 Note. RSS = residual sum of squares.

The results indicated that all of the models with a single predictor were significantly better than the null model at accounting for the variation, except for the age of first exposure to English, which was only marginally significantly better (G=3.5, p=.07). The model with the HLS’ usage scores was the best predictor of the proportion of verbal morphemes that the HLS were able to repeat correctly, thus, likehood ratio tests were performed to compare this model (Model 13) to models with two predictors, as shown in Table 41.

Table 41. Model Comparisons of Two-Predictor Models to a Model with Only the HLS’ Usage Scores as a Predictor of the HLS’ Imitation Scores (Model 13)

Model Predictors RSS G p

13 Usage 0.7

15 Usage + AoO 0.6 1.0 .32

16 Usage + # of yrs. in Korea 0.6 1.7 .20

17 Usage + Exposure 0.7 0.4 .51 Note. RSS = residual sum of squares.

The results indicate that adding other predictors to the HLS’ usage scores in a model representing the proportion of verbal morphemes that the HLS were able to repeat correctly did not significantly improve how much of the variation was accounted for in the data. The details of Model 13, with just the HLS’ usage scores as a predictor, are provided in Table 42 below.

132

Table 42. Model 13: Generalized Linear Model Predicting HLS’ Imitation Scores Using the HLS’ Usage Scores as a Predictor

Predictor B (SE) t p

Intercept 0.9 (0.0) 34.5*** <.001

Usage 0.1 (0.0) 2.5* .02

Every increase in one standard deviation point in the HLS’ usage scores was associated with about a 0.1 point increase in the number of verbal morphemes that the HLS repeated correctly. The r2 of Model 13 was 0.15, and the adjusted r2 was 0.24, which indicates that this model was able to account for 24% of the variation in the number of verbal morphemes repeated correctly by the HLS’. The residual standard error was 0.1 on 31 degrees of freedom. In general, the HLS performed at ceiling in repeating verbal morphemes, making it difficult to predict whether any aspects of their language backgrounds contributed to their results, but it appears that the proportion of verbal morphemes the HLS were able to correctly repeat in Korean was related to how often the HLS actually used Korean. This makes sense, given that the speakers who had more opportunities to speak in Korean would be better practiced in paying close attention to what they were hearing and would be better at repeating back what they had heard. The participants were grouped according to their AoOs and their levels of usage of Korean; the proportion of morphemes repeated correctly by the participants was recalculated and are summarized in Table 43.

133

Table 43. Summary of HLS’ Proportion of Morphemes Repeated Correctly in the Elicited Imitation Task according to HLS’ Usage of Korean and their AoOs

Usage

AoO Low High

Low 0.88 (0.16) 0.96 (0.23)

Mdn (IQR) 0.82 (0.19) 0.88 (0.14)

M (SD) 8/10 5/7

Number of speakers with at least 75% accuracy

High

Mdn (IQR) 0.85 (0.19) 0.99 (0.03)

M (SD) 0.80 (0.22) 0.98 (0.02)

Number of speakers with at least 75% accuracy 5/6 10/10

When the participants were grouped according to their AoOs and their usage of Korean, the HLS’ AoOs did not distinguish the speakers. Only five of the 33 speakers did not repeat at least 75% of the morphemes correctly. How often they used Korean was a better predictor of whether they would be successful at repeating the morphemes or not. The median proportion of morphemes correctly repeated was at ceiling for the speakers who used Korean frequently. AoO did not make a difference and was not a good predictor. The other measures of ability obtained from the HLS’ elicited narratives will be discussed in the following section.

5.3.3 Subordination Index, Mean Length of Utterance, and Number of Different Words

As additional measures of the HLS’ ability, the subordination index (SI), mean length of utterance (MLU), and the number of different words (NDW) used in the HLS’ elicited narratives were calculated. The HLS’ SIs were determined by counting the number of clauses in the HLS’ elicited narratives and dividing this number by the number of utterances. Clause boundaries were determined by the use of complementizers. Relative clauses with “activity verbs”, such as nol- ‘play’ in (148) below, were also counted as separate clauses, but clauses with adjectives or “descriptive verbs”, such as saylowu- ‘new’ in (149), were not counted as separate clauses.

134

(148) [Ku elun-tul-i nol-ten] game-ul That adult-PL-NOM play-PST.REL game-ACC ‘the game that the adults were playing’ (Participant #3)

(149) [Ku saylowu-n kay] That new-REL dog ‘that new dog’ (Participant #5)

In general, utterances corresponded to places where one would put a period; i.e., an utterance consisted of a string of related ideas. Speakers with higher SIs would be those who used more embedded clauses in their sentences. For the HLS’ MLUs, the total number of words the HLS uttered in Korean, not including any fillers such as icey ‘now’81, repetitions (e.g., instances of stuttering or searching for a particular word), false starts, or English words, was first counted. This total number of words was then divided by the total number of utterances uttered by the participant to represent the HLS’ MLUs. Speakers with higher MLUs would be those with longer utterances (in terms of their number of clauses and words). For the HLS’ NDWs, the total number of unique words the HLS uttered in Korean were counted, not including any fillers or repetitions or false starts. Speakers with higher NDWs would be those who were more descriptive in their sentences and who did not simply repeat the same phrases over and over again. Thus, the HLS’ SIs, MLUs, and NDWs could all be used as measures of the HLS’ productivity in Korean. The results are given in Table 44. The HLS’ SIs were fairly similar across both groups (simultaneous bilinguals’ mean: 2.4, SD: 0.7; sequential bilinguals’ mean: 2.9, SD: 0.8), but the MLU of the simultaneous bilinguals (mean: 8.6 words, SD: 2.6 words) was lower than that of the sequential bilinguals (mean: 10.7 words, SD: 3.0 words) and the control group (mean: 11.1 words, SD: 3.0 words). In general, even the simultaneous bilinguals produced embedded sentences. In terms of their NDWs, the simultaneous bilinguals had a lower mean (131.2 words, SD: 57.0 words) than the sequential bilinguals (mean: 149.7 words, SD: 60.5 words).

81 A word was considered to be a filler if it was overused and appeared at pauses in the speakers’ utterances, similar to um in English.

135

Table 44. Summary of the HLS’ Structures in their Elicited Narratives

Simultaneous bilinguals (N=15) Sequential bilinguals (N=18)

Measure Mdn (IQR) M (SD) Range Mdn (IQR) M (SD) Range

NTW 192 (165) 218 (71) 51– 478 203 (130) 254 (125) 127– 596

NDW 120 (80) 131 (57) 30– 219 135 (82) 150 (60) 81– 296

Clauses 53 (40.5) 59.3 (25.6) 14– 114 59 (30.0) 66.8 (29.0) 31 –151

Utterances 21 (11) 24.9 (11.0) 12– 56 21 (8) 23.8 (9.3) 13 – 56

SI 2.5 (0.9) 2.4 (0.7) 1.2– 3.8 2.9 (0.9) 2.9 (0.8) 1.5 – 4.4

MLU 8.2 (3.4) 8.6 (2.6) 4.3– 3.7 10.4 (3.6) 10.7 (3.0) 4.9–15.4 Note. NTW = number of total words. NDW = number of different words. SI = subordination index. MLU = mean length of utterance.

Kendall correlations (p-values unadjusted) were calculated between the measures of the HLS’ ability and aspects of their language backgrounds, as summarized in Table 45. The HLS’ self- rated scores of ability in Korean and their SIs were significantly (positively) correlated with all of their language background variables. The number of verbal morphemes correctly repeated by the HLS and the HLS’ MLUs were each significantly correlated with all aspects of the HLS’ language backgrounds except for the number of years they had spent in Korea (τ=0.23, p=.06), which was marginally significant, and their usage scores (τ=0.18, p=.13), which were not significant. The HLS’ NDWs were only significantly correlated with their exposure scores (τ=0.26*, p=.03).

136

Table 45. Correlation Table of HLS’ Language History Variables and Measures of their Ability in Korean

AoO # of yrs. in Korea Usage Exposure

τ p τ p τ p τ p

Self-rating 0.31* .03 0.34* .01 0.50*** <.001 0.30 .03

Imitation 0.26* .05 0.23· .06 0.40* <.01 0.38* <.01

SI 0.35** <.01 0.30* .02 0.28* .02 0.27* .03

MLU 0.40** <.01 0.35** <.01 0.18 .13 0.24* .05

NDW 0.21· .10 0.24* .05 0.19 .11 0.26* .03 Note. Self-rating = self-rated ability score; Imitation = imitation score; SI = subordination index; MLU = mean length of utterance; NDW = number of different words.

In general, HLS who were exposed to English at a later age, spent more time in Korea, and who had greater levels of usage of and exposure to Korean tended to also have higher scores on the various measures of their ability in Korean (see Appendix Table A10 for a summary for each participant). In order to determine whether the correlations were all meaningful or whether the HLS’ different scores of ability could be accounted for by fewer variables, each of the measures of ability were analyzed by fitting various generalized linear models to the data, which will be discussed in the sections to follow.

5.3.3.1 Subordination Index (SI)

Single predictor models were fit to the data and compared to a null model for the HLS’ SIs (see Table 46). All of the models other than the model with just the HLS’ exposure scores were significantly better than the null model at accounting for the variation in the HLS’ SIs. The model with the HLS’ exposure scores as a predictor was only marginally significantly better (G=3.7, p=.07), and the models with the smallest residual sum of squares (both 15.1) were the models with the HLS’ age of onset of exposure to English and the model with the number of years the HLS had spent in Korea as predictors of the HLS’ SIs.

137

Table 46. Model Comparisons of Single Predictor Models to a Null Model of the HLS’ SIs (Model 18)

Model Predictor RSS G p

18 (Null model) 19.3

19 AoO 15.1 8.6** <.01

20 # of yrs. in Korea 15.3 8.2** <.01

21 Usage 15.1 8.5** <.01

22 Exposure 17.3 3.7· .07 Note. RSS = residual sum of squares.

Likelihood ratio tests were also performed to compare models with two predictors to the model with the single predictor of age of onset of exposure to English (Model 19) (see Table 47), and also to the model with the single predictor of the number of years the HLS had spent in Korea (Model 20) (see Table 48). The results indicate that models with the HLS’ usage scores as an additional predictor were better at accounting for the variation in the HLS’ SIs than just using the HLS’ AoO or number of years in Korea as a predictor (G=4.6, p=.04 for both models with the usage scores as additional predictors (Models 23 and 25)); adding the HLS’ exposure scores to the models (Models 24 and 26) did not significantly improve in accounting for the data (G=0.8, p=.39 and G=0.7, p=.41, respectively), thus these models were rejected.

Table 47. Model Comparisons of Two-Predictor Models to a Model with the HLS’ AoOs as a Predictor of the HLS’ SIs (Model 19)

Model Predictors RSS G p

19 AoO 15.1

23 AoO + Usage 13.1 4.6* .04

24 AoO + Exposure 14.7 0.8 .39 Note. RSS = residual sum of squares.

138

Table 48. Model Comparisons of Two-Predictor Models to a Model with the HLS’ Number of Years in Korea as a Predictor of the HLS’ SIs (Model 20)

Model Predictors RSS G p

20 # of yrs. in Korea 15.3

25 # of yrs. in Korea + Usage 13.2 4.6* .04

26 # of yrs. in Korea + Exposure 14.9 0.7 .41 Note. RSS = residual sum of squares.

The details of Models 23 and 25 are provided in Table 49 and Table 50.

Table 49. Model 23: Generalized Linear Model Predicting HLS’ SIs Using the HLS’ AoOs and Usage Scores as Predictors

Predictor B (SE) t p

Intercept 2.3 (0.2) 13.2*** <.001

AoO 0.1 (0.0) 2.2* .04

Usage 0.3 (0.1) 3.2* .04

Table 50. Model 25: Summary of a Generalized Linear Model Predicting HLS’ SIs Using the HLS’ Number of Years in Korea and Usage Scores as Predictors

Predictor B (SE) t p

Intercept 2.3 (0.2) 11.3*** <.001

# of yrs. in Korea 0.1 (0.0) 2.1* .05

Usage 0.3 (0.1) 2.2* .04

The residual sum of squares for Model 23 was 13.1, the multiple r2 was 0.32, the adjusted r2 was 2 0.28, and the residual sum of squares for Model 25 was 13.2, F(2,30)=4.4, p=.02, the multiple r was 0.23, the adjusted r2 was 0.18. In both models, the HLS’ usage scores were better predictors of their SIs (every increase in one standard deviation of their scores was associated with a 0.3 increase in their SIs), but their AoO and number of years spent in Korea were also significant predictors (every increase in years could be associated with a 0.1 increase in the HLS’ SIs).

139

Model 23 was able to account for more of the variation in the HLS’ SIs (28%) better than Model 25 (18%). The HLS were grouped according to their AoOs and their usage scores and their SIs were compared, as shown in Table 51 below.

Table 51. Summary of HLS’ SIs according to the HLS’ Usage of Korean and their AoOs

Usage

AoO Low High

Low 1.9 (0.8) 2.5 (0.6)

Mdn (IQR) 2.2 (0.7) 2.6 (0.6)

M (SD) 4/10 6/7

Number of speakers with at least 2.0 SI

High

Mdn (IQR) 2.8 (0.7) 3.2 (0.8)

M (SD) 2.7 (0.7) 3.1 (0.7)

Number of speakers with at least 2.0 SI 5/6 9/10

With the minimum SI set at 2.0 (i.e., two clauses per utterance); nine of the HLS did not produce at least two clauses per utterance. Table 51 shows that the speakers who had low ages of arrival but who used Korean often and those who had high ages of arrival but who did not use Korean often performed similarly. The speakers who had low AoOs and did not use Korean produced the fewest median number of clauses per utterance and the speakers with the highest AoOs and used Korean frequently produced the greatest number of clauses per utterance. The HLS’ MLUs were analyzed in a similar manner and will be analyzed below.

5.3.3.2 Mean Length of Utterance (MLU)

Likelihood ratio tests were used to compare generalized linear models predicting the HLS’ MLUs using a single predictor with a null model (see Table 52). Only Model 28, with the HLS’ age of onset of exposure to English as a predictor, and Model 29, where the number of years the HLS had spent in Korea was used as a predictor instead, were significantly better than the null model at accounting for the variation in the HLS’ MLUs.

140

Table 52. Model Comparisons of Single Predictor Models to a Null Model of the HLS’ MLUs (Model 27)

Model Predictor RSS G p

27 (Null model) 284.3

28 AoO 211.0 10.8** <.01

29 # of yrs. in Korea 211.8 10.6** <.01

30 Usage 258.7 3.1· <.09

31 Exposure 260.6 2.8 .10 Note. RSS = residual sum of squares.

A single predictor model with the HLS’ usage scores was marginally significant, thus Models 28 and 29 were compared to two predictor models using likelihood ratio tests (see Table 53 and Table 54 for the results).

Table 53. Model Comparisons of Two-Predictor Models to a Model with the HLS’ AoOs as a Predictor of the HLS’ MLUs (Model 28)

Model Predictors RSS G p

28 AoO 211.0

32 AoO + Usage 206.0 0.7 .40

33 AoO + Exposure 209.2 0.3 .61 Note. RSS = residual sum of squares.

Table 54. Results of Likelihood Ratio Tests Comparing Various Two Predictor Models to a Model with the HLS’ Number of Years in Korea as a Predictor of the HLS’ MLUs (Model 29)

Model Predictors RSS G p

29 # of yrs. in Korea 211.8

34 # of yrs. in Korea + Usage 206.9 0.7 .41

35 # of yrs. in Korea + Exposure 210.4 0.2 .66 Note. RSS = residual sum of squares.

141

The results indicate that adding other predictors into the models did not account for the variation in the data significantly better. Thus, the two-predictor models were rejected.The variation in the HLS’ MLUs could be accounted for simply by looking at the HLS’ ages of first exposure to English (see Table 55) or the number of years they had spent in Korea (see Table 56), rather than their usage or exposure scores.

Table 55. Model 28: Generalized Linear Model Predicting HLS’ MLUs Using the HLS’ AoOs as a Predictor

Predictor B (SE) t P

Intercept 8.0 (0.7) 11.5*** <.001

AoO 0.4 (0.1) 3.3** <.01

Table 56. Model 29: Generalized Linear Model Predicting HLS’ MLUs Using the HLS’ Number of Years in Korea as a Predictor

Predictor B (SE) t p

Intercept 9.7 (0.5) 19.3*** <.001

# of yrs. In Korea 0.9 (0.5) 1.7· .09

In both Models 28 and 29, an increase in one year accounted for about a 0.4 increase (i.e., about a half a word longer in each utterance) in the the HLS’ MLUs, as can be seen in Table 55 and Table 56, respectively. For Model 28, the multiple r2 was 0.26, and the adjusted r2 was 0.23, 2 2 significant at F(1,31)=10.8, p<.01. For Model 29, the multiple r was 0.25, and the adjusted r was

0.23, significant at F(1,31)=10.6, p<.01. Since the Model 28 was better at accounting for the variation in the data, the speakers were grouped according to whether they had “high” or “low” AoOs and “high” or “low” use of Korean. The median and mean MLUs were recalculated for each group; the results are summarized in Table 57.

142

Table 57. Summary of HLS’ MLUs according to the HLS’ Usage of Korean and their AoOs

Usage

AoO Low High

Low 7.00 (3.65) 9.14 (1.28)

Mdn (IQR) 7.80 (2.91) 9.28 (1.69)

M (SD) 3/10 6/7

Number of speakers with at least MLU of 8.00

High

Mdn (IQR) 11.37 (3.55) 10.88 (3.50)

M (SD) 11.15 (3.33) 11.06 (2.73)

Number of speakers with at least MLU of 8.00 5/6 9/10

With the minimum MLU set at 8.00 (i.e., eight words per utterance), ten speakers had MLUs which were lower than 8.00. These speakers mainly consisted of speakers who had lower AoOs and who did not use Korean very often. Only one speaker in each of the other groups failed to produce an average of at least eight words per utterance. The median MLUs for the speakers with low AoOs had lower MLUs, thus the speakers’ AoOs were useful at distinguishing the speakers with higher and lower MLUs.

143

Table 58. Summary of HLS’ MLUs according to the HLS’ Usage of Korean and their Number of Years Spent in Korea

Usage

# of yrs. in Korea Low High

Low 7.00 (3.65) 9.04 (1.64)

Mdn (IQR) 7.80 (2.91) 9.31 (1.84)

M (SD) 3/10 5/6

Number of speakers with at least MLU of 8.00

High

Mdn (IQR) 11.37 (3.55) 10.64 (3.31)

M (SD) 11.15 (3.33) 10.89 (2.66)

Number of speakers with at least MLU of 8.00 5/6 10/11

The HLS’ NDWs were analyzed to be representative of the relatize size of their vocabularies, and were compared to the HLS’ language backgrounds as well.

5.3.3.3 Number of Different Words (NDW)

Similar to the above, single predictor generalized linear models predicting the HLS’ NDWs were compared to a null model using likelihood ratio tests (see Table 59). The results indicate that only models with the number of years the HLS had spent in Korea (Model 38) and their exposure scores (Model 40) were significant in predicting the HLS’ NDWs; the model with the HLS’ usage scores was only marginally significantly better than the null model (G=3.2, p=.08).

144

Table 59. Model Comparisons of Single Predictor Models to a Null Model of the HLS’ NDWs (Model 36)

Model Predictor RSS G p

36 (Null model) 110381

37 AoO 101528 2.7 .11

38 # of yrs. in Korea 96546 4.4* .04

39 Usage 100063 3.2· .08

40 Exposure 96491 4.5* .04 Note. RSS = residual sum of squares.

In order to determine whether adding other predictors to the model improved in accounting for the variation in the HLS’ NDWs, Model 40, with the HLS’ exposure scores as a predictor, was compared to various two predictor models, given that it accounted for the data the best (residual sum of squares: 96491 vs. the residual sum of squares for Model 38: 96546). The results are given below in Table 60.

Table 60. Model Comparisons of Two-Predictor Models to a Model with the HLS’ Exposure Scores as a Predictor of the HLS’ NDWs (Model 40)

Model Predictors RSS G p

40 Exposure 96491

41 Exposure + AoO 94120 0.8 .39

42 Exposure + # of yrs. in Korea 91218 1.7 .20

43 Exposure + Usage 95437 0.3 .57 Note. RSS=residual sum of squares.

The results indicate that adding other predictors into the model did not better account for the variation in the data in a significant way. The details of Model 40 are given in Table 61 below.

145

Table 61. Model 40: Generalized Linear Model Predicting HLS’ NDWs Using the HLS’ Exposure Scores as a Predictor

Predictor B (SE) t p

Intercept 141.3 (9.7) 14.5*** <.001

Exposure 20.8 (9.9) 2.1* .04

In the above model, we can see that an increase in one standard deviation in the HLS’ exposure scores was associated with about a 21 word increase in the HLS’ NDWs. This suggests that HLS with more sources of input, and more regular input, in Korean tended to have larger vocabularies, or at least made use of more words in their vocabularies in their elicited narratives for this experiment. The HLS were grouped in terms of the number of years they had spent in Korea and their exposure to Korean. The number of speakers with NDWs of at least 100 were also counted. The results are summarized in Table 62 below.

Table 62. Summary of HLS’ NDWs according to the HLS’ Exposure to Korean and their Number of Years Spent in Korea

Exposure

# of yrs. in Korea Low High

Low 118.5 (99.3) 90.0 (30.0)

Mdn (IQR) 142.6 (55.8) 98.7 (49.5)

M (SD) 8/10 2/6

Number of speakers with NDW of at least 100

High

Mdn (IQR) 111.0 (57.3) 154.0 (67.0)

M (SD) 120.0 (40.4) 174.9 (59.9)

Number of speakers with NDW of at least 100 3/6 10/11

Table 62 shows that only the HLS’ time spent in Korea was a good predictor of their NDWs. The speakers with high exposure to Korean (i.e., who received a lot of input in Korean, relative to the

146 other speakers) had lower MLUs if they had not spent as long in Korea; unexpectedly, the speakers with low exposure to Korean performed better than this group.

Table 63. Summary of HLS’ NDWs according to the HLS’ Exposure to Korean and their AoOs

Exposure

AoO Low High

Low 118.5 (99.3) 93.0 (46.5)

Mdn (IQR) 142.6 (55.8) 104.9 (48.0)

M (SD) 8/10 3/7

Number of speakers with NDW of at least 100

High

Mdn (IQR) 111.0 (57.3) 166.5 (73.5)

M (SD) 120.0 (40.4) 178.2 (62.1)

Number of speakers with NDW of at least 100 3/6 9/10

A summary of the predictors which were significant in accounting for the variation in the various measures of the HLS’ ability is provided in the following section.

5.4 Summary

For most of the measures of the HLS’ ability in Korean (e.g., their self-rated scores, the proportion of verbal morphemes they repeated correctly, and their subordination indices), the HLS’ usage scores were able to account for much of the variation in their results. For the mean length of their utterances, their age of onset of exposure to English was the best predictor, and the number of different words used in their elicited narratives was best predicted by using the HLS’ exposure scores. For any given measure of ability, only one or two aspects of the HLS’ language backgrounds were needed to account for the variation in the data, as summarized below in Table 64.

147

Table 64. Summary of Significance of Various Aspects of the HLS’ Language Backgrounds in Predicting their Measures of Ability in Korean

Language background

Measure of ability AoO # of yrs. in Korea Usage Exposure

Self-rated ability scores * *

Imitation score *

Subordination index * (*) *

Mean length of utterance *

Number of different words *? *

In order to determine how well the measures of ability agreed with one another, Kendall’s correlations were calculated between each pair of measures. The results are shown in Table 65 below (p-values unadjusted).

Table 65. Correlation Table of Measures of HLS’ Ability in Korean

1 2 3 4 5

τ p τ p τ p τ p τ p

1. Self-rating —

2. Imitation 0.40** .01 —

3. SI 0.30* .02 0.29* .02 —

4. MLU 0.26* .04 0.17 .17 0.70*** <.001 —

5. NDW 0.27* .03 0.24* .05 0.43*** <.001 0.56*** <.001 — Note. Self-rating = self-rated ability score; Imitation = imitation score; SI = subordination index; MLU = mean length of utterance; NDW = number of different words.

There were positive correlations between each of the different measures of the HLS’ability. All measures were significantly correlated with one another, except between the number of verbal morphemes repeated correctly by the HLS and their MLUs (τ=0.17, p=.17). There was a very high and highly significant positive correlation between the HLS’ MLUs and their SIs (τ=0.70, p<.001), likely since the longer the length of the utterance, the more likely it would be for the

148

HLS to have used more embedded structures in the utterance. In addition, there was a strong correlation between the HLS’ MLUs and their NDWs (τ=0.56, p<.001), likely given that the longer their utterance, the more likely the HLS would have been to use different descriptive words. The results indicate that, in general, the different measures of ability were in agreement on which HLS had higher ability in Korean, given that there were significant positive correlations between the different measures. In the following section, I will examine how the HLS performed with respect to Case marking on other tasks of the experiment.

149

Chapter 6 Results 6 Results 6.1 Case

6.1.1 Case Drop: Heritage Language Speakers’ Preferences

In order to address the question of whether the HLS were able to use Case markers correctly and whether they followed the Case drop hierarchy when they omitted Case markers, the HLS’ Case drop preferences and the proportion of Case markers correctly provided by the HLS were counted across different tasks of the experiment. In the preference task, the HLS’ responses on the grammatical distractors were analyzed first. The results are summarized in Table 66.

Table 66. Median and Mean Proportion of Grammatical Distractors Correctly Selected in the Preference Task

Group N Mdn (IQR) M (SD) 95% CI

Simultaneous bilinguals 15 1.00 (0.14) 0.90 (0.16) [0.81, 0.98]

Sequential bilinguals 18 1.00 (0.14) 0.93 (0.09) [0.88, 0.97]

Control 8 0.86 (0.14) 0.91 (0.07) [0.85, 0.97] Note. CI = confidence interval.

Table 66 shows that the 95% confidence intervals for the mean proportion of correct responses for each group all overlapped, which indicates that the HLS were paying attention and were sufficiently proficient in Korean to complete the task, as they did not perform differently from the control group.

In order to test the HLS’ preferences with respect to Case drop, the HLS were presented with a pair of sentences with different combinations of Case drop: “No Drop” (both the nominative and accusative Case markers were overt), “Acc Drop” (only the accusative marker was dropped), “Nom Drop” (only the nominative Case marker), or “Both Dropped” (both the nominative and accusative Case markers were dropped). The HLS were asked to determine which sentence

150 sounded more natural, as described in Section 4.3.2. No Drop was expected to be preferred over any type of Case drop at all, but if only one Case marker was dropped, Acc Drop was expected to be preferred over Nom Drop, according to the Case drop hierarchy, which predicts that the accusative Case marker will be more susceptible to Case drop than the nominative Case marker. Both Dropped was expected to be the less preferred sentence in every pair it appeared in. The responses were categorized as being correct (“one”) or incorrect (“zero”), depending on whether they matched the response expected according to the Case drop hierarchy. Table 68 shows that there were no differences between the groups in their mean proportions of correct responses and standard deviations (mean: 0.83–0.85, SD: 0.10–0.13); on average, all groups followed the Case drop hierarchy.

Table 67. Median and Mean Proportion of Correct Responses Showing Preferences of No Drop > Acc Drop > Nom Drop > Both Dropped in the Preference Task

Group N Mdn (IQR) M (SD) 95% CI

Simultaneous bilinguals 15 0.83 (0.10) 0.83 (0.11) [0.77, 0.89]

Sequential bilinguals 18 0.88 (0.08) 0.85 (0.10) [0.80, 0.90]

Control 8 0.88 (0.09) 0.84 (0.13) [0.73, 0.95] Note. CI = confidence interval.

In order to determine whether there were patterns in the participants’ responses depending on the combination of overt and null Case markers being compared, the responses were grouped according to which sentence (No Drop, Acc Drop, or Nom Drop) was expected to be preferred and which Case combination was presented as the alternative. A summary is given in Table 68.

151

Table 68. Mean Proportion of Correct Responses Showing the Expected Preferences according to the Case Drop Hierarchy in the Preference Task

Expected- Simultaneous bilingualsa Sequential bilingualsb Control groupc Alternative presented M (SD) Mdn (IQR) M (SD) Mdn (IQR) M (SD) Mdn (IQR)

No Drop

Acc Drop 0.82 (0.22) 0.75 (0.25) 0.93 (0.17) 1.00 (0.00) 0.94 (0.12) 1.00 (0.06)

Nom Drop 0.95 (0.10) 1.00 (0.00) 0.94 (0.14) 1.00 (0.00) 0.91 (0.27) 1.00 (0.00)

Both Dropped 0.95 (0.10) 1.00 (0.00) 0.93 (0.14) 1.00 (0.00) 1.00 (0.00) 1.00 (0.00)

Acc Drop

Nom Drop 0.63 (0.30) 0.75 (0.25) 0.64 (0.21) 0.75 (0.25) 0.59 (0.27) 0.63 (0.31)

Both Dropped 0.83 (0.24) 1.00 (0.25) 0.90 (0.15) 1.00 (0.25) 0.88 (0.13) 0.88 (0.25)

Nom Drop

Both Dropped 0.80 (0.17) 0.75 (0.25) 0.74 (0.28) 0.75 (0.50) 0.72 (0.28) 0.75 (0.50) a N=15. b N=18. c N=8.

Table 68 shows that overall the HLS adhered fairly well to the Case drop hierarchy and in general chose the sentences expected to be preferred from each pair of sentences. The only pair of sentences which elicited responses which were not necessarily in line with the Case drop hierarchy was the comparison of Acc Drop and Nom Drop and the comparison of Nom Drop and Both Dropped. When only one Case marker was dropped, although the responses were in favour of dropping the accusative Case marker over the nominative Case marker, the proportion of correct responses was much lower for this comparison (around 0.60 correct across all groups) than for the other comparisons (over 0.80 correct for most of the other pairs of sentences compared across all groups). The only other comparison that appeared to draw atypical responses, but only from the sequential bilinguals and the control group, was the comparison of Nom Drop and Both Dropped. For these two groups, their responses were lower (0.74 (SD: 0.28) and 0.72 (SD: 0.28) correct, respectively) than for the rest of the comparisons (excluding Acc Drop vs. Nom Drop), which were otherwise around 0.90 correct.

152

In order to determine whether any differences in the HLS’ responses were attributable to whether the HLS were simultaneous or sequential bilinguals, various mixed effects logistic regression models were fitted to the data using the lme4 package (Bates et al. 2011) in R (R Development Core Team 2011). The data was collapsed according to which type of sentence was expected to be preferred (i.e., according to the groupings of the “Expected Preference” heading in Table 68). Deviation coding was applied for the type of Case combination expected to be preferred, with No Drop (i.e., instances where No Drop was expected to be the preferred sentence) set as the reference level, such that the number of correct responses for the different types of Case combinations was compared to the number of correct responses when No Drop was the sentence that was expected to be preferred. This would allow us to see whether the strength of the HLS’ preferences for a particular type of Case combination was statistically different from the strength of their preference for No Drop. Although there were no apparent differences in the groups’ performance according to the proportion of sentences correctly selected overall in the preference task (see Table 67), in order to determine whether differences existed according to the grouping of the participants once random effects were accounted for, deviation coding was applied to the data and the control group was set as the reference level (i.e., the simultaneous and sequential bilinguals were each compared as a group to the control group, but initially not to each other). Since participants can differ in their individual preferences for particular combinations of Case drop, a random intercept was included for Participants (i.e., each individual). Since different participants may tend to have stronger or weaker preferences for particular types of Case combinations, a random slope for the effect of the type of sentence (No Drop, Acc Drop, or Nom Drop) expected to be preferred for each participant was included for the Participants (following Barr, Levy, Scheepers, and Tily (2013))82. The particular contextualizing stories provided before the test sentences may introduce some noise into the data, thus a random intercept was also included for Items (i.e., each question of the preference task). There were two versions of the preference task; they differed in the order of presentation of the two sentences being compared

82 Barr et al. (2013) argue that maximal random effects (i.e., all those required by virtue of the design of the experiment) should be included in linear mixed-effects models in order to help avoid Type I error (i.e., results which are anti-conservative). I follow their logic in including the maximal random effects in the model. Barr et al. also suggest that a random slope for items is also required if there are different within-item factors. I also tried analyzing the data by using a model predicting the particular type of sentence (No Drop, Acc Drop, Nom Drop, or Both Dropped) preferred by the participants, with random intercepts for Participants and Items, as well as random slopes for the effect of the type of sentence for each participant and for the alternative type of case presented in the item, but this model did not converge and this type of analysis was abandoned.

153 for each question, so a random intercept was also included for the version presented to the participants. Likelihood ratio tests were performed to compare the null model (Model 44) to various models with Group (i.e., whether the participants belonged to the control group or were simultaneous bilinguals or sequential bilinguals) and/or the type of sentence expected to be preferred entered as fixed effect predictors. The likelihood ratio tests were performed as one-way ANOVAs in R (see Table 69 for results).

Table 69. Likelihood Ratio Tests Comparing Single Predictor Models to a Null Model Predicting the HLS’ Correct Selections in the Preference Task (Model 44)

Model Predictor AIC Log likelihood G p df

44 (Null model) 792.30 -387.15

45 Group 796.12 -387.06 0.18 .91 2

46 Case combination preferred 777.05 -377.53 19.25*** <.001 2 Note. The control group and No Drop were set as the reference levels for these models. AIC = Akaike information criterion.

Table 69 indicates that the model with the Case combination preferred as a fixed effect predictor (Model 46) was significantly better than the null model at accounting for the variation in the data; a model with Group as a fixed effect predictor was not significantly better. In order to determine if adding Group as a fixed effect predictor to a model with the Case combination preferred as a fixed effect predictor, likelihood ratio tests were performed. The results are given in Table 70.

154

Table 70. Likelihood Ratio Tests Comparing Two Predictor Models Predicting the HLS’ Correct Selections in the Preference Task with a Model with the Case Combination Preferred as a Predictor (Model 46)

Model Predictors AIC Log likelihood G p df

46 Case combination preferred 777.05 -377.53

47 Case combination preferred + Group 780.80 -377.40 0.25 .88 2

48 Case combination preferred * Group 785.55 -375.77 3.50 .74 6 Note. The control group and No Drop were set as the reference levels for these models. AIC = Akaike information criterion.

The results indicate that there was no effect for Group; models with Group as a main effect (Model 47) or as an interaction with the type of Case combination preferred (Model 48) were not significantly better at accounting for the variation in the data than a model with the type of Case combination preferred alone as a fixed effect predictor (Model 46). Thus, the more complex models were rejected. The details of Model 46 are given below in Table 71.

155

Table 71. Model 46: Mixed-effects Logistic Regression Model Predicting the HLS’ Correct Selections in the Preference Task Using the Case Combination Expected to be Preferred as a Fixed Effect Predictor

Predictor B (SE) z p

Fixed effects:

Intercept 2.03 (0.23) 8.70 <.001

Acc Drop preferred -2.19 (0.46) -4.75*** <.001

Nom Drop preferred -2.20 (0.56) -3.90*** <.001

Random effects: Variance (SD)

Participant (Intercept) 0.53 (0.73)

Participant: Acc Drop preferred 0.89 (0.94)

Participant: Nom Drop preferred 1.61 (1.27)

Item 0.57 (0.75)

Version 0.00 (0.01) Note. No Drop was set as the reference level for this model; it is represented by the intercept.

Table 71 shows that compared to No Drop (i.e., when both the nominative and the accusative Case markers were overt in the sentence), where the participants were relatively consistent in selecting No Drop over sentences with different combinations of Case drop (i.e., Acc Drop, Nom Drop, or Both Dropped), the participants were not as likely to consistently choose Acc Drop (over Nom Drop and Both Dropped) or Nom Drop (over Both Dropped). The log odds when these were the types of sentences expected to be preferred were significantly lower than when No Drop was expected to be preferred (-2.19 and -2.20, respectively). This means that the probability of the HLS correctly selecting the preferred Case combination was 88% for No Drop, but dropped to 46% when the Case combination expected to be preferred was Acc Drop or Nom Drop. There was no effect for Version (i.e., the results were consistent across both versions of the task).

Since the effect for the type of Case combination expected to be preferred was significant, deviation coding was applied, setting the reference level to Acc Drop instead of No Drop, in

156 order to determine whether the difference between the participants’ responses to Acc Drop vs. Nom Drop were also significant. The model was fitted to the data again, and the model was still significantly better than the null model in a likelihood ratio test (G=19.27***, p<.001, d.f.=2). The details of the model are provided in Table 72.

Table 72. Model 46: Mixed Effects Logistic Regression Model Predicting the HLS’ Correct Selections in the Preference Task Using the Case Combination Expected to be Preferred as a Fixed Effect Predictor

Predictors B (SE) z p

Fixed effects:

Intercept 2.03 (0.23) 8.71*** <.001

No Drop preferred 2.22 (0.46) 4.78*** <.001

Nom Drop preferred -0.01 (0.53) 0.02 .98

Random effects: Variance (SD)

Participant (Intercept) 0.52 (0.72)

Participant: No Drop preferred 0.93 (0.97)

Participant: Nom Drop preferred 0.57 (0.76)

Item 0.57 (0.75)

Version 0.00 (0.01) Note. Acc Drop was set as the reference level for this model; it is represented by the intercept.

Table 72 indicates that there was no difference between the participants’ probability of choosing Acc Drop or Nom Drop in contexts where they were expected to be the preferred choice. There was also no difference between the groups when the participants were divided according to their ages of onset of first exposure to English, so the HLS were collapsed into a single group. Since tests with continuous variables generally have more power than those with categorical ones, instead of the grouping of the HLS (as simultaneous or sequential bilinguals), the HLS’ ages of onset of first exposure to English were used to test age effects. The data from the control group was removed, since language usage and exposure information was unavailable for the control group. A model similar to Model 46, with the type of sentence expected to be preferred as a fixed

157 predictor, was re-fitted to the data (Model 49). In order to determine whether the HLS’ usage or exposure of Korean helped predict differences in the HLS’ responses, likelihood ratio tests were performed to compare Model 49 to models with various fixed effect predictors. The results are given in Table 73 below.

Table 73. Likelihood Ratio Tests Comparing Single Predictor Models to a Null Model Predicting the HLS’ Correct Selections in the Preference Task (Model 49)

Model Predictor AIC Log likelihood G p df

49 (Null model) 646.90 -314.45

50 Case combination preferred 635.94 -306.97 14.96*** <.001 2

51 AoO 647.82 -313.91 1.08 .30 1

52 # of yrs. in Korea 647.03 -313.51 1.87 .17 1

53 Usage 647.58 -313.79 1.32 .25 1

54 Exposure 639.54 -309.77 9.36** <.01 1 Note. AIC = Akaike information criterion.

The results shown in Table 73 indicate that only the model with the HLS’ exposure scores (Model 54) and the model with the particular Case combination expected to be required as a fixed effect predictor (Model 50) were significantly better at accounting for the variation in the data than the null model. In order to determine whether any more of the variance could be better accounted for by other predictors, various two predictor models were fitted to the data and compared to Model 50 (see Table 74 for results).

158

Table 74. Likelihood Ratio Tests Comparing Two-Predictor Models to a Model with the Case Combination Preferred as a Fixed Effect Predictor of the HLS’ Correct Selections in the Preference Task (Model 50)

Model Predictors AIC Log likelihood G p df

50 Case combination preferred 635.94 -306.97

55 AoO + Case combination 637.23 -306.61 0.71 .40 1 preferred

56 Time in Korea + Case 636.45 -306.23 1.48 .22 1 combination preferred

57 Usage + Case combination 637.52 -306.76 0.42 .52 1 preferred

58 Exposure + Case combination 631.36 -303.68 6.58* .01 1 preferred Note. AIC = Akaike information criterion.

Table 74 shows that only a model with the Case combination expected to be preferred and the HLS’ exposure scores as fixed effect predictors (Model 58) was significantly better than Model 50 at accounting for the variation in the data (G=6.58, p=.01, d.f.=1). This suggests that the amount of relative exposure that the HLS had to Korean was a better indicator of how well the HLS adhered to the Case drop hierarchy in general. In order to determine whether there was an interaction for the HLS’ exposure scores and their preferences for particular types of Case combinations, a model with an interaction (Model 59) was compared to the model with two fixed effect predictors (Model 58). The results are given in Table 75.

Table 75. Likelihood Ratio Test Comparing a Model with an Interaction to a Two-Predictor Model Predicting the HLS’ Correct Selections in the Preference Task (Model 58)

Model Predictors AIC Log likelihood G p df

58 Exposure + Case combination 631.36 -303.68 preferred

59 Exposure * Case combination 621.54 -296.77 13.82*** <.001 2 preferred Note. AIC = Akaike information criterion.

159

Table 75 shows that the model with the interaction was significantly better at accounting for the variation in the data. This suggests that the HLS who received more input in Korean tended to select the No Drop sentences more consistently than they were able to select Acc Drop or Nom over the other choices. The details of Model 59 are provided in Table 76 below.

Table 76. Model 59: Mixed Effects Logistic Regression Model Predicting the HLS’ Correct Selections in the Preference Task Using an Interaction between the Case Combination Expected to be Preferred and the HLS’ Exposure Scores

Predictors B (SE) z p

Fixed effects:

Intercept 2.05 (0.23) 8.79*** <.001

Acc Drop preferred -2.11 (0.51) -4.13*** <.001

Nom Drop preferred -2.05 (0.62) -3.28** <.01

Exposure score 0.59 (0.14) 4.07*** <.001

Exposure score: Acc Drop preferred -1.08 (0.34) -3.21** <.01

Exposure score: Nom Drop preferred -1.23 (0.41) -3.01** <.01

Random effects: Variance (SD)

Participant (Intercept) 0.10 (0.31)

Participant: Acc Drop preferred 0.01 (0.12)

Participant: Nom Drop preferred 1.03 (1.01)

Item 0.61 (0.78)

Version 0.00 (0.00) Note. No Drop was set as the reference level for this model; it is represented by the intercept.

Table 76 shows how introducing the HLS’ exposure scores into the model improved the results. Every increase of one standard deviation point in the HLS’ exposure scores was associated with a higher probability (log odds) of being able to correctly select No Drop over the other cases, yet the same increase in the HLS’ exposure scores was also associated with a decrease in the probability (log odds) of selecting Nom Drop over Both Dropped. This is unexpected, because it

160 suggests that speakers who received more input in Korean were more likely to prefer sentences with Both Dropped over Nom Drop or Both Dropped and Nom Drop over Acc Drop. In particular, there was more variation in the random intercepts and slopes required for each individual participant for instances where the participants were expected to select Nom Drop over Both Dropped. This warrants further study in future research. It may be the case that dropping just one Case marker may sound pragmatically awkward in certain contexts.

6.1.2 Elicited Imitation of Nominative and Accusative Case Markers

The number of nominative and accusative Case markers that were correctly repeated by the HLS from the sentences that were selected for analysis in the elicited imitation task was also counted. All nouns provided by the participants and requiring nominative or accusative Case marking were included in the counts, even if the nouns were slightly mispronounced or substituted with another noun. If a noun was omitted from the original sentence, or if it was replaced with words such as something something in English, the data for this noun was not included (i.e., this was not counted as omission of a required Case marker, since the entire noun was missing). This resulted in 25 tokens (7.9%) being removed from the data for the simultaneous bilinguals, 17 tokens (4.5%) for the sequential bilinguals, and one token (0.6%) for the control group due to repetition errors. In addition, nouns which commonly occur with particular verbs (i.e., collocations, such as the noun contay-s-mal ‘polite-GEN-language' and the verb ssu- ‘use’) were removed from the counts. A summary of the tokens removed from the data is given in Table 77.

Table 77. Number of Tokens Removed from the Data from the Elicited Imitations

Type Simultaneous bilinguals Sequential bilinguals Control

Repetition errors 25 17 1

Collocation 16 16 8

Classifiers 34 48 19

Total 75 81 28

% of data removed 75/317 = 23.7% 81/380 = 21.3% 28/168 = 16.7%

161

Once the above data was removed, the number of Case markers correctly repeated by the participants was re-calculated; a summary is provided in Table 78. Overall, all groups repeated the Case markers correctly.

Table 78. Mean Proportion of Nominative and Accusative Case Markers Correctly Repeated in the Elicited Imitation Task

Group N M (SD) Mdn (IQR) 95% CI

Simultaneous bilinguals 15 0.86 (0.19) 0.93 (0.12) [0.76, 0.97]

Sequential bilinguals 18 0.93 (0.09) 0.97 (0.10) [0.89, 0.98]

Control 8 1.00 (0.00) 1.00 (0.00) — Note. CI = confidence interval.

Table 78 shows that all groups were successful in repeating the required Case markers back correctly. The 95% confidence intervals for the simultaneous and sequential bilinguals overlapped, which suggests that there was no difference in the performance of the two groups of HLS. Since there was no variation in the data for the control group, a comparison with the control group could not be made.

The participants’ responses were further analyzed according to the type of Case marker required (e.g., nominative vs. accusative) (see Table 79), since according to the Case drop hierarchy, accusative Case markers are more likely to be dropped than nominative Case markers. The participants’ types of errors were also calculated and grouped according to whether the Case marker had been dropped or (incorrectly) substituted with another Case marker.

162

Table 79. Mean Proportion of Case Markers Correctly Repeated in the Elicited Imitation Task

Simultaneous bilingualsa Sequential bilingualsb Controlc Response M (SD) 95% CI M (SD) 95% CI M (SD) 95% CI Nom Correctly provided 0.90 (0.20) [0.79, 1.01] 0.96 (0.08) [0.92, 1.00] 1.00 (0.00) — Omitted (Case — drop) 0.03 (0.09) [-0.02, 0.09] 0.04 (0.08) [0.00, 0.08] 0.00 (0.00) Error (wrong — Case provided) 0.07 (0.19) [-0.04, 0.17] 0.00 (0.00) — 0.00 (0.00)

Acc Correctly provided 0.85 (0.21) [0.73, 0.96] 0.92 (0.11) [0.87, 0.97] 1.00 (0.00) — Omitted (Case drop) 0.11 (0.17) [0.02, 0.20] 0.04 (0.10) [0.03, 0.09] 0.00 (0.00) — Error (wrong Case provided) 0.05 (0.08) [0.00, 0.09] 0.02 (0.05) [0.00, 0.05] 0.00 (0.00) — Note. CI = confidence interval. a N=15. b N=18. c N=8.

The summary given in Table 79 shows that the control group performed at ceiling; all members of this group repeated all of the Case markers correctly. The HLS were also overall very good at repeating the Case markers correctly. The 95% confidence intervals for the mean proportion of Case markers produced correctly by both groups of HLS for both types of Case markers overlapped, which suggests that any differences in the means between the groups was not significant. The 95% confidence intervals for the mean proportion of Case markers incorrectly substituted with another Case overlapped with zero for both types of Case markers for both groups of HLS, although there were no instances of incorrect Case markers produced in lieu of nominative Case markers by the sequential bilinguals. The 95% confidence intervals for the mean proportion of nominative Case markers dropped also overlapped with zero for both groups of HLS, but not for the mean proportion of accusative Case markers dropped. The ovelap between the groups suggests that there was no difference between the groups, but there was a difference in the mean proportions for the type of Case marker (i.e., nominative vs. accusative)

163 required. Mixed effects logistic regression models could not be run, in particular with Group as a fixed effect, since there was no variation in the data from the control group.83

Although the grouping of the HLS was not significant in differentiating their performance, in order to determine if there were differences based on the HLS’ usage and exposure scores and/or their ages of onset of exposure to English, the data from the control group was removed and the participants were collapsed together. Similar to the preference task, responses were coded as being correct or incorrect, based on whether the expected Case marker was supplied or not, and deviation coding was applied for the type of Case marker required, with instances where the nominative Case marker was required set as the reference level. Since participants can differ in their individual production of Case markers, a random intercept was included for Participants (i.e., each individual), and a random slope for the type of Case marker required (nominative vs. accusative) expected to be preferred for each participant was included for the Participants. In addition, random intercepts were included for the question the word with the Case marker appeared in, along with the version of the task that the participants received. Various single predictor mixed effects logistic regression models were fitted to the data and compared to the null model using likelihood ratio tests (see Table 80 for the results).

Table 80. Likelihood Ratio Tests Comparing Single Predictor Models to a Null Model of the HLS’ Correct Imitations of Case Markers in the Elicited Imitation Task (Model 60)

Model Predictor AIC Log likelihood G p df

60 (Null model) 295.04 -141.52

61 Case 292.55 -139.27 4.50* .03 1

62 AoO 292.95 -139.47 4.09* .04 1

63 # of yrs. in Korea 292.95 -139.47 4.09* .04 1

64 Usage 293.04 -139.52 4.00* .05 1

65 Exposure 294.48 -140.24 2.56 .11 1 Note. AIC = Akaike information criterion.

83 This was due to the separation problem that arose, given that the factor Group was able to split the data perfectly—all members of the control group repeated every Case marker back correctly.

164

Table 80 shows that all models except the model with the HLS’ exposure scores as a fixed effect predictor (Model 65) were significantly better than the null model at accounting for the variation in the data. The model with age of onset of exposure to English as a fixed effect predictor and the model with the number of years the HLS had spent in Korea were identical, thus the age of onset of exposure to English was chosen as a fixed effect predictor for ease of exposition, but the results would be identical for the number of years the HLS had spent in Korea as a predictor. The model that was the best at accounting for the variation was the model with the type of Case marker required as a fixed effect predictor (Model 61), thus this model was compared to various two predictor models using likelihood ratio tests to determine whether any of the variance attributed as a random effect to the participants could be better accounted for. The results are given in Table 81.

Table 81. Likelihood Ratio Tests Comparing Two-Predictor Models to a Model of the HLS’ Correct Imitations of Case Markers in the Elicited Imitation Task with the Type of Case Marker Expected as a Fixed Effect Predictor (Model 61)

Model Predictors AIC Log likelihood G p df

61 Case 292.55 -139.27

66 AoO + Case 290.44 -137.22 4.11* .04 1

67 # of yrs. in Korea + Case 290.44 -137.22 4.11* .04 1

68 Usage + Case 290.60 -137.30 3.95* .05 1

69 Exposure + Case 292.11 -138.06 2.44 .12 1 Note. AIC = Akaike information criterion.

165

Table 81 shows that all two predictor models, except the model with the type of Case marker required and the HLS’ exposure scores as fixed effect predictors were significantly better at accounting for the variation in the data than the model with just the type of Case expected as a fixed effect predictor. Of these models, the models with either the HLS’ AoOs or the number of years they had spent in Korea as predictors along with the type of Case required were the best at accounting for the variation in the data. In order to determine whether there was an interaction for the Case required and the HLS’ AoOs, a likelihood ratio test was performed to compare a model with an interaction and Model 66. The results are given in Table 82 below.

Table 82. Likelihood Ratio Tests Comparing a Model with an Interaction to a Two-Predictor Model Predicting the HLS’ Correct Imitations of Case Markers in the Elicited Imitation Task (Model 66)

Model Predictors AIC Log likelihood G p df

66 AoO + Case 290.44 -137.22

70 AoO * Case 292.33 -137.29 0.00 1.00 1 Note. AIC = Akaike information criterion.

A comparison of the models revealed that the interaction was not justified. Only the HLS’ AoOs (or the number of years they had spent in Korea), along with the type of Case required, were necessary as fixed effect predictors to account for the HLS’ performance. The details of this model are shown in Table 83.

166

Table 83. Model 66: Mixed Effects Logistic Regression Model Predicting HLS’ Correct Imitations of Case Markers in the Elicited Imitation Using the Type of Case Required and the HLS’ AoOs as Fixed Effect Predictors

Predictors B (SE) z p

Fixed effects:

Intercept 3.08 (0.58) 5.34*** <.001

Case (Accusative) -1.37 (0.50) -2.75** <.01

AoO 0.19 (0.09) 2.05* .04

Random effects: Variance (SD)

Participant (Intercept) 2.57 (1.60)

Participant: Accusative 0.12 (0.35)

Item 1.00 (1.00)

Version 0.00 (0.00) Note. Nominative Case was set as the reference level for this model; it is represented by the intercept.

Table 83 shows that the effect for the type of Case marker required was highly significant; the log odds for repeating the Case marker back correctly was highly significantly lower when the required Case marker was accusative Case (estimated coefficient: -1.37, SE: 0.50; z=-2.75, p<.01), which indicates that, the HLS tended to drop the accusative Case more often than the nominative Case marker. Every increase in the HLS' AoO corresponded to a 0.19 increase in the log odds of repeating the Case markers back correctly.

6.1.3 Production of Case Markers in Elicited Narratives

In order to determine whether there was a difference in the HLS’ production of Case markers when they were not modelled the correct Case markers to use, the proportion of of Case markers the participants produced correctly in their elicited narratives were also counted as a more active measure of their production of different Case markers in a more spontaneous setting. Only nouns which required use of the nominative or accusative Case marker were retained, but any pronouns and nouns which were also modified by classifiers were not included in the counts, since there is variation in the acceptability of Case marking and Case drop with nouns that are modified by

167 classifiers, depending on the particular structure that the classifier appears in. Any nouns which could have been analyzed as being incorporated into the verb (e.g., objects with the dummy verb ha- or those that appear commonly with a particular verb (e.g., son ‘hand’ and cap- ‘hold’) which might be more likely to have the accusative Case marker dropped for other reasons (e.g., frequency of the verb, or frequency of the collocation with the particular verb used)84 were removed from the counts. In addition, any coordinated nouns were removed from the counts and analyzed separately, since whether the nominative or accusative Case marker is required or whether the comitative Case marker is required depends on the Case marking of the other nouns in the coordinated phrase—only the first or last noun in the coordinated phrase can be marked with the nominative or accusative Case marker, as long as the other nouns are marked with the comitative Case marker. The overall proportion of nominative and accusative Case markers produced overtly in the required contexts is summarized below in Table 84.

Table 84. Median and Mean Proportion of Nominative and Accusative Case Markers Correctly Produced in Elicited Narratives

Group N Mdn (IQR) M (SD) 95% CI

Simultaneous bilinguals 15 0.78 (0.21) 0.69 (0.25) [0.55, 0.83]

Sequential bilinguals 18 0.77 (0.37) 0.71 (0.22) [0.60, 0.82]

Control 8 0.93 (0.04) 0.93 (0.03) [0.90, 0.96] Note. CI = confidence interval.

When the Case markers were analyzed together, the 95% confidence intervals indicated that the HLS performed significantly differently from the control group—the upper limits of the 95% confidence intervals for both groups of HLS was lower than the lower limit of the 95% confidence interval for the control group. In order to determine whether the HLS performed differently from the control group in their production of both types of Case markers, the Case markers produced were reanalyzed according to the type of Case (nominative or accusative) required. The results are summarized below in Table 85 and illustrated in Figure 1.

84 As mentioned in Section 3.2 (see example (36)), object nouns can be incorporated into the verb. This is most common with the light verb ha- ‘do’, but other nouns which frequently occur with particular verbs can be incorporated into the verb or at least do not required the accusative Case marker (Han & Rambow 2000, S.-H. Lee & Park 2007, among others).

168

Table 85. Median and Mean Proportion of Case Markers Produced in their Elicited Narratives by Type of Case Required

Simultaneous bilingualsa Sequential bilingualsb Controlc Response M (SD) 95% CI M (SD) 95% CI M (SD) 95% CI Nom Correctly provided 0.81 (0.22) [0.68, 0.93] 0.81 (0.19) [0.72, 0.90] 0.88 (0.09) [0.80, 0.96] Omitted (Case drop) 0.13 (0.22) [0.01, 0.25] 0.11 (0.14) [0.04, 0.17] 0.07 (0.04) [0.03, 0.11] Error (wrong Case provided) 0.06 (0.09) [0.01, 0.11] 0.08 (0.11) [0.02, 0.13] 0.04 (0.10) [-0.04, 0.12]

Acc Correctly provided 0.57 (0.28) [0.42, 0.72] 0.63 (0.28) [0.49, 0.77] 0.91 (0.07) [0.85, 0.97] Omitted (Case drop) 0.30 (0.24) [0.16, 0.43] 0.31 (0.28) [0.17, 0.45] 0.08 (0.07) [0.02, 0.14] Error (wrong Case provided) 0.14 (0.08) [0.09, 0.18] 0.07 (0.07) [0.03, 0.10] 0.01 (0.02) [-0.01, 0.03] Note. CI = confidence interval. a N=15. b N=18. c N=8.

169

Figure 1. Mean Proportion of Nominative and Accusative Case Markers Correctly Produced in Elicited Narratives

Although the HLS' production of both the nominative and accusative Case markers was very good on the elicited imitation task (see Table 79), Table 85 shows that there was a difference in the HLS’ production of Case markers, depending on which Case marker was required. There was no difference between the HLS and the control group in their production of the nominative Case marker; the 95% confidence intervals for all groups overlapped. However, there was a difference in their production of accusative Case markers—the 95% confidence intervals for the simultaneous and sequential bilinguals overlapped, but the upper limits of the confidence intervals were both lower than the lower limits of the 95% confidence interval for the control group, which suggests that the HLS were significantly less consistent in producing the accusative Case marker in contexts where it was required.

In order to test whether the differences between the groups were statistically significant when various random effects were taken into account, mixed effects logistic regression models were fitted to the data, as with the Preference Task and the Elicited Imitation task. The participants’ responses were classified as being correct or incorrect, according to whether the expected Case marker was correctly provided. Deviation coding was applied to the groups, with the control

170 group as the reference level (such that each group of the HLS would be compared to the control group). Deviation coding was also applied to the type of Case marker expected and the nominative Case was set as the reference level, in order to compare the results for the accusative Case marker (which was expected to be dropped more frequently) to that of the nominative Case marker. In order to control for individual differences in the production of the two types of Case markers, a random intercept was included for Participants, as well as a random slope for individual variability in the production of the nominative Case marker, compared to the accusative Case marker. Random intercepts were included for the noun that the Case marker was supposed to be attached to, the verb that the noun appeared with, and also for the picture shown to the participant that elicited the sentence containing the particular noun. Various mixed effects logistic regression models were compared to each other using likelihood ratio tests; the results are given in Table 86 below.

Table 86. Likelihood Ratio Tests Comparing Single Predictor Models to a Null Model Predicting the HLS’ Correct Production of Case Markers in their Elicited Narratives (Model 71)

Model Predictor AIC Log likelihood G p df

71 (Null model) 1340.4 -663.2

72 Group 1336.2 -659.1 8.19* .02 2

73 Case 1319.2 -651.6 23.20*** <.001 1 Note. AIC = Akaike information criterion.

Table 86 indicates that all models were significantly better than the null model at accounting for the variation in the data. Amongst the models with a single fixed effect predictor, the model with the type of Case marker required (Model 73) was the best at accounting for the variation in the data (G=23.20, p<.001, d.f.=1), but the model with Group as a fixed effect predictor was also significant (G=8.19, p=.02, d.f.=2), so Model 73 was compared to a model with Group and the type of Case marker required as main effects (Model 74) and a model with an interaction for Group and the type of Case marker required (Model 75) to determine whether the predictors were accounting for the same variation or whether each made a unique contribution. The results are summarized in Table 87 below.

171

Table 87. Likelihood Ratio Test Comparing a Two-Predictor Model to a Model with Case as a Predictor of the HLS’ Correct Production of Case Markers in their Elicited Narratives (Model 73)

Model Predictors AIC Log likelihood G p df

73 Case 1319.2 -651.6

74 Group + Case 1314.6 -647.3 8.57* .01 2 Note. AIC = Akaike information criterion.

Table 87 shows that Model 74 was better at accounting for the variation in the data (G=8.57, p<.01, d.f.=2). In order to determine if a model with an interaction could account for the data better, given the differences in the HLS’ production of nominative and accusative Case markers, likelihood ratio tests were performed to compare Model 74 to a model with the type of Case marker required and Group as an interaction (Model 75). The results are given in Table 88.

Table 88. Likelihood Ratio Test of a Model with an Interaction to a Two-Predictor Model Predicting the HLS’ Correct Production of Case Markers in their Elicited Narratives (Model 74)

Model Predictors AIC Log likelihood G p df

74 Group + Case 1314.6 -647.3

75 Group * Case 1311.5 -643.8 7.10* .03 2 Note. AIC = Akaike information criterion.

Table 87 suggests that the interaction is justified, and the effect of the type of Case marker required was different according to which group (i.e., simultaneous bilingual, sequential bilingual, or control) the participant belonged to. The details of Model 75 are given in Table 89.

172

Table 89. Model 75: Summary of a Mixed Effects Logistic Regression Model Predicting the Participants’ Proportion of Case Markers Correctly Produced in their Elicited Narratives by an Interaction for the Type of Case Marker Required and the Grouping of the Participants

Predictor B (SE) z p

Fixed effects:

Intercept 2.15 (0.24) 8.81*** <.001

Simultaneous bilinguals -1.89 (0.55) -3.41*** <.001

Sequential bilinguals -1.62 (0.54) -3.00** <.01

Accusative -1.31 (0.28) -4.73*** <.001

Simultaneous bilinguals: Accusative -1.79 (0.66) -2.70** <.01

Sequential bilinguals: Accusative -1.46 (0.65) -2.23* .03

Random effects: Variance (SD)

Participant (Intercept) 1.16 (1.08)

Participant: Accusative 0.79 (0.89)

Noun 0.69 (0.83)

Verb 0.81 (0.90)

Picture 0.11 (0.33) Note. The control group and nominative Case were set as the reference levels for this model; they are represented by the intercept.

The high intercept (2.15, SE: 0.24) of the model in Table 89 shows that the control group frequently produced the nominative Case markers correctly, which contrasts with both groups of HLS, who both produced the nominative Case marker much less frequently, as can be seen from their high negative coefficients (simultaneous bilinguals: -1.89; sequential bilinguals: -1.62). All groups produced the accusative Case marker less frequently than the nominative Case marker, but the difference was largest for the simultaneous bilinguals (-1.79 vs. -1.31 for the control group). These results support the Case Drop Hierarchy, which predicts that the accusative Case marker will be more likely to be dropped over the nominative Case marker. Importantly, it also confirms that despite the HLS’ high performance on the preference task and the elicited imitation

173 task, the HLS performed differently from the control group when they were required to produce the correct Case marker on their own. To determine whether differences in the HLS’ production of Case markers could be predicted by the HLS’ ages of onset of exposure to English or by their usage and/or exposure scores, the data from the control group was removed and mixed effects logistic regression models with various aspects of the HLS’ language experiences added as fixed effect predictors with the type of Case marker required were compared to a model with the type of Case marker required as a fixed effect predictor. The results are given in Table 90 below.

Table 90. Likelihood Ratio Tests Comparing Two-Predictor Models with a Model with Case as a Predictor of the HLS’ Correct Production of Case Markers in their Elicited Narratives

Model Predictors AIC Log likelihood G p df

76 Case 1074.4 -529.21

77 AoO + Case 1072.7 -527.34 3.74* .05 1

78 # of yrs. in Korea + Case 1072.0 -526.97 4.47* .03 1

79 Usage + Case 1070.3 -526.15 6.12* .01 1

80 Exposure + Case 1072.8 -527.38 3.64* .05 1 Note. AIC = Akaike information criterion.

Models with the additional fixed effect predictors were significantly better at accounting for the variation left unaccounted for by the type of Case required. In particular, the HLS’ usage scores helped account for more of the variation. In order to determine if any models with three fixed effect predictors were better than Model 79 (with the HLS’ usage scores and the type of Case required as fixed effect predictors), which had the lowest AIC and log likelihood, likelihood ratio tests were performed. The results are summarized in Table 91 below.

174

Table 91. Likelihood Ratio Test Comparing a Model with an Interaction to a Two-Predictor Model of the HLS’ Correct Production of Case Markers in their Elicited Narratives

Model Predictors AIC Log likelihood G p df

79 Usage + Case 1070.3 -526.15

80 Usage * Case 1070.8 -525.40 1.49 .22 1 Note. AIC = Akaike information criterion.

Table 91 shows that a model including an interaction for the HLS’ usage scores and the type of Case required (Model 81) could not account for the variation any better than the model without the interaction (Model 79). The details of Model 79 are given in Table 92.

Table 92. Model 79: Mixed Effects Logistic Regression Model Predicting the HLS’ Correct Production of Case Markers in their Elicited Narratives Using their Usage Scores and the Type of Case Required as Fixed Effect Predictors

Predictor B (SE) z p

Fixed effects:

Intercept 1.57 (0.27) 5.86*** <.001

Accusative -1.89 (0.30) -6.30*** <.001

Usage 0.61 (0.23) 2.66** <.01

Random effects: Variance (SD)

Participant (Intercept) 1.25 (1.12)

Participant: Accusative 0.89 (0.94)

Nouns 0. 69 (0.83)

Verb 0.59 (0.77)

Picture 0.16 (0.41) Note. Nominative Case was set as the reference level for this model; it is represented by the intercept.

Table 92 shows that an increase in one standard deviation of the HLS’ usage scores corresponded to a higher log odds (0.61, SE: 0.23) of correct production of Case markers (i.e., when the nominative Case was required, having a usage score that was one standard deviation

175 point higher than the mean resulted in an increase in the probability of producing the correct Case marker from 83% to 90%). When accusative Case was the type of Case marker required, the log odds of producing the Case marker was lower (-1.89 (SE: 0.30)) than when nominative Case was required (i.e., the probability of producing the correct Case marker decreased from 83% to 42%). In the elicited narratives, the HLS who used Korean more often were more consistent in producing Case markers correctly; their usage of Korean was more indicative of this ability than the age at which they first began receiving extended exposure to English.

The HLS produced the accusative Case marker less frequently than the nominative Case marker in instances where they were required, but the HLS also made errors of overproduction, where they used the nominative or accusative Case in situations where they were inappropriate. The number of overproduction errors made by the different groups is summarized in Table 93 below.

Table 93. Raw counts of Overproduction of Case Markers in the Elicited Narratives

Case Simultaneous bilinguals Sequential bilinguals Control

Nominative 12 1 1

Accusative 7 5 8

Total 19 6 9

All groups used the nominative or accusative Case marker in contexts where other Case markers were required or would sound more natural, as in the examples given below from one of the sequential bilinguals (150) and from one of the simultaneous bilinguals (151).

(150) Icey changmun-pakk-ulo kay-ka nawu-kwu, ai-ka nawu-ko. Now window-outside-to dog- NOM come.out-and child-NOM come.out-and ‘Now the dog comes out the window, and the child comes out.’ (Participant #38)

(151) Ku ntey kathi naka-ko iss-Ø-nuntey, enu tol-tam-ul wi-lo kele-se But together go.out-ko exist-PRES-but some stone-wall-ACC top-to walk-and

tases myeng-i ccwuk nalanhi ket-ko iss-Ø-eyo. five CL-NOM straight side-by-side walk-KO exist-PRES-DECL ‘But (they’re) going out together, (they’re) walking on a stone wall to the top, and five (of them) are walking straight side-by-side. (Participant #16)

In (150), ai ‘child’ is the subject of the clause with nawu-ko ‘come.out-and’, so one might expect the nominative Case marker -ka to be grammatical, but since the noun is preceded by a clause

176 with the same predicate (nawu-kwu85 ‘come.out-and’) but a different subject (kay-ka ‘dog-

NOM’), -to ‘also’ would sound more natural. The use of the nominative Case marker on ai ‘child’ is awkward in this context. In (151), enu tol-tam ‘some stone-wall’ is the object of the clause with kele-se ‘walk-and’, but the noun is followed by a postposition (wi ‘top’), so the accusative Case marker should be attached to the postposition and not the noun (i.e., the intended meaning is that the children were walking along (the top of) a stone wall). The HLS also performed errors where more than one Case marker was attached to the noun, as in (152) below, where both the nominative Case marker -ka and the particle -chulum ‘like’ are attached to the noun kay ‘dog’ even though only -chelem ‘like’ is required.86

(152) Kuntey chinkwu-tul iss-nun ay-tul tule-wa-se, ttang-eyse But friend-PL exist-PRES.REL child-PL enter-come-and, ground-at

kay-ka-chulum tung-eyse nwuwe iss-Ø-kwu. dog-NOM-like back-at lie.down exist-PRES-and ‘But the children with friends come in and are lying on (their) backs like the dog. (Participant #5)

The types of errors described above suggest that the HLS had difficulty with using Case markers appropriately. The errors did not provide evidence for a simplified Case system, unlike what Polinsky (2008) found with the speakers of heritage Russian that she studied, since a wide variety of Case markers were used by the HLS, including various locative Case markers such as -eyse ‘at/from’ and -ey ‘to’. Overall, the HLS showed varying levels of sensitivity to the presence of Case markers. They recognized the need for Case marking, but were not as consistent in their production of the markers in their elicited narratives and in the sentences that they had been asked to repeat.

6.1.4 Case Marking: Analysis of Individuals

On the preference task, the HLS’ usage and exposure scores, in addition to the particular combination of Case markers that were overt or dropped, were better predictors of the proportion of sentences that the HLS were able to correctly select. In particular, an interaction of the type of Case combination preferred and the HLS’ exposure scores were the best predictors of the HLS’

85 -Kwu is a more colloquial pronunciation of the complementizer -ko ‘and’. 86 This could be an instance of a dialectal variant of the postposition -chelem ‘like’ (Yoonjung Kang, p.c.).

177 performance. Table 65 showed that including an interaction for the type of Case combination preferred and AoO was not significantly better than a mixed effects logistic regression model with just the type of Case combination preferred (see Appendix Tables A11–A14 for summaries of the individuals’ performance on the tasks testing Case marking). The simultaneous and sequential bilinguals were classified as “low” or “high” (roughly corresponding to whether they had “low” or “high” ages of onset of extended exposure to English), and also according to their levels of exposure to and usage of Korean. Summaries are provided in Table 94 and Table 95 below.

Table 94. Summary of HLS’ Proportion of Sentences Correctly Preferred according to their Levels of Exposure to Korean and their AoOs

Exposure

AoO Low High

Low

Mdn (IQR) 0.83 (0.11) 0.88 (0.08)

M (SD) 0.78 (0.42) 0.85 (0.36)

Number of speakers with at least 75% accuracy 7/10 6/7

High

Mdn (IQR) 0.88 (0.03) 0.92 (0.04)

M (SD) 0.85 (0.36) 0.89 (0.31)

Number of speakers with at least 75% accuracy 5/6 10/10

178

Table 95. Summary of HLS’ Proportion of Sentences Correctly Preferred according to their Usage of Korean and their AoOs

Usage

AoO Low High

Low

Mdn (IQR) 0.83 (0.11) 0.88 (0.13)

M (SD) 0.79 (0.41) 0.83 (0.37)

Number of speakers with at least 75% accuracy 8/10 5/7

High

Mdn (IQR) 0.88 (0.06) 0.90 (0.04)

M (SD) 0.85 (0.35) 0.89 (0.32)

Number of speakers with at least 75% accuracy 5/6 10/10

Table 94 and Table 95 show that there was little difference in the median responses when the HLS were divided according to their usage of Korean and when they were divided according to their exposure to Korean, but importantly, there was no difference according to their AoOs. Four of the speakers with low AoOs did not correctly select at least 75% of the sentences expected to be preferred, but the median proportions of sentences correctly selected were not different from the median proportions of the other groups.

The proportion of Case markers correctly repeated by the HLS in the elicited imitation task were also re-calculated, according to whether the speakers belonged to the “low” or “high” group in terms of their AoOs and their levels of usage of and exposure to Korean. The results are given in Table 96 and Table 97 below.

179

Table 96. Summary of HLS’ Proportion of Case Markers Repeated Correctly according to their Levels of Exposure to Korean and their AoOs

Exposure

AoO Low High

Low

Mdn (IQR) 0.82 (0.22) 0.96 (0.11)

M (SD) 0.82 (0.16) 0.88 (0.20)

Number of speakers with at least 75% accuracy 7/10 6/7

High

Mdn (IQR) 0.94 (0.18) 0.99 (0.04)

M (SD) 0.84 (0.23) 0.96 (0.07)

Number of speakers with at least 75% accuracy 5/6 10/10

Table 97. Summary of HLS’ Proportion of Case Markers Repeated Correctly according to their Usage of Korean and their AoOs

Usage

AoO Low High

Low

Mdn (IQR) 0.88 (0.16) 0.96 (0.23)

M (SD) 0.82 (0.19) 0.88 (0.14)

Number of speakers with at least 75% accuracy 8/10 5/7

High

Mdn (IQR) 0.85 (0.19) 0.99 (0.03)

M (SD) 0.80 (0.22) 0.98 (0.02)

Number of speakers with at least 75% accuracy 5/6 10/10

Table 96 shows that the HLS with early AoOs and low exposure to Korean were not as consistent in repeating Case markers correctly, compared to the other groups, but their median

180 proportion of correctly repeated Case markers was still high (Mdn: 0.82, IQR: 0.22). Table 97 suggests that speakers who used Korean more often were slightly better as a group than the speakers who used Korean less often, but the median proportion of Case markers correctly produced were high for all groups. For the elicited narratives, the HLS’ production of nominative Case markers was analyzed separately from their production of accusative Case markers. The HLS were first divided into “low” and “high” groups based on their exposure and use of Korean, and their median and mean proportion of Case markers correctly produced were calculated, along with the number of speakers who produced at least 75% of the required Case markers. The results are given in Table 98 and Table 99 below.

Table 98. Summary of HLS’ Production of Nominative Case Markers in their Elicited Narratives according to their Levels of Exposure to and Use of Korean

Exposure

Usage Low High

Low

Mdn (IQR) 0.86 (0.29) 0.90 (0.25)

M (SD) 0.83 (0.19) 0.70 (0.47)

Number of speakers with at least 75% accuracy 8/12 3/4

High

Mdn (IQR) 0.92 (0.29) 0.90 (0.13)

M (SD) 0.79 (0.32) 0.91 (0.08)

Number of speakers with at least 75% accuracy 3/4 12/13

Table 98 above shows that the median proportion of nominative Case markers produced was high across the board. The group with the highest number of speakers (four) who did not produce at least 75% of the required Case markers was the group of speakers who reported low usage of Korean, as well as low input in Korean, however at least half of the group still produced at least 86% of the required Case markers (IQR: 0.29).

181

Table 99. Summary of HLS’ Production of Accusative Case Markers in their Elicited Narratives according to their Levels of Exposure to and Use of Korean

Exposure

Usage Low High

Low

Mdn (IQR) 0.65 (0.51) 0.49 (0.21)

M (SD) 0.57 (0.27) 0.53 (0.27)

Number of speakers with at least 75% accuracy 4/12 1/4

High

Mdn (IQR) 0.61 (0.50) 0.69 (0.29)

M (SD) 0.51 (0.39) 0.70 (0.24)

Number of speakers with at least 75% accuracy 2/4 6/13

Several of the HLS failed to produce the accusative Case marker at least 75% of the time in situations where it was required. The group with the lowest median proportion of Case markers produced was the group who did not use Korean very often, but who received a relatively high amount of input in Korean. This suggests that solely receiving a lot of input in Korean is not good enough.

182

Table 100. Summary of HLS’ Production of Nominative Case Markers in their Elicited Narratives according to their AoO and and their Usage of Korean

Usage

AoO Low High

Low

Mdn (IQR) 0.82 (0.22) 0.97 (0.11)

M (SD) 0.72 (0.31) 0.87 (0.24)

Number of speakers with at least 75% accuracy 6/10 6/7

High

Mdn (IQR) 1.00 (0.08) 0.88 (0.13)

M (SD) 0.93 (0.12) 0.89 (0.09)

Number of speakers with at least 75% accuracy 5/6 9/11

For the nominative Case markers, the group that produced Case markers correctly the most frequently were those who had arrived (in an English-speaking country) at a later age and who reported not using Korean very often. More than half of these speakers produced all of the nominative Case markers required in their elicited narratives (Mdn: 1.00, IQR: 0.08), and only one speaker did not produce at least 75% of the Case markers required. The other group with a very high median proportion of Case markers produced was the speakers who were born in an English-speaking country or who had arrived before the age of four and who reported using Korean frequently.

183

Table 101. Summary of HLS’ Production of Accusative Case Markers in their Elicited Narratives according to their AoOs and and their Usage of Korean

Usage

AoO Low High

Low

Mdn (IQR) 0.55 (0.55) 0.60 (0.36)

M (SD) 0.49 (0.31) 0.62 (0.22)

Number of speakers with at least 75% accuracy 3/10 2/7

High

Mdn (IQR) 0.69 (0.26) 0.75 (0.34)

M (SD) 0.66 (0.25) 0.68 (0.28)

Number of speakers with at least 75% accuracy 3/6 5/10

There was a greater difference between the groups in their production of the accusative Case marker. For the accusative Case markers, the speakers with the highest AoO performed better than the speakers with the lower AoOs, irrespective of how often they used Korean. The group with the lowest median proportion of Case markers correctly produced was the speakers with the lowest ages of arrival and who reported using Korean relatively less often (Mdn: 0.55, IQR: 0.55). The speakers were also grouped according to their AoOs and their levels of input in Korean and their production of the nominative and accusative Case markers was re-analyzed. Summaries are given in Table 102 and Table 103.

184

Table 102. Summary of HLS’ Production of Nominative Case Markers in their Elicited Narratives according to their AoOs and and their Exposure to Korean

Exposure

AoO Low High

Low

Mdn (IQR) 0.82 (0.29) 0.90 (0.10)

M (SD) 0.76 (0.24) 0.81 (0.36)

Number of speakers with at least 75% accuracy 6/10 6/7

High

Mdn (IQR) 1.00 (0.13) 0.89 (0.11)

M (SD) 0.92 (0.13) 0.90 (0.08)

Number of speakers with at least 75% accuracy 5/6 9/11

When the speakers were grouped according to the amount of input they received in Korean and according to their AoOs, again, their AoOs made the biggest difference in terms of their performance. Although all groups produced the nominative Case marker consistently, the group with low AoOs and low exposure to Korean produced the lowest median proportion of correct Case markers (0.82, IQR: 0.79), but the speakers with low AoOs who received a lot of input in Korean also produced the nominative Case markers consistently.

185

Table 103. Summary of HLS’ Production of Accusative Case Markers in their Elicited Narratives according to their AoOs and their Exposure to Korean

Exposure

AoO Low High

Low

Mdn (IQR) 0.58 (0.48) 0.60 (0.34)

M (SD) 0.55 (0.27) 0.54 (0.30)

Number of speakers with at least 75% accuracy 3/10 2/7

High

Mdn (IQR) 0.58 (0.37) 0.84 (0.24)

M (SD) 0.57 (0.26) 0.73 (0.25)

Number of speakers with at least 75% accuracy 2/6 6/10

The results for the accusative Case marker were quite different from the results for the nominative Case marker. Only the speakers with high AoOs and high levels of exposure to Korean were able to successfully produce the accusative Case marker consistently, whereas the median proportion of Case markers was much lower for all other groups (the median proportions ranged from 0.58 to 0.60). Only seven (of 23) speakers produced at least 75% of the required accusative Case markers.

6.1.5 Summary

With respect to the questions posed in Section 4.2.1, the HLS performed similarly to the control group on most of the tasks. Both the simultaneous and sequential bilinguals adhered to the Case drop hierarchy in the preference task (means for all groups were 0.83–0.86 (SD for all groups: 0.36–0.38)), which suggests that the HLS were sensitive to the differences in acceptability of Case drop, depending on whether one Case marker was being dropped or whether both the nominative and accusative Case markers were dropped. The clear preferences for No Drop vs. the other combinations of Case drop indicate that the HLS recognize that overt Case marking is preferred in Korean. The difference between Acc Drop and Nom Drop was not as clear, even for the control group, which suggests that perhaps the debate between which Case marker is dropped

186 more often (Accusative (Kim & Kwon 2004 and H. Lee 2006) vs. Nominative (Kwon 1989, Hong, K. Park, I. Chung, & J.-y. Kim 1998, and S.-H. Lee & Park 2008) (see fn. 31)) warrants more testing of monolingual speakers of Korean, in particular, of cases where only the nominative or only the accusative Case marker is dropped. The HLS’ exposure scores were better at distinguishing the performance of the HLS. This makes sense given the nature of the task—speakers who received more input in Korean would be able to discern more easily which patterns of Case drop sounded more natural.

The results of the elicited imitation task and the elicited narratives indicate that the HLS of Korean were able to produce Case markers correctly, although they were not as consistent in their production of accusative Case markers, compared to nominative Case markers in their elicited narratives. The HLS’ age of onset of exposure to English, the length of time they had spent in Korea, and their usage scores were all significant predictors of the HLS’ performance on the elicited imitation task. Since the elicited imitation task required the HLS to not only pick out the correct Case marker, but to also imitate it back correctly, the speakers who used Korean more often were able to repeat back the Case markers more consistently. With respect to the differences in the acceptability of Nom Drop and Acc Drop, the results of the preference task, and perhaps the results of the elicited imitation task as well also suggest that there is no clear preference either way. However, the data from the elicited narratives showed that the HLS mark nominative Case much more regularly than the accusative Case (i.e., they show a preference for Acc Drop). This was different from the control group, who did not exhibit this kind of difference in Case marking between the two types of Case markers. All aspects of the HLS’ language backgrounds were significant in accounting for the variation in their production of Case markers in their elicited narratives. The HLS did not make very many errors with respect to the phonological form of the Case markers required, which suggests that they acquired this correctly. These results are summarized in Table 104 and Table 105. The following section will present the results with respect to plural marking and quantificational structures.

187

Table 104. Summary of HLS’ Performance on Case-Related Tasks

Difference Experimental Simultaneous Sequential between HLS task Condition bilinguals bilinguals and Control?

Case drop In line with hierarchy Yes, but not fully Yes, but not fully No preferences (No Drop > Acc with respect to with respect to Drop > Nom Drop > Acc Drop > Acc Drop > Both Dropped)? Nom Drop Nom Drop

Difference between Weak Weak No Acc Drop > Nom Drop

Elicited Case markers produced Yes Yes Yes – Control imitation of reliably? made no errors Case markers Correct form of Case Yes Yes No markers used?

Production of Case markers produced Nom produced Nom produced Yes – Control Case reliably? more reliably more reliably produced both markers in than Acc than Acc Case markers elicited consistently narratives Correct form of Case Yes Yes No markers used?

Table 105. Summary of the Influence of the HLS’ Background Variables on their Performance on Case-Related Tasks

Experimental Task Condition AoO # of yrs. in Korea Usage Exposure

Case drop preferences In line with hierarchy (No * Drop > Acc Drop > Nom Drop > Both Dropped)?

Elicited imitation of Case markers produced * * * Case markers reliably (Nom > Acc)?

Production of Case Case markers produced * * * * markers in elicited reliably (Nom > Acc)? narratives 6.2 Plural Marking and Classifiers

In order to determine whether the HLS were sensitive to the semantic requirements and syntactic restrictions on the use of the plural marker -tul, the HLS’ responses on the preference task and

188 the elicited narration task were analyzed. In addition, the types of quantificational structures produced by the HLS were analyzed.

6.2.1 Preferences regarding Use of -Tul

In the preference task, the HLS were asked to choose between sentence pairs where one sentence had the plural marker on the noun and one did not. For specific contexts (e.g., when a demonstrative was present on the noun), the sentence with the overt plural marker was expected to be preferred; the plural marker is obligatory on plural nouns in this context. A summary of the HLS’ responses is given in Table 106.

Table 106. Median and Mean Proportion of Sentences with -tul on Specific Nouns Correctly Selected in the Preference Task

Group N Mdn (IQR) M (SD) 95% CI

Simultaneous Bilinguals 15 0.71 (0.29) 0.77 (0.14) [0.69, 0.85]

Sequential Bilinguals 18 0.79 (0.29) 0.78 (0.16) [0.70, 0.86]

Control 8 0.86 (0.04) 0.77 (0.26) [0.55, 0.99] Note. CI = confidence interval.

Table 106 indicates that the HLS performed similarly to the control group; the 95% confidence intervals for all groups overlapped (i.e., the upper bounds of each group overlapped with the lower bounds for the other groups). Thus, overall, the HLS appeared to understand that the plural marker is obligatory on nouns that are specific, although their preferences for having plural marking on specific nouns was not as strong as their preference for the grammatical distractors. The HLS were also presented with sentences where a classifier was present, and were asked to choose between sentences where one sentence had plural -tul attached to the classifer and a sentence where no plural marker was present. In Korean, the plural marker can never attach to the classifier; it is only optional on nouns. A summary of the results is given in Table 107 below.

189

Table 107. Median and Mean Proportion of Sentences without -tul on Classifiers Correctly Selected in the Preference Task

Group N Mdn (IQR) M (SD) 95% CI

Simultaneous bilinguals 15 0.86 (0.29) 0.82 (0.21) [0.71, 0.93]

Sequential bilinguals 18 0.79 (0.39) 0.78 (0.21) [0.67, 0.88]

Control 8 0.93 (0.32) 0.80 (0.26) [0.58, 1.02] Note. CI = confidence interval.

In contexts where a classifier was present, the HLS were also able to recognize that the plural marker -tul should not be attached to classifiers. The 95% confidence intervals for all groups overlapped (i.e., the upper limits for each group overlapped with the lower limits for the other groups). Comparing the results of Table 106 and Table 107, we can see that for each group, the 95% confidence intervals overlapped in the context where -tul was required and the context where it was not allowed. The median number of correct sentences selected did not differ depending on whether the sentence required -tul, as with nouns that are specific, or whether -tul was disallowed, as it is on classifiers. In order to test this statistically, the data was entered into a mixed effects logistic regression model with random intercepts for Participant, Item (the particular test item), and Version (since there were two counterbalanced versions of the task), as well as a random slope for the type of context being tested (e.g., plural -tul marking on classifiers, plural -tul marking on nouns in specific contexts, etc.) for each participant to adjust for individual differences in the strength of preferences for particular contexts. The reference levels were set to the control group for Group, and to the grammatical distractors (see Table 66 in Section 6.1.1) for the Context, and models with different fixed effects were compared to each other using likelihood ratio tests. The results are given in Table 108.

190

Table 108. Likelihood Ratio Tests Comparing Single Predictor Models to a Null Model Predicting the HLS’ Correct Selections in Various Plural Contexts in the Preference Task (Model 82)

Model Predictor AIC Log likelihood G p df

82 (Null model) 776.13 -379.06

83 Group 780.01 -379.00 0.12 .94 2

84 Context 767.09 -372.54 13.04** <.01 2 Note. AIC = Akaike information criterion.

Table 108 shows that the model with Group as a fixed effect predictor was not better than the null model at accounting for the data, but a model with Context (i.e., whether the test item was testing whether the plural marker was being attached to a noun in a specific context or whether the plural marker was not attached to a classifier) as a fixed effect predictor (Model 84) was significantly better at accounting for the variation in the data. In order to determine whether adding Group as a predictor with Context (Model 85) would result in a model that was better than the single predictor model, Model 84 was compared to Model 85, as shown in Table 109.

Table 109. Likelihood Ratio Test Comparing a Two-Predictor Model to a Model with the Plural Context as a Fixed Effect Predictor of the HLS’ Correct Selections in the Preference Task (Model 84)

Model Predictors AIC Log likelihood G p df

84 Context 767.09 -372.54

85 Context + Group 771.03 -372.51 0.06 .97 2 Note. AIC = Akaike information criterion.

191

Table 109 shows that Model 85 was not better than Model 84 at accounting for the variation in the data. A summary of Model 84 is given below in Table 110.

Table 110. Model 84: Mixed Effects Logistic Regression Model Predicting the HLS’ Correct Selections in the Preference Task Using Context as a Fixed Effect Predictor

Predictor B (SE) Z p

Fixed effects:

Intercept 2.50 (0.28) 8.92*** <.001

Contexts with Cl -0.78 (0.31) -2.53*** <.01

Contexts with specific nouns -1.21 (0.29) -4.19** <.001

Random effects: Variance (SD)

Participant (Intercept) 0.27 (0.52)

Participant: Contexts with Cl 0.39 (0.63)

Participant: Contexts with specific nouns 0.19 (0.43)

Item 0.05 (0.22)

Version 0.04 (0.20) Note. The control group and distractors were set as the reference levels for this model; the control group is represented by the intercept.

192

The negative coefficients shown in Table 110 indicate that, compared to their performance on the grammatical distractors, the HLS (and control group) were not as consistent in their responses on the usage of plural -tul. In contexts where -tul was required, the log odds of the participants selecting the correct sentence decreased by -1.21 (i.e., the probability of selecting the correct sentence decreased from 93% (based on the intercept of 2.50) to 78%), and in contexts where -tul was not allowed (i.e., on a classifier), the log odds of selecting the correct sentence decreased by -0.78 (i.e., the probability decreased from 93% to 85%). In order to determine whether there was a difference in the log odds between when the participants were choosing between sentences without -tul on the classifier and sentences with -tul on specific nouns, Model 84 was fitted to the data again with the reference level set to their responses to sentences without -tul on classifiers. The results are summarized in Table 111.

Table 111. Model 84: Mixed Effects Logistic Regression Model Predicting the HLS’ Correct Selections in the Preference Task Using Context as a Fixed Effect Predictor

Predictor B (SE) z p

Fixed effects:

Intercept 1.71 (0.30) 5.78*** <.001

Contexts with distractors 0.78 (0.31) 2.52* .01

Contexts with specific nouns -0.43 (0.29) -1.46 .15

Random effects: Variance (SD)

Participant (Intercept) 1.32 (1.15)

Participant: Contexts with distractors 0.40 (0.63)

Participant: Contexts with specific nouns 0.94 (0.97)

Item 0.05 (0.22)

Version 0.04 (0.20) Note. The contexts with classifiers (which disallowed -tul) were set as the reference level; they are represented by the intercept.

The negative coefficient for the Table 111 shows that the participants were not as consistent in selecting the sentences with -tul attached to specific nouns (vs. those without -tul) as they were

193 in selecting the sentences without –tul attached to classifiers, but this difference was not significant in affecting the probability of selecting the correct sentence (z=-1.46, p=.15). Overall, the HLS did not have difficulty in choosing the correct sentences, and thus appeared to be sensitive to the conditions on the usage of the plural -tul. In order to determine whether aspects of the HLS’ language experiences could help account for their performance, various mixed effects logistic regression models were fitted to the data with the HLS and compared using likelihood ratio tests (see Table 112 below).

Table 112. Likelihood Ratio Tests Comparing Single Predictor Models to a Null Model Predicting the HLS’ Correct Selections in Plural Contexts in the Preference Task (Model 86)

Model Predictor AIC Log likelihood G p df

86 (Null model) 628.76 -305.38

87 Context 620.15 -299.07 12.61** <.01 1

88 AoO 629.75 -304.88 1.00 .32 1

89 # of yrs. in Korea 629.32 -304.66 1.43 .23 1

90 Usage 627.60 -303.80 3.16· .08 1

91 Exposure 629.70 -304.85 1.05 .31 1 Note. AIC = Akaike information criterion.

When the simultaneous and sequential bilinguals were collapsed together, Table 112 shows that a model with the context being tested as the fixed effect predictor (Model 87) was significantly better than the null model at accounting for the variation in the data. The model with the HLS’ usage scores as a fixed effect predictor (Model 90) was marginally significantly better than Model 86 at accounting for the variation in the data. Model 87 was compared to various two predictor models using likelihood ratio tests. The results are given in Table 113.

194

Table 113. Likelihood Ratio Tests Comparing Two-Predictor Models to a Model with the Context as a Predictor of the HLS’ Correct Selections in Plural Contexts in the Preference Task (Model 87)

Model Predictors AIC Log likelihood G p df

87 Context 620.15 -299.07

92 AoO + Context 621.38 -298.69 0.77 .38 1

93 # of yrs. in Korea + Context 620.98 -298.49 1.17 .28 1

94 Usage + Context 618.24 -297.12 3.90* .05 1

95 Exposure + Context 620.86 -298.43 1.29 .26 1 Note. AIC = Akaike information criterion.

Only the model with the HLS' usage scores and the context being tested was significantly better than Model 87 at accounting for the variation in the data. In order to determine whether a model with an interaction for the HLS’ usage scores and the context being tested was better than this two predictor model, another likelihood ratio test was performed (results are given in Table 114).

Table 114. Likelihood Ratio Test Comparing a Model with an Interaction to a Model with the HLS’ Usage Scores and the Context as Fixed Effect Predictors of the HLS’ Correct Selections in Plural Contexts in the Preference Task (Model 86)

Model Predictors AIC Log likelihood G p df

94 Usage + Context 618.24 -297.12

95 Usage * Context 619.48 -295.74 2.77 .25 2 Note. AIC = Akaike information criterion.

A model with an interaction between the HLS’ usage scores and the context required was not better at accounting for the variation in the data (G=2.77, p=.25, d.f.=2). This suggests that the more complex model was not justified. The details of Model 94 are given below in Table 115.

195

Table 115. Model 94: Mixed Effects Logistic Regression Model Predicting the HLS’ Correct Selections in the Preference Task Using the HLS’ Usage Scores and Context as Fixed Effect Predictors

Predictor B (SE) z p

Fixed effects:

Intercept 2.59 (0.34) 7.58*** <.001

Usage 0.29 (0.14) 2.10* .04

Contexts with Cl -1.01 (0.33) -3.05** <.01

Contexts with Specific Nouns -1.27 (0.31) -4.08*** <.001

Random effects: Variance (SD)

Participant (Intercept) 0.36 (0.60)

Participant: Contexts with Cl 0.39 (0.62)

Participant: Contexts with specific nouns 0.19 (0.43)

Item 0.02 (0.15)

Version 0.08 (0.29) Note. The distractors were set as the reference level for this model; they are represented by the intercept.

Table 115 shows that the HLS were not as consistent in selecting the correct sentences involving -tul as they were in selecting the grammatical distractors. The probability of the HLS of selecting the correct sentence amongst the distractors was 93%, whereas this probability was lower for the sentences where -tul was not attached to the classifier (i.e., 84%) and for the sentences where -tul was attached to nouns that were specific (i.e., 79%). A usage score that was one standard deviation point above the mean was associated with a higher log odds of selecting the correct sentence (e.g., 95% vs. 93% for the distractor sentences.

6.2.2 Production of -Tul and Quantificational Structures in Elicited Narratives

In order to determine if the HLS’ actual production of -tul supported the conclusion drawn from their performance on the preference task, the number of tokens of -tul produced by the HLS in their elicited narratives in quantified contexts were counted. The tokens which were produced on

196 bare nouns were counted separately from tokens which were produced in quantificational structures (e.g., with a numeral or classifier); a summary is given in Table 116.

Table 116. Proportion of Tokens of Plural -tul Produced in the Elicited Narratives in Plural Contexts

Simultaneous bilinguals Sequential bilinguals Control

M (SD) # of tokens M (SD) # of tokens M (SD) # of tokens

Use of -tul on bare nouns (i.e., without CL or Num) 0.99 (0.11) 226/229 1.00 (0.06) 240/241 1.00 (0.00) 198/198

Use of -tul on nouns in quantified contexts 0.83 (0.39) 10/12 0.96 (0.21) 22/23 1.00 (0.00) 14/14

Overall, the HLS were accurate in their production of the plural marker -tul in plural contexts. Since there are several different ways of quantifying nouns, the number of structures produced by the HLS (including singular nouns with the numeral han-/hana ‘one’) in their elicited narratives was counted and is summarized in Table 117 below.

197

Table 117. Distribution of Quantificational Structures Produced in Elicited Narratives

Simultaneous bilinguals Sequential bilinguals Control

# of tokens/ # of tokens/ # of tokens/ Structure Proportion # of clauses Proportion # of clauses Proportion # of clauses

No noun

Num Cl 0.01 5/868 0.01 9/1191 0.00 4/989

Pre-nominal

Num N 0.01 9/868 0.02 28/1191 0.01 10/989

Num Cl N 0.00 4/868 0.01 16/1191 0.01 13/989

Post-nominal

N Num 0.01 8/868 0.00 5/1191 0.01 11/989

N Num Cl 0.02 19/868 0.02 23/1191 0.01 14/989

Total 0.05 45/868 0.07 86/1191 0.06 58/989

Table 117 shows that overall, all groups produced relatively the same proportion of quantificational structures (0.05–0.07) (see Appendix Table A15 for a summary of the individuals’ production of different quantificational structures).

Shin and Milroy (1999) called the Num N structure with a numeral preceding the noun (e.g., han swupak ‘one watermelon’ (Shin and Milroy’s (9)) a “borrowing” from English, an “error” produced by the six- and seven-year-old simultaneous and sequential bilinguals. The adult sequential bilinguals tested here produced almost three times as many tokens of this structure than both the simultaneous bilinguals and the control group. However, Sohn (1999) describes this structure as being acceptable with human nouns. The majority of the tokens of this structure produced by the HLS were with human nouns (e.g., han ai ‘one child’ (Participant #29)), but the simultaneous bilinguals produced three tokens of this structure with non-human nouns (e.g., han cwul-lo ‘one line-to’ (Participant #30), which is actually grammatical), the sequential bilinguals produced seven tokens with non-human nouns (e.g., han kangaci-nun ‘one puppy-CONTR’ (Participant #15)), and the control group produced three tokens (e.g., han pang-eyse ‘one room- at’ (Participant #20), which is actually grammatical). Shin and Milroy (1999) also labelled the other pre-nominal structures (i.e., those where a classifier was present: Num Cl N (e.g., *han kay

198 swupak ‘one Cl watermelon’ (Shin and Milroy’s (8)) as being ungrammatical, but these structures are actually only ungrammatical if the genitive Case marker is missing on the classifier, which was the case for the children tested by Shin and Milroy. The HLS tested here produced the genitive Case marker some of the time; not all tokens of these structures were ungrammatical. The simultaneous bilinguals did not have the genitive Case marker on the classifier on half of the tokens they produced (i.e., 2/4), whereas the sequential bilinguals had the correct Case marking on 11 of their 16 tokens. The control group consistently provided the genitive Case marker in these structures. These results could be interpreted as support for the influence of English on the HLS’ production of quantificational structures. Note that there is no influence from English with respect to Case marking, as the genitive Case marker is the only Case marker which is used regularly (unless the preposition of is used instead to indicate the possessor). Influence from English would be predicted to boost the HLS’ production of genitive Case marking, but this was not the case. In order to determine how many of the HLS produced classifiers at all, the number of speakers within each group who produced classifiers and quantificational structures was tabulated. The results are given in Table 118.

Table 118. Production and Position of Classifiers Produced according to Group

Simultaneous bilinguals Sequential bilinguals Control

Position of Cl # of speakers # of speakers # of speakers

No noun 1 4 1

Pre-nominal 2 4 2

Post-nominal 7 13 6

No Cl 5 3 3

No quantificational structures produced 4 3 1

Table 118 shows that not all speakers produced quantificational structures. Of the speakers who did, the majority produced classifiers and positioned these in the canonical position, after the noun. Since almost half of the simultaneous bilinguals who produced quantificational structures did not produce classifiers, there may be an age effect. In addition, the median usage score for the HLS who did not produce classifiers was -0.15 (IQR: 0.60), and their median exposure score

199 was 0.30 (IQR: 1.15), which suggests that there could also be an effect for usage. However, more data is required to investigate this further. Of the speakers who produced pre-nominal structures, only one of the simultaneous bilinguals did not also produce post-nominal structures, and only one of the sequential bilinguals produced more pre-nominal structures than post-nominal ones, thus it appears that the speakers who produced classifiers also preferred the post-nominal structure, so there is no evidence to suggest that any one variable of the HLS’ language backgrounds could be associated with this behavior and further research is required.

Unlike the sequential bilingual children tested by Shin and Milroy (1999), who had difficulty producing classifiers, over half of the simultaneous bilinguals and the majority of the sequential bilinguals tested here were able to produce classifiers fairly consistently, and in the (most common) correct word order (i.e., N Num Cl). In addition, overall, they did not incorrectly attach the plural marker -tul to the classifier. Thus, although the HLS were a bit more lenient in their acceptance of the plural marker -tul appearing on the classifier in the preference task, they were conservative in their actual production of the marker, and overall did not use it incorrectly. They used -tul correctly on specific nouns (i.e., when a demonstrative was present) as well, so as a group, they did not exhibit difficulty with the plural marker. Some of the individual differences in the HLS’ performance on the various tasks will be discussed in the following section.

6.2.3 Plural Marking and Classifiers: Analysis of Individuals

There was variation in the strength of the HLS’ preferences for sentences where -tul was (correctly) not attached to the classifier. One of the sequential bilinguals, and even one of the members of the control group, had a preference for the ungrammatical addition of -tul on the classifier, which suggests that their grammars were influenced by transfer from English. These speakers were the oldest members of their respective groups, so their grammars may have undergone attrition. Two of the simultaneous bilinguals performed at chance, which indicates that they may not have acquired the restriction against the use of -tul on classifiers. The pattern is a bit clearer in the HLS’ preference for sentences where -tul was attached to the specific noun (i.e., one with a demonstrative), although there was still some variation amongst the responses by the HLS.

200

In their elicited narratives, only one of the sequential bilinguals incorrectly used the plural

marker on the classifier, shown in (153), where -tul ‘PL’ is attached to the human classifier myeng.

(153) I cangmyen-ey-nun musun kongwen an-eyse elun-tul-hako, This scene-at-CONTR what park inside-at adult-PL-COMIT

elun sey myeng-tul-hako kangaci-hako wenlay iss-ten adult three CL.HUMAN-PL-COMIT puppy-COMIT originally exist-PST.REL

ay-tul-hako ta kong kac-ko, polo kong kac-ko nol-ko iss-Ø-supnita. child-PL-COMIT all ball take-and polo ball take-and play-KO exist- PRES-DECL ‘In this scene, adults, three adults, (the) dog, and the children who were already there are all playing with a ball, a polo ball, inside some park.’ (Participant #21)

Only two of the simultaneous bilinguals showed difficulty with the particular type of classifier required. In (154) below, Participant #25 used the classifier reserved for human nouns myeng with the animal kangaci ‘puppy’, but corrected themselves with the generalized classifier kay, which is an equally inappropriate classifier for this particular noun. This speaker also repeated this in another sentence. In (155), Participant #41 incorrectly used the classifier used for humans in reference to a dog.

(154) I sacin-ey-nun ay-tul-i kangaci twu myeng, twu kay-lang This picture-at-CONTR child-PL-NOM puppy two-CL.HUMAN two CL-COMIT

kang-ulo ka-se, kang-ul tti-ese, kangaci-hako nol-Ø-ayo. river-to go-and river-ACC run-and puppy-COMIT play-PRES-DECL ‘In this photo, (the) children are going over to the river with two dogs, running (intended: jumping over) (the) river and playing with (the) dog.’ (Participant #25)

(155) Kangaci han myeng-un kunyang ttwi-ko, kangaci han myeng-un puppy one-CL.HUM-CONTR just run-and puppy one CL.HUMAN-CONTR

eli-n ai-lul thaywu-ko young-REL child-ACC give.ride-and ‘One dog is just running, one dog is giving a child a ride.’ (Participant #41)

One of the sequential bilinguals also made the error of using myeng ‘CL.HUMAN’ in reference to

kangaci ‘puppy’, but self-corrected to mali ‘CL.ANIMAL’, as shown in (156) below.

(156) Kuliko talu-n kay han myeng-i na-o-Ø-neyyo, Also different-REL dog one CL.HUM-NOM out-come-PRES-DECL

201

han mali-ka na-o-Ø-neyyo. one CL.ANIMAL-NOM out-come-PRES-DECL ‘Also, another dog is coming out, one is coming out.’ (Participant #38)

In addition, only one of the simultaneous bilinguals incorrectly omitted -tul on a specific noun (i.e., when a demonstrative was present).

6.2.4 Summary

With respect to the questions posed in Section 4.2.2, the HLS performed like the control group. They preferred the use of -tul on specific nouns, although their preference for -tul on specific nouns was not as strong as their dispreference for the ungrammatical attachment of -tul to classifiers. This indicates there is no transfer from English in this regard, given that English always requires plural marking. Where there were differences in performance, the HLS’ usage scores appeared to be the best predictors of their performance.

With respect to their use of classifiers and the quantificational structures they chose, the HLS performed differently from the control group. They generally used classifiers correctly, and produced postnominal classifiers more often than prenominal ones. There were some tokens of prenominal classifiers produced without the required genitive Case marking—instances of the “English borrowings” evidenced in Shin and Milroy’s (1999) study of bilingual children. No statistical tests were performed on the data with respect to the HLS’ language background variables due to the low number of tokens produced. The results are summarized in Table 119 and Table 120 below. The results of the HLS with respect to the comitative Case markers in Korean will be discussed in the following section.

202

Table 119. Summary of HLS’ Performance with Respect to Plural Marking and Quantified Structures

Difference Experimental Simultaneous Sequential between HLS task Condition bilinguals bilinguals and Control?

Preference -Tul preferred on specific Yes Yes No for -tul nouns (Dem N-tul)?

-Tul dispreferred on Cl Yes Yes No (N Num Cl(*-tul))?

Production of -Tul produced reliably? Yes Yes No -tul and quantified Cl produced reliably? Weak Yes Yes—Control structures produced Cl in elicited almost all narratives the time

Correct form of Cl used? Yes Yes No

Preference for N Num Cl? Yes No—high Yes—no production of preference N Num Cl and and Num N

Table 120. Summary of the Influence of HLS’ Background Variables on their Performance with Plural Marking and Quantified Structures

Experimental task Condition AoO # of yrs. in Korea Usage Exposure

Preference for -tul -Tul preferred on specific * nouns (Dem N-tul)?

-Tul dispreferred on Cl * (N Num Cl(-*tul))?

Production of -tul -Tul produced reliably? and quantified structures in Cl produced reliably? (*)? (*)? (*)? elicited narratives Preference for N Num Cl? 6.3 Comitative Case

Using the comitative case markers -hako, -wa, and -lang correctly is another potential source of difficulty for HLS, since there are different lexical items available in English: and and with. The

203 comitative Case markers in Korean are used in the same instances that the English conjunction and and the preposition with are used, except in the Korean equivalents of phrases where the preposition with takes an inanimate object (usually an ) as its complement (e.g., played with a ball). In these cases, Korean uses the verb kaci-ko ‘take-and’ instead. The appropriateness of a particular comitative Case marker in Korean is also dependent on the register and the level of speech used.

6.3.1 Results of Preference Task

The preference task tested whether the HLS were aware of the semantic restriction on the comitative Case: the infelicity of using comitative Case markers with . When asked to choose between a sentence with the comitative Case marker in this type of context and a sentence with -lul kaci-ko ‘ACC take-and’, the HLS performed like the control group and were consistent in preferring sentences where -lul kaciko ‘ACC take-and’, rather than the incorrect comitative Case (e.g., -hako/-wa/-lang), was used correctly with inanimate . Table 121 shows that out of eight items, the median number of correct responses for all groups was 8.0.

Table 121. Median and mean Proportion of Sentences with -ul kaciko ‘ACC take-and’ with Inanimate Correctly Preferred over Sentences with Comitative Case Markers

Group N Mdn (IQR) M (SD) 95% CI

Simultaneous bilinguals 15 1.00 (0.06) 0.93 (0.14) [0.85, 1.00]

Sequential bilinguals 18 1.00 (0.00) 0.95 (0.16) [0.87, 1.03]

Control 8 1.00 (0.00) 0.97 (0.09) [0.89, 1.04] Note. CI = confidence interval.

The responses were recorded as “correct” or “incorrect”, and the data was fit to various mixed effects logistic regression models. Random intercepts were included for the Participant, Item (i.e., the particular question from the task), and Version (i.e., the version of the task that the participants were given), as well as a random slope for the type of context required for each participant. The control group was set as the reference level for Group and the grammatical distractors were set as the reference level for Context. Various mixed effects logistic regression

204 models were compared to the null model in likelihood ratio tests; the results are given in Table 122 below.

Table 122. Likelihood Ratio Tests Comparing Single Predictor Models to a Null Model

Predicting the HLS’ Selection of Sentences with -ul kaci-ko ‘ACC take-and’ in the Preference Task (Model 96)

Model Predictor AIC Log likelihood G p df

96 (Null model) 291.34 -139.67

97 Group 294.59 -139.30 0.75 .69 2

98 Context 280.34 -133.17 13.01*** <.001 1 Note. AIC = Akaike information criterion.

Table 122 shows that even when the random effects were accounted for, the HLS did not appear to perform differently from the control group—Model 97, with a fixed effect for Group, was not significantly better than the null model at accounting for the variation in the data. However, the model with Context as a fixed effect predictor (Model 98) was highly significantly better than the null model.

Table 123. Likelihood Ratio Test Comparing Two-Predictor Model to a Model Using Context as a Predictor of the HLS’ Correct Selections of Sentences with -ul kaci-ko ‘ACC take-and’ in the Preference Task (Model 98)

Model Predictors AIC Log likelihood G p df

98 Context 280.34 -133.17

99 Group + Context 283.74 -132.87 0.60 .74 2 Note. AIC = Akaike information criterion.

The details of Model 98 are given in Table 124 below.

205

Table 124. Model 98: Mixed Effects Logistic Regression Model Predicting the HLS’ Correct

Selections of Sentences with -ul kaci-ko ‘ACC take-and’ in the Preference Task

Predictors B (SE) z p

Fixed effects:

Intercept 2.78 (0.46) 6.04*** <.001

Context (-ul kaci-ko ‘ACC take-and’) 5.43 (1.86) 2.91** <.01

Random effects: Variance (SD)

Participant (Intercept) 0.26 (0.51)

Participant (-ul kaci-ko ‘ACC take-and’) 30.17 (5.49)

Item 0.77 (0.88)

Version 0.07 (0.27) Note. The grammatical distractors were set as the reference level for this model; they are represented by the intercept.

Table 124 shows that there was a high intercept (2.78), since the HLS were generally very good at selecting the correct sentence amongst the distractors. The large coefficient for the type of context tested (5.43) indicates that the HLS were even better at choosing the sentence with -ul kaci-ko ‘ACC take-and’ (instead of the comitative Case marker -hako) for in the preference task—their probability of selecting the correct sentence increased to 100% from 94% on the distractors. However, the variance in the random intercepts for the individual participants was high, which indicates there was a high degree of individual variation. The data from the control group was separated from the rest of the data and the data was re-fit to the model with various fixed effect predictors in order to determine whether aspects of the HLS’ language experiences could help account for the high degree of variation in individuals. The results are given in Table 125 below.

206

Table 125. Likelihood Ratio Test Comparing Single Predictor Models to a Null Model Predicting the HLS’ Correct Selections of Sentences with -ul kaci-ko ‘ACC take-and’ in the Preference Task (Model 100)

Model Predictor AIC Log likelihood G p df

100 (Null model) 238.51 -113.25

101 Context 230.81 -108.41 9.69** <.01 1

102 AoO 237.85 -111.92 2.66· .10 1

103 # of yrs. in Korea 236.94 -111.47 3.56· .06 1

104 Usage 239.14 -112.57 1.36 .24 1

105 Exposure 239.41 -112.71 1.09 .30 1 Note. AIC = Akaike information criterion.

Table 126. . Likelihood Ratio Test Comparing Two-Predictor Models to a Model Using Context to Predict the HLS’ Correct Selections of Sentences with -ul kaci-ko ‘ACC take-and’ in the Preference Task (Model 101)

Model Predictors AIC Log likelihood G p df

101 Context 230.81 -108.41

106 AoO + Context 231.96 -107.98 0.85 .36 1

107 # of yrs. in Korea + Context 230.89 -107.45 1.92 .17 1 Note. AIC = Akaike information criterion.

None of the models were significantly better than Model 81 (with just the type of context being tested as a fixed effect predictor) at accounting for the variation in the data. Table 106 indicates that Model 81 was the most economical model for accounting for the data. Adding the HLS’ language experiences into the model as interactions with the context of the sentence (i.e., grammatical distractors vs. sentences with inanimate ) did not account for enough of the variation to justify the level of complexity they added to the model. The details of Model 81 are given below in Table 127.

207

Table 127. Model 101: Mixed Effects Logistic Regression Model Predicting the HLS’ Correct

Selections of -ul kaci-ko ‘ACC take-and’ in the Preference Task with Context as a Fixed Effect Predictor

Predictor B (SE) z p

Fixed effects:

Intercept 2.96 (0.65) 4.55*** <.001

Context 5.29 (2.03) 2.61** <.01

Random effects: Variance (SD)

Participant (Intercept) 0.31 (0.56)

Participant (-ul kaci-ko ‘ACC take-and’) 30.99 (5.57)

Item 0. 85 (0.92)

Version 0.42 (0.65) Note. The grammatical distractors were set as the reference level for this model; they are represented by the intercept.

Table 127 shows that the model with just the HLS was very similar to the model including the control group (see Table 124 for the details of Model 79). Again, there was a high degree of individual variation, but not enough that could be more systematically accounted for by the aspects of the HLS’ language experiences. The results indicate that selecting -ul kaci-ko ‘ACC take-and’ was not a source of difficulty for the HLS. Only the comitative Case marker -hako was used in the preference task, and only sentences in level III were used.

6.3.2 Production of Comitative Case Markers in Elicited Narratives

In order to see whether the HLS were aware of the other comitative Case markers and were able to use them appropriately (e.g., given the level of speech they chose to use, and also depending on whether an inanimate was present), the conjoined structures produced in the HLS’ elicited narratives were also analyzed. The number of tokens of each type of comitative Case marker were counted for the conjoined noun phrases produced by each group. Tokens where the comitative Case marker was followed by kath-i ‘same-ADV’ or ttok-kath-i ‘exactly- same-ADV’, where the phrase could be translated as ‘just like N’, were not included in the counts, since it is a modifier of the verb, rather than a noun phrase conjoined with another noun phrase.

208

For example, in (157) below, the noun cen ‘prior’ has the comitative Case marker attached to it

and is followed by the adverbial phrase ttok-kath-i ‘exactly-same-ADV’ to convey the meaning of ‘just like before’.

(157) Ku cen-kwa ttok-kath-i ay-tul-i ta kay-lul That prior-COMIT exactly-same-ADV child-PL-NOM all dog- ACC

twi-ttala-o-kwu iss-Ø-supnita. back-follow-come-KO exist-PRES-DECL ‘Just like before, the children are all following behind the dog.’ (Participant #18)

Excluding cases like (157), the comitative Case markers which were correctly produced were counted. For each group of speakers, the proportion of each type of comitative Case marker used was calculated, and the results are summarized in Table 128 below.

Table 128. Distribution of Comitative Case Markers Produced in the Elicited Narratives

Simultaneous bilinguals Sequential bilinguals Control Type of comitative Case # of # of # of marker Proportion tokens/total Proportion tokens/total Proportion tokens/total

-hako 0.24 27/114 0.48 69/144 0.51 36/71 (neutral)

-lang/ilang 0.47 54/114 0.24 34/144 0.18 13/71 (less formal)

-wa/kwa 0.29 33/114 0.28 41/144 0.31 22/71 (more formal)

Since -hako is the least restricted comitative Case marker (e.g., it is not phonologically conditioned and is not restricted to any particular level of speech), it as expected to be the “default”. Based on the distribution of the comitative Case markers it appears that the sequential bilinguals performed similarly to the control group – approximately half of the comitative Case markers produced by these groups were -hako, whereas approximately half of the comitative Case markers produced by the simultaneous bilinguals were -lang. However, the acceptability of the different comitative Case markers is tied to the particular level of speech used, so the clauses produced by the HLS were grouped according to the level of speech used, as determined by the complementizer (see Section 3.4.1), and counted. Since the complementizers that are used with

209

subordinate clauses are not necessarily marked for a particular level of speech, subordinate clauses were grouped with the level of speech indicated by the complementizer of the matrix clause. If the utterance ended in a subordinate clause (e.g., the complementizer of the matrix clause was -ko ‘and’), all clauses in the utterance were classified as being level III87, unless the complementizer had -yo attached to it (i.e., a morpheme indicating that it was level II speech), in which case the clause was analyzed as being level II. If the utterance was incomplete and lacked a matrix verb (for example, if the sentence ended in a relative clause, as in (158) below, where

the copula i-ey-yo ‘be-PRES-DECL.II’ is missing), all clauses in the utterance were grouped as “other”. All clauses produced by each group were analyzed in this way. Table 129 below shows the distribution of the clauses produced in each level of speech, regardless of whether a comitative Case marker was produced or not.

(158) Kangaci-lang ai-tul-i kunyang nol-ko iss-nun kes. puppy-COMIT child-PL.NOM just play-KO exist-PRES.REL thing ‘The puppy and the children just playing.’ (Participant #41)

87 The complementizers in level I include -pnita and -supnita for declaratives, and -pnikka for interrogatives. These complementizers could be further broken down as -(su)pni-ta ‘I-DECL’ and -(su)pni-kka ‘I-Q’, where -(su)pni indicates level I speech, and -ta and -kka mark the declarative and interrogative, respectively, since they are the same forms used to mark the declarative and interrogative in level III speech as well. Level IV speech generally only uses the complementizer ‘-a/e/ya’, and level II looks very similar to level IV, but is distinguished by the -yo suffix (e.g., a-yo). Thus, the complementizers for subordinate clauses appear to pattern with level III complementizers.

210

Table 129. Distribution of Levels of Speech Used in Elicited Narratives by Participants

Simultaneous bilinguals Sequential bilinguals Control

# of tokens/ # of tokens/ # of tokens/ Level of speech Proportion total Proportion total Proportion total

I (e.g., -(su)pnita 0.31 270/868 0.28 330/1191 0.23 228/989 ‘DECL.I’)

II (e.g., -e-yo 0.40 348/868 0.46 551/1191 0.57 564/989 ‘DECL- II), -ko-yo ‘andII’)

III (e.g., -ta 0.19 166/868 0.22 261/1191 0.17 173/989 ‘DECL.III’, -se ‘because.III’)

IV (e.g., -a/e/ya 0.09 79/868 0.03 41/1191 0.02 23/989 ‘DECL.IV’)

Other 0.01 5/868 0.01 8/1191 0.00 1/989

Table 129 above shows that the HLS performed similarly to the control group in terms of the levels of speech they used in their elicited narratives. Level IV is only used when speaking to people who are lower in status or younger, thus levels I or II are more appropriate for colloquial speech, especially if the speaker does not know the listener very well or is unsure of their age and/or status. Contrary to other studies, which have characterized HLS of Korean as being unfamiliar with levels of speech other than level IV (which is assumed to be the most commonly used level of speech by parents addressing their children88), the HLS tested here were able to use all levels of speech. In order to see if they were also able to choose the comitative Case markers appropriate to the level of speech they used, the proportion of comitative Case markers produced in each level of speech were calculated by dividing the number of tokens of each type of comitative Case marker by the total number of clauses produced in that particular level of speech. For colloquial language, -wa is expected to be preferred with the more formal levels of speech (e.g., levels I or II, and correspondingly dispreferred with levels III and IV, unless in a written context and the level of speech used is level III). In addition, the less formal comitative Case marker -lang is expected to be dispreferred with the most formal level of speech (i.e., level

88 Anecdotally, from my own observations, level IV is commonly used by parents addressing their children in Canada (e.g., to children born in Canada), but in Korea, Y.-j. Kim (1997) notes that level II is often deliberately used by parents in order to teach their children the appropriately polite way of speaking to strangers and their elders.

211

I). A summary of the mismatches in the level of speech of the clause and the formality of the comitative Case marker is given in Table 130 below. Since every type of comitative Case marker is acceptable with level II, none of the tokens for this level was included in the table.

Table 130. Distribution of Comitative Case Markers Incorrectly Produced in Elicited Narratives

Simultaneous bilinguals Sequential bilinguals Control

Level # of # of # of of Type of tokens/# tokens/# tokens/# speech comitative Proportion of clauses Proportion of clauses Proportion of clauses

I -lang/ilang 0.04 12/270 0.02 7/330 0.00 1/228 (less formal)

III -wa/kwa 0.02 3/163 0.00 0/261 0.02 3/172 (more formal)

IV -wa/kwa 0.00 0/79 0.00 0/41 0.00 0/23 (more formal)

Table 130 shows that the HLS performed differently from the control group, but since the number of tokens was low for all groups, the data can only be analyzed qualitatively. Generally speaking, the control group did not perform errors in their choice of comitative Case markers with the various levels of speech. In contrast, although the HLS did not overgeneralize the most formal Case marker -wa with the least formal level of speech (i.e., level IV), they did produce it incorrectly with level III speech. They also incorrectly used the less formal comitative Case marker -lang with the most formal level of speech (level I). Overall, the results suggest that the HLS were familiar with all of the different comitative Case markers, but the HLS had some difficulty with the register restrictions on their usage. Since not all speakers in the group produced these errors, the number of speakers who produced these errors were analyzed separately. The results are shown in Table 131.

212

Table 131. Number of Speakers who Produced Errors with Comitative Case Markers

Simultaneous Sequential Error bilinguals bilinguals Control

-lang/ilang ‘COMIT (less formal)’ 5 2 1 with -pnita ‘DECL.I (formal)

-wa/kwa ‘COMIT (more formal)’ 2 0 2 with -ta ‘DECL.III (less formal)

Table 131 shows that more of the simultaneous bilinguals produced the error of using a comitative Case marker that was inappropriate for the level of speech they were using. This suggests that there could be an age effect. The median usage score for the HLS who produced errors was 0.03 (IQR: 0.69), and their median exposure score was 0.42 (IQR: 0.57). Since both median scores were positive (i.e., above the mean), it is unlikely that usage and exposure are related to whether speakers will commit these errors or not.

In order to determine whether the HLS were using the comitative Case appropriately in general, the conjoined structures produced by the HLS were analyzed and the proportion of correctly produced tokens of comitative Case marking (out of the number of noun phrases requiring comitative Case marking) was calculated. The results are summarized in Table 132.

Table 132. Proportion of Comitative Case Markers Provided in Elicited Narratives

Simultaneous bilinguals Sequential bilinguals Control # of # of # of tokens/ tokens/ tokens/ Response Proportion total Proportion total Proportion total

Correctly provided 0.94 116/124 0.97 143/147 0.91 77/85

Omitted (Case drop) 0.05 6/124 0.01 1/147 0.06 5/85

Error (wrong Case provided) 0.02 2/124 0.02 3/147 0.04 3/85 Note. Errors involve instances where another type of Case marker was used instead of the comitative Case marker.

Table 132 shows that the HLS generally used the comitative Case appropriately, providing it in instances where it was required. Not many tokens of inanimate were produced, but the groups differed slightly in their production of Case marking with these nouns. There were

213 instances of overproduction of the comitative Case (not included in the counts in Table 132). In the elicited narratives, three of the simultaneous bilinguals attached comitative Case markers incorrectly to an inanimate , but only once each. As in the pilot study (see section 3.4, they produced sentences such as that in (159), where the comitative Case marker was attached to the inanimate kong ‘ball’. No other HLS produced this error.

(159) #Kunyang kong-hako nol-ko isse-Ø-yo. Just ball-COMIT play-KO exist-PRES-DECL.II (They’re) just playing with the ball. (Participant #6)

However, even when the HLS used the correct verb kaci-ko ‘take-and’, they did not produce the Case marker consistently, as can be seen in Table 133 below.

Table 133. Proportion of Accusative Case markers Correctly Produced (with kaci-ko ‘take-and’) on Inanimate

Simultaneous bilinguals Sequential bilinguals Control

Proportion 0.00 0.60 1.00

# of tokens/total 0/2 3/5 1/1

The control group only produced one token of an inanimate and provided the required accusative Case marking; in contrast, the simultaneous bilinguals did not produce Case marking in these structures, and the sequential bilinguals only produced the accusative Case marker on three of the five inanimate they produced. This suggests that although the HLS were able to correctly identify the inappropriateness of comitative Case marking on inanimate in the preference task, their understanding of the semantic restriction on its usage may be incomplete. There were other differences in the performance of individuals in the preference task, which will be discussed in the following section.

6.3.3 Comitative Case: Analysis of Individuals

An examination of the distribution of the responses reveals that not all HLS performed the same on the preference task. Although the majority performed at ceiling in selecting the sentence with

-ul kaci-ko ‘ACC take-and’ on , one of the sequential bilinguals had a slight preference for the ungrammatical sentences with the conjunction (Participant #6: 3/8 correct selections), and two of the simultaneous bilinguals only had a slight preference for the correct

214 sentences where the comitative Case marker was not used (Participants #5 and 27 selected five of the eight correct sentences). These speakers appear to have overgeneralized the use of -hako

‘COMIT’ by accepting its use when attached to inanimate , even though this is ungrammatical, according to the literature. These particular simultaneous bilinguals struggled with the nominative and accusative Case marking on the elicited imitation and the elicited narration tasks as well, which indicates incomplete acquisition of the Case marking system in general. However, Participant #27 performed well on the preference task, selecting the correct sentences which corresponded to the expected preferences in Case drop.

6.4 Summary

With respect to the questions posed in Section 4.2.3, the HLS showed overall high proficiency in their ability to use comitative Case markers. They were aware that these markers were inappropriate with inanimate . In addition, when they produced the comitative Case markers, they used the correct phonological forms. Where there was some difficulty was in the choice of comitative Case markers with respect to the register of speech the HLS used. Further research is required, but the preliminary results suggest that the age of onset of English may be the most relevant variable related to this error. Since the HLS performed well on the other aspects of the usage of comitative Case markers, it appears that this was not related to any other variable of the HLS’ language backgrounds. The results are summarized in Table 134 below.

215

Table 134. Summary of HLS’ Performance with Respect to Comitative Case Markers

Difference Experimental Simultaneous Sequential between HLS and task Condition bilinguals bilinguals Control?

Preferences -Lul kaci-ko ‘ACC take-and’ Yes Yes No with respect preferred with inanimate to comitative ? Case markers

Production of Comitative Case markers Yes Yes No comitative used appropriately? Case markers in elicited Correct phonological form of Yes Yes No narratives comitative Case markers used?

Register-appropriate Yes, but also Yes, but also Yes – Control did comitative Case markers produced produced not use used consistently? errors errors inappropriate comitative Case markers

Table 135. Summary of the Influence of the HLS’ Background Variables on their Performance on Comitative Case-Related Tasks

# of yrs. Experimental task Condition AoO in Korea Usage Exposure

Preferences with -Lul kaci-ko ‘ACC take-and’ respect to comitative preferred with inanimate Case markers

Production of Comitative Case markers used comitative Case appropriately? markers in elicited narratives Correct phonological form of comitative Case markers used?

Register-appropriate comitative (*)? (*)? Case markers used consistently?

216

Chapter 7 Discussion and Conclusions 7 Summary 7.1 Findings and Discussion

This study investigated the acquisition of various aspects of the nominal domain by HLS of Korean. In order to address the question of whether HLS’ input (i.e., exposure to the HL), usage of Korean, or their ages of onset of extended exposure were better predictors of their performance, a comprehensive quantified summary of the HLS’ language experiences needed to be compiled. The results showed there was overlap in the HLS’ responses; simultaneous bilinguals (i.e., those who had been exposed to both Korean and English before the age of three) did not necessarily have low levels of usage and exposure to Korean, compared to the sequential bilinguals (i.e., those whose first language was Korean and who had been exposed to English at the age of four or later). It is well known that HLS form a heterogenous group of speakers with varying levels of proficiency in the HL. Different aspects of the HLS’ language experiences were significant predictors of various measures of the HLS’ ability in Korean. The HLS’ usage scores were useful at predicting the HLS’ self-rated ability scores in Korean, their imitation scores (i.e., the number of morphemes they were able to repeat back correctly in the elicited imitation task), and the level of embedding they used in their sentences (as measured in terms of the mean number of clauses they produced per utterance). We can interpret this to mean that speakers who used Korean relatively more often were likely more confident in their ability in Korean (or the speakers who were more confident may have used Korean more often), and were able to perceive and repeat back what was said to them in Korean, and were also correspondingly able to link more clauses together to create longer and more complex structures. The HLS’ ages of onset of extended exposure to English were useful in predicting the mean length of the HLS’ utterances, as measured by the average number of words they produced in each utterance. The HLS’ exposure scores were useful in predicting the number of different words the HLS were able to produce in their elicited narratives. The HLS’ exposure scores were not just comprised of the input they received from family members; they also included measures such as their exposure to various forms of Korean media. This suggests that even if increased input in Korean does not necessarily result in higher proficiency, more input could be useful in exposing the HLS to a

217 greater variety of words, outside of those required in the home, thus equipping them with more words to describe the world around them.

With respect to the HLS’ understanding of Case marking, the HLS showed a preference for sentences where Case marking was overt, which indicates there was no transfer from English in this regard, or there was no resort to the unmarked. Thus, the HLS seemed to recognize that overt Case marking is more common than Case drop, and they also did not have difficulty producing the correct phonological forms for each type of Case.

In terms of the different types of Case markers, there was a difference between the HLS’ imitation of nominative and accusative Case markers—the HLS were more consistent in repeating back the nominative Case markers, compared to the accusative Case markers, as predicted by the Case drop hierarchy. The control group repeated back every case marker correctly, which indicates that there was no attrition in the baseline (mothers of speakers who are HLS). In the elicited narratives, there was a difference between how often the HLS produced accusative Case markers vs. nominative Case markers, which was in line with the Case drop hierarchy, but unexpectedly there was no relative difference in Case marking and Case drop for the control group. It may be the case that describing pictures in a book or telling a story may not be the best context in which to assess this hierarchy; analyses of conversation may give more insight into this. It may also have been the case that the control group made more of a conscious effort to produce Case markers and grammatical sentences, recognizing that the experimenter was interested in analyzing Korean. The HLS followed the Case drop hierarchy—there was a difference between how consistently the speakers produced the Case markers appropriately. The HLS’ age of onset of exposure to English, the length of time they had spent in Korea, and their usage scores were all significant predictors of their performance on the elicited imitation task. Since the elicited imitation task required the HLS to not only pick out the correct Case marker, but to also imitate it back correctly, the speakers who used Korean more often were able to repeat back the Case markers more consistently. The HLS’ production of overt Case markers in their elicited narratives could be distinguished according to the HLS’ usage scores and the type of Case required, all aspects of the HLS’ language backgrounds were significant in accounting for the variation in their production of Case markers in their elicited narratives. Increased usage of Korean corresponded to a higher probability of producing the different Case markers

218 appropriately. The increase was larger for the nominative Case marker than the accusative Case marker.

One area where both the HLS and control group performed differently from expected was in their preferences for different combinations of Case drop. Although both groups preferred overt Case marking, there was no preference for Acc Drop over Nom Drop (i.e., comparisons where one sentence was missing the accusative Case marker and the other sentence was missing the nominative Case marker). On this task, the most significant predictor of the HLS’ performance was their exposure scores. The higher the HLS’ exposure score, the more likely they were to choose No Drop over the other options, but also the more likely they were to not have a clear preference for Acc Drop over Nom Drop. This suggests that other factors may be more relevant in determining the acceptability of the Drop of a particular Case marker. More careful analysis of the contextualizing stories is required to control for these factors. It may be also the case that the drop of a single Case marker is not as common as other strategies, such as dropping the entire noun (e.g., via topic drop) or by using the contrastic topic marker -nun. Much more investigation at the discourse level is warranted.

In terms of whether the HLS were able to use the semantic requirement for -tul and the syntactic restriction against -tul to constrain their acceptability and use of -tul, the results suggest the answer is yes. The HLS performed like the control group in terms of their preferences for plural marking. However, their performance was not at ceiling on the preference task. They showed greater sensitivity to the ungrammaticality of attaching the plural marker to classifers, than to the semantic requirement for -tul on nouns that are specific. This is surprising, given that Korean patterns with English in having overt plural marking on specific nouns. Any influence from English should have helped the HLS produce the correct forms. In my (2006) study, the HLS appeared to have more difficulty in recognizing that -tul is not allowed on classifiers. This suggests that the HLS’ English grammar is not influencing their HL grammar such that the HLS always require overt plural marking. The HLS’ usage scores (along with the context of the sentence) were the best predictors of their performance. Only one of the HLS produced the error of attaching the plural marker -tul incorrectly to a classifier in the elicited narratives.

In terms of the quantificational structures produced by the HLS, there were not enough tokens per participant to conduct statistical analyses, but a variety of structures were produced by all

219 groups. About a third of the structures produced by the simultaneous bilinguals were prenominal structures, and just under half of the structures produced by the simultaneous bilinguals were postnominal structures. Amongst the sequential bilinguals, half of the quantificational structures were prenominal structures (i.e., Num N or Num Cl N), and about a quarter of the responses were postnominal structures. Although Shin and Milroy (1999) describe Num N as a “borrowing” from English, the structure is acceptable with human nouns (1999), which was generally the type of noun that the sequential bilinguals placed in these structures. Also, although having a prenominal classifier, as in *han kay swupak ‘one Cl watermelon’ (Shin and Milroy’s (8)) was considered to be ungrammatical, it is only ungrammatical if the genitive Case marker is missing from the classifier. The HLS tested here produced the genitive Case marker some of the time; not all tokens were ungrammatical. Thus, there is some evidence for the possible influence of English on the types of quantificational structures used by the HLS. Also, there may be an age effect, given that almost half of the simultaneous bilinguals who produced quantificational structures did not produce classifiers. There could also be a usage effect, given that their median usage score was below zero (i.e., their scores were generally in the bottom half of the HLS tested); these speakers’ exposure scores were not much higher (i.e., their median exposure score was slightly above zero). More data is required to investigate this further.

With respect to the semantic restriction against the use of comitative Case markers with inanimate , the HLS did not have any issues. In addition, when they produced the comitative Case markers, they used the correct phonological forms. The HLS also produced a variety of levels of speech and made only a few errors in producing inappropriate comitative Case markers with particular levels of speech, thus at least some speakers were sensitive to different registers. Further research is required, but the preliminary results suggest that the age of onset of English may be the most relevant variable related to this error.

Despite the differences between the simultaneous bilinguals and sequential bilinguals in the mean ages of onset of exposure to English and the number of years spent in Korea, the HLS performed similarly with respect to several of the structures and phenomena tested. The data from the HLS tested did not show evidence for the type of reduced Case system evidenced in Heritage Russian (Polinsky 2008). The participants of this study were not pre-grouped according to a particular level of proficiency, but they mainly consisted of speakers who were relatively highly proficient in Korean. They rated themselves fairly high, even though the majority of them

220 considered themselves to be dominant in English. Young ages of arrival or ages of first extended exposure to English did not necessarily correspond to low usage of and exposure to Korean. The results of this study suggest that HLS do not necessarily perform unlike their actual baseline (i.e., the speech of their parents), but it is also not the case that the HLS are necessarily receiving “faulty” input in the form of a language under attrition, as evidenced by the performance of the control group, which was generally in line with the literature, with the exception of their strong preference for and production of overt Case marking, irrespective of the type of Case marker required.

7.2 Limitations of the Study and Areas for Future Research

There are a number of limitations of the study. As was noted in Chapter 3, the phenomena examined in this study are sensitive to several different conditions. Not all conditions were tested and counted in the elicited narratives, since the majority of nouns were animate or human, and very few pronouns were produced. Recordings of speech in a naturalistic setting, or at least modifying the instructions of the task to elicit a story, rather than just a description of the pictures, may be more successful in eliciting pronouns and other types of structures, given that a story would require the HLS to be clear in identifying the participants they introduce in their narratives and to track them (by using pronouns or demonstratives) throughout the story. Using other experimental tasks, such as picture selection tasks to test the HLS’ comprehension, may also be useful in providing further insight into the HLS’ grammars. Ionin et al. (2008) have also used written elicitation tasks effectively in order to test the effects of specificity and definiteness on L2 English learners’ use of determiners. Chung (2013) also investigated several different factors (i.e., focus, definiteness, animacy, and structural priming; also examined in monolingual corpora by H. Lee 2006, among others) which are involved in Case drop in order to determine which cues were used by HLS and L2 learners in their decision to provide or drop Case marking in Korean. She used a written elicited production task, an oral picture description task, and an oral elicited narration task, and found that HLS made use of focus and animacy cues in particular in determining their use of Case drop with subjects (i.e., nominative Case markers) and objects (i.e., accusative Case markers). The results were different across tasks and modalities, and in her written task, her focus was on very informal colloquial language. In all but one of the contexts, the proportion of objects with overt Case marking was less than .50, which is suggestive of more Case drop than what I found. The significance of the different cues, as well as the differences

221 according to the register or level of formality of the language (even in a written modality, where Case marking would otherwise be expected in more formal writings) suggest that Case drop is a phenomenon that is quite complex and highly sensitive to the interactions of different factors.

Some task effects were observed in the thesis. It is possible that the preference task may have been too easy for the speakers I tested. They may already have been aware on a metalinguistic level that Case markers are supposed to be attached to nouns in Korean, and were thus able to correctly choose the sentences with more Case markers present. In addition, the elicited imitation task had been intended to be a different way of testing the HLS’ proficiency orally, but there were ceiling effects. The test materials were read at a moderate tempo when they were recorded. The HLS with the lowest levels of ability in Korean were unable to repeat some of the sentences to the end, but overall the HLS were highly successful in the elicited imitation task. Using longer sentences read at different speech rates may help better differentiate the levels of ability amongst the HLS.

The elicited narratives of the HLS revealed errors other than those analyzed in Chapter 5. Some of the sentences were ungrammatical or difficult to understand due to the incorrect usage of particular verbs or verbal morphology. The intended meanings were recoverable since I knew which pictures the HLS were describing, but the narratives suggest that rather than the nominal domain, the verbal domain may be where HLS grammars are incomplete, and where lack of extensive input and/or usage of Korean may play an even bigger role. Anecdotally, in cases where English-dominant HLS are speaking to Korean-dominant HLS and/or adults, a significant amount of collaboration occurs in the discourse, where Korean-dominant speakers may intervene or complete the sentences of English-dominant HLS once they understand the gist of the message that the English-dominant HLS are trying to convey. Since Korean is a verb-final language, many errors in the verbal domain would likely persist if English-dominant HLS rarely need to finish their sentences. The verbal domain is also likely an aspect that is only mastered with the help of formal instruction in the language. Thus, studies of discourse between English- dominant and Korean-dominant speakers could be even more informative of HLS’ grammars, along the lines of the research being conducted in the Heritage Language Variation and Change project (Nagy 2009). This project looks at three generations of speakers, and this is another why this is important. The HLS tested in this thesis were relatively young. As described in Section 2.5, Korean media is readily available via the Internet, as well as on the radio, TV, and movie

222 theatres in Toronto and the rest of the Greater Toronto Area. It is relatively easy to find Korean- dominant bilinguals to speak to, but this was not necessarily the case even a few decades ago. Thus, it would seem as though it would be easier to preserve and maintain Korean as a heritage language now than before. This could mean that differences may exist according to the age of the bilinguals at the time of testing—not because of attrition, but because the language community when older bilinguals arrived in Canada is not necessarily as dense as that which is available to younger bilinguals who arrived in Canada more recently.

Another reason why the analysis of conversation may be quite fruitful is that HLS could recognize Case markers without being in the practice of regularly supplying them; however, it is only by being engaged in discourse in Korean, where HLS would have the opportunity to recognize that omitting Case markers can result in confusion. For example, since word order is used to identify syntactic roles in the absence of overt Case marking, if the first noun of the sentence did not have a Case marker attached to it, that noun would likely be assumed by the hearer to be the subject of the sentence. If it were not the subject, this would likely lead to confusion.

One aspect that was not addressed in this thesis is the role of dialectal variation. There are many regional dialects in Korea, and the nominal domain is an area where such variation is particularly prominent (see Sohn (1999, pp. 57-86) for a description of dialectal differences in Korean). Phonetic studies have been conducted on Kyungsang (southeastern dialect) dialects as a dialect with pitch accent (e.g., J. Jun, J. Kim, H. Lee, & S.-A. Jun 2006). Other dialectal differences do not appear to have been recorded and studied extensively. In smaller cities, dialectal differences (from the standard) are more pronounced, and the inhabitants often speak the same dialect. However, in larger cities, particularly in Seoul, speakers of many different dialects co-exist, and it is quite common to hear different regional dialects, along with the standard Seoul dialect. The situation is similar in Toronto; speakers of many different dialects co-exist. HLS who only received extensive exposure to Korean as a child are more likely to retain the dialectal characteristics of their parents’ speech. HLS who lived with their grandparents as a child often speak more like their grandmothers.

223

7.3 Contributions to the Field

The findings of this thesis are significant in light of the scarcity of research on this population of speakers. Research has been conducted on HLS who are enrolled in Korean (or another heritage language) language classes at the university level. This thesis includes speakers who may be difficult to classify as belonging to a particular level of ability and who were not enrolled in language classes at the time of testing. Their commonality was their oral and aural ability in Korean, and their grammars were compared to the language of their parents, a baseline which has not been tested in very many studies on Korean as a heritage language. The data from the control group suggests that the HLS’ baseline input may not be under attrition, similar to the findings of the Heritage Language Variation and Change project for phenomena such as pro-drop in other heritage languages (Nagy, forthcoming).

The results of this study suggest that although the age of onset of extended exposure to English plays a role in certain respects, speakers who used Korean more frequently and who received more input in Korean, regardless of the age of onset of exposure, showed more consistency in correctly producing phenomena such as Case marking in Korean. In various statistical models, the HLS’ usage scores were a significant predictor of their performance. I found that overall the HLS tested did not have a lot of difficulty with the nominal domain of Korean, and their data did not show influence from English. In areas of the Korean nominal domain that were sensitive to differing levels of acceptability, the HLS generally showed awareness of these different levels. The HLS’ production of quantificational structures indicate some vulnerability in the syntactic domain, and their production of comitative Case markers that are not appropriate for the register used indicate vulnerability at the morphosyntactic-pragmatic interface, given that register conditions the acceptability of the different type of comitative Case markers. Overall, the findings of this study suggest that the nominal domain of Korean is not an area that is very vulnerable to incomplete acquisition or attrition, although further research investigating other aspects of the nominal domain may suggest otherwise. The results are informative to the field of bilingualism in understanding which aspects of the grammar are difficult to acquire and maintain. The results also suggest that HLS’ grammars are not impaired.

This thesis also highlights the importance of integrating sociolinguistics with research on heritage language acquisition, given the importance of identifying different aspects of these

224 speakers’ language histories. This study is also useful to those interested in the preservation and maintenance of Korean as a heritage language, including instructors of heritage language schools and courses on Korean comprised of students who are HLS, given that just increased exposure to and input in the language is not necessarily enough to help develop more consistency in the production of various morphemes and grammatical structures.

225

References

Aissen, J. (2003). Differential object marking: Iconicity vs. economy. Natural Language & Linguistic Theory, 21(3), 435-483.

Au, T. K.-f., Knightly, L. M., Jun, S.-A., & Oh, J. S. (2002). Overhearing a language during childhood. Psychological Science, 13, 238-243.

Barr, D. J., Levy, R., Scheepers, C., & Tily, H. J. (2013). Random effects structure for confirmatory hypothesis testing: Keep it maximal. Journal of Memory and Language, 68(3), 255-278.

Barreña, A., Ezeizabarrena, J., & García, I. (2008). Influence of the linguistic environment on the development of the lexicon and grammar of Basque bilingual children. In C. Pérez-Vidal, M. Juan-Garau, & A. Bel (Eds.), A Portrait of the Young in the New Multilingual Spain (pp. 86-110). Clevedon, UK: Multilingual Matters.

Bates, D., Maechler, M., Bolker, B., & Walker, S. (2014). lme4: Linear mixed-effects models using Eigen and S4. R package version 1.1-7. Retrieved from http://CRAN.R- project.org/package=lme4

Benmamoun, E., Montrul, S., & Polinsky, M. (2010). Prolegomena to heritage linguistics (White paper). MS., University of Illinois & Harvard University.

Biermann, A. (1982). Die grammatische Kategorie Numerus. In H. Seiler, & C. Lehmann (Eds.), Apprehension: Das sprachliche Erfassen von Gegenstaenden I: Bereich und Ordnung der Phaenomene (pp. 229-243). Tuebingen: Narr.

Bley-Vroman, R. (1989). What is the logical problem of foreign language learning? In S. M. Gass, & J. Schachter (Eds.), Linguistic Perspectives on Second Language Acquisition (pp. 41-68). Cambridge, UK: Cambridge University Press.

Bley-Vroman, R. (1990). The logical problem of foreign language learning. Linguistic Analysis, 20, 3-49.

Borer, H. (2005). Structuring Sense. Volume I: In Name Only. Oxford: Oxford University Press.

226

Byon, A. S. (2003). Language socialisation and Korean as a heritage language: A study of Hawaiian classrooms. Language, Culture and Curriculum, 16(3), pp. 269-283. doi:10.1080/07908310308666674

Canadian Council on Learning. (2007). Lessons in learning. French-immersion education in Canada. 1-13.

Carson, J. C. (2000). The semantics of number in Malay number phrases. MA thesis, University of Calgary.

Chang, S.-J. (1996). Korean. Amsterdam, Netherlands: John Benjamins.

Cheng, L. L.-S., & Sybesma, R. (1999). Bare and not-so-bare nouns and the structure of NP. Linguistic Inquiry, 30(4), 509-542.

Chierchia, G. (1998). Reference to kinds across languages. Natural Language Semantics, 6, 339- 405.

Cho, G., Cho, K.-S., & Tse, L. (1997). Why ethnic minorities want to develop their heritage language: The case of Korean-Americans. Language, Culture and Curriculum, 10, 106- 112.

Cho, G., Shin, F., & Krashen, S. (2004). What do we know about heritage languages? What do we need to know about them? Multicultural Education Journal, 11(4), 23-26.

Cho, S. W. (1981). The acquisition of word order in Korean. MA thesis, University of Calgary.

Chomsky, N. (1995). The Minimalist Program. Cambridge, MA: MIT Press.

Choo, M., & Kwak, H.-Y. (2008). Using Korean: A Guide to Contemporary Usage. Cambridge, UK: Cambridge University Press.

Chung, E. S. (2013). Exploring the degree of native-likeness in bilingual acquisition: Second and heritage language acquisition of Korean case-ellipsis. PhD dissertation, University of Illinois at Urbana-Champaign.

227

Corbett, G. G. (2000). Number. Cambridge Textbooks in Linguistics. Cambridge, UK: Cambridge University Press.

Corder, S. P. (1967). The significance of learner's errors. International Review of Applied Linguistics in Language Teaching, 5(1-4), 161-170.

Council of Europe. (2001). from the Common European Framework of Reference for Languages: Learning, Teaching, Assessment.

Cowper, E., & Currie Hall, D. (2000). Intransitive and: Locality, movement, and interpretation. In J. Jensen, & G. Van Herk (Ed.), Proceedings of the 2000 Annual Conference of the Canadian Linguistic Association (pp. 25-36). Ottawa: Cahiers Linguistiques d'Ottawa.

Croft, W. (2003). Typology and Universals. Cambridge, UK: Cambridge University Press.

Curtiss, S. R. (1976). Genie: A linguistic study of a modern day "wild-child.". PhD Dissertation, University of California, Los Angeles.

Dixon, R. M. (1979). Ergativity. Language, 55(1), 59-138. doi:10.2307/412519

Downing, P. (1996). Numeral Classifier Systems: The Case of Japanese. Amsterdam: John Benjamins.

Erlam, R. (2006). Elicited imitation as a measure of L2 implicit knowledge: An empirical validation study. Applied Linguistics, 27(3), 464-491.

Fishman, J. A. (2001). 300-plus years of heritage language education in the United States. In J. K. Peyton, D. A. Ranard, & S. McGinnis (Eds.), Heritage Languages in America: Preserving a National Resource (pp. 81-98). McHenry, IL: The Center for Applied LInguistics and Delta Systems.

Fromkin, V., Krashen, S., Curtiss, S., Rigler, D., & Rigler, M. (1974). The development of language in Genie: A case of language acquisition beyond the "critical period". Brain and Language, 1, 81-107.

228

Fry, J. (2001). Ellipsis and 'wa'-marking in Japanese conversation. PhD dissertation, Stanford University.

Gathercole, V. C. (2007). Miami and North Wales, so far and yet so near: A constructivist account of morphosyntactic development in blingual children. International Journal of Bilingual Education and Bilingualism, 10(3), 224-247.

Grosjean, F. (1987). Vers une psycholinguistique expérimentale du parler bilingue. In G. Lüdi (Ed.), Devenir Bilingue - Parler Bilingue (pp. 115-134). Tübingen: Niemayer.

Grosjean, F. (1998). Studying bilinguals: Methodological and conceptual issues. Bilingualism: Language and Cognition, 1, 131-141.

Han, C.-h., & Rambow, O. (2000). The Sino-Korean light verb construction and lexical argument structure. Workshop TAG, 5.

Hombert, J.-M., Ohala, J. J., & Ewan, W. G. (1979). Phonetic explanations for the development of tones. Language, 55(1), 37-58.

Hong, M., Park, K., Chung, I., & Kim, J.-y. (1998). Elided postpositions in spoken Korean and their implications on center management. Korean Journal of Cognitive Science, 9(3), 35- 45.

Hyltenstam, K., & Abrahamsson, N. (2003). Chapter 17. Maturational Constraints in SLA. In C. Doughty, & M. H. Long (Eds.), The Handbook of Second Language Acquisition. Malden, MA: Blackwell Publishing Ltd. doi:10.1002/9780470756492.ch17

Ionin, T., Zubizarreta, M., & Maldonaldo, S. (2008). Sources of linguistic knowledge in the second language acquisition of English articles. Lingua, 118(4), 554-576.

Jakobson, R. (1941). Kindersprache, Aphasie, und allgemeine Lautgezetze. Uppsala, Sweden: Almqvist & Wiksell.

Jeon, M. (2008). Korean heritage language maintenance and language ideology. Heritage Language Journal, 6(2), pp. 54-71.

229

Jia, G. X. (1998). Beyond brain maturation: The critical period hypothesis in second language acquisition revisited. PhD dissertation, New York University.

Jun, J., Kim, J., Lee, H., & Jun, S.-A. (2006). The prosodic structure and pitch accent of Northern Kyungsang Korean. Journal of East Asian Linguistics, 15(4), 289-317.

Kang, Y., & Nagy, N. (2013). VOT merger in heritage Korean in Toronto. Proceedings of the Annual Meeting of the Canadian Linguistics Association 2012.

Kanno, K., Hasegawa, T., Ikeda, K., & Ito, Y. (2005). Linguistic profiles of heritage bilingual learners of Japanese. (J. Cohen, K. T. McAlister, K. Rolstad, & J. MacSwan, Eds.) Proceedings of the 4th International Symposium on Bilingualism, pp. 1139-1151.

Kaufman, D., & Aronoff, M. (1991). Morphological disintegration and reconstruction in first language attrition. In H. Seliger, & R. Vago (Eds.), First Language Attrition (pp. 175- 188). Amsterdam: John Benjamins.

Kawashima, R. (1998). The structure of extended nominal phrases: The scrambling of numerals, approximate numerals, and quantifiers in Japanese. Journal of East Asian Linguistics, 7, 1-26.

Keijzer, M. (2007). Last in first out? An investigation of the regression hypothesis in Dutch emigrants in Anglophone Canada. PhD dissertation, Vrije Universiteit Amsterdam.

Kim, H.-S. H. (2001). Issues of heritage learners in Korean language classes. In J. Ree (Ed.), The Korean Language in America: Volume 6. Papers from the Annual Conference and Teacher Training Workshop on the Teaching of Korean Language, Culture, and Literature (6th, Manoa, Hawaii, August 2-5, 2001) (pp. 257-274). Manoa, Hawaii: American Association of Teachers of Korean.

Kim, H.-S. H. (2005). Processing strategies and transfer of heritage and non-heritage learners of Korean. PhD dissertation, University of Hawai'i at Manoa.

Kim, J.-H. (2007). Binding interpretations in adult bilingualism: A study of language transfer in L2 learners and heritage speakers of Korean. PhD dissertation, University of Illinois at Urbana-Champaign.

230

Kim, J.-H., & Montrul, S. (2004a). Binding interpretations in Korean heritage speakers. Proceedings of 28th Annual Boston University Conference on Language Development (pp. 307-317). Somerville, MA: Cascadilla Press.

Kim, J.-H., & Montrul, S. (2004b). Second language influence on first language attrition: Interpretation of the Korean binding system. Proceedings from the 10th [2003] Harvard International Symposium on Korean Linguistics (Harvard (pp. 86-98). Seoul: Hanshin.

Kim, J.-H., Montrul, S., & Yoon, J. (2005). Binding interpretations by Korean heritage speakers and adult L2 learners of Korean. In A. Brugos, M. R. Clark-Cotton, & S. Ha (Ed.), BUCLD 29 Online Proceedings Supplement. Retrieved from http://www.bu.edu/bucld/files/2011/05/29-KimBUCLD2004.pdf

Kim, J.-H., Montrul, S., & Yoon, J. (2009). Binding interpretations of anaphors by Korean heritage speakers. Language Acquisition, 16(1), 3-35.

Kim, K.-h., & Kwon, J.-i. (2004). Korean particles in spoken discourse - a statistical analysis for the unification of grammar. Hanmal yenku, 15, 1-22.

Kim, K.-h., & Kwon, J.-i. (2004). The characteristics of Korean spoken particles - A statistical analysis for grammar unification. Korean Language Research, 15, 1-22.

Kim, M.-J. (2007). The absence of the adjective category in Korean. Manuscript, University of Massachusetts. Retrieved from http://semanticsarchive.net/Archive/TU4NzlkM/MinJoo%20Adjective.pdf

Kim, N.-K. (1987). Korean. In B. Comrie (Ed.), The World's Major Languages (pp. 881-898). Oxford, UK: Oxford University Press.

Kim, Y.-j. (1997). The acquisition of Korean. In D. Slobin (Ed.), The Crosslinguistic Study of Language Acquisition (Vol. 4, pp. 335-435). Mahwah, NJ: Lawrence Erlbaum Associates, Inc.

Kim, Y.-J. (2000). Subject/object drop in the acquisition of Korean: A cross-linguistic comparison. Journal of East Asian Linguistics, 9(4), 325-351.

231

Klein, W. (1998). The contribution of second language acquisition research. Language Learning, 48(4), 527-550.

Knightly, L. M., Jun, S.-A., Oh, J. S., & Au, T. K.-f. (2003). Production benefits of childhood overhearing. The Journal of the Acoustical Society of America, 114(1), 465-474.

Kondo, K. (1998). Social-psychological factors affecting language maintenance: Interviews with Shin Nisei University students in Hawaii. Linguistics and Education, 9(4), 369-408. doi:10.1016/S0898-5898(97)90006-1

Kondo-Brown, K. (2001). Bilingual heritage students' language contact and motivation. In Z. Dörnyei, & R. Schmidt (Eds.), Motivation and Second Language Acquisition (pp. 433- 459). Honolulu: Second Language Teaching and Curriculum Center, University of Hawaii.

Koo, M.-C. (2008). Grammaticalization of Korean numeral classifiers. In E. Verhoeven, S. Skopeteas, Y.-M. Shin, Y. Nishina, & J. Helmbrecht (Eds.), Studies on Grammaticalization. Trends in Linguistics. Studies and Monographs 205 (pp. 59-75). Berlin: Mouton de Gruyter.

Kwon, J.-i. (1989). Characteristic of case and the methodology of the case ellipsis. Language Research, 25(1), 129-139.

Laleko, O. V. (2010). The syntax-pragmatics interface in language loss: Covert restructuring of aspect in heritage Russian. PhD dissertation, University of Minnesota.

Lambert, W. E. (1973). Culture and language as factors in learning and education. Retrieved from ERIC database (ED096820).

Lee, C.-M. (1989). (In)definites, case markers, classifiers and quantifiers in Korean. In S. Kuno, I.-H. Lee, J. Whitman, Y.-S. Kang, S.-Y. Bak, & Y.-j. Kim (Ed.), Harvard Studies in Korean Linguistics III (pp. 469-487). Cambridge, MA: Harvard University.

Lee, C.-M. (2000). Numeral classifiers, (in-)definites, and incremental themes in Korean. In C.- M. Lee, & J. Whitman (Ed.), Korean Syntax and Semantics: LSA Institute Workshop, Santa Cruz, '91. Seoul: Thaehaksa.

232

Lee, H. (2006). Parallel optimization in case systems: Evidence from case ellipsis in Korean. Journal of East Asian Linguistics, 69-96.

Lee, I., & Ramsay, S. R. (2000). The Korean Language. Albany, NY: State University of New York Press.

Lee, J. S. (2002). The Korean language in America: The role of cultural identity in heritage language learning. Lamguage, Culture and Curriculum, 15(2), 117-133.

Lee, J. S. (2008). Korean heritage language education in the United States: The current state, opportunities, and possibilities. 6(2), pp. 1-20.

Lee, J. S., & Shin, S. J. (2008). Korean heritage language education in the United States: The current state, opportunities, and possibilities. Heritage Language Journal, 6(2), 1-20.

Lee, S. A., & Iverson, G. K. (2012). Stop consonant productions of Korean–English bilingual children. Bilingualism: Language and Cognition, 15(2), 275-287. doi:10.1017/S1366728911000083

Lee, S., & Pae, S. (1989). Hankwuk atong-uy cosa ihay paltal yenkwu. [Development of Korean Children's Comprehension of Particles.]. Unpublished manuscript, Ewha Womans University.

Lee, S.-H., & Park, Y. (2008). A corpus-based analysis of case marker ellipsis in Korean. In Current Issues in Unity and Diversity of Languages: Collection of the Papers Selected from the CIL 18, Held at Korea University in Seoul, on July 21-26, 2008 (pp. 3845- 3863). Seoul, Korea: The Linguistic Society of Korea.

Lee, S.-H., Jang, S. B., & Seo, S.-k. (2009). Annotation of Korean learner corpora for particle error detection. CALICO Journal, 26(3), 529-544.

Lee, T. (2011). Grammatical knowledge of Korean heritage speakers: Early vs. late bilinguals. (J. Rothman, & R. Slabakova, Eds.) Linguistic Approaches to Bilingualism, 1(2), 149- 174.

233

Long, M. H. (1990). Maturational constraints on language development. Studies in Second Language Acquisition, 12(3), 251-285.

Lynch, A. (2003). The relationship between second and heritage language acquisition: Notes on research and theory building. Heritage Language Journal, 1(1), 26-43.

Martin, S. E. (1992). A Reference Grammar of Korean: A Complete Guide to the Grammar. Rutland, VT: Charles E. Tuttle Company.

McQuillan, J. (1996). How should heritage languages be taught?: The effects of a free voluntary reading program. Foreign Language Annals, 29(1), 56-72.

Merino, B. (1983). Language loss in bilingual Chicano children. Journal of Applied Developmental Psychology, 4, 277-294.

Montrul, S. (2002). Incomplete acquisition and attrition of Spanish tense/aspect distinctions in adult bilinguals. Bilingualism: Language and Cognition, 5(1), 39-68.

Montrul, S. (2004). Subject and object expression in Spanish heritage speakers. A case of morpho-syntactic convergence. Bilingualism: Language and Cognition, 7(2), 125-142. doi:10.1017/S1366728904001464

Montrul, S. (2006). On the bilingual competence of Spanish heritage speakers. Syntax, lexical- semantics and processing. International Journal of Bilingualism, 10(1), 37-69.

Montrul, S. (2008). Incomplete Acquisition in Bilingualism: Re-examining the Age Factor. Amsterdam: John Benjamins.

Montrul, S. (2010). Current issues in heritage language acquisition. Annual Review of Applied Linguistics, 30, 3-23.

Montrul, S., & Foote, R. (2008). Gender agreement in adult second language learners and Spanish heritage speakers: The effects of age and context of acquisition. Language Learning, 58(3), 503-553.

234

Munnich, E., Flynn, S., & Martohardjono, G. (1994). Elicited imitation and grammaticality judgment tasks: What they measure and how they relate to each other. In E. E. Tarone, S. M. Gass, & A. D. Cohen (Eds.), Research Methodology in Second-Language Acquisition (pp. 227-243). Mahwah, NJ: Lawrence Erlbaum.

Nagy, N. (2009). Heritage Language Variation and Change in Toronto. http://projects.chass.utoronto.ca/ngn/HLVC.

Nagy, N. (forthcoming). A sociolinguistic view of null subjects and VOT in Toronto heritage languages. Lingua.

Nakanishi, K., & Tomioka, S. (2004). Japanese plurals are exceptional. Journal of East Asian Linguistics, 13(2), 113-140.

Nemoto, N. (2005). On mass denotations of bare nouns in Japanese and Korean. Linguistics, 43(2), 383-413.

O' Grady, W., Lee, O.-S., & Lee, J.-H. (2011). Practical and theoretical issues in the study of heritage language acquisition. Heritage Language Journal, 8(3), 23-40.

Oh, J. S., Jun, S.-A., Knightly, L. M., & Au, T. K.-f. (2003). Holding on to childhood language memory. Cognition, 86(3), B53-B64.

Paradis, J., & Navarro, S. (2003). Subject realization and crosslinguistic interference in the bilingual acquisition of Spanish and English: What is the role of the input? Journal of Child Language, 30, 371-393.

Park, S. (2010). The laryngeal contrast in stops in Korean-English bilingual children: an acoustic study. MA forum paper, University of Toronto.

Parodi, T., Schwartz, B., & Clahsen, H. (1997). On the L2 acquisition of the morphosyntax of German nominals. Essex Research Reports in Linguistics, 15, 1-44.

Parodi, T., Schwartz, B., & Clahsen, H. (2004). On the L2 acquisition of the morphosyntax of German nominals. Linguistics, 44(3), 669-705.

235

Pérez-Leroux, A. T., Cuza, A., & Thomas, D. (2011). Clitic placement in Spanish-English bilingual children. Bilingualism: Language and Cognition, 14(2), 221-232. doi:10.1017/S1366728910000234

Polinsky, M. (1996). American Russian: An endangered language? Manuscript, USC-UCSD.

Polinsky, M. (2006). Incomplete acquisition: American Russian. Journal of Slavic Linguistics, 14(2), pp. 191-262.

Polinsky, M. (2008). Heritage language narratives. In D. M. Brinton, O. Kagan, & S. Bauckus (Eds.), Heritage Language Education: A New Field Emerging (pp. 149-164). New York: Routledge.

Polinsky, M., & Kagan, O. (2007). Heritage languages: In the 'wild' and in the classroom. Language and Linguistics Compass, 1(5), 368-395.

R Development Core Team. (2008). R: A Language and Environment for Statistical Computing. Vienna, Austria: R Foundation for Statistical Computing. Retrieved from http://www.R- project.org

Rothman, J. (2007). Heritage speaker competence differences, language change, and input type: Inflected infinitives in Heritage Brazilian Portuguese. International Journal of Bilingualism, 11(4), 359-389.

Sánchez, L. (2004). Functional convergence in the tense, evidentiality and aspectual systems of Quechua Spanish bilinguals. Bilingualism: Language and Cognition, 7(2), 147-162. doi:10.1017/S136672890400149X

Schwartz, B. D. (2004). Why child L2 acquisition? In J. Van Kampen, & S. Baauw (Ed.), Proceedings of Generative Approaches to Language Acquisition 2003, (pp. 47-66). LOT Occasional Series, Utrecht, Netherlands.

Scovel, T. (1988). A Time to Speak. A Psycholinguistic Inquiry into the Critical Period for Human Speech. Rowley, MA: Newbury House.

236

Sherkina-Lieber, M. (2011). Comprehension of Labrador Inuttitut functional morphology by receptive bilinguals. PhD dissertation, University of Toronto.

Shin, S. J. (2002). Birth order and the language experience of bilingual children. TESOL Quarterly, 36(1), 103-113.

Shin, S. J., & Milroy, L. (1999). Bilingual language acquisition by Korean schoolchildren in New York City. Bilingualism: Language and Cognition, 2(2), 147-167.

Shin, S.-Y. (2010). The functions of code-switching in a Korean Sunday School. Heritage Language Journal, 7(1), 91-116.

Silva-Corvalán, C. (1991). Spanish language attrition in a contact situation with English. In H. W. Seliger, & R. M. Vago (Eds.), First Language Attrition (pp. 151-171). Cambridge, UK: Cambridge University Press.

Silva-Corvalán, C. (1994). Language Contact and Change: Spanish in Los Angeles. New York: Oxford University Press.

Silva-Corvalán, C. (2008). The limits of convergence in language contact. (R. Nicolaï, & B. Comrie, Eds.) Journal of Language Contact – Thema 2, 213-224. Retrieved from www.jlc-journal.org

Silverstein, M. (1981). Case marking and the nature of language. Australian Journal of Linguistics, 1, 227-247.

Slobin, D. I. (1973). Cognitive prerequisites for the development of grammar. In C. A. Ferguson, & D. I. Slobin (Eds.), Studies of Child Language Development (pp. 175-208). New York: Holt, Rinehart & Winston.

Sohn, H.-M. (1999). The Korean Language. Cambridge, UK: Cambridge University Press.

Song, M., O' Grady, W., Cho, S., & Lee, M. (1997). The learning and teaching of Korean in community schools. In Y.-H. Kim (Ed.), Korean Language in America 2 (Vol. 2, pp. 111-127). American Association of Teachers of Korean.

237

Song, S. C. (1988). Nouns and pronouns, singular and plural. In S. C. Song, Explorations in Korean syntax and semantics (pp. 299-309). Berkeley, California: Institute of East Asian Studies, University of California-Berkeley, Center for Korean Studies.

Song, S.-C. (1975). Rare plural marking and ubiquitous plural marker in Korean. Language Research, 11(1), 77-86.

Statistics Canada. (2002). 2001 Community Profiles. Released June 27, 2002. Last modified November 30, 2005. Statistics Canada Catalogue no. 93F0053X1E. Retrieved from http://www12.statcan.ca/english/Profil01/CP01/Index.cfm?Lang=E

Statistics Canada. (2007). Toronto, Ontario (Code 535) (table). 2006 Community Profiles. 2006 Census. Released March 13, 2007. Statistics Canada Catalogue no. 92-591-XWE. Retrieved from http://www12.statcan.ca/census-recensement/2006/dp-pd/prof/92- 591/index.cfm?Lang=E

Steele, J. (2002). Representation and Phonological Licensing in the L2 Acquisition of Prosodic Structure. PhD dissertation, McGill University.

Suh, E. (2006). The acquisition of Korean -tul by heritage language speakers. Manuscript, University of Toronto.

Suh, E. (2008). The structure of the nominal phrase in Korean. Manuscript, University of Toronto.

Tse, L. (2001). Resisting and reversing language shift: Heritage-language resilience among U.S. native biliterates. Harvard Educational Review, 71(4), 676-706.

Unsworth, S. (2004). Child L1, child L2, and adult L2 acquisition: Differences and similarities. In A. Brugos, L. Micciulla, & C. E. Smith (Ed.), BUCLD 28: Proceedings of the 28th Annual Boston University Conference on Language Development (pp. 633-644). Somerville, MA: Cascadilla.

Valdés, G. (2005). Bilingualism, heritage language learners, and SLA research: Opportunities lost or seized? The Modern Language Journal, 89(3), 410-426.

238

Van Deusen-Scholl, N. (2003). Toward a definition of heritage language: Sociopolitical and pedagogical considerations. Journal of Language, Identity, and Education, 2, 211-230.

Wiley, T. G. (2001). Policy formation and implementation. In J. Peyton, D. A. Ranard, & S. McGinnis (Eds.), Heritage Languages in America: Preserving a National Resource. Language in Education. Theory and Practice (pp. 99-108). McHenry, IL: Center for Applied Linguistics.

Wong Fillmore, L. (1991). When learning a second language means losing the first. Early Childhood Research Quarterly, 6(3), 323-346.

Yeni-Komshian, G. (2009). Phonological abilities of Korean-English bilinguals. In C. Lee, G. B. Simpson, & Y. Kim (Eds.), The Handbook of East Asian Psycholinguistics. Volume 3: Korean (pp. 318-330). Cambridge, UK: Cambridge University Press.

Yi, Y. (2008). Voluntary writing in the heritage language: A study of biliterate Korean-heritage adolescents in the U.S. Heritage Language Journal, 6(2), 72-92.

239

Appendices

Appendix A

Appendix Table A1. Summary of HLS’ Attitudes towards Korean and Speaking Korean

Simultaneous bilinguals (N=15) Sequential bilinguals (N=18)

Category Mdn (IQR) M (SD) Range Mdn (IQR) M (SD) Range

Self-identity 1.0 (1.0) 1.3 (0.6) 0–2 1.0 (1.0) 1.5 (0.6) 1–3

Motivation 1.0 (3.5) 2.5 (2.4) 0–7 1.0 (5.5) 2.7 (2.9) 0–8

Do not mentally prepare 2.0 (2.0) 2.1 (1.0) 0–3 2.0 (1.0) 1.8 (0.8) 1–3

Do not avoid Korean 2.0 (1.0) 2.2 (0.7) 1–3 3.0 (1.0) 2.4 (0.8) 1–3

Feel confident 1.0 (0.5) 1.3 (0.7) 0–3 2.0 (1.0) 1.8 (0.9) 0–3

Communicate clearly 1.0 (1.0) 1.3 (0.6) 0–2 1.5 (1.0) 1.5 (0.9) 0–3 Note. The responses to the HLS’ self-identity were converted numerically as follows: 0: “Canadian”, 1: “Korean- Canadian (mostly Canadian)”, 2: Korean-Canadian (“mostly Korean”), and 3: “Korean”. The HLS selected a range of reasons for learning Korean, and their responses were quantified as follows: two points for any reasons which suggested personal motivation, such as for more flexibility in acquiring/moving upward in a job, because of a desire to communicate with more Koreans, or because of a desire to understand Korean culture; one point for any reasons which suggested external motivation, such as pressure from parents or because of a need to communicate with Koreans; zero points were given for any reasons which suggested that the participants had had no choice in the matter, e.g., responses such as “I was born in Korea”. The responses were totaled for each speaker. The responses for the last four categories (i.e., “do not mentally prepare”, “do not avoid Korean”, “feel confident”, and “communicate clearly”) were converted as follows: 0: “never”, 1: “occasionally, 2: “frequently”, and 3: “never”.

240

Appendix Table A2. Summary of HLS’ Language Use in Childhood

Simultaneous bilinguals (N=15) Sequential bilinguals (N=18)

Category Mdn (IQR) M (SD) Range Mdn (IQR) M (SD) Range

Birth order 1.0 (0.0) 1.2 (0.4) 1 – 2 2.0 (1.0) 1.9 (0.9) 1 – 4

Use of Korean with parents 2.0 (0.0) 1.7 (0.6) 0 – 2 2.0 (0.0) 1.9 (0.3) 1 – 2

Use of Korean with siblings 0.0 (1.0) 0.4 (0.6) 0 – 2 2.0 (1.0) 1.5 (0.7) 0 – 2

Use of Korean with others 0.0 (1.0) 0.6 (0.8) 0 – 2 2.0 (2.0) 1.3 (1.0) 0 – 2

Family’s Korean ability 2.6 (0.3) 2.6 (0.3) 1.9–3.0 2.9 (0.3) 2.6 (0.6) 0.5–3.0

Family’s English ability 2.1 (0.5) 2.3 (0.4) 1.6–3.0 2.0 (0.7) 1.8 (0.7) 0.0–2.5 Note. The responses on the categories relating to how often the HLS used Korean with various people were converted to numerical scores as follows: 0: “never”, 1: “occasionally”, 3: “frequently”, and 4: “always”. The responses on the categories relating to the HLS’ family members’ ability to speak Korean or English were converted to numerical scores as well: 0: “not at all”, 1: “barely”, 2: “fairly well”, and 3: “fluently”, depending on whether frequency or fluency was being measured.

Appendix Table A3. HLS’ Language Use with Family and Friends

Simultaneous bilinguals (N=15) Sequential bilinguals (N=18)

Category Mdn (IQR) M (SD) Range Mdn (IQR) M (SD) Range

Korean heard from family 2.0 (0.6) 2.0 (0.6) 0.8–3.0 2.5 (0.3) 2.5 (0.4) 1.5–3.0

Korean used with family 1.8 (0.3) 1.9 (0.5) 0.8–3.0 2.3 (0.7) 2.3 (0.5) 1.0–3.0

Korean heard from friends 1.0 (0.7) 0.9 (0.5) 0.1–1.9 1.0 (1.0) 1.2 (0.6) 0.1–2.2

Korean used with friends 0.8 (0.7) 0.8 (0.5) 0.1–1.9 1.0 (1.2) 1.2 (0.7) 0.0–2.2

Friends’ Korean ability 2.1 (0.3) 2.2 (0.4) 1.6–3.0 2.0 (1.0) 2.2 (0.6) 1.2–3.0

No mixing of English with Korean 1.0 (1.5) 0.9 (0.8) 0.0–2.0 1.3 (1.0) 1.3 (0.8) 0.0–2.0 Note. The responses for all of the categories were converted to numerical scores as follows: 0: “never” or “not at all”, 1: “occasionally” or “barely”, 2: “frequently” or “fairly well”, and 3: “always” or “fluently”, depending on whether frequency or fluency was being measured.

241

Appendix Table A4. HLS’ Use of Korean in Various Situations

Simultaneous bilinguals (N=15) Sequential bilinguals (N=18)

Category Mdn (IQR) M (SD) Range Mdn (IQR) M (SD) Range

% of Korean at home 30.0 (42.5) 39.7 (29.9) 0–95 55.0 (51.3) 56.9 (34.6) 0 –100

% of Korean at school 0.0 (0.5) 1.7 (3.6) 0–10 0.0 (0.0) 5.4 (21.2) 0 – 90

% of Korean at work 0.0 (0.0) 4.2 (10.3) 0–33 0.0 (0.0) 6.9 (16.4) 0 – 50

% of Korean in social situations 20.0 (25.0) 21.4 (16.9) 1–50 10.0 (46.3) 21.9 (26.8) 0 – 80

% of Korean in phone calls 10.0 (15.0) 13.8 (11.5) 2–40 10.0 (47.5) 27.3 (32.0) 0 – 95

# of hrs. speaking Korean/wk. 4.0 (10.5) 9.5 (11.0) 1–35 17.0 (22.5) 22.3 (30.1) 0 – 120

# of hrs. listening to Korean/wk. 3.0 (13.3) 9.2 (11.5) 0–35 19.0 (30.8) 24.6 (27.4) 0.3–100

Appendix Table A5. HLS’ Passive Exposure to Korean

Simultaneous bilinguals (N=15) Sequential bilinguals (N=18)

Category Mdn (IQR) M (SD) Range Mdn (IQR) M (SD) Range

# of hrs. of TV in Korean/wk. 2.0 (3.8) 2.4 (2.3) 0 – 7 2.5 (6.8) 5.6 (8.0) 0–28

# of movies in Korean/yr. 2.0 (3.0) 3.3 (2.8) 0–10 2.0 (7.4) 8.7 (15.2) 0–60

% of music in Korean 10.0 (36.5) 20.7 (23.0) 0–70 20.0 (40.0) 31.7 (31.4) 0–95

242

Appendix Table A6. HLS’ Written Interaction in and Exposure to Korean

Simultaneous bilinguals (N=15) Sequential bilinguals (N=18)

Category Mdn (IQR) M (SD) Range Mdn (IQR) M (SD) Range

% of texts in Korean 1.0 (7.5) 7.8 (14.1) 0–50 1.0 (31.3) 20.5 (34.4) 0–95

% of emails in Korean 1.0 (3.0) 4.6 (10.2) 0–40 5.0 (18.8) 14.2 (20.8) 0–60

# of hrs. browsing in Korean/wk. 0.0 (2.0) 1.2 (1.8) 0 – 6 2.0 (4.8) 4.8 (7.9) 0–25

# of Korean social media sites 0.0 (1.0) 0.4 (0.5) 0 – 1 0.0 (1.0) 0.4 (0.8) 0 – 3

# of books read in Korean/yr. 0.0 (0.0) 0.5 (1.4) 0 – 5 0.0 (1.5) 0.9 (1.5) 0 – 5

# of hrs. reading in Korean/wk. 0.5 (1.0) 1.2 (2.6) 0–10 0.0 (1.8) 3.6 (7.5) 0–25

Figure A1. Summary of HLS’ Responses on Language Questionnaire

243

Appendix Table A7. Correlation Table for Measures Used in Usage Scores for HLS

1 2 3 4 5

τ p τ p τ p τ p τ p

1. Friends —

2. Code-switching 0.08 .58 —

3. Home 0.33** <.01 0.29* .04 —

4. Social situations 0.26* .05 0.12 .40 0.26* .05 —

5. Text messages 0.51** <.01 0.08 .57 0.26* .05 0.27* .05 — Note. Friends = number of friends that the participants speak in Korean to, code-switching = frequency of avoidance of code-switching; home = percentage of use of Korean at home; social situations = percentage of use of Korean in social situations; text messages = percentage of use of Korean in text messages.

Appendix Table A8. Correlation Table for Measures Used in Exposure Scores for HLS

1 2 3 4 5

τ p τ p τ p τ p τ p

1. Family —

2. TV 0.12 .37 —

3. Movies 0.12 .36 0.34* .01 —

4. Music 0.10 .45 0.55** <.01 0.32* 0.02 —

5. Internet 0.10 .47 0.74** <.01 0.22 0.11 0.59** <.01 — Note. Family = number of family members that speak to the participants in Korean, TV = number of hours spent watching Korean TV per week, movies = number of Korean movies watched per year; music = percentage of music listened to in Korean; Internet = number of hours spent browsing online in Korean.

244

Appendix Table A9. Summary of the HLS’ Language Histories

# of yrs. Sum of usage Sum of exposure Group Participant # Age AoO in Korea Rankings Rankings 1 2 26 0 0.5 72 57 1 3 25 0 3.0 36 54 1 5 28 0 0.25 45 22 1 11 23 0 0.3 36 39 1 13 19 0 0.25 64 92 1 16 24 0 4.0 54 49 1 18 25 3 3.0 74 94 1 24 28 0 0.17 81 88 1 25 21 0 0.04 60 35 1 26 24 0 0.4 56 79 1 27 20 0 0.37 44 87 1 29 30 0 0.12 71 11 1 30 23 0 0.04 118 72 1 31 25 3 7.0 127 99 1 41 29 3 4.5 52 59 2 1 29 4 4.08 24 14 2 4 18 4 4.63 76 31 2 6 35 5 5.67 17 35 2 7 20 5 5.17 49 58 2 8 34 9 9.5 47 8 2 10 20 9 9 135 106 2 12 20 8 8.13 83 93 2 14 20 6 8 65 105 2 15 20 9 9.9 58 136 2 17 24 9 9.75 59 65 2 21 21 7 7 56 51 2 22 19 8 8 85 57 2 28 23 9 9.17 125 127 2 32 21 10 10 76 117 2 35 18 8 8 73 80 2 37 19 8 9 139 151 2 38 20 8 9 107 93 2 40 25 7 7.4 142 120 Note. Group 1 = simultaneous bilinguals; group 2 = sequential bilinguals.

245

Appendix Table A10. Summary of HLS’ Language Histories and Measures of Ability in Korean (as High or Low)

Group Participant # AoO Usage Exposure Imitiation Self-rating SI MLU NDW 1 2 Low High Low Low Low Low Low Low 1 3 Low Low Low Low Low Low Low High 1 5 Low Low Low Low Low High High High 1 11 Low Low Low Low Low Low Low Low 1 13 Low Low High Low High Low Low Low 1 16 Low Low Low Low High Low High High 1 18 Low High High High Low Low High High 1 24 Low High High High Low High Low Low 1 25 Low Low Low Low Low Low Low Low 1 26 Low Low High Low Low Low Low Low 1 27 Low Low High Low Low Low Low Low 1 29 Low High Low Low Low High High High 1 30 Low High High High Low Low Low Low 1 31 Low High High High High High Low High 1 41 Low Low Low High High High High High 2 1 Low Low Low High Low Low Low Low 2 4 Low High Low Low High Low Low Low 2 6 High Low Low Low Low Low Low Low 2 7 High Low Low High Low Low Low Low 2 8 High Low Low Low High High High Low 2 10 High High High High High High High High 2 12 High High High High High High High High 2 14 High High High High Low High High High 2 15 High Low High Low Low High High High 2 17 High Low Low High High High High High 2 21 High Low Low Low Low High High High 2 22 High High Low High High Low Low Low 2 28 High High High High High High Low High 2 32 High High High High High High High High 2 35 High High High High Low High High Low 2 37 High High High High High High High High 2 38 High High High Low High Low High Low 2 40 High High High High High High High High Note. Group 1 = simultaneous bilinguals; group 2 = sequential bilinguals. AoO = age of onset of exposure; imitation = imitation score; self-rating = self-rating of ability in Korean; SI = subordination index; MLU = mean length of utterance; NDW = number of different words.

246

Appendix Table A11. Summary of Individual Simultaneous Bilinguals’ Performance with Case Marking on the Preference Task, Elicited Imitation Task, and the Elicited Narration Task

Nominative Accusative

Proportion of Proportion of Proportion of Proportion of Proportion of sentences Case markers Case markers Case markers Case markers correctly correctly correctly correctly correctly Participant # selected repeated produced repeated produced 2 0.92 1.00 1.00 0.70 0.92 3 0.75 1.00 0.71 0.91 0.28 5 0.58 0.33 0.38 0.60 0.18 11 0.83 1.00 0.83 0.91 0.69 13 0.88 1.00 1.00 1.00 0.80 16 0.83 1.00 0.81 1.00 0.83 18 0.88 1.00 0.97 1.00 0.60 24 0.88 1.00 0.87 0.77 0.39 25 0.96 1.00 0.89 0.91 0.82 26 0.75 0.67 0.90 0.92 0.42 27 0.92 0.67 0.00 0.22 0.00 29 0.83 1.00 1.00 0.92 0.48 30 0.67 0.83 0.90 0.91 0.69 31 0.96 1.00 1.00 1.00 0.90 41 0.83 1.00 1.00 0.91 0.68

247

Appendix Table A12. Summary of Individual Sequential Bilinguals’ Performance with Case Marking on the Preference Task, Elicited Imitation Task, and the Elicited Narration Task

Nominative Accusative

Proportion of Proportion of Proportion of Proportion of Proportion of sentences Case markers Case markers Case markers Case markers correctly correctly correctly correctly correctly Participant # selected repeated produced repeated produced 1 0.54 1.00 0.67 1.00 0.20 4 0.71 0.80 0.33 0.73 0.40 6 0.88 1.00 1.00 0.86 0.25 7 0.92 0.83 0.70 0.91 0.62 8 0.71 1.00 1.00 0.89 0.77 10 0.92 1.00 1.00 1.00 0.92 12 0.88 1.00 0.87 1.00 0.94 14 0.92 1.00 0.83 1.00 0.52 15 0.92 0.83 0.90 0.80 0.84 17 0.88 1.00 1.00 1.00 0.93 21 0.83 1.00 1.00 0.92 0.54 22 0.88 0.83 0.83 0.64 0.31 28 0.83 1.00 0.89 1.00 0.14 32 0.92 1.00 1.00 1.00 0.67 35 0.92 1.00 0.88 0.92 0.65 37 0.92 1.00 0.74 0.91 0.83 38 0.88 1.00 1.00 1.00 0.89 40 0.83 1.00 0.91 1.00 0.88

248

Appendix Table A13. Summary of Individual Simultaneous Bilinguals’ Performance (as Pass or Fail, based on 0.75 Threshold) with Case Marking on the Preference Task, Elicited Imitation Task, and the Elicited Narration Task

Nominative Accusative

Proportion of Proportion of Proportion of Proportion of Proportion of sentences Case markers Case markers Case markers Case markers correctly correctly correctly correctly correctly Participant # selected repeated produced repeated produced 2 Pass Pass Pass Fail Pass 3 Pass Pass Fail Pass Fail 5 Fail Fail Fail Fail Fail 11 Pass Pass Pass Pass Fail 13 Pass Pass Pass Pass Pass 16 Pass Pass Pass Pass Pass 18 Pass Pass Pass Pass Fail 24 Pass Pass Pass Pass Fail 25 Pass Pass Pass Pass Pass 26 Pass Fail Pass Pass Fail 27 Pass Fail Fail Fail Fail 29 Pass Pass Pass Pass Fail 30 Fail Pass Pass Pass Fail 31 Pass Pass Pass Pass Pass 41 Pass Pass Pass Pass Fail

249

Appendix Table A14. Summary of Individual Sequential Bilinguals’ Performance (as Pass or Fail, based on 0.75 Threshold) with Case Marking on the Preference Task, Elicited Imitation Task, and the Elicited Narration Task

Nominative Accusative

Proportion of Proportion of Proportion of Proportion of Proportion of sentences Case markers Case markers Case markers Case markers correctly correctly correctly correctly correctly Participant # selected repeated produced repeated produced 1 Fail Pass Fail Pass Fail 4 Fail Pass Fail Fail Fail 6 Pass Pass Pass Pass Fail 7 Pass Pass Fail Pass Fail 8 Fail Pass Pass Pass Pass 10 Pass Pass Pass Pass Pass 12 Pass Pass Pass Pass Pass 14 Pass Pass Pass Pass Fail 15 Pass Pass Pass Pass Pass 17 Pass Pass Pass Pass Pass 21 Pass Pass Pass Pass Fail 22 Pass Pass Pass Fail Fail 28 Pass Pass Pass Pass Fail 32 Pass Pass Pass Pass Fail 35 Pass Pass Pass Pass Fail 37 Pass Pass Fail Pass Pass 38 Pass Pass Pass Pass Pass 40 Pass Pass Pass Pass Pass

250

Appendix Table A15. Proportion of Types of Quantified Structures Produced in Elicited Narratives

Simultaneous bilinguals Sequential bilinguals Control

# of # of # of Structure Proportion tokens Proportion tokens Proportion tokens

No noun With Classifier (Num-Cl) 0.56 5 0.64 9 0.44 4 Other 0.44 4 0.36 5 0.56 5 Total 1.00 9 1.00 14 1.00 9

Pre-nominal Without Classifier (Num-N) 0.64 9 0.64 28 0.42 10 With Classifier (Num-Cl-N) 0.29 4 0.36 16 0.54 13 Other 0.07 1 0.00 0 0.04 1 Total 1.00 14 1.00 44 1.00 24

Post-nominal Without Classifier (N-Num) 0.22 8 0.12 5 0.25 11 With Classifier (N-Num-Cl) 0.53 19 0.56 23 0.32 14 Other 0.25 9 0.32 13 0.43 19 Total 1.00 36 1.00 41 1.00 44