A Stew with a Consonant Shift

« previous post | next post »

[This is a guest post by Lukhnos Liu]

Oden (おでん) is a popular Japanese dish. Common ingredients include fishcakes, konjac cakes (or konjac noodles), daikon, and boiled eggs, all stewed in a lightly soy- or mirin-flavored dashi broth. It is also popular in Taiwan, usually called o-lián and written as 黑輪 ("black wheel"), but I don't think I've heard anyone say the word in Mandarin (hei-lún). It is an example of how Taiwanese words are often written: the 黑 in 黑輪 does not represent the sound o͘ – the character 烏 (black, dark) does, and the character 黑 (black) is pronounced hek.

When I was younger, I thought it was called "black wheel" because one common ingredient used in Taiwan is pig's blood cake, or 豬血糕 (ti-hoeh-koé  in Taiwanese;  zhū xiě gāo in Mandarin), even though it's not sold in any cylindrical forms. You will not find that in Japan.

Of course, later I learned that it's a Japanese loanword. The excuse I had for never making the connection between oden and o-lián was they sound different, unlike, say, ライター (raidā, "lighter") and lài-tah – and perhaps the characters 黑輪 also put me off the scent.

How did oden in Japanese become o͘-lián in Taiwanese? In fact, o͘-lián is not the only word that has such change in sound. Here are two more examples:

  • ドライバー doraibā > lô-lái-bah (driver, but only screwdriver in Taiwanese)
  • 大丈夫 daijōbu > lai-chio-bù (no problem, don't worry about it)

So there seems to be a /d/ to /l/ shift from Japanese to Taiwanese. I asked Professor Mair whether there is indeed such a shift, and if there is, whether it is a Taiwanese phenomenon, or whether those words got the sound through the Kyūshū accent. A reviewer of an earlier draft of this post (who also provided two other examples above) responded:

My layman's feeling about the Taiwanese /l/ is that it is almost a plosive (i.e. the obstruction of airflow is greater than the approximant /l/ in English). Combined with the initial tongue position being quite far back (alveolar at least) it really sounds quite close to a Japanese /d/. So I don't believe it's as big a phonetic leap as it might first appear.

It is less clear whether there was any influence from the Kyūshū accent. The reviewer suggested that the relationship between /d/ and /r/ might exist, citing Yeounsuk Lee's The Ideology of Kokugo: Nationalizing Language in Modern Japan (1996, English translation 2010), in which Lee discussed how the Japanese language (nihongo), through education ordinances and textbook censorships, became the national language (kokugo) taught in schools:

The textbook placed exceedingly explicit emphasis on "correct" pronunciation. The sound and vocabulary in the first volume [of the normal elementary school reader] were "carefully introduced in order to make correct" distinctions between certain sounds that were confused in particular dialects, such as […] /d/ and /r/ in Kyushu dialect […] – an indication of a thorough and rigorous intention to standardize and unify pronunciation […] (p. 107)

I once heard that the Japanese spoken in Taiwan under the colonial rule (1895-1945) had influences from Kyūshū. According to a report, 47% of the Japanese who moved in Taiwan were from there. On the other hand, the "standard language" (hyōjungo) policy in Japan took its root when the first national textbooks came out in 1904 and 1905 (Lee, same page). Given the overlapping timing, it is difficult to say whether the sound change happened in Taiwan, or if it was more influenced by the Kyūshū accent.

My curiosity about this particular shift is personal. My late father once told me that my grandma insisted that the "max" water indicator, written in Japanese as ここまで (kokomade, "up to here") on the inside of our (presumably imported) Zojirushi water boiler, should really be pronounced /kokomare/, even though she also said that the kana で itself should still be read as /de/!

My grandma was Hakka, and I often regret that I had never learned enough to speak with her in her native tongue. I'm (perhaps wrongly) fond of a phrase she used often, which was "an lame," roughly meaning "very lousy." Only years after she passed away did I start to wonder if her /lame/ was really the Japanese だめ (dame, "useless, no good, incompetent"), yet another example of the /d/ > /l/ shift? It's no longer possible to ask her where that phrase came from, but I'm hoping this curiosity of mine keeps my memory of her alive.


  1. Bathrobe said,

    August 5, 2016 @ 8:15 pm

    ライター raidā should be raitā. Many Chinese speakers seem to have trouble distinguishing /t/ from /d/ in Japanese. The reason is that both /t/ and /d/ in Japanese are unaspirated, leaving only voicing to distinguish the two. In Chinese, /t/ and /d/ are distinguished by aspiration, not voicing, so the Japanese pair is virtually indistinguishable.

  2. ahkow said,

    August 5, 2016 @ 8:42 pm

    >豬血糕 (ti-hoeh-koé in Taiwanese; zhū xiě gāo in Mandarin)

    I'm just being pedantic here, but koé and gāo aren't cognates, right (the way it's written seems to imply that they are)? The tones don't correspond (a shang tone for Taiwanese, a ping tone for Mandarin). As far as I know koé is "standardly" written as 粿.

  3. Guy said,

    August 6, 2016 @ 12:34 am

    Isn't this shift pretty similar to the [ɾ] allophone of the /d/ phoneme in AmEng?

  4. Iain said,

    August 6, 2016 @ 8:04 am

    What about durian and 榴莲 (liúlián)?

    I always thought 榴莲 was a strange choice of characters for what is clearly a phonetic loan from Malay. Hokkien was traditionally the primary language for ethnic Chinese in the Malay peninsula, so this is another example of Hokkien /d/ to /l/?

  5. dainichi said,

    August 6, 2016 @ 8:55 am

    > both /t/ and /d/ in Japanese are unaspirated

    I don't agree about /t/. Wikipedia(https://en.wikipedia.org/wiki/Japanese_phonology#Consonants) says:

    "Voiceless stops /p t k/ are slightly aspirated: less aspirated than English stops, but more so than Spanish."

    which agrees with my impression.

    I speak Japanese natively. Once I said "ramen" (the noodles) with a Japanese pronunciation to a Dane, and he asked me why I said "damen".

  6. Jichang Lulu said,

    August 6, 2016 @ 10:52 am

    Possibly relevant: udon うどん, 乌冬面 wūdōngmiàn on the Mainland but 烏龍麵 wūlóngmiàn in Taiwan.

  7. Carl said,

    August 6, 2016 @ 11:44 am

    My experience with Americans learning Japanese is that they often confuse the Japanese R/L sound in, e.g., râmen with an American D-sound. I guess it works the other way around though.

  8. Ken said,

    August 6, 2016 @ 1:58 pm

    My Taiwanese parents would order "uron" (udon) at Japanese restaurants. They also pronounce the Japanese word for person as "hi'-to" (two full syllables with the stress on the first), which I understand is from the western region of Japan. The standard Tokyo pronunciation is more like "sh'to" with the stress on the second syllable.

  9. Victor Mair said,

    August 6, 2016 @ 2:08 pm

    From Jichang Lulu:

    @dainichi, Bathrobe

    I think you are both right, in that Japanese voiceless stops are usually described as 'slightly aspirated', and the degree of aspiration lies somewhere between the aspirated and unaspirated stops of Mandarin.

    "We found that Japanese VOT for /p/, /t/, and /k/ was 30.0, 28.5, and 56.7 ms, respectively…"

    Riney et al. (2007) "The intermediate degree of VOT in Japanese initial voiceless stops", Journal of Phonetics 35(3):439-443 (abstract)

    Actually it's more complicated, with regional and generational variation and an overall trend towards longer VOTs (i.e. more aspiration), and perception of the lenis-fortis contrast relying on something else than VOT (Wilson & Hashimoto). Mieko Takada 高田三枝子 has written several papers and a book on VOT in Japanese.

    Takada is also a co-author of this study, that talks of an ongoing shift towards less heavy voicing of /b, d/, g/. These evolved from Old Japanese nasal + /p, t, k/, with traces of prenasalisation still found in some dialects in the 20th century.

    Incidentally, one example of that historical change is むぎ mugi 'wheat, barley' < OJ


    , which as discussed on the cereal thread is possibly related to Korean, Tungusic and (more controversially) Chinese cereal words.

  10. Victor Mair said,

    August 6, 2016 @ 3:23 pm

    From Lukhnos Liu:

    @Bathrobe, ライター should indeed be raitā, not raidā. Thanks for pointing out the typo.

    @ahkow, I should have used 豬血粿 for ti-hoeh-kóe – kóe is indeed 粿, and 糕 is ko. It is ironic that I wrote about how the 黑 in 黑輪 is not really the o͘ in o͘-lián, and in the next paragraph I just used 豬血糕 without realizing it has exactly the same issue…

    In addition, I should have made my pinyin uses more consistent. It should be hēilún instead of hei-lún for 黑輪, and zhūxiěgāo instead of zhū xiě gāo for 豬血糕.

  11. David Marjanović said,

    August 7, 2016 @ 3:31 pm

    I don't agree about /t/. Wikipedia(https://en.wikipedia.org/wiki/Japanese_phonology#Consonants) says:

    "Voiceless stops /p t k/ are slightly aspirated: less aspirated than English stops, but more so than Spanish."

    which agrees with my impression.

    My impression is that the Japanese /p t k/ are unaspirated fortes, like those of French, standard Italian, those southern German accents that have a fortis/lenis distinction at all, Hungarian and all of Slavic. (Such sounds are indeed absent from Spanish.) My exposure to Japanese has been far too limited to notice any variation, however, and I've only begun to read the interesting paper by Wilson & Hashimoto.

  12. JS said,

    August 8, 2016 @ 1:33 pm

    @David Marjanović
    I assume your impression derives from a native language with unaspirated /p t k/? Native English speakers of course map the Japanese sounds to their voiceless aspirated obstruents. Perhaps not surprisingly, the numbers cited by Jichang Lulu (30.0/28.5/56.7 ms) indicate that the truth is somewhere between these two, with aspiration plainly involved in Japanese but VOT noticeably shorter than in English (60/70/80 via Wikipedia).

    And if anything Spanish /p t k/ are still less aspirated than in Japanese? Wikipedia gives VOTs of 5, 10, and 30 ms for these sounds in Spanish, and I'm guessing Italian is similar.

  13. dainichi said,

    August 9, 2016 @ 9:52 am

    Looking at Wikipedia's information on Taiwanese phonology


    it seems it has a 3-way voiced-unvoiced/unaspirated-aspirated distinction for bilabial and velar stops, but not for alveolar stops where the voiced one is missing. If the Japanese voiced stops are otherwise interpreted as voiced stops in Taiwanese, that could explain why the interpretation of /d/ would use /l/ instead of the in isolation perhaps more obvious unaspirated /t/. If the Japanese unvoiced stops are interpreted as the unaspirated unvoiced stops in Taiwanese, that would further explain why unvoiced /t/ wouldn't be available for the mapping of Japanese /d/.

    Does anybody know?

  14. dainichi said,

    August 9, 2016 @ 10:01 am

    Correction: Close to the end I mean "unaspirated /t/", not "unvoiced /t/". /t/s are usually not voiced…

    Hm… looking at examples of Japanese loanwords in Taiwanese elsewhere in the same Wikipedia article, I see o͘-tó͘-bái from ōtobai (so unaspirated t for the Japanese t), but pháng from pan (I assume ph means the aspirated bilabial stop), so it looks like it's more complicated than what I speculated about above.

  15. Victor Mair said,

    August 11, 2016 @ 11:55 am

    From Jichang Lulu:

    As dainichi said, Taiwanese Min has voiced /b, g/ but no voiced /d/. There are studies on perception of J. /d-/ by Taiwanese learners of Japanese. This one by Masako Okubo 大久保雅子 has the advantage of being online (and having an English abstract). Section 5.2.4 (page 5 of the linked pdf) and figures 7 and 8 on the next page concern misperception of /d-/. In initial position before /e/, /d/ is misperceived more than half of the time, mostly as /r/, less often as /n/.

    This paper (in Japanese with Chinese abstract) by Tan Le-kun 陳麗君 (notice the spelling of the name) discusses Japanese loanwords in Taiwanese, finding regular correspondences J. /d/ > Tw. /l/, J. /t/ > Tw. /th/. Examples are oden, doraibā. It's worth reading in its entirety as several other correspondences (not all so regular: e.g. J. /b/ > Tw. /b/ or /m/) are treated.

    I think several factors conspire to make the mapping /d/ > /l/ almost inevitable in these loanwords. First, the stop repertoire of Taiwanese, as dainichi said, especially when Mandarin interference was less likely.

    Second, a possible awareness of /d/ as an variant or allophone of /l/ elsewhere in Min. This is something Iain's comment perhaps was alluding to. Throughout Min, stops are mostly like in Taiwan (three series of stops except there's no /d/); but not everywhere. If I remember correctly, Xiamen has or had until recently /d/ (as an allophone or /d/ I think). Some other Min topolects have a /d/ phoneme (Putian, an implosive in Hainan), but those are less relevant. Xiamen would seem quite relevant though. Taiwanese speakers with exposure to Xiamen /d/ corresponding to Taiwan /l/ would naturally treat a Japanese /d/ the same way.

    One more conspiring factor, which I'm more sceptical about. Many Chinese speakers realise Mandarin /l/ as a flap (some also merge /n/ into that sound). Based only on personal experience, people who do that can have a wide range of topolectal backgrounds. I think I have heard that flap from people from Taiwan as well. If we agree that a flap is perceptually close to a voiced stop, and suppose a flap allophone of /l/ existed in Taiwan at the time of borrowing, hearing Japanese /d, r/ as /l/ could be even more natural.

  16. Chris Button said,

    August 11, 2016 @ 10:03 pm

    @ dainichi

    It's interesting you mention that Taiwanese has voiced "b" and "g" but no "d". I would have expected it to be harder to maintain voicing in the "g" with its back articulation than in "d".

  17. Jichang Lulu said,

    August 12, 2016 @ 12:38 am

    @Chris Button

    The received wisdom about gaps in the (phonemic) voiced stop inventories is indeed "/g/ implies /d/ implies /b/". Example: Dutch (which went /g/ > /ɣ/ (in some varieties then merged into /x/)). WALS has only data for missing /g/.

    But as stated above, not just Taiwan but the bulk of Min has missing /d/.

    Khalkha Mongolian has no voiced [b] or [d]; only the velar (with uvular allophone) and palatalised velar stop is voiced. Moreover, Khalkha contrasts bilabial and alveolar lenes (voiceless) against fortes (aspirated), but there's no velar fortis; so we can't blame the gap on the phonology. The situation is different in other varieties of Mongolian.

  18. Chris Button said,

    August 12, 2016 @ 2:28 pm

    @ Jichang Lulu

    Do you happen to know any of the historical background behind the devoicing of /d/ in Min? It looks interesting as a rare development if /g/ retains voicing.

    On the devoicing of /g/, my studies of Northern Chin (NC) languages in Northwest Burma threw up some interesting developments. Tibeto-Burman *g- devoiced to merge with /k-/ in all the languages I studied, yet in many of the languages /g/ existed as a phoneme but as a result of original /r-/ shifting to /g-/ and then in one case going even further to merge with the velar nasal /ng-/ so as to retain the voicing.

  19. Jichang Lulu said,

    August 14, 2016 @ 4:55 am

    @Chris Button

    I'm the last person you should be asking about Min. For example, my earlier comment appears to suggest that Putian has a /d/. It doesn't. Other forms of Min do have a /d/, just not Putian.

    I guess sometimes you just want the /d/ so badly that you see it where it isn't. So take anything I say about Min with a grain of salt.

    The missing /d/ in most of Min isn't just the result of devoicing. While voiced stop initials are reconstructed for Proto-Min, most topolects (including Xiamen, Taiwan) have devoiced them wholesale. For example, 茶 'tea' is reconstructed with a voiced *d- in Proto-Min, but Xiamen (and I think all of Min) has a t- initial. Devoicing killed all the voiced stops, not just /d/.

    Modern Xiamen and Tw. /b, g/ come from nasals instead. E.g. 米 can start with a b-, 五 with a g-. You'd expect that denasalisation to give /d/ from /n/, but instead you get /l/. There are two possibilities: either /n/ > /d/ > /l/ (denasalisation, then merger into /l/) or just /n/ > /l/ (just the merger). I don't know of any evidence for the first scenario; in particular, the [d] I vaguely remember Xiamen has (or used to have) is supposed to be an allophone of /l/. I'd be interested to read about this if anybody knows a reference.

    (Those 'denasalisations' and 'merger' occur under complex, topolect-dependent conditions. E.g. /b, n/ don't always become /m, l/; they all coexist.)

    In short: Taiwan missing /d/ vs extant /b g/ isn't due to devoicing vs preservation; devoicing killed the old voiced stops, and /d/ just failed to arise from denasalisation. I don't know how relevant any of this is for Northern Chin.

    Examples of such (full or partial) denasalisation can be found in many other places. Within Sinitic, it happens in Jin, and presumably in the source of Japanese Kan'on readings such as 男 dan. One non-Sinitic example: Korean word-initial nasals /m n/ can be [b d] or [m n] in seemingly free variation.

    Actual /d/ in Min has other origins. In Hainan you do have (implosive) /b d/, but those come from voicing of /p t/. So (I think) in Hainan 茶 has a d- initial, but not because it has preserved Proto-Min *d-; it was first devoiced and then voiced again. 刀 and 东, which were voiceless, are also voiced in Hainan.

    Finally, some Northern Min topolects do have voiced initials (stops or sonorants) that correspond to some older voiced stops. One possible explanation is borrowing, e.g. from Wu, which preserves MC /b d g/. Another explanation is preservation of certain earlier Min voiced initials. Norman's Proto-Min has six series of stops, three of which are voiced: plain, aspirated and 'softened' (prenasalised?). The voiced initials in Northern Min seemingly correspond to some extent to Norman's softened voiced stops. Perhaps certain cases could be due to Wu influence. Discussions of NMin voiced initials often refer to Jianyang 建阳, which Norman studied in the 60s. The 'voiced' initials are sonorants in Jianyang: there's no d-. But elsewhere in NMin, for example in Shibei 石陂 the reflexes are actual voiced stops, including d-. Examples: 茶 *d- (plain voiced) > t- (voiceless) everywhere; but 铜 *-d- (softened) > Southern t-, Jianyang l-, Shibei d-.

    This is discussed in many places. For Proto-Min softened nasals, Jianyang and Shibei, what could be better than getting it from the racoon's mouth? (Jerry Norman's Manchu name means 'racoon'.) I can provide some other references on Min initials, if I find them. I don't think I've seen a systematic discussion of n-> [> d-?] > l-, or of Xiamen's purported [d].

  20. Chris Button said,

    August 14, 2016 @ 3:13 pm

    @Jichang Lulu

    So how about the following general scenario (not particular to Min):

    /b/, /d/, /g/ are all devoiced to merge with /p/, /t/, /k/, or alternatively nasalised to merge with /m/, /n/, /ng/, with the whole process starting with the loss of /g/ as the least able to hold voicing. The voiced phonemes are then reintroduced by sound changes involving other consonants (e.g. /l/ >/d/ or /r/ > /g/ etc…) to fill the vacated slots in the phonemic inventory. This would happen at different stages so /g/ could re-emerge first (say from /r/) to give an inventory at a certain time period of /p/, /t/, /k/ and /g/ (with no /b/ or /d/) which on the surface looks contrary to the notion that /g/ is less able to hold voicing, but in fact does not violate the basic principle.

  21. Jichang Lulu said,

    August 16, 2016 @ 2:27 pm

    @Chris Button

    That scenario does indeed subsume both Southern Min and (if I'm understanding it correctly) those Northern Chin languages that went *r- > g-. Inasmuch as the g => d => b rule has an articulatory basis, I would still argue that Taiwan and Khalkha are genuine counterexamples. If [g] (or uvular [G]) is harder to produce than other voiced stops, it should be less likely to exist at any given time. Then again, 'likely' is the key word there. The rule predicts a propensity, which, judging by the WALS data I linked to above, seems statistically confirmed (it would have been less so if the Min data point had been Xiamen instead of Fuzhou, but overall I don't think their data could have an anti-missing-/d/ bias). I agree that these examples don't falsify g => d => b; they just show how it can be overruled.

    For what it's worth: here's an article on (older) Xiamen's [d]-ish /l/, by Hu Fang who calls it a prenasalised stop.

RSS feed for comments on this post