ISCA Archive SLaTE 2023
ISCA Archive SLaTE 2023

Learner and Linguistic Factors in Commercial ASR Use for Spoken Language Practice: A Focus on Form

Elizabeth Bear, Stephen Bodnar, Xiaobin Chen

While there has been increased interest in the use of commercial automatic speech recognition (ASR) technology for second/foreign language (L2) practice, few studies have targeted morphosyntactic form. This study evaluates the performance of five major commercial ASRs in regard to learner factors---native language (L1), proficiency, and gender---and linguistic factors---grammaticality and morphosyntactic structure. We analyzed the recordings of 167 participants of 26 L1 backgrounds reading aloud a series of grammatical and ungrammatical sentences of seven structure types. Our results revealed an interaction between L1 and the particular ASR and a three-way interaction between ASR, grammaticality, and structure type. The findings suggest that, while the use of commercial ASRs for spoken L2 practice is promising, certain structure and error types remain challenging for ASR recognition, which may have implications for the targeting of those structures in L2 activity design.