Analyzing Read Aloud Speech by Primary School Pupils: Insights for Research and Development

Reading software based on Automatic Speech Recognition (ASR) has been proposed as a possible supplement to traditional classroom instruction to help pupils achieve the required level of reading proficiency. However, the knowledge required to develop such software is not always available, especially for languages other than English. To this end, we analyzed a corpus containing speech material from Dutch native primary school pupils who read texts aloud at their mastery reading level. We investigated reading strategies, reading miscues, a novel reading miscue index and their relationship with AVI level (reading level) and gender. We found a significant effect of AVI level on reading miscue index, but did not find a decrease of reading miscue index as AVI level increased. Pupils mostly used lexical reading strategies, which seem to increase when AVI level increases. Miscues most frequently concerned low-frequency words with at least two syllables, and omitted and inserted words were generally high frequent, unstressed function words. These results provide insights that help design the content of reading interventions and that can contribute to developing and improving ASRbased reading software. We discuss the results in view of current trends in education and technology, and their implications for future research and development.