著者

Akiyo HIRAI Yusuke KONDO Ryoko FUJITA

出版者

The Japan Association for Language Education and Technology

雑誌

外国語教育メディア学会機関誌 (ISSN:21857792)

巻号頁・発行日

vol.58, pp.17-41, 2021 (Released:2021-08-18)

---

This study examines the accuracy of an automated speech scoring system. The system graded English language learners’ retelling performances according to five features, and its scores were compared to those given by both non-native and native English-speaking (NNES and NES) raters. The results show that, of the five features, words per second was the most consistent predictor of both NNES and NES evaluations. However, the NNES rater tended to pay more attention to exact word similarities between the speech utterances and the original text, while the NES raters focused more on similarities of meaning and gave credit to rephrased expressions. Additionally, the correspondence between the automated scores and those given by human raters was moderate (exact agreement = 48% to 65%; rs = .48 to .52), though less than that between the NNES and NES scores (rs = .70). These results indicate that the automated scoring system for retelling performances may be applicable to low-stakes tests if the speech transcription of learners’ utterances is obtained.