refers to the Tangraphic Sea analysis of the tangraph spotlighted in "-u x 2 ^ 2":


TT0434 言詞 EXPRESSION ngwu 1.1 =

left of TT0443 漱口 RINSE-THE-MOUTH khyï 2.29 +

looks like LANGUAGE (Nishida radical 154) + an element unique to this tangraph (function unknown)

right of TT3420 口 MOUTH lya 2.18

looks like PERSON (Nishida radical 204) + MOUTH (Nishida radical 255)

TT0434 looks like a combination of

LANGUAGE (Nishida radical 154) +

反 MOUTH (Nishida radical 255)

It sounds like 語 'language' which would have been pronounced as ?*nggü in Tangut period northwestern Chinese. If TT0434 is a loanword, that would explain why it is the only instance of a velar-initial -wu syllable in Gong's reconstruction of Tangut: it was an attempt to imitate a foreign syllable that was absent from Tangut. Was -wu really more like *[ü]? Are there other loanwords with the correspondence Tangut -wu : TPNWC -ü?

The answer to the latter question appears to be 'no'.

TPNWC ?*-ü was transcribed with tangraphs ending in Gong's -yu 1.2/2.2/1.3/2.3, not -wu 1.1/2.1 (Nishida 1964: 42-43) and TPNWC ?*-ü was borrowed as Tangut -yu 1.2/2.2/1.3/2.3, not -wu 1.1/2.1 (Gong 1982: 762).

Moreover, TPNWC ?*ngg- was borrowed as Tangut g- , not ng- (Gong 1982: 722-23).

Therefore if TT0434 言詞 EXPRESSION were borrowed from TPNWC ?*nggü, it should have been gyu 1.2/2.2/1.3/2.3, not ngwu 1.1. (The syllables gyu 1.2, gyu 1.3, and gyu 2.3 did exist. The absence of gyu 2.2 is probably a random gap in the system.)

I conclude that the similarity between TT0434 言詞 EXPRESSION ngwu 1.1 and TPNWC ?*nggü may be coincidental. Or is it?

Could ngwu 1.1 be related to the Old Chinese ancestor *nga' of TPNWC ?*nggü? Although the initials match, OC *-a' corresponds to Tangut -a, not Tangut -wu 1.1 (Gong 1995: 79-80):

'bitter': 苦 OC *kha' : TT1791 kha 2.14

'door': 戶 OC *gwa' : TT2468 Ga 1.17

(Tangut has no gw-. Did earlier *gw- become Tangut G-? OC *gw- became Middle Chinese *G- or *Gw- depending on the following rhyme.)

After nonemphatic initials, OC *-a' corresponds to Tangut -ya:

'father': 父 OC *ba' : TT1999 wya 1.19

'you': 汝 OC *na' : TT2882 nya 2.17

Could ngwu 1.1 be derived from a lost *ngya 1.19 via vowel and medial alternation? Gong (1988) does not list any alternations of -wu 1.1 with -ya 1.19.

ngwu 1.1 is, however, involved in two other alternations that I will investigate in my next entry. -U x 2 ^ 2

The chart in "-u x 2" left out one type of Tangut -u syllable: -wu syllables:


Is there any modern spoken Asian language that distinguishes between -u and -wu after a consonant? I don't know of any. (English does has such a distinction after s: i.e.., soup vs. swoop and soon vs. swoon. Are there any other examples?)

Although Gong reconstructed -u for both 1.1/2.1 and 1.4/2.4, -w- never occurs with -u 1.4/2.4. This suggests that

1.4/2.4 had some quality that clashed with -w-


the absence of -w- in 1.4/2.4 is simply due to chance, since 1.4/2.4 is less common than 1.1/2.1.

All but two -wu syllables (ngwu 1.1 and 1.2) begin with dentals or alveolars.

-wu is almost never preceded by grave initials (labials and velars). Why was ng- the only velar initial followed by -wu?

Gong's Tangut reconstruction has no instances of ngu without -w-.

Sofronov's ngu 2.1 without -w- always corresponds to Gong's ngwu 2.1 (15 instances: TT0179, 0279-0281, 0286, 0292, 1281, 2194, 2368, 3310, 3903, 3908, 4202, 5339, 5343).

The only tangraph whose reading Sofronov reconstructed as ngwu 1.1 was

TT0434 言詞 EXPRESSION ngwu 1.1

Li Fanwen (1986) reconstructed all of the above as ngu without -w-.

One can arrange these reconstructions in a continuum if one ignores tones:

Li Fanwen: ngu-onlySofronov: ngu and ngwuGong: ngwu-only
ngu x 16ngwu x 1, ngu x 15ngwu x 16
ngu 1.1ngu 2.1ngwu 1.1ngu 2.1ngwu 1.1ngwu 2.1

Which of these reconstructions, if any, is correct?

All agree that the 16 tangraphs in question cannot be totally homophonous. Homophones lists 15 in one homophone group and one (TT0434) as the sole member of another homophone group.

There are two interrelated issues:

1. What was the distinction between TT0434 and the fifteen others? Was the distinction purely tonal (1.1 vs. 2.1), or was it also segmental (i.e., did TT0434 had a -w-)?

2. Was there a -w- in one or both homophone groups?

Fanqie evidence points to a -w- in TT0434:

TT1814 nwu 1.1 < TT3682 nyi 1.11 + TT0434 ngwu 1.1

TT5398 ngwe 1.8 < TT0434 ngwu 1.1 + TT5043 thyiy 2.33 (not a -we 1.8 syllable!*)

The -w- in TT1814 and TT5398 must be from TT0434, as there is no -w- in TT3682 or TT5043. (But there are also fanqie in which the -w- of TT0434 ngwu 1.1 is ignored**.)

Do any of the 15 2.1 tangraphs represent an initial + -w- or a -w- + final sequence in fanqie for other -w- tangraphs? I cannot find any fanqie containing those tangraphs.

For now, I prefer Sofronov's reconstruction of ngu for the 15 2.1 tangraphs. ngwu is an unusual sequence and it would be highly unusual if Tangut had 16 instances of ngwu and zero instances of ngu.


*Why would ngw- + -yiy 2.33 = ngwe 1.8? This suggests that rhymes 2.33 and 1.8 were segmentally similar. Although Gong reconstructed 1.36/2.33 as -yiy, Clauson (1964: 65) reconstructed this rhyme as -e and Nishida (1964: 52) reconstructed this rhyme as -eH. I am inclined to agree with them, as the Tibetan transcriptions and Tangut transcriptions of Sanskrit clearly point toward -e. There are no Tibetan transcriptions of 1.8, but the transcriptions of its rising tone counterpart 2.7 contain i, not e (Sofronov 1968 II: 9). They support Nishida's reconstruction of 1.8/2.7 as -I, which would be close to the e of 2.33.

Gong's reconstruction of Tangut has no ngywiy 1.36/2.33 (phonetically closer to ngwe?). Did -ywiy 1.36/2.33 shift to -we 1.8/2.7 after velars? This would predict that that there should be no velar-initial -ywiy 1.36/2.33 syllables, but there are:

kywiy 1.36 x 1, 2.33 x 2

gywiy 1.36 x 3

The absence of ngywiy 1.36/2.33 (and khywiy, xywiy, Gywiy 1.36/2.33) may be due to chance.

**In these fanqie, TT0434 ngwu 1.1 stands for ng-, not ngw-:

TT1203 nga 1.17 = TT0434 ngwu 1.1 + TT2746 la 1.17

TT1707 nga 1.17 = TT0434 ngwu 1.1 + TT2746 la 1.17

TT2522 ngə 1.27 = TT0434 ngwu 1.1 + TT0272 lə 1.27 -U x 2

(Pronounced 'you too'.)

Gong's Tangut reconstruction has -u for two pairs of rhymes:

-u 1.1 / 2.1

-u 1.4 / 2.4

The Chinese, Tibetan, and Sanskrit data for both rhyme pairs all point to -u. The distinction between the pairs may have been absent from those three languages.

One might guess that the two -u are in complementary distribution: i.e., they occur after different initials. If so, each -u might have reflected the preceding initial: e.g., one -u could have been rounded after labials. However, there are instances of the same initial after both types of rhymes:

r-r- must be followed by a retroflex vowel
ch-palatals must be followed by -y-

The initials can be grouped into four categories:

1. Never occur before 1.1/2.1 or 1.4/2.4: r- and palatals

2. Occur only before 1.1/2.1: p-, ph-, b-, t-, th-, n-, l-, lh-, ts-, tsh-, dz-, s-, z-, g-, '-

3. Occur only before 1.4/2.4: m-, d-, G-

4. Occur before 1.1/2.1 and 1.4/2.4: k-, kh-, x-

More initials can precede 1.1/2.1 than 1.4/2.4. Therefore 1.1/2.1 was the 'default' -u rhyme and 1.4/2.4 might have been a slightly different -u-like rhyme: e.g.,

Nishida (1964): -uH

Sofronov (1968): -uC with an unknown final consonant

Shi et al. (1983): -uu

Li Fanwen (1986): -uo

There is nothing unique about the category 3 intials (m-, d-, G-) that distinguishes them from the category 2 labials, dentals, and velars. All the category 4 initials are voiceless velars, but I don't know why those two particular qualities would allow them to appear before both 1.1/2.1 and 1.4/2.4. Did the unknown special quality of 1.4/2.4 have nothing to do with initials?

Next: -u x 2 ^ 2: Double u. -IU R MISSING

(Pronounced 'you are missing'.)

Earlier tonight, I noticed that Gong's Tangut reconstruction (GTR) has no -iur - or -i- before any u - even though it has ia, ie, and io. GTR has a constraint against high and central vowels after i.

GTR does, however, have the following -i-less -u rhymes:




GTR has no long tense vowels, so the absence of -(y)ụu is expected.

GTR does have long retroflex vowels, but as I noted last night, it has no *-(y)uur (because it became *(y)wïïr?).

Strangely, GTR has -yu assigned to two pairs of rhymes (1.2/2.2 and 1.3/2.3) after -u (1.1/2.1). Why? Usually the rhymes after a -V rhyme are -iV rhymes which are then followed by -yV rhymes: e.g.,

-o 1.49/2.42

-io 1.50/2.43

-yo 1.51/2.44

So I would have expected 1.2/2.2 to be -iu, not -yu. Was 1.2/2.2 originally -iu before it merged with -yu 1.3/2.3?

-u 1.1/2.1

-yu 1.2/2.2 (< ?*-iu)

-yu 1.3/2.3


In "A Fiery Feminine Hand", I pointed out that HAND (Nishida radical 236) had two phonetic values, ryur 2.70 and zyur 2.70. This indicates that - if Gong's initials are correct - r- and z-syllables can be written with the same phonetic.

In Homophones, z- and zh-syllables were grouped with l-, lh- and r-syllables in chapter IX (liquids) rather than with s-syllables in chapter VI (alveolars) and sh-syllables in chapter VII (palatals). This makes me suspect that there was something l(h)- and/or r-like about Tangut z- and zh-: e.g., z- or zh- might have been an l-like voiced alveolar lateral fricative lzh- or an r-like voiced retroflex fricative Zh-.

Middle Chinese *z- and *zh- are from Old Chinese *sl- and *ml-. I wondered if Tangut z- and/or zh- originated from earlier laterals or rhotics, so I looked for gDong-brgyad rGyalrong cognates of Tangut words with initial z- and zh- in Guillaume Jacques' "Essai de comparaison des rimes du tangoute et du rGyalrong":

TT2384 LEOPARD zerw (< ?*rzek) 2.78 : rG kU rtsëG

TT2385 LEOPARD (Grinstead's EAST!) zerw (< ?*rzek) 2.78 : rG kU rtsëG

TT3358 LONG zyir (< *?zri) 2.72 : rG kU zri

TT0834 JUNIPER-TREE zhyiw (< *-k) 1.46 : rG shëG

Is this instance of Tangut zh- from a voiced prefix plus original *sh-?

rG r corresponds to the retroflexion of the Tangut vowel and is not evidence for deriving Tangut z- from *r-. There is no lateral or rhotic in rG shëG.

Are there any cases of lyur written with HAND? No. Moreover, according to Gong's reconstruction, there was no lyur in Tangut. Is the absence of lyur accidental or meaningful?

Since rGyalrong initial, medial, and final r often (not always) correspond to Tangut retroflexion, Tangut lVr syllables could have come from *rlV, *lrV, and/or *lVr. The clusters *rl- and *lr- would be quite awkward, and it's possible that one and/or the other was simplified to *l- or *r- before Tangut developed retroflex vowels:

*rlV > *lV > lV (no retroflexion)

*lrV > *rV > rVr (automatic retroflexion of r-initial syllables)

If so, maybe pre-Tangut had *rlyu and *lryu but not *lyur, and the former two became lyu and/or ryur.

Tangut has 28 retroflex rhymes (46 including tones). Does l(h)- occur in all of them, or only in some of them (e.g., only before high vowels)?

1.90 (twice!)-yor
2.81 (twice!)

(I have corrected l(h)yoor 1.94 to l(h)oor 1.94.)

Out of 46 x 4 = 184 theoretically possible lVr(G) syllables, it seems that only ten actually exist. These ten syllables contain the following vowels:

i(i)rï(ï)r(no u[u]r)
(no a[a]r)

There are no cases of a(a)r or u(u)r preceded by l(h)-. This may imply that pre-Tangut *lra, *rla, *lar, *lru, *rlu, *lur lost their lateral initials. u and especially a are common vowels, and it would be strange if pre-Tangut had no syllables combining *l and *r with those two vowels.

Tangut had no əər. If pre-Tangut had such a vowel, it may have merged with short ər.

Tangut did have an eer after only two initials: m- and d-. Neither initial has much in common with the other, so there is no special reason for this limited distribution. Hence the absence of leer is probably due to chance.

Why are there no instances of Tangut l(h)or? lwər - the only instance of a retroflex schwa after a lateral could be from an earlier *lor. Tangut did have an -or, but this rhyme never occurred after dentals or the liquid-like z-.

1.89: kor, Gor, wor

2.80: Gor, tsor, 'yor, ror

Tangut did, however, have r after t-, th-, d-, z- as well as l-. Thus perhaps *-or > -wər after dentals and alveolar l.

(Alveolar ts- can occur with both -or and -wər, so I can't generalize the rule to include all alveolars.)

The absence of -wər and -or after tsh-, dz-, n-, and lh- may simply be due to chance.

I am surprised that there are no syllables like so(o)r or swər, since s- is more common than tsh- or dz-. (Why no swəər with a long vowel? See above.)

I wonder if lhywïïr could be from an earlier *lhyuur. Tangut had no -(y)uur. Did *-(y)uur break to *(y)wïïr just as *-or may have broken to -wər after dentals and alveolar l? There are only four syllables ending in -wïïr:

kywïïr 1.92 (< ?*kyuur)

pywïïr 1.92, pywïïr 2.85 (< ?*pyuur)

lhywïïr 1.92 (< ?*lhyuur)

As the table above indicated, Tangut did have a -(y)ur, but this rhyme never occurred after dentals and l(h)-. (It did, however, occur after z-: zur 1.75, zur 2.69.) I guessed that pre-Tangut *-(y)ur became -(y)wïr after dentals and l(h)-, but there are no dental-initial syllables like twïr. I wonder if chhwïr and jwïr are partly or wholly from pre-Tangut *thru and *dru (but where's the chwïr that would be from *tru?). A FIERY FEMININE HAND

In the previous post,

TT1410 VENUS ryur 2.70

looked like NOT + HAND and wasn't a hand.

TT0567 zyur 2.70

looks like FIRE-TRIGRAM (Chn 離; Nishida radical 256) + FEMININE (Nishida radical 018) + HAND (Nishida radical 236).

What does it mean? If the answer isn't in your short-term memory (hint!), scroll over the blank text below:


The left side of TT0567, FIRE-TRIGRAM, is clearly semantic.

Although the Tangut word for HAND was lạ 1.63, the graphic element HAND has the phonetic values zyur 2.70 and ryur 2.70 when it is a component in another tangraph:

TT3239 SWEEP zyur 2.70 (with GRASS [Nishida radical 144] on the left)

TT1410 VENUS ryur 2.70 (with NOT [Nishida radical 041 on the left)

TT2784 (a surname) ryur 2.70 (with CLAN [Nishida radical 164] on the right)

TT5736 BECOME-SOUR ryur 2.70 (with 造 MAKE [Nishida radical 056] on the right)

Did the Tangut B word for HAND sound like zyur 2.70 or ryur 2.70?

I have no idea what FEMININE is doing in TORCH.

(Is FEMININE somehow derived from 母 'mother' or a squished version of 女 'woman'?)

I don't see why TORCH simply isn't FIRE-TRIGRAM (semantic) + HAND (phonetic). I don't know of any tangraph with that combination of elements in that order. But a tangraph with those elements in the opposite order does exist:

TT2647 BURN nywï 2.61

It's usually not clear to me why some fiery tangraphs have FIRE-TRIGRAM and others have FIRE, or why tangraphs have those fire elements on the left or on the right. However, TORCH could not have been written as FIRE + HAND because that combination had already been used to write another word:

TT5499 熱 HEAT / 燙 SCALD tshywa 1.20

Nonetheless, HAND + FIRE was still available, though it might have been confused with HAND + WATER:

TT2803 彎 ?CURVED khywï 1.30

The right-hand versions of FIRE and WATER look very similar: FIRE is (フ x 2) x 2 and WATER is フ x 3. Compare:

TT3920 (syllable of a name) tsya 1.20 (PERSON + FIRE)

TT4037 (a place name) lyi 2.9 (EARTH + WATER)

(I can't find any minimum pair of FIRE and WATER graphs sharing the same left side.) NOT A HAND

The next few Tangut posts will very loosely link back to a Chinese word I've discussed in more than one post last week. See if you can guess where this is going.


looks like NOT (Nishida radical 041) + HAND (Nishida radical 236).

So is it

- a semantophonetic compound for a negative word that sounds like

TT1545 HAND lạ 1.63 (the function of the vertical line [Nishida radical 039] is unknown)

- a semantic compound for something that is not a hand (and what would that be)?

Maybe its Tangraphic Sea analysis would help:



left of TT1344 LOSE lyạ 2.57 (Shi et al. [2000: 293]: 鑽 DRILL, 穿 PENETRATE; not in Homophones, so no Li Fanwen [1986] gloss available) +

[bottom] right of TT0567 火炬 TORCH zyur 2.70



Scroll over the blank space below to find out what TT1410 meant and how Gong reconstructed its reading:

TT1410 明星 (Chn 'bright star'; according to Shi et al. [1983: 515, 2000: 169]) VENUS ryur 1.76

Nevsky (1960 I: 231) glossed it as название звезды 'name of a star', вечера 'evenings'.

Tangraphic Sea (F81A72) defined it as a word for a star.

Both r- and z-syllables are in the same chapter of Homophones, so it's probable that TORCH is an abbreviated phonetic in VENUS.

I have no idea what LOSE, DRILL/PENETRATE, or 刂 NOT have to do with VENUS. The left side almost works as an fanqie initial, but not quite, as l- + zyur 2.70 is not ryur 1.76.

According to Nevsky, VENUS was a disyllabic word

TT2515 1410 VENUS syiy 1.36 ryur 1.76

(Grinstead [1972: 143] didn't gloss TT1410 but did gloss TT2515 as VENUS.)

I don't know if either half can appear by itself. According to Nevsky (1960 II: 607), TT2515 was used to transcribe the last syllable of Tibetan kra ga se (a name?). I don't know of any other uses of either tangraph.

This word is attested in Tangraphic Sea (F81A72). I don't know how it differs from

TT0434 0207 GOLD kiẹ 1.66 STAR gyịy 1.61

the word for 'Venus' in Pearl in the Palm glossed in Chinese as 金星 'gold star' = Venus. YESTERDAY'S OAK

In "tsak lak rak?", I wondered why

昨 OC *s-lak > MC *dzak 'previous day'

柞 OC *s-rawk > MC *dzak 'oak'

(as reconstructed by Sagart 1999: 67-69)

were written with a phonetic 乍 for OC *tsak-like syllables.

One possibility is that both words had an initial *dz- when they were first written with 乍. This implies that the sound changes *sl-, *sr- > *dz- predate the first attestations of the graphs 昨 and 柞. If early OC still had *sl- and *sr-, then the graphs 昨 and 柞 could not have appeared in the oldest texts.

Is that prediction correct? To the best of my limited knowledge, it might be. Neither Karlgren (1957) nor Schuessler (2007) list either 昨 or 柞 as appearing in bronze inscriptions or the even earlier oracle bone inscriptions. (Schuessler does not list 柞 at all.)

Nonetheless, I still have reservations about Sagart's reconstructions of those two words.

First, I would really be convinced if there were early spellings of those words with unambiguous liquid phonetics. Sagart (1999: 67, 68) linked them to 昔 *s-hlak 'last night' and 櫟 *r(e,a)wk 'oak' (vowel uncertain).

It is not certain whether 昔 had a liquid. Schuessler (2007: 523) reconstructed it as *sak or *syak. If either of Schuessler's reconstructions is correct, then 昔 could not be related to 昨, and there would be no reason to reconstruct a liquid for 昨. 昔 ?*s-hlak, ?*sak, ?*syak might have shared nothing more than a rhyme with 昨, which I prefer to reconstruct as *dzak.

(Then again, what if 昨 *dzak is from *N-[t]-s[y]ak, a prefixed version of 昔 *s[y]ak? Maybe 昨 and 昔 are related, but not in the way that Sagart proposed. For the development of *dz- from *N-s-, cf. the affricated pronunciation of the coda of prince [prInts] in English.)

*r(e,a)wk 'oak' almost certainly had an *r. Nonetheless it does not have an alternate MC reading *dzak which would confirm Sagart's proposal to derive MC *dz from OC *r-clusters in some cases.

Second, 櫟 *r(e,a)wk 'oak' has a *-wk phonetic but Sagart's 柞 OC *s-rawk has a *-k phonetic (乍 OC *rdzaks 'suddenly'). I don't know of any other cases of *-wk words written with *-k phonetics. In Shijing, there is only one example of interrhyming between *-wk and *-k(s) words not written with 乍. 乍-words clearly belong to the Shijing *-ak rhyme category. *-ak and *-awk were distinct until Han times.

Third, Sagart himself (1999: 239) reconstructed OC *sr- as a source of MC *Sh- (a retroflex fricative) as well as MC *dz-. Many, including myself, believe that OC *sr- became MC *Sh- (though my 'emphasis' may correspond to other phonetic features in different reconstructions.) When would OC *sr- become MC *Sh- and when would it become MC *dz-? One could account for the two outcomes by assigning them to different types of dialects. But is a nondialectal explanation also possible?

Next: A sr-lution?

Tangut fonts by Mojikyo.org
All other content copyright © 2002-2007 Amritavision