Language users and learners are sensitive to distributional information in their environment, which enables them to extract regularities that occur in the language input that they are exposed to. This process is referred to as statistical learning. While the statistical learning phonotactic literature thoroughly investigates the learning of overall phonotactics in specific languages, little is known about cases where different phonological systems coexist within a single language. The Japanese lexicon is generally classified into four lexical strata according to the etymological status of each word (Itô & Mester, 1995, 1999, 2001). Although each stratum includes the internal phonological similarity in the Japanese language as a whole, there are also distinctive phonological properties. A recent study suggests that language users should be able to learn phonotactics of each sublexicon based on the same kind of statistical probabilities that computers analyse from language users’ accumulated lexicons (Morita, 2018). This thesis examines whether second-language (L2) learners can learn the loanword phonotactics/phonology of Japanese through experience of using and/or passive exposure to Japanese lexical stratification. Using two loanword phonological regularities (categorical and gradient rules) as a case study, two fully-crossed perceptual experiments involving English- speaking learners of Japanese, native speakers of Japanese, and English-speaking monolinguals are presented.

The first experiment explores listeners’ phonotactic/phonological knowledge of nativised loanwords in Japanese using a well-formedness task which shows the adaptation of English final consonants in monosyllabic words. Listeners judge whether the pronunciation they hear is how the word would be pronounced if it was a Japanese word, rating how confident they are on a scale of 1-5. This study shows that L2 learners learn categorical rules, but not gradient patterns. This study also confirms that loanword phonotactics and overall phonotactics make separate contributions to perceived well-formedness. L2 learners access and make use of the sublexicon-specific probabilities of Japanese during the task. The second perceptual experiment is designed to support the findings in the first experiment, by testing for discrimination of non-native consonantal contrasts. Even under high memory demand, L2 learners show the ability to discriminate non-native consonantal contrasts (i.e., CVCV/CVCCV) effectively enough to support findings in the first experiment.

These results suggest that L2 learners can implicitly detect the statistical structure of a language’s sublexicon phonology over the course of acquiring a natural language. However, while native speakers of Japanese learn a gradient rule, L2 learners of Japanese do not. A potential explanation for the differences in gradient rule learning is that the vocabulary size of the target language might play a crucial role. This remains an open question.

In addition, the present work provides a basis for future investigation into whether L2 learners of Japanese, whose native language is other than English, are able to learn Japanese loanword phonotactics/phonology. L1 English-L2 Japanese speakers might gain advantage in perceiving the English input which inevitably overlaps with the phonological form of the host language.