Sino-Japanese n- / d- initial interchange

October 1, 2025 @ 7:39 am · Filed by under Borrowing, Historical linguistics, Phonetics and phonology

« previous post | next post »

In his remarks on "Stay hyDRAEted", Alec Strange noted that you can't avoid reading dorei no remonēdo ドレイのレモネーど  (intended to be "Drae's Lemonade") as "slave lemonade" (dorei / ドレイ / 奴隷 ["slave"]).  Coming at 奴隷 from the Sinitic side, my instinct is to read 奴隷 as beginning with an n- (or in a few cases l-), so it would have nothing to do with "Drae's".

(Wiktionary)

So I started to ask around how is it that Japanese has a d- initial for 奴隷 ("slave") and Sinitic has an n- initial?

David Spafford:

No idea, linguistically, but it’s not an odd one off. 

努力 ("make an effort; strive; endeavor") is pronounced doryoku. [VHM:  not like nǔlì as in Mandarinì

Nathan Hopson:

Do is the standard ondoku 音読* learned in middle school, ヌ is considered hyō-gai 表外 (outside of the table of standard readings).
https://kanji.jitenon.jp/kanjic/1204

*[VHM:  the reading of a Japanese word written in kanji that is roughly based on the pronunciation of the kanji characters in the originating Sinitic topolect at the time the word was introduced into Japanese (Wiktionary)]

VHM:  phonologically speaking, ヌ is pronounced with an n- initial, not a d-

Etymology

Simplified in the Heian period from the man'yōgana kanji 奴.

Pronunciation

Syllable

ヌ • (nu) 

    1. The katakana syllable ヌ (nu). Its equivalent in hiragana is ぬ (nu). It is the twenty-third syllable in the gojūon order; its position is ナ行ウ段 (na-gyō u-dan, “row na, section u”)

(Wiktionary)

N.B.: nǔlì 努力 ("effort"); núlì 奴隸 ("slave").

The following remarks by John Whitman are intended for historical linguists::

The alternation btw d- and n- with 奴 reflects the general alternation between kan’on 漢音 [VHM:  Japanese kanji readings borrowed from Chinese during the Tang dynasty (7th-9th cc.)] and go’on 呉音 [VHM:  the earliest form of on'yomi 音読み (Japanese readings of Chinese characters), preceding the kan'on 漢音 readings; both go'on and kan'on exhibit characteristics of Middle Sinitic (MS) in Sino-Japanese. The kan’on 漢音 for 奴 is do, but the go’on is nu, identical to the usually reconstructed MS nu for 奴. In this case, the go’on reading is relatively unusual on the Japanese side, but it occurs e.g. in the reading 奴婢 (nuhi ぬひ), the category of slaves in the Ritsuryō 律令 Nara Period legal system.

The d-~n- alternation is standard when there is an opposition btw kan’on and go’on readings involving original MS /n/, for example 男性 dansei ‘male’ vs 男体 nantai ‘male body’. The alternation between 女性 zyosei < dyosei ‘female’ vs女体 nyotai ‘female body is the same thing.

This reflects a change in Sinitic, not Japanese. Some northern MS dialects in roughly Tang times depalatalized MS /m/, /n/, /ng/. South Coblin has a detailed study of this, looking not just at the phenomenon in Japanese kan’on but at Tibetan and intra-Sinitic Buddhistic readings. A mystery, unresolved as far as I know, is how this made it into Japanese kan’on 漢音 but not Sino-Korean, which are both held to have been borrowed around the same time, roughly mid-late Tang, perhaps a bit later in the SK case. One possibility is that the denasalizing region included Chang’an; the Koreans were savvy and in constant contact enough with China to understand that the denasalized pronunciation was substandard, even if associated with the capital region, while the Japanese clerics who imported the readings were less with it, or more superliteralist.

In modern Japanese, kan’on readings are vastly less marked, and almost always used in neologized kango 漢音. Go’on readings have a strong association with Buddhism. The 呉 wu2 designation refers most likely not to any region of China (such as Southeastern China/Suzhgou~Shanghai region), but to the Korean peninsula. The kun 訓 vernacular reading of 呉, kure, refers to Korea and is probably the same word as Korean 고려 Koryŏ [koryə] or possibly Kuryŏ [kuryə], what you get if you subtract the flattering 高 from 高句麗, as non-Korean texts often do. This reflects the fact that both Sinography and Buddhism were originally imported to Japan from Paekche.

John added this P.S.:

OK! I think a lot of the people who read your blog – especially the Chinese historical specialists – already know this better than me. But if you get comments, I have other things to say – first regarding the historical Korean sources of go’on, and then regarding an interesting nasal stop denasalization process that is going on right now in contemporary Korean.

I expect that we will receive significant comments from Koreanists.

Meanwhile, every time I see "dorei no remonēdo ドレイのレモネーど", I can't help but think of "dorai no remonēdo ドライのレモネーど" ("dry lemonade"), but that's completely my problem.  Power of reverse suggestion.

 

Selected readings

October 1, 2025 @ 7:39 am · Filed by under Borrowing, Historical linguistics, Phonetics and phonology


20 Comments

  1. Thomas said,

    October 1, 2025 @ 10:09 am

    Let this be the comment that asks to what extend contemporary Korean currently replaces 네 (ne) with denasalized de.

  2. Je said,

    October 1, 2025 @ 10:40 am

    The denasalization of Middle Chinese nasal initials is also common in Hokkien. In this example, we have 奴 lô͘, where Hokkien /l/ is phonetically something in the range of [d~l~ɾ]. Similarly we have
    文 H bûn, J bun (kan) / mon (go) < MC mɨun
    人 H jîn, J jin (kan) / nin (go) < MC ȵin
    業 H gia̍p, J gyō (kan) / gō (go) < MC ŋɨap

    Hokkien (and Min languages in general) is in fact quite similar to Japanese in many other aspects (e.g. the MC retroflexes like /ʈ/ become dentals /t/ in both Hokkien and Japanese, so 程 is H thêng, J tei / jō[<dyau], cf. Cantonese cing4 and Mandarin chéng).

    Coincidentally, in modern Korean, nasal initials /m/ and /n/ are also often denasalized word-initially, becoming [b] and [d]. But there is no merger of voiced plosives (/b-, d-/) and nasal sounds (/m-, n-/), since the voiced initials became devoiced word-initially, so 門 문 mun is [bun], but 分 분 bun is [pun]. The devoiced initials may have some degree of aspiration, but still don't merge with the aspirated initials (/pʰ-, tʰ-/), since the voiced ones (both [devoiced] plosives and [denasalized] sonorants) trigger the low pitch in the syllable, while the voiceless ones trigger the high pitch.

  3. Chris Button said,

    October 1, 2025 @ 11:24 am

    The LMC nasals were prenasalized stops in the standard dialect that was the source of the kan-on readings.

  4. Michael Watts said,

    October 1, 2025 @ 6:04 pm

    The 呉 wu2 designation refers most likely not to any region of China (such as Southeastern China/Suzhou~Shanghai region), but to the Korean peninsula.

    For the benefit of anyone who might find this post and wonder what this disclaimer is about, I should note that the "Suzhou~Shanghai region" is where "Wu (呉) Chinese" is spoken today.

    I'm curious about the standard Cantonese transcriptions of 奴. I was under the impression that the shift of n- to l- in Cantonese was basically complete.

  5. AG said,

    October 1, 2025 @ 8:42 pm

    FWIW, I also kept seeing it as "dry lemonade".

  6. Jongseong Park said,

    October 2, 2025 @ 5:06 am

    A mystery, unresolved as far as I know, is how this made it into Japanese kan’on 漢音 but not Sino-Korean, which are both held to have been borrowed around the same time, roughly mid-late Tang, perhaps a bit later in the SK case.

    Voiced stops such as [b] and [d] cannot occur word-initially in Modern Korean, while Japanese has no such constraints. If a similar situation held at the time of the borrowings of Sino-Korean and kan'on, it would account for the denasalized initials of early Northwest Chinese being mapped to voiced stops in Japanese but not in Korean.

  7. Chris Button said,

    October 2, 2025 @ 8:41 am

    Early Mandarin might have bypassed the prenasalized stops stage of LMC to keep the nasals from EMC. Another option is that it didn't bypass LMC in that regard but reverted them back to nasals. That would be a perfectly reasonable sound change, albeit a circular one.

  8. David B Solnit said,

    October 2, 2025 @ 1:04 pm

    Typo in Prof. Whitman's comment: "Some northern MS dialects in roughly Tang times depalatalized MS /m/, /n/, /ng/." Should be "denasalized MS /m/…" as it is subsequently, but it gave me pause.

  9. Michael Watts said,

    October 2, 2025 @ 1:13 pm

    Voiced stops such as [b] and [d] cannot occur word-initially in Modern Korean, while Japanese has no such constraints. If a similar situation held at the time of the borrowings of Sino-Korean and kan'on, it would account for the denasalized initials of early Northwest Chinese being mapped to voiced stops in Japanese but not in Korean.

    But if this were the explanation, I would have expected Japanese voiced stops to correspond to Korean voiceless stops, not Korean nasal stops.

  10. Victor Mair said,

    October 2, 2025 @ 3:58 pm

    From South Coblin:

    As regards:

    “A mystery, unresolved as far as I know, is how this made it into Japanese kan’on 漢音 but not Sino-Korean, which are both held to have been borrowed around the same time, roughly mid-late Tang, perhaps a bit later in the SK case. One possibility is that the denasalizing region included Chang’an; the Koreans were savvy and in constant contact enough with China to understand that the denasalized pronunciation was substandard, even if associated with the capital region, while the Japanese clerics who imported the readings were less with it, or more superliteralist.”

    I think this is close to the correct view. In my opinion, the Koreans seem to have preferred the Guandong 關東 pronunciation of the Eastern Capital 東京, i.e., Luoyang, rather than that of the Guanzhong 關中 area, where Chang’an was located. Exactly why they went this route I don’t know. Both forms of pronunciation were viewed as acceptable in Tang times. By Song times, it was the Guandong area of Bianjing 汴京, i.e., today’s Kaifeng, that had become general and standard.

  11. John Whitman said,

    October 2, 2025 @ 7:28 pm

    Michael Watts: Voiced stops also did not occur word-initially in Old Japanese (8th c.). They are still very rare in native (Yamato) words in Modern Japanese; most of the b- d- g- initial words that will come to your mind are Sino-Japanese.

    The denasalization was really a Chinese-internal phenomenon, as Professor Coblin's and others' work has shown, borrowed by the Japanese in kan'on.

    As Korean-speaking readers of this blog will know, Korean is currently undergoing denasalization of /m/ and /n/ in phrase-initial position. This was noticed as early as the 1920s by the English phonetician Daniel Jones. A Japanese linguist once told me in all seriousness that Sino-Korean for 馬 'horse' is (it is /ma/). I think what he had heard was the denasalized pronounciation of native /mal/ 말 'horse', which sounded to his ears, and probably would to you too if you are a native English or Japanese speaker and heard the word at the beginning of a sentence, as [bal].

  12. John Whitman said,

    October 2, 2025 @ 7:38 pm

    Jongseong Park: Voiced stops also did not occur word-initially in Old Japanese (8th c.). They are still very rare in native (Yamato) words in Modern Japanese; most of the b- d- g- initial words that will come to your mind are Sino-Japanese or other loans.

    As Je comments, Korean is currently undergoing denasalization of /m/ and /n/ in phrase-initial position. This was noticed as early as the 1920s by the English phonetician Daniel Jones. A Japanese linguist once told me in all seriousness that Sino-Korean for 馬 'horse' is (it is /ma/). I think what he had heard was the denasalized pronounciation of native /mal/ 말 'horse', which sounded to his ears, and probably would to you too if you are a native English or Japanese speaker and heard the word at the beginning of a sentence, as [bal]. As Je also points out, this results in no mergers, because the voiced allophones of /p t k/ do not occur in phrase-initial position. Prosodic position seems to be important for the conditioning of contemporary Korean denasalization. This makes one wonder whether similar constraints were at work in the LMC denasalization process.

  13. Jongseong Park said,

    October 2, 2025 @ 10:08 pm

    @Michael Watts: But if this were the explanation, I would have expected Japanese voiced stops to correspond to Korean voiceless stops, not Korean nasal stops.

    My reasoning was that denasalized nasals like [n͊] would still have been interpreted as allophones of nasal phonemes like /n/. They wouldn't have been heard as voiceless stops such as [t] if voiced stops such as [d] were not available as candidates.

    @John Whitman: Voiced stops also did not occur word-initially in Old Japanese (8th c.).

    Thanks for pointing this out. Would it still have been possible that the constraint was weaker in Old Japanese so that voiced stops became permitted in loans, whereas this didn't happen in Korean?

    It might be worth considering that Old Korean likely did not a fully contrasting aspirated series of stops at least in the early stages before it developed from clusters involving /h/. But the aspirated stops ㅌ /tʰ/ and ㅍ /pʰ/ occur widely as initials in Sino-Korean (although ㅋ /kʰ/ is present only in one reading). So these sounds might have been borrowed despite having no native equivalents, maybe because similar sounds were already occurring allophonically due to cluster simplification. Perhaps something similar was happening with Old Japanese allowing initial voiced stops in their own borrowings from Chinese, maybe because voiced stops were already occurring as medial allophones. This is all just my personal speculation and I know nowhere enough about Old Japanese to judge.

    By the way, Korean speakers are in no way conscious of the denasalization of initial /m/ and /n/. Even I cannot really hear it for myself despite all the evidence presented. Not that I'm doubting this phenomenon, of course.

    As early as the 17th century, you have Dutch geographer Nicolaas Witsen's book on Inner Eurasia, Noord en Oost Tartaryen, transcribing numerous (though not all) Korean words that start with /n/ with initial d such as Deüye for "four" (cf. Modern Korean 네 ne) or Doen for "eye" (cf. Modern Korean 눈 nun). So there is reason to believe that this denasalization in Korean is not just a contemporary phenomenon.

  14. Chris Button said,

    October 2, 2025 @ 10:26 pm

    I thought I might bring up some phonetic points here.

    The EMC nasals /ŋ/, /n/, /m/ didn't simply denasalize to /g/, /d/, /b/ in LMC. Rather, they developed intrusive stops as [ᵑg], [ⁿd], [ᵐb] to preserve their acoustic distinctiveness in onset position. Those prenasalized stops can then lose their nasality.

    However, that shift introduces a phonetic force working in the other direction. Obstruents are inherently voiceless, so /g/, /d/, /b/ may develop prenasalization as [ᵑg], [ⁿd], [ᵐb] to retain their voicing. Those prenasalized stops can then become fully nasal sonorants.

    So, while there may well have been an LMC dialect with [ŋ], [n], [m] and an LMC dialect with [ᵑg], [ⁿd], [ᵐb], there might possibly have just been a singular circular shift of EMC [ŋ], [n], [m] to LMC [ᵑg], [ⁿd], [ᵐb] back to [ŋ], [n], [m].

  15. Jongseong Park said,

    October 3, 2025 @ 4:23 am

    The voiced stops in Japanese native vocabulary seem to have been prenasalized earlier, having arisen from older combinations of nasals and (voiceless) stops (somewhat similar to what happened in Modern Greek). This explains why they are only found medially in native words. But just as Modern Greek has initial voiced stops in borrowings such as ντολμάς [dolˈmas] from Turkish dolma, Old Japanese may have introduced initial voiced stops through borrowings.

    So at the time of borrowing of kan'on readings, could it be possible that both EMC and OJ had prenasalized stops that were naturally matched to each other? If so, then Chinese either reverted to nasals due to a sound change or because the varieties that used stops were replaced by ones that kept earlier nasals, while Japanese kept these as voiced stops that gradually lost the prenasalization. I think some dialects of Japanese in fact still preserve traces of prenasalization. Incidentally, Southern Min voiced stop initials can also be prenasalized in some dialects.

  16. John B. Whitman said,

    October 3, 2025 @ 8:54 am

    Jongseong Park: You are exactly right. The medial (phonemic) voiced stops in OJ (and pJR) were the result of nasal/obstruent clusters, and were prenasalized (and thus didn't occur initially, as there were no onset clusters). Some would say that nasalization was their distinctive property and that all obstruents were voiced intervocalically, like modern Korean and modern Japanese Tōhoku dialects.

    This does raise the possibility that you suggest and that Chris Button's comment supports: that Japanese "voiced" /b d g/ were chosen to transcribe MC nasals because the latter (in the relevant varieties) were prenasalized stops, and so were OJ/EMJ /b d g/. In other words, the denasalizing variety of MC was a good phonetic match with OJ/EMJ /b d g/, as you suggest.

    It's worth noting that the earlier go'on SJ readings used the OJ "voiced" stops to represent MC voiced initials. It is widely thought that specialists (J 音博士onhakase) from Paekche played a part in the formation of the go'on readings. Professor Seungjae Lee, emeritus of Seoul National University, has gone so far as to argue that the existence of a voicing distinction in go'on shows that the Paekche language had a voicing distinction. I am dubious about this, because efforts to prove a voicing distinction for earlier Korean have not been successful, and I do not think the evidence shows that the variety used in historical Paekche was that different from what we can reconstruct from Middle Korean. Paekche was a multiethnic state, and it is possible that the Paekche writing specialists who brought Sinography to Japan were Chinese speakers. To them, voiced and nasalized medial /b d g z/ in OJ may have sounded like the best fit for MC voiced obstruents.

  17. Jonathan Smith said,

    October 3, 2025 @ 10:47 pm

    Re: Southern Min, these situations are allophonic splits conditioned by features of the syllable final. In core Hokkien, old (say "proto-Coastal Min") e.g. *m- remains phonetic [m-] only before nasal(ized) vowels, elsewhere (usual representation) [b-]. In say Teochew, old *m- is [m-] before nasal vowels AND nasal coda rimes, elsewhere [b-]. Plus other situations…

    The (historically totally separate) "Old Northwest Chinese" situations seem to be similar (Teochew-like?) allophonic splits at least in some varieties/data, not simply all m- realized [(ᵐ)b] etc. See e.g. Coblin's (1991: 2.1) discussion, which notes Maspero's identification of such a split based on the Amoghavajra (mid 8th c.) transcriptions.

    Note Kan-on actually reflects a split of some such kind: 'name' 'fate' 'bright' etc. are mei not bei. Cf. e.g. Taiwanese [miã5] 'name', [miã7] 'fate' (contrast [biŋ5], [biŋ7] in some late northern Chinese loans.) Who knows how faithfully the Kan-on situation reflects some aspect of the relevant NWC dialect(s). A lot of Sino-Japanese (like Sino-Korean) is tinkered with after all… I have read somewhere that bei and the like were at some point concocted for 'bright' etc., but can't recall where.

    —

    More randomly re: allophony, modern representations by native speakers of S. Min languages do often seem to use a single onset symbol… you might think "m" but e.g. I have a book which uses "bb" for /[m] ~ [b]/ (because "b" = /p/), thus e.g. bbniá for 'name' above ("n" = vowel nasalization)… bizarre for the learner but interesting window onto (one) native sense of what is going on here.

    Finally it actually isn't 100% true that e.g. Hokkien varieties have "complete complementary distribution" (Bodman 1985 and all other sources) [b] + oral vowels vs. [m] + nasal(ized) vowels. One finds odd instances of m + oral vowel in old dictionaries and modern languages due to various historical factors.

  18. Chris Button said,

    October 4, 2025 @ 12:48 pm

    I should probably add that I think there is a good phonetic reason for when the shift does not occur before an -ŋ coda in Late Middle Chinese, which is then reflected in kan-on.

    Based on one of John Ohala's (1993) nasal theorems — "The further back a nasal consonant is articulated, the less 'consonantal' it is" — the acoustic effect of an -ŋ coda can then be that of a fully nasalized rhyme, whose syllable-spanning nasal prosody then affects the articulation of the onset.

    In that regard, it's worth noting how the LMC codas -m and -n are preserved in Kan-on (albeit -m later merges with -n), but the rhymes with LMC -ŋ are borrowed as nasalized rimes (i.e., without a final -ŋ) that then lose their nasalization.

  19. Michael Watts said,

    October 4, 2025 @ 9:46 pm

    The voiced stops in Japanese native vocabulary seem to have been prenasalized earlier, having arisen from older combinations of nasals and (voiceless) stops (somewhat similar to what happened in Modern Greek). This explains why they are only found medially in native words. But just as Modern Greek has initial voiced stops in borrowings such as ντολμάς

    This made me curious what happened with μπουγάτσα, which appears to have descended honestly from πογάτσα.

  20. Rodger C said,

    October 5, 2025 @ 10:03 am

    Michael Watts: I don't know Modern Greek, but judging from the words' appearance, I'd guess independent borrowing from the same non-Greek word, maybe in different lects.

RSS feed for comments on this post