Metacognition and Learning in Spoken Dialogue Computer Tutoring

We investigate whether four metacognitive metrics derived from student correctness and uncertainty values are predictive of student learning in a fully automated spoken dialogue computer tutoring corpus. We previously showed that these metrics predicted learning in a comparable wizarded corpus, where a human wizard performed the speech recognition and correctness and uncertainty annotation. Our results show that three of the four metacognitive metrics remain predictive of learning even in the presence of noise due to automatic speech recognition and automatic correctness and uncertainty annotation. We conclude that our results can be used to inform a future enhancement of our fully automated system to track and remediate student metacognition and thereby further improve learning.