太田真理、「音韻的・意味的要因が連濁に与える影響：連濁データベースとロジスティック回帰分析を利用した研究」、『音韻研究』、日本音韻論学会、18、85–92、2015

Abstract: In the present study, I examined phonological and semantic factors that predict the occurrence of rendaku, or sequential voicing. I targeted 29,385 compound words with clear rendaku occurrences from the rendaku database. Using logistic regression analyses, I confirmed that phonological and semantic factors proposed in previous studies were actually effective for predicting the rendaku occurrence. Moreover, I specified that a geminate consonant in the second element is an effective factor for predicting rendaku occurrence. The logistic regression model accurately predicted the rendaku occurrence for 80% of the compound words. The present study demonstrates that multivariate analyses are effective tools for understanding complex phonological phenomena such as rendaku.