Mair's hypothesis on tonal repetition

« previous post | next post »

In his major July 1 speech celebrating the 100th anniversary of the CCP, one of Xi Jinping's many pronouncements featuring "blood" was this:

yù xuè fèn zhàn

浴血奋战/ 浴血奮戰

lit., "bathe blood energetic / rousing / fierce battle / fight"

i.e., "fight hard in bloody battles; a bloody fight"

It is hard to pronounce four 4th tones in a row.  Indeed, in normal speech, it is virtually impossible to do so.  When four 4th tones occur in succession, some sort of natural sandhi will arise to obviate that condition.  Most people I know who pronounce this quadrisyllabic expression will convert the third syllable to a light first tone.

Aside from the consonants, vowels, and tones, when one begins the study of Mandarin, one of the first things one learns about the pronunciation of the language is that you cannot have two 3rd tones in succession:  the first one has to become a 2nd tone, e.g., Nǐ hǎo –> Ní hǎo 你好 ("How are you?").

It is physically uncomfortable to repeat two low, dipping syllables that are at the bottom of your register.  To avoid that, one will automatically convert the first of the two 3rd tone syllables to a 2nd tone.  This amounts to a law of Mandarin pronunciation that everyone accepts.

The succession of syllables with identical tones other than the 3rd is also proscribed by phonological constraints, though not prescribed as with the 3rd tone.

Two first tones don't seem to present any difficulty requiring tone sandhi, e.g., māmā 媽媽 ("mom; mama"), though I do often hear it pronounced as māma.

In general, aside from the 3rd tone, which reaches — with a stretch, almost to the breaking point — the extreme bottom of the vocal register —  it seems that only when groups of three or more identical tones occur that tone sandhi is invoked.  My impression is that this is the result of physiological restraints of the vocal tract.

I should note that these more or less ad hoc types of sandhi occur, different speakers may adopt different patterns, especially for longer sequences of the same tone repeated.

During the 60s and 70s, when Rulan Chao Pian was teaching Mandarin at Harvard, she had a fairly thick booklet detailing such patterns of sound changes in the spoken language.  These were extracted and adapted from various publications and notes made by her father, the distinguished linguist, Y. R. Chao.  Harvard students would spend several weeks studying these sound patterns and would consult the booklet later for reference purposes.  I used that guide in a summer school Mandarin course I taught at Harvard one year, and I remember that it included many different types of tone sandhi, each of which is real, but few of which are ever taught systematically, if at all.

Let's take as a case study my own Chinese name, Méi Wéihéng 梅維恒.  I've never heard anyone pronounce it with the canonical tones, except perhaps neophyte foreigners who would bob their head up and down while doing so.  Instead, native speakers of Mandarin will almost always pronounce my name as Méi Wēihéng, even if they've never met me before.

Ditto for the Song period poet Méi Yáochén 梅堯臣 (1002-1060).

All that said, I will now state the Mair hypothesis as simply as possible:
When a succession of three or more syllables with identical tones occurs in a spoken utterance, sandhi will arise to modify one or more of the tones, except for the repetition of third tone, which requires only two syllables with the same tone to invoke sandhi.
This is only a hypothesis and remains to be proven (or disproven, or modified).  Bear in mind, moreover, that it is based solely on my own observation of the behavior of native speakers.

Selected readings


[Thanks to Mark Metcalf]


  1. Mark Liberman said,

    July 6, 2021 @ 8:38 am

    There's a relevant linguistic literature, under the heading of the "Obligatory Contour Principle"…

  2. Victor Mair said,

    July 6, 2021 @ 10:14 am

    James Unger asks:

    Do the changes of, e.g., 2 2 2 to 2 1 2 you describe appear mostly in normal or fast speech but seldom in slow, deliberate speech? If so, you may want to look at Samuel E. Martin’s ideas about juncture in Japanese, which he used to explain variations in Tokyo pitch-accent patterns over multiple phrases.

    My reply:

    Exactly as you say, Jim. Only in normal or fast speech, but seldom in slow, deliberate speech.

  3. Neil Kubler said,

    July 6, 2021 @ 10:45 am

    In 3-syllable words, the middle syllable often changes to Tone 1 in rapid speech. For example, 加拿大 "Canada" said slowly is Jiānádà but at a normal rate of speech, it becomes Jiānādà.

  4. Jerry Packard said,

    July 6, 2021 @ 12:51 pm

    Tonal assimilation and dissimilation (more precisely, tonal _endpoint_ assimilation and dissimilation) had been noticed for quite awhile before Will Leben formalized it as the OCP (the Obligatory Contour Principle cited by myl) in his 1973 MIT dissertation. Assimilation is the unmarked (usual) process in phonology and dissimilation is marked (less common). The OCP was formulated by Leben as a way of explaining why adjacent tonal segments would dissimilate (like Victor’s 4444 > 4414, where 4th tone HL becomes HH 1st tone) when they might be expected to assimilate (like Victor’s Méi Wéihéng > Méi Wēihéng and Neil’s Jiānádà > Jiānādà, both of which show 2nd tone LH > 1st tone HH). These processes are affected by speech rate (as Jim Unger points out) and phrase parsing (as seen in the work of Richard Sproat and Chilin Shih).

  5. David C. said,

    July 6, 2021 @ 8:50 pm

    The pronunciation of 血 throughout the speech also didn't go unnoticed. The expected pronunciation is xuè when used in a literary expression (文讀), rather than the colloquial reading (白讀) xiě. And definitely not xuě. See here for an explanation.

  6. Claw said,

    July 7, 2021 @ 12:02 pm

    Presumably a hypothesis on the physiological constraints of tonal production should apply to other Chinese topolects as well (not to mention tonal languages in general). For comparison, Cantonese is generally regarded as exhibiting no tone sandhi, and as far as I've been able to perceive, speakers are easily able to produce utterances of sequential syllables having the same tone without tone sandhi. For instance, the following are four-character idioms whose syllables are all spoken with Cantonese's 陽平 tone, which is pronounced as a low tone at the bottom register very similar to Mandarin's third tone:

    從難從嚴 chùhng nàan chùhng yìhm
    危言危行 ngàih yìhn ngài hàhng
    含糊其詞 hàhm wùh kèih chìh
    微乎其微 mèih fùh kèih mèih
    文如其人 màhn yùh kèih yàhn
    遙遙無期 yìuh yìuh mòuh kèih

    While Cantonese's 陽平 tone does not have the characteristic rise at the end of Mandarin's third tone, the rise is generally also not pronounced in Mandarin except when spoken in isolation or at the end of an utterance. More specifically, Mandarin's third tone is cited as having a contour of 214, but is generally pronounced as simply 21 intersyllabically, which matches the tone contour of Cantonese's 陽平 tone.

    Here's an example in Cantonese where sequential syllables are spoken with a rising 13 contour (陽上 tone):

    有理冇理 yáuh léih móuh léih

    I can't think of as many examples for this tone, but again, Cantonese speakers can seemingly produce that utterance easily with no tone sandhi.

  7. Richard Warmington said,

    July 8, 2021 @ 12:13 am

    With the 4th tone, there is tone sandhi even with a succession of only two syllables, according to Y. R. Chao's "Mandarin Primer".

    "When a 4th Tone is followed by 4th Tone, the first does not fall quite to the bottom, but only to the middle." [p. 26]

    Also, on the same page, Chao states a rule about successions of three syllables.

    "If in a three-syllable word or phrase the first syllable is a 1st or 2nd Tone, the second is a 2nd Tone, and the third syllable is any except the neutral tone, then the second syllable (which is in the 2nd Tone) is pronounced in the 1st Tone. An example of this change is dongnan-feng 'southeast wind,' [pinyin dōngnán-fēng] which is 1st-2nd-1st, changing to 1st-1st-1st."

    At the end of Part II (Foundation Work), Chao provides one example for each of the 100 possible sequences of three syllables. (4 possible tones for the first syllable, 5 for the second, and 5 for the third, resulting in 4x5x5=100 combinations)

    Of relevance to Mair's Hypothesis, the list includes examples of the following tone sequences: 1-1-1, 2-2-2, and 4-4-4. No sandhi is indicated for 1-1-1. Naturally, 2-2-2 changes to 2-1-2 in accordance with the rule Chao described. In the case of 4-4-4, both of the first two 4th tones are subject to sandhi and fall "only to the middle".

    Finally, Chao indicates that a 3-3-3 sequence (as in 我也有) changes to 2-1-3.

  8. Tom said,

    July 8, 2021 @ 2:31 am

    @Claw: Isn't there a ton of tone sandhi in Cantonese? Even just with reduplicatives (basedd on my beginner's knowledge of Canto), I can think of many examples.

    Like, 姐 ("older sister") is ze2 on its own, but when my Canto-speaking family refers to my daughter (in contrast to her younger brother) it's as 姐姐 ze4 ze1. My son, the little brother, is called 弟弟 dai6 dai1 (different tones, same character). In fact, the tone sandhi is the reason my in-laws said I should stick to Mandarin: it all gets too complicated in Canto!

  9. Claw said,

    July 8, 2021 @ 10:51 am

    @Tom: While Cantonese features tone changes, the phenomenon you describe is not considered tone sandhi. Tone sandhi is motivated by the phonetic environment around syllables, whereas tone changes in Cantonese are lexically motivated. In other words, there are no phonetic rules that determine when these tone changes occur in Cantonese; rather they occur in a limited set of words in specific semantic domains. They can be analyzed as being derived from an addition of a tonal morpheme, which is itself thought to originate from a contraction of a diminutive suffix such as 子 or 兒. In this regard, Cantonese changed tones are similar to the phenomenon of erhua in Mandarin, which also occurs in certain words and is not phonetically motivated.

    For more information, see:

    (That latter link at the end does also describe a different process in Cantonese that could be considered tone sandhi, but it is contested and is generally considered obsolete in modern Cantonese anyway.)

  10. David C. said,

    July 8, 2021 @ 7:26 pm

    Thank you Claw. The discussion of the high-level tone and the high-falling tone was fascinating, and a distinction that I was not consciously aware of. I had trouble articulating the difference at first, and it clicked for me when I tried to say 廣州 (Guangzhou; gwong2 zau1) in an exaggerated Guangzhou accent. I have noticed it in the past but it's nice to finally read an explanation of this phenomenon.

RSS feed for comments on this post