Beyond responding fast or slow: Improving cognitive models of memory retrieval using prosodic speech features

Adaptive learning systems have improved the process of fact and word learning by tailoring learning procedures to the needs of individual students. Recently, as a result of developments in automatic speech recognition technology, learning systems that do not require typed- or keypress input but allow for spoken learning have become increasingly popular. Here, we explore the possibility of improving such speech-based learning systems by exploiting automatically scored, supra-segmental prosodic speech features such as pitch dynamics, speaking speed and loudness. We demonstrate that prosodic speech features are associated to learner performance, and that they can be used to improve the adaptive model estimations of learner performance on future trials. Our results have both theoretical and practical implications, as they elucidate how speaker certainty is reflected in prosodic speech features and contribute to the further development of educationally relevant speech-based adaptive learning systems.