A Multimodal Generative AI Tutor for Teaching Phonological Processes and Feature Geometry: Enhancing Metaphonological Awareness, Articulatory Accuracy, and Transfer in L2 Learners

This study created and assessed a multimodal generative AI tutor that provides explicit instruction on phonological processes and feature geometry to second language learners. This study combined a multilingual model, text to speech synthesis, interactive articulatory visuals and adaptive rule induction. This was incorporated into the classroom for 8 weeks. 90 L2 English learners were asked to take part in the mixed-methods quasi-experimental design. Learners were separated into two groups: AI tutoring and traditional explicit instruction. The tutor administered tests to measure learners’ knowledge of processes (assimilation and feature spreading), phonological awareness, accuracy in controlled production and knowledge transfer into free speech. The results revealed significantly larger gains in the AI group for explicit knowledge and metaphonological awareness. Proper pronunciation added value significantly, for process-based errors within the AI condition, which notably notable transfer to spontaneous speech. Learners conveyed that the instructional framework were highly useful, though an insignificant number noted occasional AI hallucinations in the rule explanations. The findings portrayed that multimodal generative AI can link theory in phonology with real human speech pronunciation. Furthermore, it proposals a tutor system that can be reprocessed and shows that teaching phonological structures function better when delivered in an suitable multimodal way. Likewise, analysis showed that gains were strong for phonological processes where features cut across ordered nodes. Looking at how learners interacted with the system, those who used the feature tree tools often scored higher on transfer tasks. In all, working dynamically with structure helps more, than getting generative clarifications.

Keywords: generative explanations, Artificial Intelligence, phonological processes, explicit instruction, production accuracy.