A probabilistic model of loanword accentuation in Japanese

Abstract This paper presents a probabilistic model of loanword accentuation in Japanese, analysing a corpus of 3,017 English-based loanwords. Through corpus analysis and computational modelling, the study reveals that Japanese loanword accentuation involves two distinct types of faithfulness effects, alongside markedness effects. First, there is a significant influence of the stress patterns of English source words and the epenthetic status of loanword syllables. This challenges the common assumption that accents driven by faithfulness are merely sporadic exceptions, highlighting instead a probabilistic interplay between faithfulness and markedness. Second, this study uncovers faithfulness to Japanese speakers’ implicit knowledge of the English stress system. Rather than merely imitating the stress patterns of individual English words, Japanese speakers develop an internalised theory of the English stress system and mimic what they believe is the correct pronunciation according to their internalised theory.