A Quantitative Study of Voiced Velar Nasalization in Japanese

This paper presents a corpus-based analysis of voiced velar nasalization (VVN) in the standard (Yamanote) dialect of Japanese, in which a voiced velar plosive /g/ becomes nasalized in prosodic-word-medial position. To the best of our knowledge, this is the first quantitative study of the phenomenon, and confirms the impressionistic observation reported in the previous literature; at the same time, our study finds that these generalizations are stochastic. Looking more closely at the determinants of this variation reveals intriguing ways in which phonological grammar and lexicon interact, as well as the role of frequency in shaping phonological variation. Outside of frequency, we also examine factors such as prosodic length, and the effects of the segmental context on VVN.