Modern reconstructions of Old Japanese: Miyake and Frellesvig

After writing all those posts summarizing some of Miyake and Frellesvig’s research, I figured I was missing their actual reconstructions! Without further ado, Old Japanese reconstructed phonetics:

Phoneme Miyake Frellesvig
Initial Medial
[p] [p],[ɸ] [b],[β]
t [t]² [t]² [d]²
k [k] [k],[x] [ɡ],[ɣ]
w [w]³ [w] [w]
y [j]³ [j] [j]
s [s]⁴ [s],[ts]⁵ [z],[dz]⁵
z [nz]⁴ [nz],[ndz]⁵
b [mb]⁶ [mb],[mβ]
d [nd] [nd]
g [ŋɡ] [ŋɡ],[ŋɣ]
ǐ [i], [ji] [ji]
ï [ɨ],[ɨj] [wi],[uj]
u [u]⁷ [u]⁷
ě [e],[je] [je]
ë [əj] [ej],[aj]
ǒ [o] [wo]
ö [ə] [o]
  1. /p/ corresponds to modern /h/.
  2. Unlike modern J., /t/ and /d/ as [t] and [nd] even before /i/ and /u/.
  3. /w/ was a glide realized not only before /a/, but also /i/, /e/, /o/; /y/, before /a/, /u/, /e/, /o/.
  4. Miyake allows [s] and /z/=[nz] might have been palatalized to [ɕ] and [nʑ] before /e/ and /i/ (not just /i/ as in modern Tokyo Japanese).
  5. The status of /s/ and /z/ as sibilants ([s],[z]) or affricates ([ts],[dz]) is still disputed; Miyake proposes sibilants, Frellesvig variation.
  6. Miyake says the prenasalized consonants ([nd], [nz], [mb], [ŋɡ]) could have been reduced ([d],[z],[b],[ɡ]) by nasalizing the preceding vowel. Frellesvig believes all such vowels were nasalized.
  7. [u] proper, not [ɯ] as in modern J.

Each row lists an Old Japanese (OJ) phoneme, followed by Miyake and Frellesvig’s reconstructions. I’m not putting asterisks all over the place; everything here is tentative. Vowels with kan/otsu distinctions are noted in modified Mathias-Miller style (inverted circumflex = kan, umlaut = otsu). The phonetic symbols are standard IPA; readers unfamiliar with them are pointed to Wikipedia, where each symbol has its own page with descriptions and recordings. Phonemes not listed are assumed to be the same as in modern standard Japanese (MSJ). For consistency, I substituted Miyake’s use of “y” to denote a palatal glide for IPA [j], as well as Frellesvig’s non-syllabic diacritics to [w] and [j].

Many OJ words don’t observe the kan/otsu distinction (i.e. they use A-grade and B-grade phonograms in free variation); both Miyake and Frellesvig conclude that these neutral (or “C-grade”) rhymes were realized freely as either A- or B-grade (e.g. o could be pronounced like ǒ or like ö).

Of special note is Frellesvig’s description of obstruents. Japanese obstruents can be divided in two classes, traditionally called 清音 sei-on “clear sounds” and 濁音 daku-on “muddy sounds”, which in modern Japanese correspond to unvoiced and voiced, respectively. It is generally accepted that daku-on were prenasalized until quite late, and in agreement with this, Miyake’s reconstruction also has it that “voiced obstruents were prenasalized”. Frellesvig also agrees with this consensus, but goes further: he believes prenasalization, not voicing, was the distinctive feature of daku-on in OJ, because sei-on could also be voiced on occasion—in fact, whenever they were in the middle of a word (word-medial). In other words, sei-on were voiceless initially, and voiced medially, while daku-on were always voiced and prenasalized. Therefore he doesn’t translate sei-on and daku-on as “voiced” and “voiceless”, but as “tenues” and “mediæ”:

Modern Japanese OJ
Initial Medial
Tenuis = sei-on /h/ = [h],[ɸ],[ç] [p],[ɸ] [b],[β]
/t/ = [t],[tɕ],[ts] [t] [d]
/k/ [k],[x] [ɡ],[ɣ]
/s/ = [s], [ɕ] [s],[ts] [z],[dz]
Media = daku-on /b/ [mb],[mβ]
/d/ = [d],[dʑ],[dz] [nd]
/g/ = [g],[ŋ] [nɡ],[nɣ]
/z/ = [z], [ʑ] [nz],[ndz]

His contrastive examples:

/hata/ “flag” [pada]
/hada/ “skin” [pãnda]
/hana/ “flower” [pãna]
/tanabata/ “Vega” [tãnãmbada]

The hypothesis of medial voicing came from Wenck 1959. It’s supported by evidence from surviving dialects and from historical sound changes, including onbin.

I believe that, as of 2012, these reconstructions are the most respected among those of Western scholars, and have largely displaced earlier reconstructions with “eight-vowel systems” (based on central vs. peripheral vowels without glides). Miyake’s research was published in 2003, and it’s hard not to be impressed by the throughness and care in the selection and analysis of data. One interesting feature of his reconstruction is that he quantified phonogram use statistically; frequent phonograms were considered safer data than those that only appear once or twice. Frellesvig’s book is from 2010; he considers his conclusions “in overall compatible” with Miyake’s, proposing his description as narrower; he also praises Miyake’s work as “an extremely useful and well-documented study with an impressive command of all the relevant materials”.

And yet, I’d be wary of taking any reconstruction as gospel. Who knows what kind of novel discovery or insight will be made in the future.


  • Bjarke Frellesvig, A History of the Japanese Language, and apud:
    • Günther Wenck, 1959. Japanische Phonetik, vol. IV. Wiesbaden: Otto Harrassowitz.
  • Marc Hideo Miyake, Old Japanese: A Phonetic Reconstruction.

One thought on “Modern reconstructions of Old Japanese: Miyake and Frellesvig

Leave a Reply

Your email address will not be published. Required fields are marked *