Gavin Bui

As one of the pioneering applications of machine learning for quantitative data analysis in applied linguistics, this study investigates the predictive power of linguistic features on English as a Foreign Language (EFL) writing quality. To overcome the limitations of traditional linear regression, specifically multicollinearity and non-normal data distributions, we employed a Multilayer Perceptron (MLP) neural network to analyse 96 essays from Chinese secondary school students. The model evaluated the relative predictive importance of various lexical, syntactic, fluency, and accuracy features. Demonstrating exceptional fit, the neural network accounted for approximately 80% of the variance in overall writing scores with no evidence of overfitting. Sensitivity analyses revealed that lexical diversity was the most robust predictor of writing quality (100% normalised importance), followed by grammatical accuracy (60.4%). Conversely, syntactic complexity, text length, and lexical sophistication exhibited comparatively minimal influence. These findings underscore the critical role of vocabulary diversity in evaluating adolescent EFL writing, while successfully establishing neural networks as a powerful, innovative methodological tool for future applied linguistic research.

This study investigates the effects of task interactivity (monologic versus dialogic tasks) and willingness to communicate (WTC) on the speech fluency of 88 Hong Kong undergraduate English as a second language learners. Participants, categorised into high- and low-WTC groups based on a WTC questionnaire, completed either a monologic or a dialogic collaborative “spot-the-picture-difference” task. Fluency was assessed through speech rate, frequency of filled pauses and unfilled pauses at difference locations, and repairs. Results indicated that learners with high WTC demonstrated significantly faster speech rates and produced fewer repairs than those with low WTC. Monologic tasks elicited significantly more mid-clause pauses and slightly more filled pauses compared to dialogic tasks. Crucially, a significant interaction effect revealed that the disparity in speech rate between high- and low-WTC learners was more pronounced during monologic tasks. This suggests that the detrimental effect of lower WTC on speed fluency is particularly marked when learners engage in independent speaking, while dialogic tasks may offer a more supportive environment for learners with lower WTC to maintain a fluent speech rate. These findings highlight the complex interplay among affective factors, task design and second language oral performance, with implications for language pedagogy and assessment.

This research explores the dynamics in perceptions of L2 English learners in task-based speaking interactions with human interlocutors in comparison to AI (chatbot “Doubao”) communication partners over seven weeks. Findings from three rounds of interviews indicated enthusiastic initial adoption of the AI talk pal, driven by its ability to stimulate output, its ease of access, and its lower affective barriers to communication. However, with continued use, their fervent zeal had cooled to being “lukewarm” and finally “cool” due to AI’s lack of social presence, speech prosody, responsive interaction, and appropriate feedback, whereas interaction with real communication partners was consistently preferred for their authentic nature. In light of AI’s limitation, learners adapted to develop prompt literacy which enabled them to deliberately employ AI for particularised tasks, such as increasing fluency and acquiring vocabulary. Based on the findings, a dynamic model for AI acceptance in L2 task interaction is proposed to advance our understanding of the role of AI in taskbased learning and teaching. It suggests that a hybrid pedagogy, which sequences AI and human interaction to leverage their complementary strengths, is essential for effective task-based language learning.

This study investigated the engagement of 60 Hong Kong English-as-a-second-language (ESL) learners in two types of tasks, each with two levels of complexity (-elements/simple versus +elements/complex). The learners formed self-initiated pairs and completed two descriptive tasks (-reasoning demand) and two narrative tasks (+reasoning demand) in a counterbalanced order in synchronous video-based computer-mediated communication (SvCMC) across two separate meetings. Immediately after each task performance, the learners were interviewed about their emotional experiences during the tasks. Quantitative analysis of their spoken discourse revealed that learners were more cognitively engaged in the simple descriptive task than in the complex one. However, the number of elements did not seem to affect learner engagement in the narrative tasks. Descriptive tasks engaged learners behaviorally in task performance, while narrative tasks encouraged social engagement. The participants generally found both types of tasks emotionally engaging, but more participants experienced positive emotions during the descriptive tasks than during the narrative tasks. These findings suggest that task complexity and task type should be carefully considered when designing interactive oral tasks in online teaching and learning contexts.

This mixed-methods study investigated the impact of musical lyrics and working memory capacities on second language (L2) English reading comprehension among adult Chinese native speakers. Participants were 57 adult ESL learners with advanced L2 proficiency, divided into high and low working memory capacity groups based on a pre-test. They completed reading comprehension tasks while listening to the same melody with either Chinese or English lyrics. The results revealed that participants exhibited better reading comprehension, hence less distraction, when exposed to Chinese lyrics compared to English lyrics, contrary to initial expectations. Interestingly, working memory capacity did not significantly mediate the effects of background music on reading comprehension, suggesting that other factors such as linguistic familiarity and cognitive load may play a more prominent role. The interview data further highlighted participants’ perceptions, indicating that music, particularly songs with lyrics in their native language, often distracts from reading tasks. This finding showed discrepancies between some adult learners’ perception and their actual L2 reading performance. This study contributes to the understanding of how background music and its lyrics influence L2 reading and provides insights into the complex interplay between linguistic familiarity, cognitive resources, and reading performance.