Automatic Speech Recognition Technology for Practicing and Testing Reading Aloud in L1 and L2 (73432)
Session Chair: Manashi Gogoi Dutta
Saturday, 11 November 2023 11:45
Session: Session 2
Room: Sri Nakron
Presentation Type: Paper Presentation
Reading is a learned skill that children acquire after intensive practice that also requires guided reading aloud. Since this is often difficult to organize in traditional classrooms, researchers have been studying the usability of Automatic Speech Recognition (ASR) technology in learning to read programs, primarily for reading in English. For languages other than English less research has been conducted and fewer resources are available to develop pedagogically sound ASR-based programs for learning to read.
In the project 'Dutch ASR-based Reading Tutor' we investigated how reading software with Automatic Speech Recognition (ASR) technology for learning to read in a language other than English can best be developed and implemented in the school practice and at home. Together with the partners Zwijsen Publishers and NovoLanguage we developed the first prototype of an online Reading Tutor for Dutch and used it to conduct research on reading development. The increasing diversity and multilingualism in Dutch primary schools mean that more children with different home languages have to learn to read in Dutch and this poses additional challenges to ASR technology. In our research a total of 752 first graders from 44 Dutch schools practiced with the Reading Tutor and received feedback on their reading in different forms. Since the Reading Tutor is equipped with logging capabilities, the logfiles containing the speech recordings and information about the children’s behaviour during practice and their reactions to the different forms of feedback can be used for research on reading development. In addition, to study the effectiveness of practice and feedback with the Reading Tutor a test of reading proficiency was administered to all children and recorded online. The recordings of the children reading were first evaluated by 51 teachers and subsequently analysed through different ASR systems.
In this contribution we present results about the affordances of different ASR systems for learning to read, both for providing practice in guided reading aloud and for developing more efficient and objective tests of reading proficiency. We discuss the performance of traditional and more recent ASR systems for reading practice in L1 and L2 and outline possible lines of future research.
Abstract Summary
Reading is a learned skill that children acquire after intensive practice involving guided reading aloud. Since reading aloud is not always feasible in classrooms, researchers have been considering Automatic Speech Recognition (ASR) technology to support reading programs, especially for English. In the project 'Dutch ASR-based Reading Tutor' we investigated how ASR technology can contribute to reading development in 752 first graders, for a language other than English.
In this contribution we present results about the affordances of different ASR systems for learning to read, both for providing practice in guided reading aloud and for developing more efficient and objective tests of reading proficiency. We discuss the performance of traditional and more recent ASR systems for reading practice in L1 and L2 and outline possible lines of future research.
Authors:
Wieke Harmsen, Radboud University, Netherlands
Yu Bai, Radboud University, Netherlands
Ferdy Hubers, Radboud University, Netherlands
Catia Cucchiarini, Radboud University, Netherlands
Helmer Strik, Radboud University, Netherlands
About the Presenter(s)
Ms Wieke Harmsen is a University Doctoral Student at Radboud University in Netherlands
See this presentation on the full schedule – Saturday Schedule
Comments
Powered by WP LinkPress