Introduction

There has been a renewal of interest in the study of loanword phonology since the recent development of constraint-based theories. Such theories readily express target structures and modifications that foreign inputs are subject to (e.g., Paradis and Lebel 1994, Itô and Mester 1995a,b). Depending on how the foreign sounds are modified, we may be able to make inferences about aspects of the speaker's grammar for which the study of the native vocabulary is either inconclusive or uninformative. At the very least we expect foreign words to be modified in accordance with productive phonological processes and constraints (Silverman 1992, Paradis and Lebel 1994). It therefore comes as some surprise when patterns of systematic modification arise for which the rules and constraints of the native system have nothing to say or even worse contradict. I report a number of such ‘emergent’ patterns that appear in our study of the adaptations of French words by speakers of Japanese (Shinohara 1997a,b, 2000). I claim that they pose a learnability problem. My working hypothesis is that these emergent patterns are reflections of Universal Grammar (UG). This is suggested by the fact that the emergent patterns typically correspond to well-established cross-linguistic markedness preferences that are overtly and robustly attested in the synchronic phonologies of numerous other languages. It is therefore natural to suppose that these emergent patterns follow from the default parameter settings or constraint rankings inherited from the initial stages of language acquisition that remain latent in the mature grammar.