Abstract
It has been noted that vowel backness is largely preserved in English-to-Mandarin loanwords, while vowel height is not. This asymmetry contradicts the native phonological patterns in that vowel backness for mid vowels is non-phonemic in Mandarin. This study attempts to re-examine this issue by carrying out similarity rating experiment and transcription task for the mapping of English vowels [ɛ, ɚ, ɔ, æ] to Mandarin vowel system. Moreover, the acoustic similarities between the vowels in English and Mandarin are calculated by measuring the Euclidean distances in formant space for each English-Mandarin vowel pair in the stimuli. Further implementations of linear regression analysis revealed that the variance in the rating results is largely explained by the acoustic similarity, supporting the effectiveness of our rating results in the phonetic model. By comparing our experiment results to Lin (2008a)’s loanword corpus study, this study mostly supports the vowel adaptation principle in loanword phonology proposed in Lin (2008a), namely the feature preservation constraint of vowel backness is dominant over that of vowel height in loanword adaptation. However, there are still some discrepancies between our findings and Lin’s corpus study. For instance, [æ] in English is mostly translated to [a] in Mandarin in the corpus, but our experimental results reveal that [ei] in Taiwan Mandarin has the highest perceptual similarity to [æ] in English. This discrepancy may be attributed to the fact that the combinations of C-[ei] sequences is more limited than C-[a] sequences in Mandarin; this distributional asymmetry plays a key role in loanword adaptation but is not a factor in our perceptual experiment. Therefore, the inconsistency between perceptual similarity and loanword phonology may be accounted for by the phonotactic or distributional constraints in the recipient language -- Mandarin. Putting together, the findings of this study supports the hypothesis that the vowel selection in loanword phonology is not determined by a single phonological constraint or solely the acoustic similarity, but by factoring in all the related attributes in different models, including phonetics, phonology, orthography, etc.