Contact address:

Nishimura (2003) first pointed out that in Japanese loanwords, voiced geminates devoice optionally when they co-occur with another voiced obstruent i.e. when they violate OCP(voice) (e.g. /baggu/ → [bakku] ‘bag’). This devoicing of geminates has been used to make several theoretical claims in the recent phonological literature. However, these claims have so far largely been based on intuition-based data provided by Nishimura (2003) and Kawahara (2006). Kawahara (in press) addressed this problem by conducting a rating study. The first aim of this study, building on Kawahara (in press), is to further support the empirical foundation of these theoretical claims by way of a large-scale rating study. The current study shows that (i) the OCP and geminacy each affect naturalness rating of devoicing, and (ii) that there is nevertheless something special about the combination of the OCP and geminacy. The second aim is to test an assumption behind the recent literature on this phenomenon. The assumption is that this devoicing pattern is monolithic—i.e. all voiced geminates uniformly undergo devoicing in a certain phonological environment. The current experiment shows that this assumption is too simplistic. In particular it shows (i) speakers rate the devoicing of affricates as natural as that of stops, (ii) speakers find devoicing of items that merge with other lexical items less natural, (iii) speakers rate devoicing as more natural when there are multiple triggers, (iv) speakers find devoicing of [dd] more natural than that of [gg], and (v) speakers find devoicing of more frequent items more natural.