Felicia : Development of a Voice-Based AI Companion Supporting Reading Literacy among Young Learners

Sammanfattning

The decline in children’s reading proficiency globally highlights an urgent and pressing need for targeted, individualized reading literacy practice. In Sweden, where recent PISA results reveal a regression back to the unexpectedly low 2012-level reading scores, this issue becomes even more critical. There is an immediate demand for scalable, classroom-compatible literacy interventions that can bridge the gap effectively. While one-to-one human tutoring remains the most proven method to significantly improve reading skills, resource limitations within classrooms frequently render such personalized approaches unfeasible on a large scale. To address this gap, this study explores how voice-based AI technologies, specifically automatic speech recognition (ASR), text-to-speech, and large language models (LLMs), can be designed to support teachers in their reading instruction and young learners in their reading training by replicating key instructional and motivational features of human tutoring. Through a user-centered co-design process with teachers (n=6) and second grade pupils (n = 43), an AI-assisted reading tutor named Felicia was developed. Felicia engages in pair-reading with children, offering personalized feedback, and visual scaffolding support on word-level reading errors. The system was evaluated through a mixed-methods study with 32 pupils, using the pre- and post-intervention reading engagement surveys, observational data, and the system’s interaction logs. Results indicate significant short-term increases in cognitive and affective reading engagement, with the study participants responding positively to the tutor’s immediate feedback and turn-taking design. This work contributes design insights for educational AI systems for reading literacy training and demonstrates the potential of socially responsive, speechbased AI tutors to support teachers’ provision of personalized early literacy instruction at scale. Future work should investigate long-term effects on the reading engagement of learners over time, the impact on reading proficiency, and the broader development of human-AI support in diverse classroom environments.