The second item in Guillaume Jacques' 2006 list of Tangut-Japhug rGyalrong comparisons is

Tangut 5113 1wji 'to do' (stem 1), 36211wjo 'to do' (stem 2)  : Japhug kɤ-pa 'to close'

also cf. other rGyalrong forms: e.g., Somang ka-pa 'to do'

(more in #1133-1135 at Nagano and Prins' database)

See my first "Eat-ymology" post for an explanation of stems 1 and 2.

That post was about a parallel pair of stems:

Tangut 4517 1dzji 'to eat' (stem 1), 4547 1dzjo 'to eat' (stem 2)  : Japhug kɤ-ndza 'to eat'

The parallelism is not as apparent if one looks at rhyme numbers and/or my reconstructions:

Stem 1 2
'to do' vɨi 1.10 vɨo 1.51
'to eat' dzi 1.11 dzio 1.51

Guillaume uses Gong Hwang-cherng's reconstruction in which rhymes 1.10 and 1.11 are both -ji. I reconstruct them differently. I also reconstruct different allophones of 1.51 after different initials. The class II initial v- is followed by Grade III -ɨ- but not Grade IV -i-. It was somehow antipalatal in a way that medial -w- is not. Gong reconstructed w in both initial and medial position and did not reconstruct a Grade IV distinct from Grade III.

I can see why Gong reconstructed rhymes in those two grades identically. There are a few minimal pairs involving them: e.g.,


0932 ʔɨi 1.10 'many' (only with that meaning in dictionaries?) : 3119 ʔi 1.11 'many'

Those pairs were probably the reason why the Tangut split rhymes (e.g., into 1.10 -ɨi and 1.11 -i). When no such pairs were present, the Grade III/IV distinction was subphonemic, and there was no split. Hence there was only one rhyme 1.51 -ɨo/-io. The frontness of the first half of the diphthong was predictable.

I think the Grade III/IV distinction was absent from pre-Tangut. I reconstruct the pre-Tangut sources of 'to do' as

stem 1: *CI-pa > *CI-pja > *CI-β- > *vi > vɨi 1.10

stem 2: *CI-pa-w > *CI-pjaw > *CI-βj- > *vjo > vɨo 1.51

I do not know if *-ja and *-jaw became *-i and *-jo before or after lenition. My guess is that such shifts predated the loss of stop codas that might have blocked the raising of *-ja to *-i:

Brightening stage 1 *-ja *-jaw *-jaC
Brightening stage 2 *-i *-jo *-jaC
Final coda loss *-i *-jo *-ja
GV > VV diphthong reanalysis -i -io -ia

Some of the changes above could be viewed in terms of a drag chain:

*-jaC > *-ja > *-i

Tangut syllables with lenited initials must have once had presyllables conditioning intervocalic lenition:

*presyllable + labial > *β- > v-
*presyllable + dental > l-

*presyllable + alveolar > *z- > ɮ-

*presyllable + palatal > - > ʐ-

*presyllable + velar > ɣ-

Lenition must have preceded brightening because there are words such as 'to eat' with brightening but without lenition. 'To eat' must have lost its brightening presyllable before 'to do':

Gloss 'to eat' 'to do'
Pre-Tangut *NI-dza *CI-pa
Brightening *NI-dzja *CI-pja
Presyllable to prenasalization *Ndzja *CI-pja
Lenition *Ndzja *CI-β-

If brightening had followed lenition, 'to eat' should have been *ɮi 1.11 (stem 1) / *ɮio 1.51 (stem 2) with *ɮ- from a lenited *-dz-. EAT-YMOLOGY

Guillaume Jacques' 2006 list of Tangut-Japhug rGyalrong comparisons begins with

Tangut 4517 1dzji 'to eat' (stem 1), 4547 1dzjo 'to eat' (stem 2)  : Japhug kɤ-ndza 'to eat'

Guillaume used Gong's 1997 reconstruction of Tangut.

Those Tangut words in my reconstruction are 1dzi (without -j-) and 1dzio (which could be rewritten as dzjo).

Stem 2 (in bold below) was used before the first and second personal singular suffixes when the object is in the third person. Otherwise stem 1 was used:

Subject \ object of 'eat' ... me ... us ... thee ... you ... him/her/it/them
I ... (no 'I eat me', etc.)


We ...


Thou ...

 (no 'You eat you', etc.)
You ...


He/she/it/they ...





The seventeen slots in that table have only six forms. I list my reconstructions when they differ from Gong's.

1. bare stem 1 (3rd person subject and object)

2. stem 1 + 2ŋa (first person singular object)

3. stem 1 + 2nja (= my 2nia; second person singular object)

4. stem 1 + 2nji (= my 2ni; nonthird person plural subject and/or object)

5. stem 2 + 2ŋa (first person singular subject + 3rd person object)

6. stem 2 + 2nja (= my 2nia; second person singular subject + 3rd person object)

Reconstructing the history of the Tangut and Japhug words for 'to eat' involves dealing with issues 2 and 3 from my last post.

Cognates of the Tangut word such as Japhug kɤ-ndza, Written Tibetan za-ba, and Written Burmese cā generally have a. The high front vowel of Tangut -ji is assumed to be the product of 'brightening' (Matisoff 2004).

In 2009, Guillaume derived -ji in 'eat' from *-ja. I don't know whether he still does in his new book. This is phonetically plausible. However, it raises the question of where the *-j- in *-ja came from. How far back can it be projected? Did languages such as Japhug, Tibetan, and Burmese lose it? Or is it a Tangut-internal innovation?

Gong (1994: 42) thought Old Chinese and Tangut retained Proto-Sino-Tibetan *-j- whereas Tibetan and Burmese generally lost it. Perhaps he would have said Japhug had lost it in this word. (There are no *affricate-j clusters in Guillaume's (2004: 331-332) Proto-rGyalrong reconstruction. Did *ndzj- simplifiy to *ndz-?)

On the other hand, I did not reconstruct -j- in Tangut. I proposed that the brightening of *a to -i was due to a high-vowel presyllable:

*CI-dza > *CI-dzja > 1dzi

*CI-dza-w > *CI-dzjaw > 1dzio

The problem with this hypothesis is the absence of external evidence for *CI-. Could the Japhug reflex of *CI- be n-; i.e, was *CI- something like *[ni]? If so, perhaps the presyllable was absorbed into the initial in both Japhug and Tangut:

Pre-Proto-rGyalrong *ni-dza > Proto-rGyalrong *ndza >

Japhug -ndza

Somang -zá

Zbu -ndzeʔ, -ndziʔ (with brightening conditioned by the front vowel of *ni-?), -ndzʌʔ

Tangut: *ni-dza > *ni-dzja > *ndzi > 1dzi

I have followed Gong and Arakawa who reconstructed Tangut voiced obstruent initials without prenasalization, but others such as Nishida (1964) and Sofronov (1968) would disagree. The most recent scholar in favor of complex voiced obstruent initials is Tai (2008:

[...] there are regular use of prescripts in front of voiced obstruents [in the Tibetan transcription of Tangut], suggesting that there should be a pre-initial consonant [in Tangut], which is probably a weak nasal or glottal sound.

I followed Guillaume who reconstructed *ndz- at the Proto-Tangut (= my pre-Tangut) level in 2009. GUILLAUME JACQUES' ESQUISSE DE PHONOLOGIE ET DE MORPHOLOGIE HISTORIQUE DU TANGOUTE NOW IN PRINT

That was the best news I'm likely to hear all week. This month too. Maybe even this year.

Unfortunately I haven't seen the book yet. Google Books has no preview for it. Nonetheless I am confident that I will be impressed. I have seen Guillaume's previous work on Tangut and rGyalrong and look forward to see how he has build upon it. I am particularty interested to see his treatment of the following topics:

1. What shared innovations distinguish his proposed Macro-rGyalrongic group from the rest of Qiangic or - if  Macro-rGyalrongic is his term for Qiangic - the rest of Sino-Tibetan?

Tonight I found the 2011 dissertation of Marielle Prins (whose rGyalrongic database I constantly use) which states  on p. 21 that there is "an absence of common innovations" in Qiangic. Prins proposed that

the similarities between the Qiangic languages may be caused by diffusion rather than be genetic in nature. [...] It is more likely that the shared features of these languages are the result of contact induced structural convergence, and that the Qiangic group should be considered an areal language group rather than a group of genetically related languages. (p. 22)

I wonder what Guillaume would say about that.

I am not sure whether Prins is denying that the Qiangic languages are related at all, or if she is just rejecting Qiangic as a subgroup. The latter position need not entail a complete absence of a genetic relationship: e.g., Qiangic could consist of languages from multiple Sino-Tibetan branches which have converged. Is Prins' Qiangic like my Altaic (completely unrelated languages tha have converged) or like the Balkan languages (which are from different branches of Indo-European)?

2. I assume Guillaume is still using Gong's 1997 reconstruction of Tangut which has three grades of rhymes (his III corresponds to my III and IV):

Grade Gong Gong's source Arakawa This site
I -Ø- *-Ø- -Ø- -Ø- + lowering of high vowel
II -i- *-r- -j- -ɤ-
III -j- *-j- long vowel -ɯ-
IV -i-

Gong's Tangut -j- and Old Chinese *-j- were retentions from his Proto-Sino-Tibetan *-j-. On the other hand, Guillaume does not reconstruct Old Chinese *-j-. How does he account for Gong's Tangut *-j-?

3. Pre-Tangut *a was raised and fronted ('brightened'; Matisoff 2004) to various degrees. I have tried to explain the multiple reflexes of *a by reconstructing presyllables with front vowels:

*CI-Ca > Ci

*CE-Ca > Cie

More recently I have hypothesized that some 'brightening' might have been conditioned by a suffix *-j: e.g.,

1749 *kwa-j > 1kwe 'hoof'

cf. Ersu nkhuɑ⁵⁵ 'id.'; more cognates at STEDT

How does Guillaume explain 'brightening' in Tangut?

4. Gong reconstructed long vowels that do not correspond to long vowels in Tangut transcriptions of Sanskrit. I am now agnostic about those vowels and reconstruct them with an abstract symbol ' to differentiate them from their much more common '-less counterparts (short vowels in Gong's reconstruction). The zero-' distinction does not seem to correspond to anything in rGyalrong; both types of Tangut vowels correspond to the same Japhug rGyalrong vowels (Jacques 2006): e.g.,

Tangut rhyme Gong This site Japhug
37 -jij -ie -i, -e, -o
40 -jiij -ie'

What does Guillaume think is the source of vowel length in Gong's reconstruction? Does that length reflect a disticnction lost in Japhug?

5. I reconstructed *-H as the source of the Tangut second ('rising') tone; syllables without *-H developed the Tangut first ('level') tone. This type of tonogenesis has parallels in Chinese, Tibetan, and Burmese, but not. Southern Qiang (Evans 2007). What is Guillaume's account of the origin of Tangut tones? A *E(YE)-GRADE ROOT?

If Tangut 1new 'breast', 2niu 'to drink milk', and 2niụ 'to give milk' are from the root *√n-w that I proposed in my last post, there ideally should be other sets of -ew ~ -iu words. Unfortunately, I still haven't gotten around to looking for them, but today it did occur to me that Tangut

4684 1me 'eye'

and Old Chinese 目 *muk 'eye' might share a root *m-kʷ:

*m-e-kʷ (e-grade) > *mew > 1me

(See this series of posts on Tangut *labial-w syllables: 12.7.23 / 12.7.28 / 12.7.29).

*m-kʷ (zero-grade) > 目 *muk

This word is widespread in Sino-Tibetan (STEDT roots #33, 681, 682). It often has an -i-: e.g., Tibetan mig 'eye'. Was there an i-grade? (I have borrowed the terms e-grade and zero-grade from Indo-European studies. There is no such thing as an i-grade in Indo-European, but perhaps it existed in Sino-Tibetan.) Or was the root *mʲ-kʷ with an initial palatalized consonant that was vocalized as -i- in the zero grade in many languages?

One might want to resurrect an old-fashioned reconstruction *mjuk for Old Chinese 目 *muk and view its *mj- as a reflex of *mʲ-, but I have never seen any evidence for *-j- in modern Chinese languages, and there is no trace of *-j- in Sinoxenic:

Taiwanese bak (colloq.), bok (lit.)

Cantonese muk

Mandarin mu (in earlier reconstructions, *mj- > w-, not m-)

Sino-Vietnamese mục (not *dục < *mjuk or *mʲuk)

Sino-Korean mok

Sino-Japanese moku, boku

One might also be tempted to regard 覓 Middle Chinese *mek 'to seek' as being from an e-grade *m(ʲ)ekʷ like Tangut 1me 'eye', but the earliest attestation of the word that I can find is in Yupian (c. 543 AD), so I do not know if it should be reconstructed at the Old Chinese level. It could be a later unrelated innovation that has nothing to do with m-words for 'eye'. A *N-W WORD FAMILY?

The Tangut word

4834 2niụ < *S-nuH 'to give milk'

from my last two posts is a causative derivative of

4614 2niu < *S-nu 'to drink milk'

and I think those two words are related to


2123 1new 'breast' =

left of 3588 1new 'radish' (phonetic) +

left and center of 5275 2nɪʳ 'breast' (semantic).

Tangut *-w can be from *-k or *-w. If 2123 1new 'breast' is from *new and not *nek*, then it and the niu-words may share a root *√n-w:

pre-Tangut prefix root consonant 1 vowel root consonant 2 suffix gloss
e-grade Ø n- -e- -w Ø breast
zero-grade Ø -H to drink milk
S- to give milk

The *-w of the zero-grade root would have been pronounced as a vowel [u].

Could the grade hypothesis account for the vocalic diversity of these cognates?

Old Chinese 乳 *Cɯ-noʔ 'nipple, milk' could be from an o-grade *n-o-w. (The prefix could be *pɯ- if 孚 *phu is phonetic.)

If the above scenario is correct, are there other cases of *Cew ~ *Cu alternations in Tangut, and what is the significance of the different grades?

Li (2008: 832) regarded 5275 2nɪʳ 'breast' as a loan from Chinese. The only similar Chinese word I know of is 奶 'breast, milk'. But I cannot find any attestations of 奶 before the Qing Dynasty. If 奶 had existed in northwestern Middle Chinese, it would have been pronounced *nəjˀ, and if Tangut speakers added a *T-prefix, the resulting *T-nəjˀ would have developed into 2nɪʳ. The Old Chinese source of *nəjˀ would be *Cʌ-nəʔ which might have come from an even earlier *Cʌ-nəw-ʔ: i.e., a schwa-grade form of *√n-w. Perhaps the pre-Tangut prefix directly reflects the Old Chinese prefix if the latter had survived in the colloquial speech of the northwest during the Middle Chinese period:

OC *Tʌ-nəw-ʔ > *Tʌ-nəʔ > MC *T(ʌ)-nəjˀ > pre-Tangut *T(ʌ)-nəjˀ > Tangut 2nɪʳ

However, all that is highly speculative.

2nɪʳ could be an unrelated lookalike from a pre-Tangut source such as *Cʌ-nirH.

In any case, I cannot think of a way to derive 2nɪʳ from *√n-w within Tangut. If it is ultimately from that root, it would have to be a Chinese loanword.

*One might be tempted to reconstruct *-k since Maru has nuk⁵⁵ 'breast, milk', but Maru -k is an innovation. A BOVINE DYNASTY? (PART 2)

Guillaume Jacques (2010) equated the second half of ngo.snuHi, the Tibetan transcription of the name of the mythical first Tangut emperor, with Tangut

2niụ < *S-nuH 'to give milk'

In 2008, Guillaume rejected the temptation to go further and equate Tibetan s- with pre-Tangut *S- (his *s-):

Une hypothèse plus audacieuse pourrait être de voir dans ce s- une notation du préfixe causatif *s- qui doit se reconstruire pour ce verbe. En tangoute, nju.² [= 2niụ]  est dérivé de nju² [= 2niu]  (#4614) 'boire du lait'; le préfixe causatif *s- a disparu, laissant comme seule trace la 'voix tendue' notée par un point en dessous de la voyelle (Gong 1999). Cette hypothèse, toutefois, est très improbable dans la mesure où elle supposerait que soit conservée dans la graphie tibétaine une prononciation du tangoute plus ancienne que le système reconstruit à partir des dictionnaires du XIIème siècle, et donc antérieure d’au moins quatre cent ans aux textes tibétains eux-mêmes.

He regarded the Tibetan s- as "un simple artifice orthographique" since

dans le tibétain central du XIVème siècle, les consonnes préinitiales étaient déjà probablement confondues, voire amuies

but I wonder if ngo.snuHi reflects a nonstandard 14th century Tangut dialect which preserved pre-Tangut *S-. Tangut may have been internally diverse, and this dialect may have been to 12th century standard Tangut what modern Cantonese (which preserves final stops) is to Tangut period northwestern Chinese (which lost final stops) or what Ladakhi (which preserves some s-clusters: examples here and here) is to 14th century central Tibetan.

The -Hi may reflect a -j which was another trait of this 14th century Tangut dialect. Summing up the differences between the two types of Tangut and their common parent pre-Tangut:

Word cow to give milk
Pre-Tangut *ŋwə(-j)-H *S-nu(-j)-H
Standard Tangut 2ŋwɪ 2niụ
Later nonstandard Tangut ŋwə or ŋ(w)o snuj
Tibetan transcription ngo snuHi

The standard dialect had a *-j suffix in 'cow' absent from the nonstandard dialect. Conversely, the nonstandard dialect had a *-j suffix in 'to give milk' absent from the nonstandard dialect.

It is remotely possible that the -iụ of standard 2niụ could be a metathesis of *-u-j rather than a breaking of *u. But even if that were true - and I don't think it is - many or even most -iu could not be from *-u-j, as -iu regularly corresponds to Japhug rGyalrong < Proto-rGyalrong *-u (Jacques 2004: 143, 2006: 16-17). Moreover, if such a metathesis had occurred in standard Tangut, I would expect Chinese *-uj or perhaps even *-wi to correspond to Tangut -iu in very early loans. No such loans have been identified.

In any case, *-j cannot go back very far because probable cognates lack it, and nothing else leads me to believe that Tangut preserved a *-j lost elsewhere.

Next: A *n-w word family? A BOVINE DYNASTY? (PART 1)

Guillaume Jacques (2010) equated 2339, the first syllable of the Tangut imperial surname

2ŋwɪ 1mi,

with its homophone (and near-homograph) 0395 2ŋwɪ 'cow':


The shared center and right components are phonetic. The surname tangraph has 'sage' on the left, whereas 'cow' has the center of 'bear' according to Precious Rhymes of the Tangraphic Sea:


0395 2ŋwɪ 'cow' = 5605 2riẽ 'bear' + 2139 2ŋwɪ 'a kind of bird'

Without looking outside Tangut, I could reconstruct the pre-Tangut source of 2ŋwɪ 'cow' as

*Cʌ-ŋwiH (if the -w- is original) or

*Pʌ-ŋiH (if the -w- is from a presyllable)

with a low presyllabic vowel to condition the lowering of *i. However, it is unlikely that the root vowel was once *i.* Probable external cognates such as Old Chinese 牛*ŋʷə 'cow' and Written Burmese nvāḥ (< *ŋwaH?*; many more here) point to a nonfront vowel. This word was borrowed into southwestern Tai as *ŋuaA 'ox'**.

I used to reconstruct the rhyme of 2ŋwɪ 'cow' as -əi. Could 2ŋwɪ or 2ŋwəi be from *ŋʷə-i-H?

The name of the mythical first Tangut emperor was transcribed as ngo.snuHi in Tibetan. Guillaume Jacques identified that as Tangut

0395 4834 2ŋwɪ 2niụ 'the cow gives milk /  [someone] fed milk by the cow'

whose meaning was rendered in Tibetan as

ba-la Ho.ma Hthung-ba

cow-DAT milk drink-NMLZ

'he who drinks milk from the cow'

ngo might be a transcription of a nonstandard Tangut *ŋwə without my proposed suffix *-i. There is no character for schwa in the Tibetan script, so Tibetan o might represent a schwa. It is also possible that a pre-Tangut *ŋwə could have become *ŋo in that dialect (whereas *-wə did not fuse into -o in standard Tangut).

*Was *ŋw- > *nw- a regular change in Proto-Lolo-Burmese? I can't remember if my unpublished reconstruction from twenty years ago had either cluster. Matisoff's (1972) reconstruction has only one word with *ŋ(w)- which has a variant initial *mw- (not *nw-!).

**Do variants with w/v- and h- reflect different sources of borrowing? See Gedney's list of forms in Hudak 2008: 95. Unfortunately I could not find the word in Pittayaporn's  2009 dissertation on Proto-Tai. It may have been excluded because it could not be reconstructed at the Proto-Tai level.

Tangut fonts by Mojikyo.org
Tangut radical and Khitan fonts by Andrew West
Jurchen font by Jason Glavy
All other content copyright © 2002-2014 Amritavision