Measurement of Pronunciation Difficulty using Support Vector Regression
Takehiko Yoshimi, Ryukoku University, Japan ; Katsunori Kotani, Kansai Gaidai University, Japan
EdMedia + Innovate Learning, in Amsterdam, Netherlands ISBN 978-1-939797-42-1 Publisher: Association for the Advancement of Computing in Education (AACE), Waynesville, NC
This study examined a method of measuring language learners’ difficulty in pronouncing certain sentences by evaluating the measurement accuracy between different learning algorithms, the contribution of the features, and the influence of training data size. Although these types of influences should have been clarified in a machine learning approach to the measurement of pronunciation difficulty, they have not been considered in previous studies. This study found that (1) the support vector regression method outperformed the multiple linear regression method; (2) the greatest influence on measurement accuracy was the learners’ features (TOEIC scores); and (3) the measurement accuracy decreased when the training data size was reduced to 20% of the entire training data.
Yoshimi, T. & Kotani, K. (2019). Measurement of Pronunciation Difficulty using Support Vector Regression. In J. Theo Bastiaens (Ed.), Proceedings of EdMedia + Innovate Learning (pp. 729-733). Amsterdam, Netherlands: Association for the Advancement of Computing in Education (AACE).
© 2019 Association for the Advancement of Computing in Education (AACE)