Abstract
English and Mandarin Chinese differ in the voice onset times (VOTs) of /b/ and /p/. Hence the way bilinguals perceive these sounds may show 'tuning' to the language-specific acoustic structure of a bilingual's languages (a discrete model), or a shared representation across languages (a unitary model). We investigated whether an individual's early childhood exposure influences their model of phoneme perception across languages, in a large sample of early English-Mandarin bilingual adults in Singapore (N = 66). As preregistered, we mapped identification functions on a /b/-/p/ VOT continuum in each language. Bilingual balance was estimated using principal components analysis and entered into GLMMs of phoneme boundary and slope. VOT boundaries were earlier for English than Mandarin, and bilingual balance predicted the slope of the transition between categories across both languages: Those who heard more English from an earlier age showed steeper category boundaries than those who heard more Mandarin, suggesting early bilinguals may transfer their model for how phonemes differ from their earlier/stronger languages to later/weaker languages. We describe the transfer model of discrete phoneme representations and its implications for use of the phoneme identification task in diverse populations.