A branch of natural language processing in computer science, ASR uses speech processing software to recognize words and in some applications, speech sounds in a speaker’s spoken input. The Reading Tutor’ ASR function is based upon Carnegie Mellon University’s open-source Sphinx-II speech analyzer, which uses the well-known probabilistic “Hidden Markov Model” approach to interpreting spoken input.
Published in Chapter:
A Computer-Based Reading Tutor for Young Language Learners
Kenneth Reeder (The University of British Columbia, Canada), Jon Shapiro (The University of British Columbia, Canada), Margaret Early (The University of British Columbia, Canada), Maureen Kendrick (The University of British Columbia, Canada), and Jane Wakefield (The University of British Columbia, Canada)
Copyright: © 2008
|Pages: 30
DOI: 10.4018/978-1-59904-895-6.ch010
Abstract
This chapter describes the first year of research on the effectiveness of automated speech recognition (ASR) for ESL learners in the early school years. The aim was to learn how such technology can enhance literacy learning as an element of L2 development, using prototype research software entitled the Reading Tutor (RT). In addition to assessing learners’ gains in reading scores, the attitudinal dimension of speech recognition technology was investigated in an effort to explain the effectiveness of the software. We found that both heritage language (L1) and level of English proficiency were linked to students’ reading gains with the RT. Further, the RT was shown to be equally effective to a more time-intensive volunteer tutoring program. A positive affective impact of the RT was demonstrated in the interview data but not in two widely used attitudinal scales. An Appendix describes the technical implementation of the project.