An ASR-enabled Reading Tutor: Investigating Feedback to Optimize Interaction for Learning to Read

Abstract

An ASR-based Dutch Reading Tutor (RT) was developed and applied to further investigate the impact of different forms of feedback as opposed to no feedback on reading aloud by Dutch first graders. The total of 752 first-grade students of Dutch practiced with the RT during fluency exercises in which they had to read words twice and received automatic feedback (implicit or explicit) or no-feedback. The results show that lower reading accuracy at the first attempt was accompanied by a slowdown in reading speed at the second attempt, even in the no-feedback condition. This trade-off between reading accuracy and speed resulted in higher accuracy scores at the second attempt across the board, with the best results in the explicit feedback condition. The results also show that such an ASR-based RT can be employed as a research instrument to obtain detailed insights into reading development. In turn these can also contribute to optimizing the design of RTs.