Exploring Mobile-assisted High Variability Phonetic Training and ASR-based Articulation Practice for EFL Children: Impacts on L2 Perception, L2 Production, Phonological Working Memory and Learner Experience = 모바일 기반 고변이 음성 지각 훈련과 자동 음성 인식 기반 조음 연습 탐색:영어를 외국어로 배우는 아동의 제2언어 지각, 제2언어 발화, 음운 작업 기억 및 학습자 경험에 미치는 영향

다국어 초록 (Multilingual Abstract)

With the recent advancement of AI-based technologies, there has been a growing effort to effectively integrate automatic speech recognition (ASR) tools into second language pronunciation instruction. ASR enables real-time pronunciation feedback with personalized practice. While effective for production, High Variability Phonetic Training (HVPT) has been recognized as an effective technique for improving L2 speech perception and, to a lesser extent, production (Thomson & Derwing, 2015). HVPT has the potential to bridge research and classroom practice (Levis, 2016) and is accessible and cost-effective through computer- or mobile-assisted platforms (Thomson, 2012). However, its application to authentic educational settings, particularly among young learners, remains underexplored. The current study addresses this gap by examining the effects of mobile-assisted HVPT combined with ASR-based articulation practice on Korean elementary English as a foreign language (EFL) learners’ L2 perception, L2 production, and phonological working memory.
Twenty-four Korean elementary students (mean age = 8.5) participated in six mobile-based HVPT sessions over two weeks. The training incorporated talker variability, ASR, and real-time corrective feedback. To analyze level-specific effects, learners were grouped by two proficiency levels using mock TOEFL Primary listening test scores.
L2 perception was measured via a forced-choice identification task in vowel pairs. The results showed significant improvement from pretest to both posttest and delayed posttest. Generalized linear mixed-effects modeling revealed that learners were 3.48 times more likely to correctly identify vowels after training. The /eɪ–ɛ/ pair showed the highest gain, which was statistically significant. In contrast, the /ɑ–ʌ/ pair remained persistently difficult, showing minimal and non-significant perception enhancement. The session and vowel interaction was significant, confirming unequal training benefits across contrasts. Stimulus-level analysis indicated higher accuracy for familiar lexical items. The learners’ proficiency level exerted no significant influence.
L2 production was assessed through free spontaneous speaking and elicited imitation tasks with the carrier sentence. The results increased significantly from pretest to posttest, with sustained effects at the delayed posttest. Vowel /ɪ/ was consistently the most challenging to produce, while /ɑ/ and /ʌ/ were most accurately produced. Linear mixed-effects modeling revealed significant main effects of session, task, and vowel, indicating sustained production gains with notable vowel-specific variation. No significant effect of proficiency level was found. Perception and production were moderately correlated at pretest and strongly correlated at posttest, especially among low-proficiency learners. The relationship weakened at the delayed posttest.
The phonological working memory (PWM) was evaluated with the nonword repetition task (NWR) and digit span task. For NWR, participants’ performances significantly improved, with longer syllable sequences predicting lower scores. In the digit span task, digit length was a strong negative predictor of accuracy, and a significant interaction was found between session and digit length at the posttest. However, no significant main effect of session or proficiency level was observed. Learner experience was probed through a brief post-training survey. The responses indicated high engagement and positive attitudes toward mobile-based pronunciation training. Open-ended responses highlighted interest in gamified features, user-friendly design, and sustained access to speech learning tools.
Overall, the current study provides empirical evidence for integrating HVPT and ASR in a mobile platform: the approach yields durable gains in L2 perception, L2 production, and PWM while sustaining learner motivation. The current study offers actionable guidelines for the development and classroom deployment of developmentally appropriate, technology-enhanced pronunciation tools for elementary EFL settings.

국문 초록 (Abstract)

최근 인공지능 기반 기술의 발전에 따라 자동 음성 인식 도구를 제2언어 발음 교육에 효과적으로 통합하려는 노력이 증가하고 있다. 자동 음성 인식은 실시간 발음 피드백과 개인 맞춤형 연습을 가능하게 한다. 자동 음성 인식은 발화 능력 향상에 효과적인 반면, 고변이 음성 훈련은 제2언어 음성 지각 향상에 효과적인 기법으로 인정받아 왔으며, 발화에도 일정 부분 효과가 있는 것으로 알려져 있다(Thomson & Derwing, 2015). 고변이 음성 훈련은 연구와 교실 현장을 연결해줄 수 있는 잠재력을 가지며(Levis, 2016), 컴퓨터나 모바일 기반 플랫폼을 통해 접근성과 비용 효율성 측면에서도 장점을 가진다(Thomson, 2012). 그러나 실제 교육 현장, 특히 아동 학습자를 대상으로 한 적용은 아직 충분히 탐색되지 않았다. 본 연구는 이러한 공백을 보완하고자, 모바일 기반 고변이 음성 훈련과 자동 음성 인식 기반 조음 연습을 결합하여 한국 초등 영어 학습자의 지각, 발화, 음운 작업 기억에 미치는 영향을 탐구하였다.
24명의 한국 초등학생(평균 연령 = 8.5세)이 2주간 6회의 모바일 기반 고변이 음성 훈련 훈련 및 자동 음성 인식 기반 조음 연습에 참여하였다. 수준별 효과 분석을 위해, 학습자들은 모의 토플 프라이머리 듣기 점수를 기준으로 숙달도에 따라 2개 그룹으로 나뉘었다.
강제 선택 식 지각 과제를 통해 측정된 지각 결과는 사전 검사 대비 사후 검사 및 지연 사후 검사에서 유의미한 향상을 보였다. 일반화 선형 혼합효과모형 분석 결과, 훈련 후 학습자가 모음을 정확히 식별할 확률이 3.48배 높아진 것으로 나타났다. /eɪ–ɛ/ 모음 쌍에서 가장 큰 향상이 나타났으며, 통계적으로 유의미하였다. 반면, /ɑ–ʌ/ 모음 쌍은 지속적으로 어려움을 보여 지각 향상이 미미하고 유의하지 않았다. 세션과 모음 간 상호작용 효과는 유의하였으며, 이는 모음 대조쌍별로 훈련 효과가 다름을 보여준다. 자극 단위 분석에서는 익숙한 어휘일수록 정확도가 높았다. 학습자의 숙달도는 유의한 영향을 미치지 않았다.
발화 점수 또한 사전 검사 대비 사후 검사에서 유의미하게 향상되었고, 지연 사후 검사에서도 효과가 유지되었다. /ɪ/ 모음은 지속적으로 가장 발화하기 어려웠으며, /ɑ/와 /ʌ/ 모음은 가장 정확하게 산출되었다. 선형 혼합효과모형 분석 결과, 세션, 과업 유형, 모음에 대한 주요 효과가 모두 유의하였으며, 이는 지속적인 발화 향상과 함께 모음별 차이가 있음을 시사한다. 숙달도는 유의한 영향을 보이지 않았다. 지각과 발화 간에는 사전 검사에서 중간 수준의 상관관계가, 사후 검사에서는 특히 저숙련 학습자에서 강한 상관관계가 나타났다. 그러나 지연 사후 검사에서는 그 관계가 약화되었다.
음운 작업 기억과 관련하여, 실어 반복 과제에서 학습자의 수행 능력이 유의미하게 향상되었으며, 음절 수가 길수록 점수가 낮아지는 경향이 나타났다. 숫자 기억 과제에서는 숫자 길이가 정확도의 강한 부적 예측 요인이었으며, 사후 검사에서 세션과 숫자 길이 간 상호작용 효과가 유의하였다. 그러나 세션이나 숙련도의 주요 효과는 유의하지 않았다.
훈련 이후 실시된 간단한 설문을 통해 학습자의 경험을 조사한 결과, 모바일 기반 발음 훈련에 대해 높은 몰입도와 긍정적인 태도를 보였다. 개방형 응답에서는 게임화된 기능, 사용자 친화적 디자인, 지속적인 접근 가능성에 대한 흥미가 강조되었다.
종합적으로, 본 연구는 고변이 음성 훈련과 자동 음성 인식을 모바일 플랫폼에 통합하는 것이 제2언어 지각, 발화, 음운 작업 기억에 지속적인 향상을 가져오며, 학습자의 동기 유지에도 기여함을 실증적으로 보여준다. 본 연구는 초등 영어 교육 현장에 적합한 기술 기반 발음 훈련 도구 개발 및 수업 적용을 위한 실천적 지침을 제공한다.

목차 (Table of Contents)

• ABSTRACT i
• TABLE OF CONTENTS iv
• LIST OF TABLES viii
• LIST OF FIGURES x
• LIST OF ABBREVIATIONS xii
• CHAPTER 1. INTRODUCTION 1
• 1.1 Background of the Study 1
• 1.2 Purpose of the Study 6
• 1.3 Research Questions 7
• 1.4 Organization of the Dissertation 8
• CHAPTER 2. LITERATURE REVIEW 9
• 2.1 Theoretical Framework for L2 Speech Learning 10
• 2.1.1 Speech Learning Model 10
• 2.1.2 Speech Learning Model-revised 13
• 2.1.3 Perceptual Assimilation Model (PAM) 15
• 2.2 Vowels in English and Korean 18
• 2.3 High Variability Phonetic Training 25
• 2.3.1 HVPT and Its Effects on Perception 25
• 2.3.2 Diverse Effects of HVPT by Proficiency Level 29
• 2.3.3 Transfer of HVPT Effects to Production 30
• 2.3.4 Generalizability of HVPT Effects 33
• 2.3.5 Retention Effects of HVPT 35
• 2.4 Articulation Practice and the Development of Phonological Working Memory 36
• 2.4.1 Definition and Components of Phonological Working Memory 36
• 2.4.2 Articulation Practice and Phonological Working Memory in L2 Learning 38
• 2.4.3 Phonological Working Memory as Both Predictor and Outcome in L2 Speech Training 40
• 2.4.4 Measuring Phonological Working Memory in Young L2 Learners 41
• 2.5 Technology-assisted Pronunciation Instruction 44
• 2.5.1 Mobile-assisted Pronunciation Training 44
• 2.5.2 Automatic Speech Recognition in Pronunciation Learning 46
• 2.6 Learner Experience and Integrative Models of Technology Acceptance 49
• 2.7 Summary of the Literature Review 52
• CHAPTER 3. METHODOLOGY 54
• 3.1 Research Design 54
• 3.1.1 Participants of the Study 54
• 3.1.2 Setting of the Study 55
• 3.2 Experimental Design and Materials 56
• 3.2.1 Research Instruments 59
• 3.2.1 Target vowel for the experiment 69
• 3.2.3 Intervention Tools Used for HVPT and Articulation Practice 70
• 3.3 Procedures 75
• 3.3.1 Intervention Procedures 75
• 3.3.2 Post training Survey 79
• 3.4 Data Collection and Analysis 80
• 3.4.1 Quantitative Analysis 80
• 3.4.2 Learner Experience Survey and Analysis 84
• CHAPTER 4. RESULTS 86
• 4.1. Improvement in L2 Vowel Perception 86
• 4.1.1. Descriptive Analysis of L2 Vowel Perception Enhancement 86
• 4.1.2 GLMM Analysis of Vowel Perception Accuracy Enhancement 93
• 4.1.3 Stimulus-level Variability in Vowel Perception Accuracy 104
• 4.2 Improvement in L2 Vowel Production 107
• 4.2.1. Descriptive Analysis of L2 Vowel Production Enhancement 107
• 4.2.2. LMM Analysis of Vowel Production Intelligibility Enhancement 122
• 4.3 Relationship between Perception and Production 133
• 4.4 Enhancement of Phonological Working Memory 140
• 4.4.1. PWM Measured by Nonword Repetition Tasks 140
• 4.4.2. PWM Measured by Digit Span Tasks 147
• 4.5 Learner Experience of Mobile-assisted HVPT and ASR-based Speaking Practice 154
• 4.5.1 Learner Perceptions from Survey 154
• 4.5.2 Learner Reflections on HVPT and Articulation Practice 157
• 4.6 Summary of Key Findings 161
• CHAPTER 5. DISCUSSION 163
• 5.1 Instructional Effects of HVPT on L2 Vowel Perception 163
• 5.1.1 Overall Gains and Retention Patterns in L2 Perception 163
• 5.1.2 Effects by Vowel Contrast and Stimulus-level Variability 166
• 5.1.3 Outcomes by Learner Proficiency 169
• 5.2 Instructional Effects of HVPT with ASR-based Training on L2 Vowel Production 172
• 5.2.1 Overall Gains and Retention Patterns in L2 Production 172
• 5.2.2 Effects by Vowel and Production Task 176
• 5.2.3 Outcomes by Learner Proficiency 180
• 5.3 Correlational Insights from Perception and Production 182
• 5.3.1 Correlational Changes across Sessions 182
• 5.3.2 Outcomes by Learner Proficiency 184
• 5.4 Instructional Effects on Phonological Working Memory 185
• 5.4.1 Overall Gains and Retention Patterns in PWM 185
• 5.4.2 Effects by Input Length and Task Features 186
• 5.4.3 Outcomes by Learner Proficiency 188
• 5.5 Pedagogical Implications Based on Key Findings 190
• 5.5.1 Prioritizing High-Impact Vowel Contrasts 190
• 5.5.2 Designing MAPT for Young Learners 191
• 5.5.3 Integrating PWM Training into Pronunciation Instruction 193
• 5.6 Theoretical and Methodological Contributions 195
• 5.6.1 HVPT as a Bridge between SLA Theory and Practice 195
• 5.6.2 Validation of Perception–Production Link 196
• 5.6.3 Expanding Methodological Practices for L2 Research on Children 197
• 5.7 Learner Experience and Implications for MAPT Design 198
• CHAPTER 6. Conclusion 201
• 6.1 Major Findings 201
• 6.2 Implications 204
• 6.2.1 Pedagogical Implications 204
• 6.2.2 Theoretical Implications 205
• 6.2.3 Methodological Implications 206
• 6.2.4 Practical and Technological Implications 207
• 6.3 Limitations of the Study and Directions for Future Research 208
• REFERENCES 213
• APPENDICES 230
• 국문초록 252