Research on non-native speech perception often highlights how native-language phonological categories shape the perception of non-native sounds, while studies of lexical recognition emphasize factors such as lexical frequency and phonological neighborhood density—typically overlooking phoneme-level confusability. This study bridges these approaches by examining how native Japanese listeners recognize English words varying in lexical frequency and neighborhood density under quiet and speech-shaped noise conditions (SNR +6 dB). On each trial, participants heard a word and typed what they perceived. A two-tailed t-test showed that high-frequency words were recognized significantly more accurately than low-frequency words in both conditions (67.7% vs. 22.5%, p .001). Overall accuracy declined in noise (48.3% vs. 38.1%). Noise more significantly affected coda consonants in /CVC/ words than onsets or vowels. Fricatives (/f, ʃ, θ, ð/) had low identification rates. Noise also significantly impaired recognition of initial consonants in onset clusters. No effect of phonological neighborhood density was found, possibly due to limited English vocabulary among participants. Lexical frequency emerged as the most robust predictor of recognition accuracy. Reduced performance with consonant clusters and codas likely reflects the influence of Japanese phonotactic constraints, which limit such structures in the native language.
Nozawa et al. (Wed,) studied this question.