A corpus-based computational analysis of high-front and -back vowel production of L1-Japanese learners of English and L1-English speakers

Martin Schweinberger, Yuki Komiya

Friday, December 16th, 2022, 4.15pm – 4.45pm


This study combines acoustic phonetics with computational and applied corpus linguistics to analyze and compare the production of the monophthongal vowels /ɪ/, /iː/, /ʊ/, and /uː/ in the speech of 148 L1-Japanese learners (JPN) and 107 L1-speakers of English (ENS) based on The International Corpus Network of Asian Learners of English (ICNALE). The study aims to ascertain if JPN merge spectrally close vowels by calculating Bhattacharya coefficients. In addition, the study uses mixedeffects linear regression to determine if JPN compensate for the potential mergers by exaggerating durational contrasts between spectrally similar vowels. The results of the analysis confirm that JPN exhibit high degrees of overlap for both /ɪ iː/ and /ʊ uː/. Their L1 English peers, however, also exhibit substantive overlap for /ʊ uː/. With respect to duration, the analysis shows that JPN extend the duration of all vowels and exaggerate the difference between /ɪ iː/ and /ʊ uː/ to compensate for the lack of qualitative differences between short and long vowel pairs. This study represents the first corpus based acoustic analysis of JPN vowels in spontaneous speech.