Although I can't find any examples of the second half of

lɨa R19 2.16 ni R11 2.10 'good omen' (see the last post)

by itself or in combination with other morphemes, the first half also occurs in

lɨa R19 2.16 jie R37 1.36 'auspicious sign'

(a calque of Chn 瑞相; jie and 相 mean 'look; appearance'; lɨajie and 瑞相 are translations of Skt puurvanimitta 'omen')

I don't know if lɨa can occur by itself. Current Tangut dictionaries do not distinguish between tangraphs that can represent words and tangraphs that can only form parts of words.

lɨa has a near-homophone that is the first half of

liạ R67 1.64 l(d)ị R70 2.60 'misfortune'

lɨa and liạ could share a common root *la:

*Cɯ-la-H > *Cɯ-lɨaʰ > lɨa R19 2.16 'luck'

*si-la > *si-lia > *slia > *zlia > *llia > liạ R67 1.64 'misfortune'

I was tempted to claim that *si- was a negating prefix, but I know of no other case of a tense-vowelled morpheme being the antonym of a lax-vowelled near-homophone. Could the similarity between lɨa and liạ be coincidental?

Could one or both be cognate to Chn 斜 'slant' (< Old Chinese ?*sla; not attested before Late Old Chinese?): 'slant toward(/away) from one's favor' > '(un)lucky'?

*si-la looks like Old Chinese 邪 ?*sla 'awry; crooked; depraved' (cognate to 'slant'; its phonetic is 牙 * r-ŋa, implying a complex intial like *slŋ-).

If the initial of the second syllable of 'misfortune' is l- rather than ld-, it could be a partial reduplication of lia, or derived from *la via 'ultrabrightening' of *a to i (conditioned by a presyllable with *i?).

No analysis for l(d)ị is available, but it looks like


TT2302 'disaster, calamity' =

Li Fanwen radical 072 'meat' (why?) +

Li Fanwen radical 112 'die'

Tangraphic Sea analyzes liạ as


TT4695 liạ R67 1.64 'calamity; disaster' =

frame of TT4696 liạ R67 1.64 'north' (phonetic)* +

right of TT3744 ŋĩ R16 1.16 'disaster; calamity'

looks like

Li Fanwen radical 118 'person' +

Li Fanwen radical 207 'grass' +

Li Fanwen radical 112 'die'

but I'll present the real analysis next time.

If both halves of

lɨa R19 2.16 ni R11 2.10 'good omen'


Li Fanwen radical 110 'birth'

then it's fitting that both halves of its near-antonym

liạ R67 1.64 l(d)ị R70 2.60 'misfortune'


Li Fanwen radical 112 'die'

though not all morbid tangraphs contain it. Compare

TT3144 si R11 2.10 'die'

TT3131 riʳ R84 2.72 'die'

right side is the source of the right side of TT5123 'kill'

TT3153 lhi R11 2.10 'die'

TT3150 dzəu R1 2.1 'sick horse' (i.e., a dying horse?)

Li Fanwen radical 177 'horse' is on the right


TT5480 siə R31 1.30 'die'

TT5476 ɣɛʳ R83 1.78 'die'

shares its right side with TT3131 'die' above

which lack the radical 'die'.

*The only part of

TT4696 liạ R67 1.64 'north' (phonetic)

whose function is known is Li Fanwen radical 118 'person' in the center at the bottom.

According to Tangraphic Sea, 'person' is from

TT4479 nɨaa R21 1.21 'black'

Black is associated with the north in China.

The other two parts of 'north' are from

TT4694 rəʳ R90 2.76 'shadow; reflection' I'M REALLY LUCKY TO BE BORN IN THE PAST

(unlike you guys born in the future!)

Each of the tangraphs for the Tangut word

lɨa R19 2.16 ni R11 2.10 'good omen'


Li Fanwen radical 110

which presumably represents 'birth' instead of ... its other meaning.

No analyses are known for either tangraph, so I can only try (and fail) to make sense of their parts:


TT5609 lɨa R19 2.16 'auspicious; good omen; luck' =

Li Fanwen radical 063 (meaning unknown) +

Li Fanwen radical 110 'birth/feces'*

most of TT4141 muoʳ 'true; original; one's own (baby)'

(neither the first nor the third element appear in tangraphs for la-like syllables, so they are probably not phonetic)


TT2966 ni R11 2.10 'omen; auspicious; lucky' =

Li Fanwen radical 110 'birth/feces' +

Li Fanwen radical 118 'person' +

Li Fanwen radical 261 'the past'

(derived from the top and bottom left of Chn 前 'before' or 昔 'the past'?)

The latter also looks like


Li Fanwen radical 110 'birth/feces' +

TT3557 ŋa R17 2.14 (a surname)

written as 'person' + 'the past'

(The surname Nga is homophonous with TT3818 'I' - hence the title.)

but its right side is probably phonetic from

TT1607 nie R37 2.33 'the past'

written as Li Fanwen radical 003 'hand' +

Li Fanwen radical 090 'tilt' +

Li Fanwen radical 261 'the past'

I cannot account for 'birth/feces' and 'person'.

If Tangut B existed, its known relatives, if any, should have words for 'luck', 'birth', and 'feces' sharing a common syllable.

*10.4.2:16: Could Li Fanwen radical 110 this be cryptophonetic in

TT5609 lɨa R19 2.16 'auspicious; good omen; luck'?

TT1081 liẹ and TT4340 lhi 'give birth to a child' are phonetically similar to lɨa, though they lack radical 110:

They don't even look like each other. WHAT PLUS HAND EQUALS BIRTH?

The answer is too gross to be in the title of this post.

In "The Silent Ink Solution", I mentioned that the only R12 tangraph with Tibetan transcriptions is

TT2754 wəəi R12 1.12 'give birth to; bear'

I don't know of any external cognates; the closest etymon I can find is Proto-Tibeto-Burman ŋ-(w)aaj 'make love', but the phonetic and semantic match is poor.

It looks like


Li Fanwen radical 221 'hand' + Li Fanwen radical 110

Nishida (1966: 242) identified LFWR110 as 糞 'feces' (which looks like 米 'rice' atop 異 'strange').

Why would 'hand' and 'feces' combine to signify 'give birth'?

The Tangraphic Sea analysis of 'give birth' is


TT2754 wəəi R12 1.12 'give birth to; bear' =

right of TT4463 ʃwɨa R19 1.19 'birth' < ?*Pɯ-ʃa +

vaguely similar to Middle Chinese 生 *ʂɨæŋ 'give birth'*

left of TT2967 dʒɨe R36 2.32 'give birth to'

also with 'feces'!

The analyses of its components include it:


TT4463 ʃwɨa R19 1.19 'birth' =

left of TT4462 no R51 2.42 'property' (are children property?) +

left of TT2754 wəəi R12 1.12 'give birth to; bear'


TT2967 dʒɨe R36 2.32 'give birth to' =

right of TT2754 wəəi R12 1.12 'give birth to; bear' +

right of TT3949 zwị R70 2.60 养 'nurture' (Li Fanwen 1986: 472), 生、产 'give birth to' (Shi et al. 2000: 297)

Notice that the right sides of TT2967 and TT3949 don't match in the Mojikyo font!

I have found three variants of TT3949 with different right sides:

1. In Li Fanwen (1986: 472, 760) and Shi et al. (2000: 297) with radical 328 which is also on the right of TT2967

2. In Sofronov's handwriting in Li Fanwen (1986: 644)** and Sofronov (1968 II: 364) with radical 265 as in the Mojikyo font above

3. In Li Fanwen (1997: 1087) with a bizarre right side that looks like 攵 atop 攵.

This tangraph is not in Kychanov and Arakawa 2006.

Unfortunately, Han Xiaomang doesn't list this tangraph at all, so I can only guess that the first variant is standard but cannot verify the existence of the other two - they may be ghost graphs accidentally created in modern times.

Why is 'feces' in tangraphs for birth

(I discuss the second tangraph below.)

as well as

TT2965 wụ R62 2.52 'dung; excrement'

TT2964 pəĩ R15 1.15 'dung; excrement'

Perhaps Li Fanwen radical 110

represents ... things coming out of the body, alive or otherwise. Its shape may be derived from Chn 生 'give birht'.

*Middle Chinese 生 *ʂɨæŋ 'give birth'* or a later form was borrowed into Tangut as

TT3736 ʃɛ̃ R42 1.41 'give birth to, bear'

The tangraph looks like


TT2754 wəəi R12 1.12 'give birth to; bear' + Li Fanwen radical 118 'person'

but was analyzed as


all of TT2754 wəəi R12 1.12 'give birth to; bear' +

left of TT3949 zwị R70 2.60 养 'nurture' (Li Fanwen 1986: 472), 生、产 'give birth to' (Shi et al. 2000: 297)

with the same source tangraphs as


TT2967 dʒɨe R36 2.32 'give birth to'!

**This page was supposed to be in Sofronov (1968 II: 262) but a duplicate of p. 141 was printed instead. The correct page did not appear in print until 18 years later! THE SILENT INK SOLUTION

While writing the R8-14 table in the last post, I was looking at the lists of Tangut transcriptions with those rhymes in Gong's "類林西夏文譯本漢夏對音字研究".

Only one R12 tangraph represented Chinese syllables in the Tangut translation of Leilin:

TT0155 bəəi R12 2.11 'eyebrow; brow'

cognate to Old Chinese 眉 *rməj 'eyebrow'? - but b and *m don't match!

more likely borrowed from northwestern Late Middle Chinese *mbi (transcribed in Tibetan as Hbi) with later vowel bending conditioned by a lost low vowel prefix:

*Cʌ-mbii > bəəi?

for Tangut period northwestern Chinese

默 'silent' and 墨 'ink', both *mb- < Middle Chinese *mbək

R12 is the long counterpart of R8 -əi (Grade I). I've vacillated between reconstructing R12 as -əəi with a long schwa or as -əii with a long i. I've favored -əəi because it looks less exotic and I want Tangut to look as 'normal' as possible.

I left out the rhyme of Tangut period northwestern Chinese *mb- 'silent' and 'ink' because it's uncertain. The pre-Tangut Tibetan transcription of 'silent' is Hbug; no transcription for 'ink' is available. Modern northwestern dialects have mei, mɛi, or for 'silent' (and presumably also for 'ink'). If these forms are indigenous and not imported from the east (cf. Hphags-pa Chinese mue for 'silent' and 'ink' [Coblin 2007: 126]), perhaps they developed like this:

NW Late Middle Chinese *mbək cf. Kan-on boku < *mboku in 8th century Japanese pronunciation
Pre-Tangut period NW Chinese ?*mbwəɣ cf. the Tibetan transcription Hbug
Tangut period NW Chinese ?*mbwəj or ?*mbwej ?*mbwə Tangut bəəi looks more like ?*mbwəj or ?*mbwej; Tangut had no labial-w-clusters, so *mbw- was borrowed as b- (phonetically [mb]?).
Modern dialects Generally mei except for Pingliang mɛi and Lanzhou (see next column) Lanzhou Could be from the east? Cf. Beijing mo < *mɤ.

I don't have any compelling Chinese-internal evidence for i(i) as the main vowel of 'silent' and 'ink'. If R12 were -əii rather than -əəi, I would not expect R12 to correspond to the rhyme (*-əj or *-ej) of 'silent' and 'ink'. An R12 -əii might be used to approximate Chinese -i syllables instead of *-əj or *-ej syllables.

'Silent' and

每 'every'

NW Late Middle Chinese *mbʌj

Pre-Tangut period NW Chinese ?*mbwʌj (cf. the Tibetan transcription Hbe)

Modern dialects: mei except for Pingliang mɛi

were used to transcribe R12 syllables (Sofronov 1968 II: 12).

In "西夏語中的漢語借詞", Gong identified only one R12 loanword from Chinese:

TT1402 twəəi R12 2.11 'pair; couple'

written as 刂 'not' + TT2369 'one; single'

from 對 'pair':

NW Late Middle Chinese *twəj

Pre-Tangut period NW Chinese ?*twəj (transcribed in Tibetan as twaHi and dwe)

Tangut period NW Chinese ?*twəj or *?twej

Modern dialects: mostly tuei except for tui in Karlgren's early 20th century Xi'an data

Like 'silent', 'ink', and 'every', 對 'pair' also never had i(i) as its main vowel.

The Chinese evidence points to -əəi or possibly -eei for R12. Yet the Tibetan transcriptions of R12 all have -i (Tai 2008: 208) instead of -aHi or -e. If R12 had a long mid vowel followed by a short high vowel, why would Tibetans 'A' and 'C' zero in on the final vowel? It doesn't help that all six available Tibetan transcriptions (wi, wwi, d-wi [x 2], b-wi, and ñi [why a nasal?]) are of a single R12 syllable

TT2754 wəəi R12 1.12 'give birth to; bear'

I hesitate to draw firm conclusions from such a small sample.

Could the Tibetan transcriptions represent a nonstandard Tangut dialect which had shifted -əəi closer to -i (e.g., -ɨɨi)? THE W-IAA-RD ORDER OF THE AA-RHYMES (PART 1: THE SIBILANT TEST)

Most of the rhymes in Tangraphic Sea are organized by vowel , grade, and length: e.g., R8-14:

Gong's rhyme group II: i-rhymes Vowel length My grade Rhyme number My reconstruction
Short I 8 -əi
II 9
III 10 -ɨi
IV 11 -i
Long I 12 -əəi
II 13 -ɪɪ
(III/)IV 14 -ii

R14 -ii presumably is a merger of Grade III *-ɨii with Grade IV -ii.

One would expect a similar order of a-rhymes:

Gong's rhyme group IV: a-rhymes Vowel length My grade Rhyme number My reconstruction
Short I 17 -a
II 18
III 19 -ɨa
IV 20 -ia
Long I 21 -aa
II 22 -ææ
III 23 -ɨaa
IV 24 -iaa

However, several scholars think that R21-24 are in an unexpected order:

Gong's rhyme group IV: a-rhymes Vowel length Nishida's grade Gong's grade Arakawa's grade My grade Rhyme number My reconstruction Number of examples (including homophones)
Short I? I I I 17 -a 234
II? II II II 18 76
III? III IIIa III 19 -ɨa 64
II? IIIb IV 20 -ia 94
Long III IV III 21 -ɨaa 34
I I I I 22 -aa 22
II II II II 23 -ææ 32
I III III IV 24 -iaa 23

The reason for this order is unknown.

Why can't I just reconstruct R21-24 as Grade I-IV instead of III/I/II/IV?

If merged Grade III/IV rhymes are ignored, each grade has characteristic initials.

Grades I and IV are generally the only grades that contain Homophones chapter VI initials (ts-, tsh-, dz-, s-) and the similar Chapter IX initial z-. (I know of only a few exceptions*.)

Conversely, Grades II and III are generally the only grades that contain Homophones chapter VII initials (tʃ-, tʃh-, dʒ-, ʃ-). (What about ʒ-?**)

The distribution of sibilants in R17-18 and R21-22 are quite different. It's clear that R21 and R22 do not share grades with R17 and R18.

My grade Short vowel rhyme Sibilants My grade Long vowel rhyme Sibilants
I R17 ts-, tsh-, dz-, s-, z- not I/IV R21 tʃ-, tʃh-, dʒ-, ʃ-; dz-, z- (rare)
II R18 tʃ-, tʃh-, dʒ-, ʃ- not II/III R22 ts-, dz-, s-
III R19 tʃ-, tʃh-, dʒ-, ʃ-, ʒ-; ts-, tsh- (rare) not I/IV R23 tʃ-, tʃh-, dʒ-, ʃ-
IV R20 ts-, tsh-, dz-, s- not II/III R24 dz-

So far, I can only say that R21 cannot be Grade I or IV and R22 cannot be Grade II and III. In part 2, I'll look at more evidence to narrow down the possibilties.

*Known oddities are

1. TT3089 tsɨa R19 1.19 (Grade III) instead of the expected tsia R20 1.20 (Grade IV)

2. TT3062 tshɨa R19 2.16 (Grade III) instead of the expected tshia R20 1.20 (Grade IV)

3-4. TT0019 and TT3621 dzɨaa R2 1.21 (Grade III) instead of the expected dziaa R24 2.21 (Grade IV)

5.. TT4137 zɨaa R21 2.18 (Grade III) instead of the expected ziaa R24 2.21 (Grade IV)

6 (?). Li Fanwen 1997 lists 5839 = TT5464 as zɨi R10 2.9 (Grade III) but Precious Rhymes of the Tangraphic Sea lists it as R70 2.60, a merged Grade III/IV rhyme.

**10.1.0:41: One might expect ʒ- to have a similar distribution, but it occurs in Grade IV as well as II and III:

TT0583 and 4376 ʒɪ R9 2.8 (Grade II)

TT0007 ʒɨi R10 2.10 (Grade III)

TT0005 ʒi R10 2.10 (Grade IV)

This may imply that ʒ- is somehow different from the other alveopalatals (e.g., was it retroflex ʐ-?). If it was an r-like sound, its placement in the liquids chapter makes sense. WAA-T'S WIAA-RD ABOUT THE AA-RHYMES?

When counting the number of syllables in the Tangut a-rhyme group last night, I noticed that aspirates rarely preceded some aa-rhymes:

Gong's grade I II III I III
Arakawa's grade IV III II
My grade III IV III or IV?
Rhyme R22 R23 R21 R24 R88 R89
Gong (1997) -aa -waa -iaa n/a -jaa -jwaa -jaa -jwaa -aaʳ n/a -jaaʳ n/a
Arakawa (1999) -a' -wa' -ja' -jaa -jwaa -aa' -waa' -aʳ' -jaʳ'
My reconstruction -aa -waa -ææ (-wææ) -ɨaa -wɨaa -iaa -wiaa -aaʳ (-waaʳ) -ɨ/iaaʳ (-ɨ/iwaaʳ)
Frequency (counting homophones separately) 1 1 6 n/a 7 6 1 1 2 n/a 1 n/a
Frequency (not counting homophones separately) 1 1 3 4 4 1 1 1 1

R88 and R89 are rare, so I shouldn't be surprised by their low figures..

However, I did not expect the R22 and R21 figures to be so low compared to R23 and R24. Why would Grade I Ch(w)aa and Grade IV Ch(w)iaa be less common than Grade II Chææ and Grade III Chɨaa? Cha R17 is very common, so why is its long vowel counterpart Chaa R22 so rare?

Could this all be sheer chance?

The pattern of Grade I < Grade II and Grade IV < III is also applicable to those rhymes if non-aspirates are included:

My grade I II III IV
Rhyme R22 R23 R21 R24
My reconstruction -(w)aa -ææ -(w)ɨaa -(w)iaa
Frequency (aspirates; counting homophones separately) 2 6 13 2
Frequency (non-aspirates; counting homophones separately) 20 26 21 21
Total frequency 22 32 34 23
% of aspirates 10% 19% 38% 9%

Contrast the above pattern with the Grade I > Grade II and Grade IV > III pattern exemplified by the a-rhymes:

My grade I II III IV
Rhyme R17 R18 R19 R20
My reconstruction -(w)a -(w)æ -(w)ɨa -(w)ia
Frequency (counting homophones separately) 234 76 64 94


My presyllabic hypothesis predicts that any given Tangut vowel should have more 'unbent' than 'bent' reflexes: e.g.,

the reflexes of *Ca (no presyllable) and *Cʌ-Ca (low vowel presyllable)

should outnumber

*Cɤ-Ca (mid vowel presyllable conditioning raising) and

*Cɯ-Ca (high vowel presyllable conditioning raising)

But here are the actual statistics for a-rhymes in Tangut (including homophones):

Presyllable Pre-Tangut short vowel: *Ca Pre-Tangut long vowel: *Caa
none or R18 (Grade II): 76 æ or Cɑɑ R23 (Grade II): 32
*Cɤ- Ca [ɐ]? R17 (Grade I): 234 Caa [ɐɐ]? R22 (Grade I): 22
*Cɯ- Cɨa R19 (Grade III): 64 Cɨaa R21 (Grade III): 34
Cia R20 (Grade IV): 94 Ciaa R24 (Grade IV): 23

R17 Ca is three or more times more common than any other Ca-type syllable in the first and largest rhyme cycle. This probably means that at least some Ca directly reflect pre-Tangut *Ca. It's unlikely that *Cɤ-Ca would outnumber the sum of *Ca and *Cʌ-Ca.

I could rewrite the above table as

Presyllable Pre-Tangut short vowel: *Ca Pre-Tangut long vowel: *Caa
? or R18 (Grade II): 76 æ or Cɑɑ R23 (Grade II): 32
none or *Cʌ- Ca R17 (Grade I): 234 Caa R22 (Grade I): 22
*Cɯ- Cɨa R19 (Grade III): 64 Cɨaa R21 (Grade III): 34
Cia R20 (Grade IV): 94 Ciaa R24 (Grade IV): 23

with only two types of presyllables (at least until the origin of Grade II is worked out):

*Cʌ- (nonhigh vowel; does not affect following *a)

*Cɯ- (high vowel; following *a bends upward to ɨa ~ ia)

Both types are also in my Old Chinese reconstruction

Just now I realized that there could have been a three-way distinction between presyllables based on palatality as well as height:

*Cʌ- (nonhigh nonpalatal vowel) > conditioned Grade I

*Cɯ- (high nonpalatal vowel) > conditioned Grade III

*Ci- (high palatal vowel) > conditioned Grade IV

Such a triangular system existed in Pacoh as described by Richard Watson in 1964 (Mon-Khmer Studies 1: 144):

*Ca- (low nonpalatal vowel)

*Cu- (high nonpalatal vowel)

*Ci- (high palatal vowel)

Sample derivations:

*s(ʌ)-la > TT1545 lạ R66 1.63 'hand' (Grade I; cf. Written Tibetan lag-pa 'hand, arm')

*Cɯ-naa > TT4479 nɨaa R21 1.21 'black' (Grade III; cf. Written Tibetan nag-po 'black')

*Ci-maa > TT3746 miaa R24 1.23 'fruit' (Grade IV; cf. gDong-brgyad rGyalrong sɯ-mat 'fruit')

Since minimal pairs involving Grades III and IV such as

TT3089 tsɨa R19 1.19 < *Cɯ-tsa 'broil; roast' (with a verbal prefix?)

written with 'fire' on left

TT5505 tsia R20 1.20 < *Ci-tsa 'warm'*

looks like 'fire' + 'waist/bird' + 'sage'

are hard to find, perhaps there was only one high vowel presyllable *Cɯ- that conditioned Grade III or IV depending on the following consonant: e.g.,

*Cɯ-wa > wɨa R19 (Grade III; there is no Grade IV wia R20)

*Cɯ-Pa > Pia R20 (Grade IV; there is no Grade III Pɨa R19)

(*P- = any bilabial; I suspect *w was labiodental *[v]; ww-, b-, bh- in Tibetan transcriptions may be attempts to write Tangut w [v])

If pre-Tangut had both *Cɯ- and *Ci-, it is highly unlikely that *w- was never preceded by *Ci- and that *P- was never preceded by *Cɯ-. It is far more likely that *w- and *P- could both be preceded by a single presyllable *Cɯ-.

Instances of initials followed by the 'wrong' grade may have been conditioned by a prefix or presyllable added to *Cɯ-: e.g.,

*Cɯ-tsa > tsia R20 'warm' (Grade IV)

*C(V)-Cɯ-tsa > tsɨa R19 'broil; roast' (only instance of ts- with Grade III R19)

Next: waa-t's wiaa-rd about the aa-rhymes?

*Both tsɨa 'broil' and tsia 'warm' share the same root *tsa. Cf.

Written Tibetan tsha-po < *ts- 'hot'

Proto-Tibeto-Burman *tsa 'hot' [Matisoff 2003: 616, 654])

and within Tangut:

TT0407 zaʳ R85 1.80 < *rʌ-tsa 'hot; bitter' (with medial lenition before late presyllable loss)

TT0408 tsaʳ R85 1.80 < *r(ʌ)-tsa 'chili' (with a nominal prefix?)

written as 'hot; bitter' + 'grass'

TT5498 tshia R20 1.20 < *ki-tsa 'burn' (with a verbal prefix?)

written with 'fire' on left

TT5499 tshwia R20 1.20 < *p-ki-tsa or *k-pi-tsa 'roast; toast' (with a verbal prefix?)

written as 'fire' + 'hand'

Tangut fonts by Mojikyo.org
Tangut radical font by Andrew West
All other content copyright © 2002-2008 Amritavision