Effective pronunciation training is crucial for EFL learners' language proficiency and communication skills. Difficulties in English vowel pronunciation, particularly diphthongs, are common among Arab EFL learners due to differences between Arabic and English phonological systems. High-Variability Phonetic Training (HVPT) has shown promise in consonant pronunciation, but its application to vowels requires further investigation. This study examines the impact of HVPT, with and without captions (using YouGlish as a multimodal tool), on Saudi EFL learners' pronunciation of English diphthongs. The research questions investigate the interaction between modality (captions) and variability (HV vs. LV) on pronunciation accuracy and student perceptions of HVPT using YouGlish. The study aims to fill a gap in existing research focusing primarily on consonants and explore YouGlish’s effectiveness as a multimodal pronunciation tool.

The literature review examines existing research on speech production training, highlighting different approaches like explicit vs. incidental and single-modal vs. dual-modal training. It discusses the use of videos in EFL, noting their potential to engage learners but also the risk of passive consumption without structured instruction. The review focuses on the use of videos and captioning in improving pronunciation and vocabulary acquisition, citing studies showing both positive and mixed results based on learner proficiency level and task focus. Studies on HVPT and LVPT are reviewed, highlighting the benefits of HVPT for perception and production, but also noting the potential for increased cognitive load. The study positions itself within the framework of Cognitive Load Theory (CLT), considering the mental effort required to process information under different training conditions.

This study used an experimental design with 56 Saudi undergraduate female EFL learners, divided into four groups: LV no captions, HV no captions, HV with captions, and LV with captions. Participants completed a pre-test assessing their pronunciation of three English diphthongs (/ɔɪ/, /aʊ/, and /aɪ/) using 15 minimal pairs. They then underwent 3 training sessions using YouGlish videos, with the HV groups exposed to multiple speakers and the LV groups to a single speaker. Post-tests, generalization tests (with new minimal pairs), and delayed tests were administered. A survey measured participants' perceptions of YouGlish's effectiveness in improving pronunciation, using a 5-point Likert scale. Data analysis involved descriptive statistics, independent-sample t-tests, and two-way repeated measures ANOVA tests using SPSS version 27. Pronunciation test scores were evaluated by three applied linguistics Ph.D. professors to ensure objectivity. A pilot study was conducted to refine the word lists and ensure task validity.

Descriptive analysis showed that the LV no caption group had a mean score of 11.25, the LV caption group 13.19, the HV no caption group 12.55, and the HV caption group 9.91. Repeated measures ANOVA revealed significant main effects for captioning, variability, and their interaction. Pairwise comparisons showed significant differences between LV no caption and HV no caption groups and between HV caption and LV no caption groups. The LV with captions condition showed the highest improvement, followed by HV no captions, LV captions, and finally HV captions. The ANOVA test on the perception survey (Cronbach's alpha = 0.80) showed no significant difference in the overall positive perceptions of YouGlish among the four groups, although the LV caption group reported the highest mean score (4.31).

The findings suggest that both HVPT and LVPT were effective in improving diphthong pronunciation, irrespective of captioning. The LV no caption condition performed better than HV no caption group indicating that in the absence of captions, focused training on a consistent model is more effective. The LV with captions condition showed the best results, supporting the beneficial impact of captions in focused training according to the CLT. Conversely, the addition of captions in the high variability condition led to a decrease in performance, possibly due to increased cognitive load. This contradicts some previous studies and highlights the complex interaction between variability and modality. The generally positive perception of YouGlish across all conditions suggests its potential as a supplementary tool for pronunciation training. Future research could explore the optimal balance between variability, modality, and learner proficiency levels and the interaction of explicit instruction with HVPT.

This study demonstrates the effectiveness of both HV and LV pronunciation training for EFL learners, while also highlighting the nuanced role of captioning. The optimal approach depends on the specific training condition and suggests the importance of considering cognitive load. YouGlish shows potential as a valuable supplementary tool. Future research should investigate the interaction of these factors with explicit instruction and different proficiency levels.

The study's focus on female Saudi undergraduate students limits the generalizability of the findings. The specific diphthongs chosen might not fully represent the range of pronunciation challenges faced by all EFL learners. The use of a specific online platform (YouGlish) may limit the broader applicability of the conclusions. The duration of training could be expanded in future studies to assess long-term retention.