Abstract:

Proficiency of English as a Second Language (ESL) has been measured using a computer-based method that examines ESL learners’ language use. This method statistically estimates proficiency by calculating the association between proficiency test scores and linguistic features derived from the learners’ language use. Linguistic features explain the performance of learners’ language use, and these features demonstrate a strong correlation with test scores of proficiency tests such as the Test of English for International Communication (TOEIC).

Computer-based proficiency measurements crucially support ESL learning and teaching. ESL teachers can measure their students’ proficiency at the end of each class, and these short-term assessments can enhance students’ motivation and reveal their learning achievements. Among the various types of language use, a measurement method that examines learners’ performance in reading aloud is acknowledged for its cost-effective administration. Read-aloud performance demonstrates both passive and active processes under which ESL learners recognize lexical and syntactic elements for articulating speech sounds. Reading aloud-based proficiency measurement requires a few minutes to read aloud a short passage, while proficiency tests require a longer test time, for example, Duolingo in an hour and TOEIC in a few hours.

The read-aloud performance involves principal factors such as fluency and accuracy in reading aloud. Fluency indicates how closely ESL learners’ temporal features of speech sound resemble those of English speakers, and it can be automatically assessed through speech rate. Accuracy aligns ESL learners’ speech sounds with those of English speakers and traditionally requires manual evaluation. Thus, the evaluation of speech sound accuracy increases administration costs. However, this cost issue can be resolved by introducing an automatic accuracy evaluation, although its validity must be confirmed.

Hence, this study aims to verify a reading aloud-based measuring method that automatically evaluates the accuracy and fluency of read-aloud performance. General linguistic features, such as sentence and word length, were extracted using natural language processing tools. Phonetic features, including pitch and formant distributions, were extracted using an acoustic-phonetic processing tool, Praat. This study establishes the validity of the automatic evaluation of reading aloud accuracy. Experimental results confirm the effectiveness of automatic evaluation methods, suggesting the suitability of a reading aloud-based L2 proficiency measuring method for short-term ESL proficiency assessments.

Keywords:

Second language proficiency, reading aloud, acoustic-phonetic processing.