Send message to Author

Paper short abstract

This paper re-evaluates Wenck (1959)’s hypothesis of intervocalic obstruent voicing in Old and Early Middle Japanese through comparison with Old Tuscan evidence, suggesting that voicing may be gradient and non-systematic, resulting from a probabilistically conditioned irregular sound change.

Paper long abstract

To make the well-known lenition pathway /-p-/ > /-w-/, traditionally labelled hagyō tenko, phonologically plausible, Wenck (1959: 81) proposed that /p/ was allophonically voiced word-internally in Old and Early Middle Japanese, and that hagyō tenko proceeded through successive weakening stages [b > *bβ > *β > w > ∅], with segmental loss blocked before /a/. Wenck further suggested that voicing extended to other obstruents in intervocalic position, adducing typological parallels from Korean and Northeastern Japanese dialects. From a sound-change perspective, this proposal offers a convincing basis for the lenition processes affecting the syllables involved in the so-called onbin changes. It has been supported, albeit with differing emphases, by Hayata (1977), Takayama (1992), Frellesvig (1991, 1995, 2010), and Hamano (2000), with partial agreement by Unger (2004). Nevertheless, many Japanese scholars continue to interpret hagyō tenko as a process of fricativisation, typically reconstructed as *p > *ɸ > w (Mabuchi 1971; Shibatani 1990, among others). Vovin (2020) likewise rejects systematic intervocalic voicing, appealing primarily to man’yōgana orthography, which does not consistently reflect such voicing. Beyond the orthography, a blanket voicing hypothesis raises two further issues: (i) it does not explain why lenition seems to be arrested in syllables that did not undergo onbin, and (ii) it entails a later fortition reversal, namely intervocalic devoicing in Late Middle Japanese, that is cross-linguistically uncommon and therefore difficult to justify. This paper reassesses these competing accounts in light of Romance historical phonology. In many Romance varieties, Latin intervocalic obstruents underwent voicing and subsequent lenition, yielding trajectories comparable to Wenck’s chain. Crucially, however, Old Tuscan did not develop an exceptionless intervocalic voicing rule; instead, weakening was selective, gradient, and sensitive to consonant type, prosodic position, and lexical diffusion (Canalis 2014, 2015). Treating intervocalic voicing as probabilistic rather than categorical thus offers a comparative model in which non-automatic, non-systematic outcomes may be expected. Applied to Japanese, this perspective supports a scenario in which voicing and lenition spread differentially across consonant types and environments, allowing the same intervocalic segment to lenite in some words but not others, without positing uniform system-wide voicing or an implausible later fortition change.