If <i>rendaku</i> isn’t a rule, what in the world is it?
The morphophonemic voicing phenomenon in Japanese known as <i>rendaku</i> is highly irregular, but several factors are believed to make <i>rendaku</i> more or less likely. This paper reviews some experiments intended to test the psychological reality of three such factors: Lyman’s Law, the semantic relationship between the two elements in noun + verb compound nouns, and salient semantic or phonological resemblances between novel compounds and existing compounds. The evidence suggests that each of these factors has at least a detectable effect on responses in experimental situations. Any realistic overall account of <i>rendaku</i> will have to incorporate a signiﬁcant degree of intractable irregularity, but it will also have to be consistent with the intuition of naïve native speakers that <i>rendaku</i> is predictable.