Multiple Linear Regression of Combined Pronunciation Ease and Accuracy Index

Abstract

This study proposes an index for measuring the ease and accuracy of pronunciation that is derived by combining the ease and the accuracy of learners’ pronunciation, assesses the reliability and validity of the combined index, and develops a measurement method. The ease is a scale score for the ease of pronunciation that learners subjectively judged, and the accuracy demonstrates learners’ pronunciation performance. A previous study proposed an index regarding ease of pronunciation, and other previous studies independently proposed indices related to accuracy. These two types of indices should be combined because they have different aspects and compensate for each other. To develop the proposed measurement method, a learner corpus of pronunciation is compiled, and index reliability and validity are assessed using the classical test theory. The assessments demonstrate that the proposed index is moderately reliable and not valid. The proposed measurement method is developed using multiple linear regression analysis. The dependent variable is an index consisting of both pronunciation ease, which is subjectively judged by learners of foreign languages, and pronunciation accuracy, which is defined as the similarity between a reference sentence and learner pronunciation. The independent variables are defined as sentence linguistic features and learner features, such as foreign language proficiency test scores. The experimental results demonstrate that the measured ease and accuracy of pronunciation has moderate correlation with the observed ease and accuracy ($r = 0.69$), and a significant contribution is observed in all the linguistic and learner features ($p < 0.05$).