Linguistic ghosts of Sino-Japanese: go-on, kan-on, tôsô-on and other misnomers

Sometimes people ask from what variety of Chinese did each of the traditional kanji “readings” came. Short answer, according to Miyake:

  • Late Old Chinese, Early Middle Chinese probably through Sino-Paekche → go-on;
  • Chang’an Late Middle Chinese → kan-on;
  • Song/Yuan Late Middle Chinese onwards → tôsô-on.

Long answer: it’s complicated.

Chinese writing and Japanese readers

First, let’s recall some basic facts. The Chinese writing system (hànzì) is morphosyllabic—each character represents a syllable, and almost always also a morpheme—with both uses tailored specifically for the Chinese languages; they weave a complex network of phonetic and semantic hints that only really make sense in a Chinese context. It is singularly ill-suited to write Japanese, a completely different language in all possible regards; but it’s the first writing system that the Japanese learned (via Koreans), and during most of history the Chinese civilization had such enormous prestige that other systems they learned never really caught on.

So the problem the Japanese faced was how to use this writing if it was not adequate for their own language. Their first and most straightforward solution was to simply learn Chinese, and indeed until very recently the Chinese language was a major literary and utilitarian medium—but only written Classical Chinese; spoken Chinese was (probably) never a major established language in Japan, except for specialists. This is an important point! We might reasonably expect that the Japanese would make changes to Chinese writing so that it could better represent Japanese, and that did in fact happen; but we might be surprised to learn the extent that Japanese language itself changed in reaction to Chinese, and furthermore, that the influence happened basically through the written medium (contradicting somewhat the tendency of modern linguistics to dismiss writing as of secondary importance). It’s as if the language adapted so that it could be better written with hànzì. Thus Fischer goes as far as claiming Japan is essentially a nation of readers—that “Japanese culture” as we know it arose as a direct consequence of large-scale reading. He might be exaggerating, but you get the point.

Totally available

The influence of Chinese writing on Japanese worked on all levels—phonetic, morphosyntactic, even pragmatic (e.g. it seems that the native morpheme iro “color” came to have sexual connotations (as in iro-otoko color-man = “lady-killer, sexy male”) because it was equated with Chinese 色 “color”, which had such implications in Chinese culture). But the biggest influence, by far, was in the lexicon. To call this “borrowing” is to underestimate its scope; up to 60% of a typical Japanese dictionary is comprised of Chinese “borrowings”, or of natively-coined composite words made of Chinese-borrowed morphemes (though a typical utterance will have a much smaller proportion; a study found NHK broadcasts to hit 18%, and NHK-speak is more formal—therefore sinified—than everyday speech). The mechanism operating here is that of “total availability”: a writer could, in principle, graft any Chinese word—hànzì and all—into his Japanese-language text, and the use of Chinese terms for something that could very well be expressed in native Japanese wasn’t frowned upon; rather, it would only sound more erudite and prestigious. What’s more, the typical literate reader could read Chinese anyway, so there was no reason not to use it (in later periods such as Edo, when literacy expanded to a wider public, liberal use of furigana “subtitles” helped in decoding texts, so that they kept borrowing unusual hànzì freely and playfully). I believe the morphemically segmented nature of Chinese writing must have made it natural to think of the borrowings in terms of morphemes, and to recombine them in novel ways—or even to mix them with Japanese morphemes in composite words.

The Japanese weren’t alone in this interesting situation. Parallel processes happened with two other linguistic groups that were in a similar cultural position regarding China: the ancestors of what we now call Korea and Vietnam. Martin coined the word “Sino-Xenic” to denote the “vast bodies of borrowed language forms” from Chinese into these three languages.

Since the main mechanism of borrowing was classical-text reading followed by written imitation, we could wonder if the Sino-Japanese (SJ) vocabulary might not be mostly restricted to writing and never actually come alive in speech (like e.g. the Latin abbreviations that I have a bad habit of sprinkling in English text). And indeed, SJ reached such a level of homophony that too much of it would make speech unintelligible (see below). But an impressively large number of words gradually came to be used in conversation (—and to be internalized by new speakers as part of their native language), which raises an important question: how did the Japanese pronounced the foreign words? In fact the established reading conventions were quite regular, even artificial; the Japanese (and Koreans and Vietnameses) sent scholars to China to study their language and philology, and these scholars apparently devised rational correspondences between the Chinese initials and finals (as described in fǎnqiè dictionaries, rhyme tables and such) and native phonemes. Therefore, one can think of Sino-Xenic systems like go-on as many-to-one functions designed to map Classical Chinese readings to nativized phonemic representations. The rationality of such systems is complicated by three things: there’s more than one of them for each language; they each were subject to historical change specific to their mother language, distinct from further changes in Chinese; and there was no one model “Chinese”, to begin with.

Sino-Xenic are like onions

Up until now we have been pretending Chinese is a single static reference point, but in fact the Chinese themselves were already in a state of diglossia: the written Classical Chinese had fossilized in a language of learning, like Classical Latin, Classical Arabic, or Quenya, while its spoken counterpart changed significantly and diverged into many varieties. Even among Sinitic speakers, each linguistic community would have different traditions for how to pronounce the classical works, and it’s on these conflicting traditions (probably even some Korean ones) that the Japanese based their own, which would later alter their very language. The processes went something like this:

  1. The Japanese learn “how to pronounce Chinese texts” from some source;
  2. These pronunciations are consciously adapted into Sino-Japanese readings (on-yomi) with regular phonetic correspondences;
  3. The resulting words, morphemes, and even some new phonetic features of this reading tradition are incorporated in the Japanese spoken language, changing it significantly;
  4. As the Japanese language is subjected to natural changes, all the on-yomi “readings” change with it;
  5. While 3–4 are still ongoing, another iteration starts at 1.

The Japanese have traditionally classified their Sino-Japanese readings into a number of strata: go-on “Wu sounds”, kan-on “Han sounds”, and tôsô-on “Tang and Song sounds”. But one should never lose sight of the evolution of Sino-Japanese forms themselves, together with the Japanese language. So when we think of the relationship of, say, go-on readings with Chinese, we have to think not only which Chinese, but which go-on reading; early go-on and kan-on (as suggested e.g. by the historical kana spelling (rekishiteki kanazukai)) had a number of Sinitic features (such as glides) that were lost later and are not reflected in their current counterparts. At first they even had available the Old Japanese (OJ) so-called“eight vowels” system, and presumably must have employed the extra rhymes to try to represent different Chinese finals. The most curious of things is that some of the current go-on and kan-on readings don’t even exist. What? Let’s look more closely at the categories:

Very early borrowings

Some words now thought to be Chinese borrowings are not even perceived as Sino-Japanese by speakers, and usually aren’t regarded as such; accordingly, there’s no traditional category in which to file modern uma “horse” or ume “plum”, which were believed to be kun’yomi or “native readings” (translations assigned to characters), even though they’re likely of Sinitic origin. Miyake calls these “pre–Sino-Japanese loans”.


Miyake believes go-on was derived from Late Old Chinese (LOC) in the earliest and Early Middle Chinese (EMC) in the newer layers, probably as filtered through Sino-Paekche. This corresponds roughly to the Six Dynasties period.

The traditional term is not necessarily evidence that the origin of these readings was the Wu region in the south; “Wu sounds” was a disparaging Chinese expression for all non-Qin pronunciations by advocates of the new Chang’an standard.

Kan-on and the Reading Wars

With the unification of China by the Sui and the subsequent Tang dinasty centered in Chang’an, Chinese sociolinguistics changed progressively. At first the old standard of Wu was perfectly acceptable, but in time the northwestern dialect of Chang’an (Chang’an Late Middle Chinese) rose to prominence and the old kind of recitation came to be considered “provincial and substandard” (Pulleyblank). The Yamato envoys of the Tang era learned this and hurried back to correct the “wrong” pronunciation of their fellows.

Surprisingly enough, they suceeded; the new pronunciations became the de facto standard to this day. We have historical evidence of the battle, such as governmental edicts mandating Buddhist clergy to adapt to the new sounds—which, oddly enough, Yamato scholars called Han sounds (kan-on) instead of “Qing sounds” as was their name in China (which would be sin-in in kan-on). Decrees or not, Buddhism was the one major vocabulary area where kan-on didn’t caught. As Miller (p. 104) memorably puts it:

The good pious souls who had been taught to trust in the future bliss of issai shujô “all sentient beings” in gokuraku “paradise” and their salvation from jigoku “hell” through faith in the jihi mugen “unlimited compassion” of the Bodhisattva, and all this in go-on, could now hardly be expected almost overnight to shift both their faith and their pronunciation to the issetsu shûsei, kyokugaku, chigyoku, and shihi bukan—the kan-on equivalents of the terms above. It had been hard enough for the Japanese converts to Buddhism to learn what all this sort of thing was about in the first place, without now having to learn to say it all differently.


“Japanese sounds”. These are go-on vocabulary items (other than Buddhist terms) that were already so naturalized that they resisted kan-on assimilation; e.g. niku “meat” or netsu “fever”, which in kan-on would be jiku and zetsu. They’re simply an older layer of go-on, and were formed in the same manner. Advocates of kan-on might have used this term, “Japanese sounds”, for go-on in general, in order to bring more prestige to the “Chinese sounds” they brought from the continent.

New kan-on (shin kan-on)

A term coined by Tôdô Akiyasu to design late Tang and Northern Song readings that were brought by scholars even after the Heian court had officially closed relations with the Tang. These words (like hosa for go-onbosatsu”) never became popular and didn’t survive for long.


“Sounds of Tang” (tô-on) and of Song (sô-on); an umbrella category for any reading imported after shin kan-on, the most important group of which was the specialized vocabulary of the new Zen sect. These readings never made any major inroads in mainstream Japanese, though a few words managed to get in (e.g. 椅子 “chair” as tôsô-on “isu” and not go- or kan-on “*isi → *ishi”).

Notice that by this time the Tang dinasty had already fallen, so that the term tôsô-on is misleading; its fundamental sources are late Northern Song and early Yuan. Also, go-on probably wasn’t Wu, kan-on wasn’t Han, and wa-on wasn’t Japanese :)

Fake readings

In current Japanese practice, every character is assumed to have a kan-on and at least one go-on reading. That’s a bit funny, because many go-on readings were lost after the victory of kan-on. Even kan-on lost some readings to history, and had others that didn’t come to Japan at all; but a typical kanwa jiten (character dictionary) will have kan-on readings for everything. What happens is that modern Japanese lexicographers developed a practice of inferring the missing readings, extrapolating them from analogy with other characters and from fǎnqiè. This wouldn’t be that bad if it wasn’t for the fact that they typically don’t mark the unattested creations at all, so that the reader has no way of knowing whether a given reading is real or made up; Miyake says that “I have yet to see two SJ dictionaries whose go-on readings match completely”.

Miller (again, memorably) called such coinages “linguistic ghosts”, “generated on the basis of readings for other characters, and so on [no pun intended?] in a vicious circle”. Miyake calls them pseudo–go-on and pseudo–kan-on.

What this all looked like

I don’t have sources at hand about what kind of transformations the various Sinoxenic “functions” applied to their sources—I don’t even know if this is reconstructed with certainty; the ghost mismatches suggests not. Here are a couple random examples from Miyake and Miller:

  • Middle Chinese (MC) finals -p, -t, -k have individual representations in go- and kan-on, but in tôsô-on they all become a final -tu, following their change to a glottal stop in Chinese languages. This -tu isn’t intended to represent an actual [tu], but it’s the ortographic indicator of germination (then an innovation in Japanese). So, the kan-on words tiku “bamboo” and seki “stone” both become situ (modern shitsu) in tôsô-on; used as prefixes, they become sippei (=situpei) “bamboo staff to hit distracted Zen acolytes” and sikkui (=situkui) “mortar, stucco”.
  • Modern kan-on /oo/ corresponds to all of Late Middle Chinese (LMC) rhymes *-əw, *-aw, *-aaw, *-aŋ, *-aaŋ, *-aayŋ, *-aawŋ, *-əwŋ, *-ap, and *-aap. It gets a tad little better if we use the historical kana, which for what is now /oo/ had three different forms ou, au, and ahu, each corresponding to a distinct set of LMC finals; so that au corresponds “only” to the six finals *-aw, *-aaw, *-aŋ, *-aaŋ, *-aayŋ, *-aawŋ.
  • Infamously, kan-on s- correspond to all Chinese dental, retroflex, and palatal fricative and affricate initials; go-on, not much better, represents all that with s- and z-.

Miyake (p. 103) gives the structure of SJ syllables as:


  • C₁: initial
  • G: glide
  • V: vowel
  • C₂: SJ counterpart of MC consonant, being one of:
    • (a) the mora nasal n
    • (b) the vowels {i, u}, forming diphtongs like [ai], [ui], [au] (→ modern [oo]) or [eu] (→[yoo])
    • (c) one of the CV sequences {hu, ti/tu, ki/ku}, or mu = [-m]

(MC *-m and *-n had distinct OJ equivalents, but they merged into the moraic /N/ by late Heian; prior to the merge, MC *-m was written as “mu”.)

Next week I’ll blog on what Miyake calls “systems of sinographic readings” and the phonogram-based transcription of Old Japanese, which should also illuminate a few facets of the Sino-Japanese phenomenon.

The Chinese words in this post are reconstructions based on Sarostin and Pulleyblank’s, as cited by Miyake (who likes them a lot). Keep in mind that this is still a debated area & there are a number of competing theories. Main sources of this post are:

  • Marc Hideo Miyake, Old Japanese: A phonetic reconstruction, chapter 5.
  • Roy Andrew Miller, The Japanese Language, p. 101–112.

11 thoughts on “Linguistic ghosts of Sino-Japanese: go-on, kan-on, tôsô-on and other misnomers

  1. I beg forgiveness of the Linguistic Gods for kind of conflating together morphemes and their phonetic realizations with character readings. I find it hard to keep them clearly distinct in this context; the way the Japanese have traditionally thought and discussed about things like go-on and kan-on are as “readings for kanji”, not “competing strata of Chinese borrowings transformed by conventional phonetic operations followed by historical change”.

  2. «Parallel processes happened with two other linguistic groups that were in a similar cultural position regarding China: the ancestors of what we now call Korea and Vietnam.»

    Parallel processes happened with the ancestor of Hokkien too, among others.

  3. @番 : Thanks for your comment!

    You’re right, the influence of the Middle Kingdom must have spread all through East Asia and beyond, changing vocabularies and phonologies all over.

    But there’s something that puts Korean, Japanese and Vietnamese in a particular subgroup. It’s the conjunction of two facts: first, they’re not Sinitic languages at all, but are completely different kinds of language; and, second, the Chinese influence there was mostly through writing, not speech. (The Japanese call this kanji bunkaken, the hànzì cultural sphere.) It’s this written transmission to incompatible languages that generated the “systems of sinographic reading” of Sino-Xenic, and their productive, combinatorial morphemes. I think a large part of this derives from the practice of interlinear glossing (Japanese kun’doku, Korean hundok).

    I’m not at all knowledgeable about Sinitic languages, but since Hokkien is in the same family as Mandarin and friends, and furthermore they’re geographic neighbors, we can expect oral transmission of whole words (and, due to being from the same family, contamination of semantics in cognates). Even Mongolian, a non-Sinitic language, mainly got whole words, because of oral transmission (cf. Greg Pingle’s answers here).

  4. I know (and knew) what You mean, sir. But that’s exactly what I’m trying to say. The sinicization of the forerunner of Hokkien-Teochew took place much the same way as did the sinicization of the forerunner of Vietnamese, if we go at least as far back as, say, 1100 AD. It just went farther and deeper in the «end». I recommend John Phan’s work on the history of the Vietnamese language, among others. I’m not good with citations.

    «But there’s something that puts Korean, Japanese and Vietnamese in a particular subgroup. It’s the conjunction of two facts: first, they’re not Sinitic languages at all, but are completely different kinds of language;»

    Typologically, Korean and Japanese and co. is way different from the mainstream of the E/SE Asian main. No doubt about that.

    But Hokkien-Teochew and Vietnamese is far from being «completely different kinds of language». To this I’d add Cantonese and what some call «Northern Zhuang» — two other typologically similar tongues with full-blown «kanji-plus» scripts. The conventional wisdom that «Hokkien-Teochew IS a Sinitic language» is based on assumptions that don’t stand up on closer inspection. (I mean, what IS Sinitic? Isn’t it based on political definitions, when we get down to the bottom of it? Diachronically, how is Hokkien-Teochew Sinitic but not Vietnamese? Again, I’m leaving Korean and Japanese out of this.)

    «I’m not at all knowledgeable about Sinitic languages, but since Hokkien is in the same family as Mandarin and friends, and furthermore they’re geographic neighbors,»

    Geography’s got a lot to do with it, in fact. Your assessment of «Chinese» (however defined) is more or less true AFAIK with regards to the languages spoken by most Han (again, a politically defined tribe), including throughout the Yangtze lowlands (You could call these the «and friends» of Mandarin). Kind of counterintuitively, though (at least for me, at the outset), the hilly seaboard from about 台州 on down was only periodically intimate with the heartland of China, giving rise to historico-linguistic phenomena much like or just like what we see in Vietnamese. Excluding the mountains themselves, the Hokkien-Teochew region happened to be «the place of maximum geographic isolation from the heartland of China» along this seaboard (not b/c of distance, but b/c of terrain, and heartlanders’ aversion to saltwater); the West River Valley and upper Pearl Delta happened to be «the zone of most consistent contact with the heartland of China». (Hence the irony of Canton City Cantonese being thought of as the quintessential Deep South China tongue. In fact it’s on the short list of «most Mandarized» south-of-台州 seaboard tongues.)

  5. That’s all news to me. Do you mean that Hokkien had little contact with the “mainstream” Sinitic languages except through writing, and that they acquired individual morphemes primarily indirectly, through hànzì? Or do you mean that Vietnamese was actually influenced by spoken contact, putting it in a class like Hokkien and Mongolian and unlike Japanese and Korean?

    As for how is it that Hokkien is “Sinitic” but Vietnamese isn’t: I thought Vietnamese had been established as an Austroasiatic language (i.e. as a cousin to Mon and Cambodian Khmer) by traditional historico-comparative methods (for example, like this); i.e. that the linguistic consensus was that it’s not a relative to Mandarin—they don’t descend from a common ancestor; whereas Hokkien was established, through the same methods from historical linguistics, to have a common ancestor with Mandarin, Cantonese, Wu etc. In other words, Hokkien and Mandarin both descend from Proto-Sino-Tibetan, whereas Vietnamese and Khmer descend from Proto-Austroasiatic.

    But, again, I just trusted what I thought to be the sinological consensus on this.

    I took only a cursory look on John Phan’s dissertation preview, Lacquered Words, but I see that he accepts Vietnamese as “unquestionably” Austroasiatic and “completely unrelated to Sinitic” (p. 11–12). But he also claims that Vietnamese was unlike Japanese and Korean in that Vietnamese speakers actually knew spoken Sinitic too –“unlike Sino-Korean or Sino-Japanese, Late Sino-Vietic resulted from a bilingualism in Sinitic and Vietic languages that flourished in the area of northern Vietnam throughout the Tang dynasty” (p. 10). I find that thesis to be deeply interesting (it goes directly against Miyake’s lumping of Vietnamese with Japanese/Korean as “written transmission” Sino-Xenic), and I’ll certainly read his dissertation when it’s available. Thanks for the recommendation.

  6. Better late than never, I figure!

    «As for how is it that Hokkien is “Sinitic” but Vietnamese isn’t: I thought Vietnamese had been established as an Austroasiatic language (i.e. as a cousin to Mon and Cambodian Khmer) by traditional historico-comparative methods (for example, like this); i.e. that the linguistic consensus was that…»

    True, linguists tend to “agree” that Vietnamese is Austroasiatic and Hokkien is Sino-Tibetan. Going by the tree model, this makes sense. The problem is over-application of the tree model. Linguistics is stuck in the 19th century in this regard. Subhash Kak’s “On the Classification of Indic Languages” (available online as a free PDF) is on point as to this problem in general. On the Sinitic side, Mantaro Hashimoto ideated as far back as the 70s that the southern Sinitic languages were non-Sino tongues that Sinicized slowly over time. Contrast this with the “orthodox” belief that Sinitic languages have been DIverging. Pan Wuyun (潘悟云) — maybe among others by now, I don’t know since I don’t “command” the literature well — also seems to believe something similar. For instance, he seems to believe (I just found the paper, still need to read it) that Hailamese (a sister language of Hokkien) was an “inter-language” (中介語) arising when the pre-Sino dwellers of Hailam tried to learn the “Hokloid” language of the incoming Sino settlers. This inter-language then replaced the incoming Hokloid language itself locally. This is interesting to compare with what Phan thinks happened maybe a few centuries earlier just a short ways west in what’s today northern Vietnam.

    At the very least these theories make us ask this question: with all the fine-grained understanding of language contact that linguists now have from studying the last few hundred years of language contact, why do we still almost exclusively use the tree model in long-range historical linguistics?

    Tree model is a nice little toolbox, but it doesn’t apply well when languages are in constant contact over long periods of time. I think R.M.W. Dixon also had a lot to say about this.

    Our over-applied all-tree-model-all-the-time type of scholarship has its roots in the 19th century along with other “patrilineal” isms and ologies, such as the orthodox Darwinist view of genetics…

    Don’t get me wrong: in terms of typology and vocabulary, there’s (clearly?) a “Sinitic” type of language. Modern Hokkien clearly fits that mold much better than Vietnamese. But looking at just the last 450 years of written and concretely attested spoken Hokkien, it’s clear that Hokkien has been converging to a Sinitic mold spearheaded by modern and late imperial Mandarin. Is it possible that this is just some kind of anomaly, and Hokkien was actually diverging from Sinitic before that? We have to believe that, if we want to believe the “tree model consensus” on Hokkien. What seems more likely to me is that the “Sinitic” package of core vocabulary and typology is more of a cultural or areal thing than a “genetic” thing.

    What would Vietnamese look like today if Vietnam had been under Chinese administration for the last half millennium? Would it still have “Southeast Asian-type” pseudo-pronominal system? And, is core Swadesh vocabulary really off limits to borrowing? If not (and apparently it isn’t), what would it mean for the tree model if a non-Sino language could “become Sino” through borrowing?

    My two yen.

    «But, again, I just trusted what I thought to be the sinological consensus on this.»

    Careful with that. John Cikoski roasted Sinology-as-we-know-it nicely in the long preface to NOTES FOR A LEXICAL OF CLASSICAL CHINESE (available online as a free PDF).

  7. Just came across this conversation (and thanks for the interesting article on Onyomi – I was not aware that many of them were reconstructed or “fake”, … there goes some of my sound correspondences research, sniff …)

    About the relationship of Hokkien to the relatives of Mandarin: a straight tree model is indeed too simplistic, but does anybody really propose this? The way I see Hokkien, we have:

    (1) A base stratum of words for which no cognates with Mandarin (or Wenyanwen) can be found: kin-á (child), báh (meat), tsa-bó (girl), beh (to want), suí (beautiful), …

    (2) A very old layer of Han words showing archaic features: soã (mountain), óh (to learn), tshiáh (to eat), …

    (3) A more modern layer of pronunciations, imported from the North much later and producing a set of literary readings for almost every characters: san (mountain), hók (to learn), sít (to eat/food)

    Add to that a sprinkle of alternative colloquial and literary pronunciations and you get something far more complex than the “tree model”. With that being said, though, the everyday language is much better described as a variant of archaic Han with a non-Han substrate and a literary Han layer on top than a non-Han language which just happens to be using 90% Han vocabulary and grammar :-D. My 5c.

Leave a Reply

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