When converting Gong's reconstruction of

TT2506 gjiʳw R94 2.79 'drop, spot, a bit'

to my system for "Singular Characters", I had to decide whether to reconstruct a front or a central vowel. I chose a central vowel since R94 only occurred after velars, and velars occurred before nonpalatal Grade III R46 -ɨw, not palatal Grade IV R47 -iw. There are only three R94 syllables:

kɨʳw R94 1.88 (four tangraphs)

Li Fanwen (1986: 184) stated there are five 1.88 tangraphs, but I can only find four in Tangraphic Sea, Precious Rhymes of the Tangraphic Sea, and his version of Homophones.

kɨʳw R94 2.79 (five tangraphs)

gɨʳw R94 2.79 (two tangraphs)

These can be traced back to a pre-Tangut *r(ɯ)-KVw(-H) and/or *r(ɯ)-KVk(-H).

- the vowel retroflexion comes from an earlier *r somewhere in the syllable

- the *r has to be pre-initial since

- medial *-r- + a high vowel would have become R46 or R47

- final *-wr, *-rk, and *-kr are absent from Tangut's neighbors and hence were probably absent in pre-Tangut

- V could be *i, *e, or

- K stands for a velar stop *k or *g or velar cluster (e.g., *ŋg-) that fused into g-

- there may have been a high ultrashort vowel after *r- which conditioned raising of nonhigh vowels

- Tangut -w may be from *-k or *-w

- -H is the source of the rising tone

the correlation of g- with the rising tone is probably accidental

I would expect R94 to follow non-velar initials as well. Its mid-vowel counterpart R93 -eʳw is preceded by nonlabial, nonpalatal, and nonglottal consonants other than k- and g-:

Homophones chapter III V VI IX
Initial t- d- k- g- ŋ- ts- r- z-
R93 -eʳw teʳw deʳw ŋeʳw tseʳw reʳw zeʳw
R94 -ɨʳw kɨʳw gɨʳw

One might be tempted to claim that R93 and R94 were originally a single pre-Tangut rhyme that split into two depending on the initial. However, I can't understand why acute initials (and velar ŋ-!) are associated with mid vowels whereas velar k- and g- (but not velar ŋ-!) are associated with high vowels. I could sweep the height problem under the rug by interpreting the R93/R94 distinction purely in terms of palatality:

R93 -iʳw (palatal Grade IV)

R94 -ɨʳw (nonpalatal Grade III)

However, that still does not answer the question of why ŋ- behaves like an acute initial. Moreover, a palatal interpretation adds a new problem: why does Grade IV R93 precede Grade III R94? Normally Grade IV rhymes follow their Grade III counterparts.

Next: Other reconstructions of R93 and R94. ST-ILL ALONE

Guillaume Jacques pointed out to me that

TT2507 tiẹ R64 1.61 'only; unique'

is cognate to the gDong-brgyad rGyalrong root sti 'alone'. This forces me to abandon my attempt to link tiẹ to Tangut lew 'one' since lew is from *tek, whereas sti is from *-k-less proto-rGyalrong *ste, *sti, or *stij. (The Somang cognate of sti is tə-ɕtɕî without a -k which would have been retained from proto-rGyalrong.)

Perhaps the proto-rGyalrong form was *ste, since the pre-Tangut ancestor of tiẹ may have been *sɯ-te. However, I lack the gSar-rdzong cognate (if any) which would determine the vowel of the proto-rGyalrong reconstruction. WHAT HAPPENED TO PRE-TANGUT *S(V)-CVW?

(This entry expands on a note I wrote last night.)

Pre-Tangut *S(V)-CV became Tangut CṾ. One might expect pre-Tangut *S(V)-CVw to become Tangut CṾw, but in fact there are no Tangut syllables with tense vowels followed by -w in Nishida's, Gong's, or Arakawa's reconstructions*. This gap has two possible explanations:

- The acute initial prefix(es) that I reconstruct as *S(V)- never happened to occur before *CVw-roots in pre-Tangut. This is unlikely since *CVw-roots have no special phonological or semantic characteristics that distinguish them from all other roots: e.g., onsets that would be 'hostile' to *S(V)- or membership in a closed word class 'hostile' to affixation.

- *S(V)-CVw became something other than CṾw in Tangut.

The second explanation is far more likely. I think that Tangut had a constraint against tense vowels in words with 'long roots' containing a long vowel or a final -w (the only glide coda in my reconstruction):

Root type Non-*S-derivatives *S-derivatives
Pre-Tangut Tangut Gloss Pre-Tangut Tangut Gloss
short short vowel *ka ka equal *S(ɤ)-ka kạ to make equal
long long vowel *K(ɯ)-puu phuu cover over *S(ɯ)-puu pụ cap, crown; cover over
-w (partly from *-k) *Cɤ-tek lew one *Sɯ-tek tiẹ only; unique

A simple rule could account for all *S(V)-derivatives:

*S(V)-CṾ̣(X) > CṾ

in which X is anything that makes a short root long (either vowel length or *-w).

However, the above example involving a -w (< *-k) root is the only one known to me, and it may turn out to be invalid if tiẹ is from an unrelated root *taj.

Perhaps *S(V)-CVw merged with syllables ending in tense labial vowels: e.g.,

*S(V)-Ciw, *S(V)-Cu > Cụ

*S(V)-Caw, *S(V)-Cew, *S(V)-Cəw, *S(V)-Co > Cọ

The discovery of -Vw ~ -ụ/-ọ alternations will confirm this hypothesis.

*In Sofronov's reconstruction which has instead of -w, there are rhymes like -ạɯ, etc. in the second small rhyme cycle reconstructed by others with retroflex vowels. He did not reconstruct -Ṿɯ-type rhymes in the first small rhyme cycle commonly assumed to be tense. SINGULAR CHARACTERS: LHEEW IS NOT ALONE

Neither TRECD nor Grinstead 1972 list any tangraphs with

TT2501 lheew R48 2.41 'one; only'

on the right side. However, Sofronov listed six tangraphs with TT2501 on the left:

Tangraph Tangut Telecode My reconstruction Rhyme Gloss (Li Fanwen 1997) Function of TT2501

TT2502 liuu R7 1.7 carve, engrave phonetic

TT2503 ɣwə̃i R15 1.15 complete, overhaul semantic

TT2504 tjwaʳ R87 1.82 only, single, alone semantic

TT2505 ʑiẹ R64 2.54 a surname; decoration* unknown; rhymes with TT2507 tiẹ if tones are disregarded, but the initials don't match: t- lenited to l-, not ʑ-.
TT2506 gɨʳw R94 2.79 drop, spot, a bit semantic
TT2507 tiẹ R64 1.61 only; unique semantic; probably not phonetic, even if this word is cognate to TT2501 because the lenition of *t- predates the invention of tangraphy; this word is discussed in more detail in the previous entry

*5.28.1:57: For a moment, I thought that ʑiẹ could be from pre-Tangut *Sɯ-ɕɨe, a loan from Middle Chinese 飾 *ɕɨk 'decorate' with a Tangut prefix attached. However, the rhymes don't match: MC *-ɨk should correspond to Tangut R70 -ị or R72 -ɨ̣, not R64 -iẹ.

Tibetan transcriptions of the northwestern reflex of MC *-ɨk sometimes have -e- instead of -i- or -ï- Coblin 1994: 419-421):

sheg (homophonous with 飾 'decorate')



(These -e- probably represent or rather than a front vowel *e.) However, if 飾 'decorate' were something like *ɕek with a front vowel in the northwest, its final *-k would have become -w after a Tangut front vowel: *Sɯ-ɕɨek > ʑɨẹw.

But Tangut has no -w rhymes with tense vowels. Did such rhymes merge with non-w rhymes? Could ʑɨẹ be from ́́*ʑɨẹw? Was final *-w lost due to a constraint against bimoraic tense syllables? Tense vowels cannot be long in Tangut. WHAT PLUS EIGHT EQUALS ONE?

TT1632 jaʳ R87 1.82 'eight'

is beneath an unknown element (̣'horned ユ') in

TT2501 lheew R48 2.41 'one; only'

(no analysis available)

which must be cognate to

TT1075 lew R44 1.43 'one'

The lh- of lheew could be from an earlier *kl-. The functions of *k- and the vowel lengthening in lheew are unknown.

lew may be from *Cɤ-tik (cf. Written Tibetan gcig < *k-tik 'one') with a nonhigh presyllabic vowel that conditioned root vowel lowering as well as root initial lenition:

*Cɤ-tik > *Cɤ-təik > *Cɤ-tek > *Cɤ-dek > *Cɤ-lek > *Cɤ-lew > lew

(The relative chronology of changes is unknown; the actual order probably differs.)

Since I think *K-t- became th-, I suspect that *K- was added to 'one' after its initial was lenited to l-.

Another possibility is that lheew originally had the same *Kə- prefix as lew:

*Kɤ-tiik > *Kɤ-təəik > *Kɤ-teek > *Kɤ-deek > *Kɤ-leek > *K-leek > *K-leew > *X-leew > lheew

However, the *Kɤ- prefix of lheew fused with the root initial, whereas it was lost before lew.

I have previously assumed that these words for 'one' were cognate to

TT2507 tiẹ R64 1.61 'only; unique'

(with TT2501 lheew R48 2.41 'one; only' on the left plus the right side of kwiẹ̃ R65 2.55 'self; alone')

but the rhymes don't match, since tiẹ is presumably from pre-Tangut *Sɯ-te (or *Cɯ-S-te?) without a coda. Although lew and lheew could be from a pre-Tangut root *te(e)k with a mid vowel, a final stop in 'one' is present among Tangut's distant relatives:

Written Tibetan gcig < *k-tik 'one' (with the same prefix as pre-Tangut *Kə-tik?)

Written Burmese tac < *tik 'one'

Old Chinese 隻 *tek 'single'

Hence *-k must be quite old. I know of no other examples of open syllable ~ -w alternation that could justify positing *-k-deletion as a derivational process in pre-Tangut.

5.28.00:34: On the other hand, maybe the root of Sino-Tibetan 'one' is *ti, and *-k is a suffix. Pre-Tangut *Sɯ-te could be cognate to Written Burmese thiḥ 'single; alone' and Jingpho tai 'single' which lack reflexes of *-k (forms from Schuessler 2007: 614).

Matisoff (2003: 262) traced Jingpho tai 'single' back to Proto-Tibeto-Burman *taj ~ *tan 'single, one, whole, only', a root distinct from his PTB *t(y)ik 'one'. He cited Karlgren's Old Chinese 單 *tân (*tar in my reconstruction) 'single' whose coda "might in turn go back to an even earlier *-r" and Written Tibetan thor-bu (< *tor-; defined by Nitartha as 'scattered, dispersed, fragmentary, single, separate'), implying a Proto-Sino-Tibetan *tVr.

If Matisoff is correct, Tangut l(h)e(e)w and tiẹ are from unrelated roots *tik/*tek and *taj. The correspondence of Late Middle Chinese *-aj to Tangut R34 -e (see Gong, "西夏語中的漢語借詞", p. 770) suggests that pre-Tangut *-aj monophthongized to -e shortly before the Tangut period:

*Sɯ-taj > *Sɯ-tɨaj > *Stɨaj > *ttɨạj > *tɨạj > tiẹ GREEN FIVE

TT0452 gəəu R5 2.5 'five'

is cryptophonetic (via its native synonym ŋwə R28 1.27 'five') in

TT5405 ŋwəʳ R90 1.84 'seven; seventh' (first syllable; presumably of the ritual language word)

with Li Fanwen radical 043 on the left. Nishida (1966: 244) identified this radical as 靑 'blue; green'. I wonder if colors had numerical values in Tangut culture.

The Tangraphic Sea analyzed the first half of 'seven(th)' as


TT5405 ŋwəʳ R90 1.84 'seven; seventh' (first syllable) =

left of TT5412 kạ R66 1.63 'seven; seventh' (second syllable [of the ritual language word?]; 'green' twice around a vertical line)

left of TT0459 ŋwəʳ R90 2.76 'blue; dark; green'

The tangraph for the second syllable of 'seven; seventh' was analyzed as


TT5412 kạ R66 1.63 'seven; seventh' (second syllable) =

left of TT5407 ŋwəʳ R90 1.84 'blue; dark; green' +

right of TT5405 ŋwəʳ R90 1.84 'seven; seventh' (first syllable)

TT5407 must be cognate to TT0459; the latter has a rising tone from a pre-Tangut suffix *-H.

The etymology of

ŋwəʳ-kạ 'seven; seventh' (ritual language?)

is unknown. It may go back to a pre-Tangut *r-ŋwə-ska containing ŋwə R28 1.27 'five' (cf. Khmer prampii 'seven' = pram 'five' + pii 'two'*) but *r-...-ska does not resemble any Tangut word for 'two' or 'second'.

5.26.17:49: I didn't want to upload this entry without commenting on the other disyllabic (ritual language?) numerals which can be extracted from Guillaume Jacques' list of month names:

monosyllabic disyllabic (before kaʔo 'month') derivation
one lew (none?) n/a
two nɨɨ rɨʳ-lọ ?
three sọ lheʳ-gu ?
four lɨɨʳ dzia-ŋwəʳ (2nd syl homophonous with 1st half of 'seven'!) ? + prefix + 'five'
five ŋwə tɕɨʳ-ləu ?
six tɕhiw ʑiw-wəi prefix + 'six' + ?
seven ɕiạ ŋwəʳ-kạ (1st syl homophonous with 2nd half of 'four'!) prefix + 'five' + ?
eight jaʳ nɨɨ-lɨɨʳ 'two' plus 'five'
nine gɨɨ lɨɨʳ-ŋwə 'four' plus 'five'
ten ɣạ nɨɨ-ŋwə 'two' times 'five'
eleven ɣạ lew ŋwə-tɕhiw 'five' plus 'six'
twelve ɣạ nɨɨ ̣̣̣(none?) n/a

There are no disyllabic words for 'one' or 'twelve' in this list since 'first month' is literally 'new year month' and 'twelfth month' is literally 'the month that became cold' (< 'become cool' plus a directional/perfective prefix 'toward the speaker').

The disyllabic words for 'four', 'six', and 'seven' contain basic numerals plus unknown affixes, whereas those for 'two', 'three', and 'five' cannot be derived from the basic numerals and are completely obscure.

*5.26.17:59: pəmpɨl in normal speech. The -l seems to be an irregular reflex of earlier *-r which normally has been lost in speech but is still preserved in writing (pii 'two' is spelled BIIR). Are there are any other instances of -l < *-r in Khmer?

5.26.18:10: I found what appear to be two such instances searching for "CVl" in SEAlang's online Khmer dictionary:

yul 'droop' (spelled YUR)

ʔol 'pouring heavily (of rain)' (spelled ʔUR) NEGATIVE FIVE

TT1359 ləəi R12 2.11 'black; dark'

is the first tangraph in TRECD because TRECD's ordering is based on right-hand elements, beginning with horizontal lines.

Its left-hand element is the common semantic element 刂 'negative' (Li Fanwen radical ) but its right-hand element seems to be unique.

TT1359 has a variant that appears in Nevsky (1960 I: 236), Sofronov (1968 II), and Grinstead (1972) without the lower right horizontal line:

The right-hand element of this variant is

TT0452 gəəu R5 2.5 'five'

a borrowing from northwestern Chinese 五 *ŋgu 'five'. (The basic native Tangut word for 'five' is ŋwə R28 1.27.) TT0452 is used to write Chinese loanwords containing 'five'.

TT0452 looks like a horizontal line atop

TT1632 jaʳ R87 1.82 'eight'

Why would 'five' be written as a line over 'eight'? (Could TT0452 - with or without a horizontal line on the bottom - be modelled after the old sinograph for 'five'?)

And what does 'negative' + 'five' have to do with 'black'? If 'five' in 'black' is an error for the unique element 'five' + horizontal line, what does that unique element mean? Is that unique element an abbreviation for another tangraph with 'five' presumably meaning 'light' ('not' + 'light' = 'dark')? Does the lower left horizontal line represent omitted strokes? (Grinstead [1972: 58] noted that "Many Chinese abbreviated cursive forms end with a dot, often to show that strokes have been omitted.")

The most likely source of the right side of 'black' that I could find was

TT0458 ruʳ R80 2.69 'clean'

which isn't a perfect semantic fit, since 'not' + 'clean' should be 'dirty', not 'black' or 'dark'.

Unfortunately, I cannot confirm this guess since 'black', like most rising tone tangraphs, was presumably analyzed in the lost second volume of Tangraphic Sea.

5.26.00:19: An even less likely source of the right side of 'black' is

TT0459 ŋwəʳ R90 2.76 'blue; dark; green'

in which gəəu R5 2.5 'five' may be cryptophonetic (via its native synonym ŋwə R28 1.27 'five'). I doubt that 'dark' is from 'not' plus 'dark'.

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