Identity Avoidance and Lyman's Law

Rendaku is a morphophonological process whereby the initial consonant of the second member of compounds becomes voiced. One famous factor that blocks rendaku is Lyman's Law: when a second element already contains a voiced obstruent, rendaku is blocked. This blockage of rendaku due to Lyman's Law is almost exception-less in the contemporary Japanese, and thus has been treated as if it applies uniformly to all forms that fit the structural description. However, our current experiment shows that this uniformity assumption does not hold. Concretely, the experiment reveals a hitherto unnoticed generalization: among those structures that violate Lyman's Law, there are some that are more disfavored than others. More specifically, Japanese speakers disfavor structures with two adjacent identical CV moras with a voiced obstruent onset (e.g. /. dadana/) more than structures that merely contain two voiced obstruent onsets (e.g. /. do. gara/). In addition to this new descriptive discovery in Japanese phonology, this paper makes three contributions to general linguistic theory: (i) the importance of experimentation in linguistic research; (ii) the role of grammar that cannot be deduced from lexical patterns; (iii) parametrization of the locality of dissimilatory effects within a single language.