AI Twin: Enhancing ESL Speaking Practice through AI Self-Clones of a Better Me

Paper Title

AI Twin: Enhancing ESL Speaking Practice through AI Self-Clones of a Better Me

Publication Info

• Topic area: AI-assisted language learning with a focus on emotional and motivational engagement.
• Keywords: ESL, AI self-clone, Ideal L2 Self, language learning, engagement, motivation, voice synthesis, implicit feedback, conversational AI, second language acquisition.

Background and Problem

• Problem / challenge: Existing AI-assisted ESL speaking tools often rely on explicit corrective feedback, which can disrupt conversational flow, discourage learners, and negatively impact motivation and emotional engagement.
• Significance: Addressing emotional and motivational barriers in language learning is crucial for fostering learner confidence, reducing anxiety, and promoting sustained practice.
• Motivation and related work: Prior systems have focused on accuracy and correction but have largely neglected affective and motivational support. Implicit feedback strategies like recasts have shown promise but remain ambiguous. The Ideal L2 Self framework highlights the motivational power of aligning learning tools with learners' aspirational identities.

Solution

• Proposed approach: AI Twin—a personalized AI self-clone that rephrases learner utterances into more fluent English and delivers them in the learner’s own voice, aligning with their aspirational Ideal L2 Self.
• Novelty:
  1. Introduction of AI Twin as a personalized self-clone embodying the learner’s Ideal L2 Self.
  2. Empirical evaluation showing that implicit rephrasing fosters greater emotional engagement than explicit feedback.
  3. Design implications for integrating self-representative AI into educational technologies to enhance motivation and engagement.
• Procedure and key techniques:
  • Learners register their voice to create a personalized AI clone.
  • Spoken input is transcribed using ASR, reformulated by an LLM, synthesized in the learner’s cloned voice, and passed to an AI interlocutor for conversational practice.
  • Feedback is embedded through rephrasing, preserving conversational flow and reducing anxiety.

Results

• Concrete findings: AI Twin significantly increased emotional engagement (mean score: 4.83) compared to explicit feedback (mean score: 4.03, p < .001). Cognitive and behavioral engagement showed no significant differences across conditions.
• Advantage over baselines: AI Twin elicited higher emotional engagement than explicit feedback and comparable engagement to non-personalized rephrasing (AI Proxy). Participants described AI Twin as more enjoyable, motivational, and reassuring.
• Experiments / evaluation:
  • Within-subject study with 20 adult South Korean ESL learners (ages 24–36, CEFR levels A1–C1).
  • Three feedback conditions: Explicit Feedback, AI Proxy, AI Twin.
  • Mixed-method design combining engagement questionnaires and semi-structured interviews.
• Limitations and future work:
  • Single-session design limits evaluation of long-term learning effects.
  • Exclusively focused on Korean adult learners; results may not generalize to other populations.
  • Future work should explore longitudinal studies, hybrid feedback designs, and applicability across diverse cultural and linguistic contexts.

Summary

AI Twin introduces a personalized AI self-clone that rephrases learner utterances into fluent English and delivers them in their own voice, aligning with their Ideal L2 Self. In a study with 20 ESL learners, AI Twin significantly enhanced emotional engagement compared to explicit feedback, fostering motivation and reducing anxiety. Participants described the system as immersive, enjoyable, and supportive of their aspirational identity as proficient English speakers. While preferences varied based on engagement and perceived learning, AI Twin highlights the potential of self-representative AI to create affectively supportive and motivationally aligned learning technologies. Future work should explore hybrid designs and broader applicability across diverse learner populations.