We analyse hybrid compound words in Japanese, where a hybrid compound is one formed from stems that belong to more than one lexical stratum. In native-foreign compounds, where one stem belongs to one of the three native strata: Yamato, Sino Japanese and mimetic, as identified by Itô and Mester (1995), and the other to the foreign stratum, we observe that violations of phonological wellformedness constraints in the foreign stem are significantly less probable than in pure foreign words. These observations are explainable through gradient phonotactic probability, where the probability of a phoneme is determined by the whole sequence of phonemes that precedes it. We shall argue that this observed phonotactic behaviour of hybrid compounds is best explained by the hypothesis that both lexical strata distinctions and phonotactics are graded rather than categorical.
