AI-Enhanced Live-Streamed English Language Learning: Designing and Evaluating an Intelligent Platform for Chinese Learners

Open Access
Article
Conference Proceedings
Authors: Nan Jiang, Chengcheng Ma, Peter Childs
Abstract

Knowledge Sharing Live Streaming (KSLS) on Chinese commercial platforms such as Douyin (TikTok) and BiliBili is a popular choice for English language learning, offering attributes such as accessibility, low cost, and large-scale reach. However, a fundamental paradox exists: while effective language learning requires active engagement, real-time feedback, and meaningful interaction, KSLS is inherently a passive, one-to-many format that struggles to support these essential pedagogical elements. With AI technologies, particularly large language models, there is an opportunity to explore if AI can bridge this gap. A focus group study was conducted with 18 participants experienced in live-streamed English lessons. Thematic analysis identified two primary pain points: (1) uncertainty about pronunciation accuracy due to the absence of immediate corrective feedback, and (2) limited contextual speaking practice, as self-organised groups frequently collapsed without structure or topic guidance. Informed by these findings, an AI-integrated platform was designed with two key innovations: an AI-driven pronunciation feedback module providing real-time visual and auditory assessments, and AI-led conversational practice in breakout rooms where AI assumes four differentiated roles (mentor, expert, partner, servant) along two dimensions of closeness and guidance derived from social actor theory. The design was evaluated through Wizard-of-Oz experiments. For the pronunciation module, a within-subject study found that 63% of learners preferred multimodal (visual + auditory) feedback and 78% preferred sentence-level immediate feedback. For the AI roles, a between-subject study showed that the mentor role achieved the highest overall satisfaction (M=4.3/5), while the partner role scored highest on comfort (M=4.4/5). A subsequent deployment study using AI-powered chatbots further confirmed significant differences across conditions in engagement (F=10.521, p<0.05), vocabulary acquisition (F=81.213, p<0.05), and speaking proficiency across fluency, grammar, lexicon, and pronunciation. This study contributes to human-computer interaction and technology-enhanced language learning by providing empirical evidence of the KSLS paradox in English education, a theoretically grounded design framework, and experimental validation of AI integration in live-streamed language learning.

Keywords: Live-Streamed English Learning, AI-Enhanced Language Education, Pronunciation Feedback, Conversational Practice, Human-Computer Interaction

DOI: 10.54941/ahfe1008235

Cite this paper
Downloads
2
Visits
4
Download PDF

More from this volume

"They Will Move": Confirmation Bias, Expectation Frames, and Premature Monitoring Drop Off in COLREG Encounters →
View all articles in Human Factors in Design, Engineering, and Computing →