Some more thoughts on the onbin changes

Seems like I'm not quite done with the onbin sound changes yet. As I learn more about it, I start to have more sympathy for the approach that Frellesvig took. Rather than going in detail trying to understand exactly what happened, he simply showed what the resulting phonological system is. The clear disconnect between the written language and the spoken language that happens around middle Japanese times, and the modern dialects being hopelessly mixed indeed makes it rather difficult to say anything definitive about what the "actual" development was.

Let me just add a little something to further mystify us. Frellesvig has an interesting abstraction on page 138 of his 1995 book. He classifies the onbin changes into 4 main groups:

(i) The n-group (only ni and nu syllables)
(ii) The Labial group (pi/u, bi/u, mi/u)
(iii) The g-group (gi/gu)
(iv) The k-group (ki/ku)

ialects can differ on whether the reflexes of any of these groups are consonantal (Q or N) or vocalic (U/I/Ũ/Ĩ) which he abstracts to C and V. He then posits Three varieties (I, II and III)

To the n-group I think we can safely add t and r which like n always have a consonantal outcome.

Frellesvig observes that variety I aligns quite nicely with the Western dialects whereas II and III align mostly with the eastern dialects. In modern day Kyoto the B group is broken up, where m and b have consonantal outcomes while p has a vocalic outcome, but we see in Rodrigues' Arte da Lingoa de Iapam that in the 17th century at least one western dialect still had vocalic outcomes for all of them.

As can be seen the B and G groups are groups that show some amount of disagreement, and in fact the K group also shows disagreement within the Tokyo dialect. As you can see in my previous blog post, both the K and G groups occasionally have consonantal outcomes, e.g. *piki-paru > hipparu 'to pull tight' ('to pull'+ 'to stretch'). I've had trouble finding good examples of consonantal outcomes for G, but kanbashii (besides koobasii) 'fragrant' apparently developed from *kagupasi-.

But the strange thing is, there seems to be a split here. In the verbal morphology, the outcome of K/G is always vocalic, regardless of variety, so *kaki-te > kaite 'writing'; whereas in compounds the K is always consonantal, and G seems generally consonantal.

The reverse is true for the pi/pu syllables: In the verbal system (in the Tokyo dialect, not the Western dialects) the outcome is consonantal: *kapite > katte'buying', but I have been unable to find any examples of pi/pu giving consonantal outcomes anywhere outside of the verbal system, while the vocalic outcome U is rather common, it's especially visible in compounds with the historical *pito 'person':

sira-pito > sirooto 'amateur' (white-person)
kuro-pito > kurooto 'professional (black-person)
ma-pito > mooto 'you (to a lower class person)  (genuine-person)

The same is true of *bi/bu which also mostly is found vocalic in compounds (even though it is consonantal in the Tokyo verb).

aki-bito > akyuudo 'merchant' (trade-person) (besides akindo, one of the few cases I've found with a consonantal realization)
mesi-bito > mesyuudo 'prison' (summon (?)-person)
naka-bito > nakoodo 'intermediary' (middle-person)
waka-bito > wakoodo 'young person' (young-person)
kari-bito > karyuudo 'hunter' (hunt-person)
kuni-bito > kunyuudo 'native' (land-person)
oti-bito > otyuudo 'defeated soldier' (fall-person)
kura-bito > kuroodo 'keeper of imperial archives' (storehouse-person) (besides kurando)

For *mi/mu however, it is especially common to find consonantal reflexes:

yumi-te > yunde 'bow hand'
kami-ta> kanda 'god field'
kami-kaki> kangaki'god fence'
kami-tati> kandachi'upper class'
kami-nusi (or kamu-nusi) > kannusi'Shinto priest' ('god' + 'lord')
kami-tukasa (or kamu-tukasa) > kandukasa 'Person who administered religious rituals within the dazaifu jurisdiction (ritsuryo system)​' (god + official)
pumi-paru> funbaru'to brace one's legs' ('to step' + 'to stretch')
pumi-kiri> fungiri'determination' ('to step' + 'to cut')
pumi-tukeru> fundukeru'to trample on' ('to step' + 'to attach')

In fact, I can think of only two cases of vocalic reflex of *m. First is the place name hyuuga which judging from its spelling, must come from *pi-muka'sun-turning'. As this place is not exactly close to Tokyo, it seems fair to assume that the vocalic reflex is part of the local dialect, but not normal for the dialects that feed into standard Japanese. The second one is less easily explained: *ko-miti > koodi 'small road'.

So for the Tokyo dialect we actually get a rather striking 'mirrored' distribution. The labial stops have vocalic reflexes in derived forms, consonantal reflexes in the verbal forms. And velars have consonantal reflexes in derived forms, but vocalic reflexes in the verbal forms.

How should we explain this? I have no idea, very happy to hear suggestions though.