Can we use rendaku for phonological argumentation?

This paper addresses the general issue of the quality of phonological data, using rendaku as a case study. Rendaku is a widely-discussed voicing process that accompanies compound formation in Japanese. The issue of the quality of phonological data has been discussed from time to time from various perspectives throughout the history of phonological research, and we are recently witnessing renewed interest in this topic. This paper takes up rendaku as a case study to address this issue. Rendaku has been used to argue for many theoretical devices, but it is rarely acknowledged in the theoretical literature that rendaku involves extensive lexical irregularity. The specific question addressed in this study is whether it is appropriate to use rendaku for phonological argumentation. The answer that this paper proposes is yes, but the more crucial lesson is that theoretical phonology should evaluate the quality of phonological data more explicitly. The current discussion offers a first step toward establishing general guidelines about what kind of evidence can be used to decide whether a pattern under question is phonological or not.