PREDICTING ENGLISH PROFICIENCY WITH READ-ALOUD PERFORMANCE AND LINGUISTIC DIFFICULTY OF SENTENCES

K. Kotani1, T. Yoshimi2

1Kansai Gaidai University (JAPAN)
2Ryukoku University (JAPAN)
Background:
Learners’ RaP (read-aloud performance) helps to predict their EP (English proficiency), oral and written proficiency. This is because learners’ RaP reflects their linguistic processing ability.
Previous studies have found the correlation between RaP and EP. These studies examined the correlation of the RaP that was measured solely with the accuracy of the read-aloud.
However, the RaP should also be measured with a learner’s confidence for the read-aloud. In addition, the EP should be predicted by considering linguistic features of read-aloud sentences, because the RaP depends on the linguistic difficulty of sentences . The remaining questions from the previous studies are: How can EP be predicted with RaP measured with the accuracy and the confidence compositely?; Do the linguistic features contribute to the prediction of EP?

Purposes:
To fill the gaps in the literature, this study addresses the following research questions:
•To what extent can EP be predicted with the composite RaP and linguistic features?
•What are the effective predictors of EP?

Method:
To answer these research questions, this study adopted multiple linear regression analysis to predict learners’ EP, and to analyse the effects of predictors with standardized regression coefficients. The prediction accuracy was verified in a leave-one-out cross validation test.
The data to be analyzed included 750 instances of speech sound that 50 Japanese learners of English read aloud 15 sentences.
Learners’ EP was determined using their scores of the Test of English for International Communication. RaP was assessed with three evaluation criteria compositely. Learners’ confidence was derived by converting the subjective judgment according to a five-point Likert scale. Read-aloud accuracy was calculated by dividing the number of words correctly read aloud by that in the corresponding sentence. Speech rate was calculated by dividing the number of words by the duration of read-aloud.
The linguistic features included sentence length, mean word length, the number of multiple-syllable words, and word difficulty. These linguistic features were automatically derived from sentences in the read-aloud material.

Results:
A significant regression equation (F (6, 743) = 27.71, p < 0.05) was yielded with an adjusted squared correlation coefficient R^2 of 0.18. This means that the effect size of this squared correlation coefficient was medium.
A medium correlation (r = 0.41) between the predicted and observed EPs was found in the cross-validation test. A small correlation in the ease of read-aloud (r = 0.17), and medium correlations in read-aloud accuracy (r = 0.33), and speech rate (r = 0.33) were found.
Statistically significant effect (p < 0.05) was found in read-aloud accuracy, speech rate and the sentence length, but not in the ease of read-aloud, mean word length, number of multiple-syllable words, and word difficulty.

Discussion:
The reason why the effect size of the composite evaluation criteria was larger than that of the single evaluation criteria would be that the composite evaluation criteria could compensate the weakness of other criteria.
In comparison with read-aloud accuracy and speech rate, the ease of read-aloud was determined by learners, which may have caused the rater reliability.