Gong reconstructed palatal medials for both Tangut Grade II and III: -i- and -j-. This fits well with the fact that alveopalatal (hereafter 'palatal') initials

tɕ- tɕh- dʑ- ɕ- (from Homophones chapter VII [palatals])

ʑ- (from Homophones chapter IX [liquids])

occurred after Grade II and III rhymes but not after Grade I rhymes which lacked palatal medials.

Li Fanwen (1986: 45) also reconstructed a palatal nasal ɲ- before rhymes that Gong reconstructed as Grade II and III.

How can this distribution be reconciled with my reinterpretation of Grade II as pharygealized? I thought there would be a conflict because palatal initials in Late Old Chinese developed only in nonpharyngealized syllables. If Tangut was like LOC, it shouldn't have had palatal initials in Grade II, though it obviously did. But perhaps Tangut was like Cairene Arabic, which has both nonpharyngealized and pharyngealized "palatals" (Youssef 2006: 13):

nonpharyngealized ʃʒj

I am confused by Youssef's use of IPA nonpalatal symbols for his "palatals". For example, it's not clear to me whether his ʃ represents a postalveolar [ʃ] or a alveopalatal [ɕ]. (I assume the 'sh' sound is not a true palatal [ç].)

According to Youssef,

"... all [Cairene Arabic] consonants except [q] have both plain and pharyngealized counterparts. The difference can be clearly heard next to low vowels: all consonants are plain next to [a] and pharyngealized next toʕ], including the pharyngeals themselves*."

Similarly, perhaps all Tangut initial consonants had two or three allophones depending on the following rhyme:

Grade I: plainGrade II: pharyngealizedGrade III: palatalized
Palatals(absent)ɕʕ-ɕ- (no need for a superscript j since these initials were inherently palatal)

These three types of consonants were not distinguished in fanqie.

I have wondered if the pharyngealized palatals of Grade II were postalveolar or retroflex: e.g., [ʃʕ] or [ʂʕ]. If they were, Grade II and Grade III fanqie should have nonoverlapping sets of palatal initial spellers: e.g., the speller for [ʃʕ] or [ʂʕ] (Grade II) should not be used in Grade III fanqie, and vice versa. But I was able to find mixed-grade spellers for palatal-initial tangraphs of both grades:

Initial (ignoring pharyngealization/palatalization)tɕ-tɕh-dʑ-ɕ-ʑ-
Grade II speller for Grade III tangraphTT3303TT3582TT4073(no Grade II ɕ-spellers?)TT2295?**
Grade III speller for Grade II tangraphTT0838, 2928TT5428TT4990TT0962, 2773, 3699TT1564

(I have excluded Li Fanwen's [1986] ɲ- since I used a fanqie database based on Gong's reconstruction which lacks that initial.)

If there was a difference between Grade II and III palatals, it must have been very subtle (e.g., [ɕʕ] vs. [ɕ] rather than [ʂʕ] vs. [ɕ]).

*I am skeptical about the existence of 'plain' pharyngeals.

**Although Gong reconstructed TT2295 as lhie, it appears as an initial speller for Grade III tangraphs that Gong reconstructed with ʑ-, implying that TT2295 should be reconstructed as ʑie 2.8 (or ʑeʕ in my revision). But others have reconstructed TT2295 with ʑ-like initials:

Sofronov (1968 II: 328): z̀ê 2.8

Li Fanwen (1986: 119) ʑi 2.8

(I presume the zi in Li Fanwen [1986: 478] is an error for ʑi.)

However, TT2295 ?ʑie 2.8 was listed in Homophones as an isolated tangraph without homophones (55A41) instead of the ʑie 2.8 homophone group (52B37-38). Perhaps TT2295 had different initials in the dialect of the author of the Tangraphic Sea fanqie and the dialect of the author of Homophones. THE GRADE II M-I-STERY: THE PHAR-I-NGEAL SOLUTION (PART 2)

If Tangut Grade II words had pharyngealized vowels, one might also expect pharyngealized initials, as in my reconstruction of Old Chinese:

'Emphatic' syllables: *CʕVʕ

Nonemphatic syllables: *CV

Jonathan Evans' (2006) account of pharyngealized vowels in Hongyan Qiang does not mention pharyngealized initials, so perhaps that language really does combine nonpharyngealized initials with pharyngealized vowels. (But maybe it has pharygealized allophones of consonants before pharyngealized vowels:/CV ʕ/ > [CʕVʕ]?)

Pharyngealized initials would not appear in Tangut Grade I or Grade III which had nonpharyngealized vowels. Therefore they would be spelled by Grade II tangraphs in fanqie, and Grade II fanqie should never contain Grade I or III initial spellers. However, 'mixed-grade' initial spellings do exist: e.g.,

TT5057 ɣeʕ R35 1.34 (Grade II) =

TT0875 ɣjɨ 1.29 (Grade III) + TT1328 kheʕ R35 1.34 (Grade II)

TT0875 is also the initial speller for TT3615 ɣuʕ R4 1.4, which I regard as Grade II. Some R4 1.4 tangraphs had Grade I initial spellers: e.g.,

TT5761 kuʕ R4 1.4 (Grade II?) =

TT4077 R28 1.27 (Grade I) + TT4725 ɣuʕ R4 1.4 (Grade II?)

Here's a clearcut case of a Grade II tangraph with a Grade I initial speller:

TT5476 ɣiʕr R83 1.78 (Grade II) =

TT5576 ɣur R80 1.75 (Grade I) + TT3928 kiʕr R83 1.78 (Grade II)

(TT5476 'die' may be derived from TT3928 'harm' (Grinstead [1972: 147], Nevsky [1960 II: 35]: вредить [what is his second Russian gloss?]): *CV-kiʕr? Could TT5576 ɣur 'corpse' also be part of the same word family whose root might have been *rq-, *qr-, or *q-r?)

1.11.00:55: But Shi et al. (2000: 172) defined TT3928 as 魘 'nightmare' which is harder to relate to 'die' and 'corpse'.

TT5476 was also the initial speller for its Grade I initial speller.)

There are also examples of Grade II tangraphs with Grade II initial spellers: e.g..,

TT3861 ɣiʕw R45 1.44 (Grade II) =

TT3409 ɣiʕ R9 1.9 (Grade II) + TT5271 kiʕw R45 1.44 (Grade II)

But even its Grade II initial speller had a Grade III initial speller: TT5688 ɣju 1.3.

This table sums up the possible initial spellers for tangraphs of each grade, based on a very quick check of random fanqie:

Grade I tangraphGrade II tangraphGrade III tangraph
Grade I initial spelleryesyesyes (e.g., TT2890, 4139)
Grade II initial spelleryesyesyes (e.g., TT3928, 5388)
Grade III initial spelleryesyesyes (in most cases?)

Did Tangut have only one set of consonants which appeared in all grades?

I had a difficult time quickly finding Grade I and II initial spellers for Grade III tangraphs. This might imply that GIII initials were somehow different from GI and GII initials (e.g., GIII initials were palatalized?), yet I could easily find GI and GII initial spellers for GIII tangraphs. I don't know what to make of this apparent asymmetry. (I say "apparent" because maybe there are more GI and GII initial spellers than I think, or 'mixed-grade' fanqie are rarer than they seem to be at a glance.) Ideally, I should fill the above table with statistics rather than 'yes' or 'no'. DOES 'TRUTH' TRULY HAVE A UVULAR INITIAL?

Arakawa reconstructed


as qje R35 1.34 in "Study of the Tangut version of "Prajñāpāramitā-hrdaya-sūtra with a commentary" preserved in Russia" (2006). I was surprised to see an initial q-, since q- was not in the list of initials in Arakawa's "A Study of the Tone of Tangut" (1999). (But it is in Kotaka Yuuji's list of Arakawa's initials.)

In "Tangut Fragments preserved in the University of Tokyo" (2003), Arakawa reconstructed


as qjị R69 2.59 with a uvular initial.

Kotaka's list of Arakawa's reconstructions contains three more q-initial words:

TT0134 BREATH qjị R69 2.59

TT0224 OLD MAN qo R51 1.49

TT3772 碍 ?OBSTACLE  qjị R69 2.59

According to Kotaka, Arakawa's q- corresponds to Nishida's ɣk-, though Nishida (1966: 336) reconstructed TT0224 with ɣw-. There is no agreement on how to reconstruct the initials of the five aforementioned q-words which are all from Homophones' eighth chapter of 'throat sounds':

ArakawaNishidaGongMy guessTibetan transcriptionTangut period NW Chinese transcriptionNotes
q-ɣk-ɣ-?ɣ-Hg-, dg(h)-, bg-*ji k(h)-Homophones groups VIII.28, 52, 53
ɣw-ʔ-?ɣw-(none)*ŋgw-TT0224 in Homophones group VIII.84

q- doesn't match the voiced initials in the Tibetan and Chinese transcriptions.

I chose ɣ(w)- because

g- is in Chapter V (velars)

Tibetan gh- could be an attempt to transcribe ɣ-

ɦ- would have been transcribed in Tibetan as H- or h-, not Hg-, dgh-

However, ɣ- doesn't match the Chinese transcription *ji k(h)-. I wondered if these could be fanqie spellings for something like ?*je, but all those transcriptions represented Grade II syllables -

Homophones groupRhymeArakawaNishidaGongMy guessTibetan transcriptions (Nishida 1964)Chinese transcriptions (reconstructions of the rhymes of the second tangraphs are uncertain)
VIII.28R35 1.34qjeɣkjɛɣiej?ɣeʕ [ɣɛʕ]Hgi, Hge (Li Fanwen 1986: 110 also listed dghe, bge)夷皆 *ji k-
VIII.52R9 1.9?qjiɣkɪěɣie?ɣiʕ [ɣɪʕ]Hgi, dghi夷隔 *ji k-
VIII.53R69 2.59qjịɣkɪ̣ěɣiẹ?ɣịʕ [ɣɪ̣ʕ]dghi, Hge
VIII: isolated tangraphs without homophonesR45 1.44?qjewɣkewɣiew?ɣịʕw [ɣɪ̣ʕw](none)夷隔 *ji k-, 夷客 *ji kh- (which may have ended in *-j and did not end in *-w!)
R63 2.53?qjẹɣkjɛ̣ɣiẹj?ɣẹʕ [ɣɛ̣ʕ]夷 耿 *ji k-
R83 1.78?qjirɣkɪěrɣier?ɣiʕr [ɣɪʕr]夷 格 *ji k-

(table added 1.4.1:01; Arakawa reconstructions with question marks are my guesses)

- and j- is associated with Grade III rather than II in Gong's reconstruction (and in my tentative revision of it).

1.4.1:04: All but one example of Arakawa's q- occurs before j-, and if Arakawa's q- is always equivalent to Nishida's ɣk-, then it frequently precedes j-. Both Arakawa's qj- and Nishida's ɣk- are unlikely initial clusters. Does any language have them? THE GRADE II M-I-STERY: WHERE IS GRADE II -U?

In the previous post, I asked,

why is there no /uʕ/ = [ʊʕ] [in my revision of Gong's Tangut reconstruction]?

I suspect that the missing /uʕ/ is Gong's second -u (rhyme 4: 1.4/2.4). Gong's reconstruction has no way to distinguish minimal pairs such as

TT4139 ku 1.1 'therefore' (R1: Gong's first -u rhyme)

TT5761 ku 1.4 'phoenix' (R4: Gong's second -u rhyme)

I would distinguish them as

TT4139 ku 1.1 'therefore'

TT5761 kuʕ 1.4 'phoenix'

R1 is much more common than R4. It occurs after a larger variety of initials than R4 and can be preceded by -w-, unlike R4:

RhymeLabialsDentalsAlveolar fricatives/affricatesLiquidsVelarsGlottals
R1 + no medialp-, ph-, b- (but not m-!)t-, th- (but not d-!), n-ts-, tsh-, dz-, s-, z-l-, lh-k-, kh-, g-, (but not ŋ-!), x-, (but not ɣ-)ʔ-
R1 + -w-(no labial-w-clusters in Tangut)thw-, dw-, nw-tshw-, sw-lw-, lhw-ŋw-(ʔw- if TT4635 belongs to R1)
R4 (no medial)m- (only one example each in rhymes 1.4, 2.4)d- (12 examples)nonek-, kh- (but not g-), x-, ɣ-none

It is therefore most likely that R1 was a simple -u, whereas R4 had some other extra feature: e.g., pharygealization.

The fact that R4 occurs almost exclusively after velar initials fits with my pharyngeali

zed hypothesis if those velars came from uvulars that conditioned pharyngealization:

*qu > *quʕ > kuʕ R4

*qhu > *qhuʕ > kuʕ R4

*ɢu > *ʁuʕ > ɣuʕ R4

*χu > *χuʕ > xuʕ R4


*ku > *ku > ku R1

*khu > *khu > khu R1

*gu > *gu > gu R1

*xu > *xu > xu R1

TT3619 ɣuʕ R4 1.4 < ?*ɢu 'head' seems to be cognate with Old Chinese 后 *g [ɢoʕʔ] 'ruler' < 'head' (as suggested by Guillaume Jacques in "Об этимологии слова 后").

The trouble is that ɣuʕ R4 1.4 also seems to be cognate with

Japhug rGyalrong tɯ-ku

Somang rGyalrong tə-kó

Zbu rGyalrong tə-kuʔ

< all Proto-rGyalrong?*ko

whose root initial is velar, not uvular.

The Zhongu dialect of Tibetan which may preserve earlier uvulars also has a velar initial: ŋgo (cognate to mgo in Written Tibetan which lacks uvulars).

(But Zhongu does not have ɴɢ-, so perhaps ŋgo was once *ɴɢo. Also, the Qiangic language Pumi does have a uvular initial: qho. Could something have caused a Proto-Sino-Tibetan uvular initial to front to a velar in Proto-rGyalrong?)

The vowel correspondence of uʕ R4 1.4 with

Proto-rGyalrong *o

(also supported by a second cognate in Jacques 2003:

'pine': Japhug kɯ-ɴɢu, Zbu kə-ɴɢwə < *-o : Tangut ku R4 1.4 < ?*quʕ)

Zhongu and Written Tibetan -o

Pumi -o

(i.e., a pharyngealized high vowel corresponding to a nonpharyngealized mid vowel) has a parallel in Qiang:

Hongyan Qiang iʕ : Mawo Qiang e

(One would expect Hongyan uʕ to correspond to Mawo o, but in fact Hongyan u and uʕ both correspond to Mawo u.)

This high : mid pattern is reminiscent of the lowering of pharyngealized high vowels in Old Chinese:

Early Old Chinese *iʕ > Late Old Chinese *-ej

Early Old Chinese *uʕ > Late Old Chinese *-ow

If R4 is supposed to reflect an earlier uvular, where did the m- and d-initial R4 words come from? Did they originally have initial clusters with uvulars? But surely there were clusters with uvulars and other initials as well. One could try to ignore the two m-words as errors, but 12 examples of d-initial R4 words cannot be easily dismissed.

Furthermore, if Grade II was pharyngealized, then R4 is a Grade II rhyme, and Grade II rhymes should have initial distribution patterns similar to R4's: i.e., they should occur almost exclusively after velars. But they don't: e.g., there is no puʕ, but there is a pịʕ 1.66 (TT4473 'frog'; Gong's piẹ 1.66).

There is a remote chance that 'frog' once had a uvular in it: ?*q-pi; cf. Japhug qaɕpa 'frog'. (The correspondence of Japhug a to Tangut i is regular.) Does the tense pharyngeal vowel (does such a vowel exist in any living language?) reflect an earlier *qp-cluster? Jacques (2003) found only three other Japhug cognates for Tangut Grade II tense vowel words. Two out of three of the Japhug cognates have uvulars, but such a small sample is not meaningful:

'boil': kɤ-sqa : TT5496 ɣịʕ 1.66 (Gong ɣiẹ) < ?*CV-sqi

(*CV-prefix needed to explain *q- > ɣ-lenition)

'willow': qa-ʑmbri : TT3469 bịʕ 1.66 (Gong biẹ) < ?*q-bi

but note also TT0983 ʕ 1.28 'willow' whose vowel is not tense

'ear of grain': kɯ-ɕnom : TT1704 nọʕ 2.63 (Gong niọ) < ??*q-no

but there is no way to tell whether the pre-Tangut prefix was uvular or not

There is no special correlation between Japhug uvulars and Tangut Grade II. Japhug words with uvulars can be cognate to Tangut Grade I and III words: e.g.,

'needle': ta-qaβ : TT2597 ɣa 1.17 (Grade I)

'boil': kɤ-sqa : TT0581 ɣjii 1.14 < ?*CV-qi (Grade III; cf. Grade II TT5496 ɣịʕ 1.66 'boil' above - a rare mixed Grade II/III word family)

Conversely, 9 or 11* out of 18 Japhug cognates of Tangut Grade II lack uvulars (e.g., 'ear of grain' above). If there is any connection between uvulars and Grade II (or even just R4), it is not straightforward.

*22:14: This depends on whether one counts Japhug words ending in uvulars as 'having uvulars'. I doubt that uvular codas had anything to do with Tangut Grade II. Out of the 23 Tangut cognates for Japhug words ending in < Proto-rGyalrong *-q, only two belong to Grade II:

'plowshare': qraʁ : TT2573 khaʕ 2.15 (Gong khia)

'mud': tɤ-rcoʁ : TT4256 tɕoʕr 1.90 (Gong tɕior)

The Tangut reflex of an earlier *-q seems to be zero rather than Grade II.

22:52: The distribution of velar initials before R1 and R4 has unexpected imbalances:

R1 -u18101600
R4 -uʕ (Gong: -u)1030 (since *ɢu would have become ɣuʕ)0215

First, the only R1 ku word is TT4139 'therefore'. All other ku are R4. It is unlikely that kuʕ (or pre-Tangut *qu) would outnumber ku by a ratio of ten to one. I would expect kuʕ R4 to merge with *ku R1 as ku R1, not the other way around. TT4139 'therefore', being a common word, might be an archaism that escaped the merger.

Second, there are more ŋwu words than any other kind of velar-u(ʕ) words. And there are no other velar-wu(ʕ) words at all, except for five rising tone words with tense vowels: one kwụ R61 2.51 and four khwụ R61 2.51. I have no explanation for this.

The absence of ŋuʕ R4 implies that pre-Tangut had no uvular initial *ɴ-. (Does any language have ɴ- as a single initial, as opposed to clusters like Japhug ɴq- and ɴɢ-?)

Finally, if my lenition hypothesis is correct, I would expect pre-Tangut *CV-Ku (K = any velar obstruent) to become ɣu R1. Yet Tangut has no ɣu R1 and 15 ɣuʕ R4. It is unlikely that pre-Tangut only had *CV-Qu (Q = any uvular obstruent) but no *CV-Ku. If *CV-Ku merged with *CV-Qu, I would expect the result to be ɣu R1 rather than ɣuʕ R4. ɣ- is not exclusively a Grade II initial; see above for examples of Grade I and III words with ɣ-. THE GRADE II M-I-STERY: THE PHAR-I-NGEAL SOLUTION (PART 1)

This is my current phonemic reinterpretation of Gong's reconstruction of the Tangut vowel system. My alterations are in bold.

GradecyclelengthGong's vowels
I: base vowelslaxshortieəaou
II: pharyngealized, lowered vowels
(Gong's iV diphthongs)
III: raised vowels preceded by -j-laxshortjijejajoju

(Gaps that may be random are labelled with 'none'. Systematic gaps are labelled 'none!'.)

The changes can be grouped into three categories:

1. The biggest change is the reinterpretation of Grade II in terms of pharygealization. In phonetic terms, the basic Grade II vowels might have been:

Phonemic/iʕ//eʕ/ʕ//aʕ//oʕ/(why is there no /uʕ/ = [ʊʕ]?)
Gong's reconstructionieiejiaio(there is no iu in Gong's reconstruction)

2. High and mid nonback vowels are in complementary distribution in Gong's reconstruction:

GradeFront vowelCentral vowel

There is no j-less i or ɨ in Gong's reconstruction.

My phonemic reconstruction ignores height which is automatically predictable on the basis of the presence of pharygealization and -j-, which are in complementary distribution:

GradeFront vowelCentral vowel
I/i/ [ɪ]/ə/[ə]

My intepretation of R8-R9 as /i/ and /iʕ/ fits the transcriptive evidence better:

RhymeGongMy reconstructionTranscribed in Tibetan asUsed to transcribe Sanskrit
8-e-i(no data)(not used, perhaps because Tangut /i/ was phonetically [ɪ] [which was also Nishida's reconstruction]; Sanskrit e is always long: [ee])
9-ie-iʕ-i (not -ye)(not used; Sanskrit has no -ie or -iʕ)

3. I have reinterpreted Gong's ej / iej / jij as e / e / je:

GradeGongMy reconstruction

The transcription evidence for R36-R37 (Gong's -jij and -jij) tends to indicate a mid vowel:

RhymeGongMy reconstructionTranscribed in Tibetan asUsed to transcribe Sanskrit
36-jij-je-e (mostly), -i (rarely)-e (mostly), -i (rarely)
37-e (mostly), -u, -o (rarely; reflecting an -o suffix added to -e roots?)(not used)

A raised allophone of e would be close to [ɪ]. THE GRADE II M-I-STERY: THE BRAHMI EV-I-DENCE

Pulleyblank (1984) reconstructed long vowels in Late Middle Chinese Grade II with a preceding *-j- after velar initials. This *-j- corresponds to the *-i- that Gong reconstructed after all initials in Grade II of Tangut period NW Chinese.

However, there is no -y- corresponding to Pulleyblank's *-j- and Gong's *-i- in the Khotanese Brahmi transcriptions of Grade II syllables in a NW LMC text (Emmerick and Pulleyblank 1993):

SinographKB transcriptionPulleyblank's LMC reconstructionTPNWC (Gong-style reconstruction)Notes
hva:*xwaaă*xiwaKB h-:- represented initial [x] (E&P 1993: 32)
ka', keyi, ke'yi*kjaaj*kiejKB ' represented "some sort of prosodic feature such as breathiness" (E&P 1993: 55) - but no breathiness is reconstructible for this Chn word
keyiborrowed into Tangut as TT5813 kie 1.9
borrowed into Tangut as TT1882 kiẹj 2.53; 戒 and 界 were homophones in Chn, so I don't know why they have different rhymes in Tangut - were they borrowed from different Chn dialects and/or at different points of time?; the tense vowel of TT1882 is a trace of a Tangut prefix
pa'hä:*paajk*piejKB -h-: may represent [x] (E&P 1993: 4); it seems to reflect a late LMC *-x which disappeared in TPNWC
phihä:, phehä:*pɦaajk*phiej
katä*kjaaw*kioKB ä = [ə]; I don't understand why KB -tä is a common transcription for Chn *-w (and to a lesser extent, the Chn rising tone < *-ʔ - cf. the use of Jpn ッ tsu for [ʔ] as a coda)
ha:bhuuä*ŋjaaw*ŋgioKB h-:- reflects a denasalized Chn initial (E&P 1993: 40): possibly *ɣ-?; E&P (1993: 45) interpreted -bhuuä as [βu]
haṃ':ŋä*xɦjaawŋ*xio(w)borrowed into Tangut as TT4425 ɣiow 1.55; the voiced initial probably reflects borrowing before EMC *ɣ- devoiced to *x- rather than lenition following a lost presyllable: *CV-x- > *ɣ-
parä*paat*piaLMC *-t had weakened to late LMC *-r which disappeared in TPNWC; cf. Sino-Korean 팔 phal < pharʔ, borrowed from an eastern (?) dialect of late LMC
cha*tʂhaaă*tʃhiaKB ch = [tʃh]; this transcription cannot tell us whether NW Chn had *-j-/-i- because it would be understandable if NW Chn *tʃhja or *tʃhia were transcribed as cha instead of chya

(Khotanese aa does not correspond to Pulleyblank's *aa, but this is because Khotanese aa was [ɔ], not [aa].)

Since TPNWC postdates the dialect in the Khotanese Brahmi transcriptions, one could argue that late LMC *a (phonetically [æ]?) broke to TPNWC *ia. (Cf. Jpn ya for Eng [æ] after velars: cat > kyatto.) But is it really necessary to reconstruct an *-i- in TPNWC? It corresponds to zero in the Tibetan transcriptions of the Tangut rhymes corresponding to Chinese Grade II rhymes:

TPNWC Grade II rhyme vocalism according to GongGong's Tangut Grade II rhymeTibetan transcription
*iViVV (not yV, iyV, or iHV; iV is not possible in Tibetan script)

I prefer to assume that Grade II in both TPNWC and Tangut had some characteristic which did not sound like a Tibetan i or y and which was ignored in Tibetan transcription. THE GRADE II M-I-STERY

Gong's reconstruction of Tangut has three 'grades':

Grade I: no -i-/-j- medial

Grade II: medial -i-

Grade III: medial -j-

Medial -w- can be found in all three grades: e.g.,

GradeNo medial -w-Medial -w-

I am skeptical about Gong's interpretation of Grade II as -i(w)-. Nonetheless, I will continue to use -i(w)- to symbolize Grade II, even though I do not take that representation literally.

-j- appears to be an infix that generally occurs before u. (Could the alternation that Gong reconstructed as -u ~ -ju really have been -u ~ -y?)

There are no alternations indicating that -i- was also an infix. Gong did not find any alternations between Grade II and non-Grade II rhymes. I only know of one such alternation:

TT1791 kha 2.14 'bitter (plant)'

TT0608 khe 1.8 'bitter'

TT0705 khie 1.9 'bitter'

TT2642 khieej 1.38 'eighth Heavenly Stem' (whose Chinese name can also mean 'bitter')

(the first three from Jacques [2003: 9]; cognate to Written Tibetan kha < *qh- [cf. Zhongu and other Tibetan dialects with uvulars] and Old Chinese 苦 *kha [qhɑ])

Gong's article "Phonological Alternations in Tangut" only lists 12 alternations between Grade II words:

Alternations 1-2: voiced stop ~ voiceless aspirated stop

did nasal prefixes voice stops?

why are there no voiced stop ~ voiceless unaspirated stop alternations?

1. #2

TT3828 bie 2.8 'to release, to open'

TT2804 phie 2.8 'to release, to open, to untie'

2. #7

TT0811 bia 1.18 'split, explode'

TT3822 phia 1.18 'split, divide in two'

Could these four words be cognate? I would need to find other examples of -e ~ -a alternation to be sure.

Alternations 3-6: voiceless unaspirated stop ~ voiceless aspirated stop

3. #30 + 41 (also contains -ie ~ -io alternations)

TT0610 kie 2.8 'hate, dislike'

TT0607 khie 1.9 'hate, dislike'

TT4440 kio 2.43 'hate, dislike' < ?*kie-o-H

TT4443 khio 1.50 'hate, dislike' < ?*s-kie-o

I assume that there were two waves of *s-prefixation:

- the first resulted in aspirates: *s-C- > Ch-

- the second resulted in tense vowels: *s-CV > *CṾ

4. #32

TT3023 kia 1.18 'firm, secure'

TT4627 khia 2.15 'firm, secure, sincere'

5. #39

TT5271 kiew 1.44 'collapse, crumble down'

TT2796 khiew 1.44 'destroy, pull down'

6. #42

TT3582 tɕioow 1.57 'to steal, to rob'

TT1727 tɕhioow 1.57 'to steal, to rob'

Alternation 7: zero ~ -w-: #63

TT3409 ɣie 1.9 'strength, force, power'

TT2761 ɣiwe 1.9 'strength, force, power'

Alternation 8: -e ~ -o: #95 (also see #41 above and #259 below)

TT5312 wier 1.78 'value, love'

TT5316 wior 1.90 'value, love'

Alternations 9-12: level ~ rising tones:

9. #216

TT2619 niəj 1.41 'to be unclean'

TT4536 niəj 2.36 'dirt, mucus'

10. #247+259 (also contains -iej ~ -io alternations; see TT2642 for another example of -j suffixation in Grade II)

TT0353 kiej 1.34 'compel to come' < ?*kie-j

TT0345 kiej 2.31 'compel to come' < ?*kie-j-H

TT1432 kio 1.50 'compel to come' < ?*kie-o

TT1929 kio 2.43 'compel to come' < ?*kie-o-H

11. #261

TT5547 tɕiow 1.55 'to assemble, to amass'

TT3892 tɕiow 2.48 'many'

12. #269

TT2350 ʔwiọ 1.71 'round, yard, enclosure, hall, college'

TT0792 ʔwiọ 2.63 'garden, orchard'

I have followed Sofronov and Shi et al. (2000) in reconstructing w. If that is correct, these words may be borrowed from Late Old Chinese 圓 *wien 'round' and 園 'garden'*wuon (cf. Sino-Japanese on < won), though a Chinese origin cannot account for the glottal initial (a trace of an earlier prefix *ʔV-?)

(I have discarded #281 since the surnames in it may or may not be cognate.)

For vowel length alternations in Grade II, see Gong, "A Hypothesis of Three Grades".

It seems that Grade II is an inherent quality of Tangut roots that cannot be altered by derivational processes.

Comparison with Japhug rGyalrong has not been able to elucidate the nature of Grade II. Guillaume Jacques found only 18 gDong-brgyad rGyalrong cognates for Grade II words:

gDong-brgyad rGyalrong rhymeProto-rGyalrong sources of DBG rhymesNumber of cognates in Tangut grades
-a*-a(k), *-ɐ7629
-aʁ*-aq, *-ɐq, *-ɔq1514
-e*-aj, *-ej2214
-ɤɣ*-ɐk, *-ɔk, *-ek505
-ɤr*-ɐr, *-ɔr302
-ɤs*-ɐs, *-ɔs100
-ɤt*-ɐt, *-ɔt, *-et213
-ɤβ*-ɐp, *-ɔp, *-ɯp104
-i*-e, *-i(k)5336
-iɯ(no source listed in Jacques 2004; maybe *-ju?)100
*-ɯ, *-u(k)15215
-ɯɣ*-ɯk, *-uk, *-iq101
-ɯm*-om, *-ɯm, *-um, *-im004
-ɯr*-ɯr, *-ur, *-ir203
-ɯs*-ɯs, *-us, *-is102
-ɯt*-ut, *-it204
-ɯβ*-up, *-ip000
Total: 28710418165
% of total36%6%58%

Why are there so few Grade II words? Conversely, why are there so many Grade III words? One would expect more Grade I -(w)- words than Grade III -j(w)- words. The Grade III figure is inflated by the raising of pre-Tangut *a to ji, but there may be other causes for the imbalance.

I briefly considered the possibility that most Grade II words were Chinese loanwords that would not have gDong-brgyad cognates. Although some Grade II words correspond to Chinese words in the second grade of Chinese phonology* -

TT3651 pia 1.18 'scar' < 疤

TT2167 bioo 1.53 'cat' < 貓

TT3312 kiwã 1.25 'barrier' < 關

TT5469 kia 1.18 'price' < 價

TT1882 kiẹj 2.53 'world' < 界

does the tense vowel reflect a Tangut prefix added to a Chinese borrowing?

TT5813 kie 1.9 'precept' < 戒

TT3736 ɕiəj 1.41 'be born' < 生

TT2381 ɕioo 2.46 'double' < 雙

TT4161 ɕia 1.18 'sand' < 沙

TT1391 ɕiã 1.25 'mountain' < 山

TT3620 xiwa 1.18 'flower' < 花

TT2975 ɣiej 1.34 'shoe' < 鞋

TT4425 ɣiow 1.55 'surrender' < 降

TT3861 ɣiew 1.44 'study' < 學

TT0730 xiəj 2.36 'apricot' < 杏

- others are not: e.g., the various 'bitter' words (which would have been kho or khu if they had been borrowed from Middle Chinese or Tangut period NW Chinese).

None of the tangraphs that Grinstead glossed as 'Skt.' (i.e., for Sanskrit transcription) belong to Grade II. Therefore the distinguishing characteristic of Grade II was probably absent from Sanskrit.

The Tibetan transcriptions of Grade II rhymes have nothing in common:

Rhyme numberGong's reconstructionArakawa's reconstructionTibetan transcription from Nishida (1964)
13-iee-ji'-i, -iH, -u (! - reflecting an -o suffix?)
23-iaa-ja'-ar (the final -r is unexpected for the first cycle of nonretroflex vowels; did this transcrbe an alternate reading ending in -ar?)
35-iej-je-i, -e
52-io-jo-uH, -oH
57-iow-jõ-o, -ong
69-iẹ-jị-i, -a (! - error for -i?)
76-iə̣j-ẹ2-e? (or is this a transcription for R79 1.74, which Nishida grouped together with R76 2.65?)
83-ier-jir-i, -u (! - reflecting an -o suffix?)

Gong's -i- and the -j- that Arakawa reconstructed in most of Gong's Grade II rhymes should correspond to Tibetan -y-, but in fact Tibetan -y- was never used to transcribe Grade II. This suggests that Tangut Grade II (and by extension, Tangut period NW Chinese Grade II which mostly arose from pharyngealized consonant clusters with *r?) had a distinguishing feature that could not be written in the Tibetan alphabet. The presence of Grade II in the tense and retroflex vowel cycles indicates this feature must have coexisted with tenseness and/or retroflexion.

I have thought this feature might have been vowel velarization (as in some rGyalrongic languages) or pharyngealization (as in Hongyan Qiang, which has pharyngealized retroflex and pharyngealized nonretroflex vowels; see Evans 2006). Does any language have pharyngealized tense vowels?

A less exotic possibility is a class of vowels that were lower/fronter than their closest Tibetan counterparts: e.g.,

Gong's Grade II diphthongMy guessAlternate guessTibetan transcription
ieɪ?iʕ (close to [ɪ])i
iaæ? - the value of Grade II *a in southern Chinese; it might have been backer in the NW: ɑ?aʕ (close to [ɑ])a
ɘ/ɤ? (an upper mid central or back unrounded vowel) ɨʕ (close to [ɘ], [ɤ])i, e
ioɔ? (but this conflicts with the u in the Tibetan transcriptions!)oʕ (close to [ɔ])u, o

Note that pharyngealized vowels are lowered than their plain counterparts.

*The reconstruction of the second grade in Tangut period NW Chinese is uncertain. Gong reconstructed the second grade of Tangut and TPNWC as -i-. INFIXES FROM SUFFIXES?

I believe that Sagart's (1999) Old Chinese infix *-r- originated as a prefix:

*r-C- > *Cr-

Even after this metathesis occurred, new *Cr-forms could have been created from *C-roots by analogy with*C- ~ *Cr-alternations originating from earlier *C- ~ *r-C-alternations.

I wondered if the apparent Tangut infix -w- in word families such as

TT3089 tsja 1.19 'hot'

TT5505 tsja 1.20 'hot'

TT5499 tshja 1.20 'hot'

TT5498 tshjwa 1.20 'make hot, to heat'

could have had a similar origin:

*P-C- > Cw-

(*P- = an unknown labial prefix - or the vowel *u-?)

This hypothesis would exclude *P- as a possible source of tense vowels.

Could the Tangut infix -j- also have originated from a prefix?

*T-C- > Cj-

(*T- = an unknown acute prefix - or the vowel *i-?)

This hypothesis would exclude *T- as a possible source of tense vowels. I presume that *T- was not *s-, which is one likely source of tense vowels (see Gong, "西夏語的緊元音及其起源").

Gong ("The Phonological Reconstruction of Tangut") found alternations indicating a suffix *-j:

R8 -e ~ R34 -ej

R9 -ie ~ R35 -iej

R10 -ji ~ R36 -jij

R11 -ji ~ R37 -jij

(Gong hypothesized R12 -ee ~ R38 -eej on the basis of their placement in Tangraphic Sea, but did not find any examples)

R13 -iee ~ R39 -ieej

R14 -jii ~ R40 -jiij

Was *-j only added to roots ending in front vowels? There are no alternations like

R28 ~ R41 -əj

R30 -jɨ ~ R43 -jɨj

and there are no rhymes like -uj -oj -aj. Did *-j metathesize after nonfront vowels?

*-uj > -ju

*-oj > -jo

*-aj > -ja

Could this suffix *-j be the source of the -j- infix before nonfront vowels?

Similarly, there are no rhymes like -uw -ɨw -əw -aw.

Tangut nucleus + glide coda distribution patterns

coda \ nucleusieɨəauo

Did *-w metathesize after nonfront vowels other than *o?

*-uw > -wu

*-ɨw > -wɨ

*-əw > -wə

*-aw > -wa

These hypotheses predict that what appears to be glide infixation should be found after nonfront vowels. -j-alternation occurs almost entirely before nonfront vowels as expected, whereas -w-alternation occurs mostly before front vowels:

Vowels following -Ø- ~ -j- alternations (disregarding tenseness/retroflexion):

u: 12 cases

a: 1 case

o: 1 case

e: 1 case (-ẹ ~ -jị; -jẹ does not exist in Gong's reconstruction)

Nonfront vowels: 14 cases (almost all u)

Front vowel: 1 case

(I have discarded Gong's set #56 since both members [TT1608 and TT4931] have -j- and may even be homophonous.)

Vowels following -Ø- ~ -w- alternations (disregarding tenseness/retroflexion):

u: 1 case

e: 3 cases (including -ej x 2)

i: 6 cases (including -ij x 2)

a: 1 case

ɨ: 1 case

o: 1 case

There is also one case of -jiir ~ -jwɨɨr which could be grouped with i

Nonfront vowels: 4-5 cases

Front vowels: 9-10 cases

The very different distribution of -j- and -w-alternations could imply that they had different origins.

The *-j metathesis hypothesis might account for the -Ø- ~ -j- alternations, though it does not explain why pre-Tangut would have more *-u-j than *-o-j or*-a-j.

The *-w metathesis hypothesis cannot account for the high frequency of -Ø- ~ -w- alternations before front vowels. I would expect alternations like

(I = front vowel, A = nonfront vowel)

-I ~ -Iw < *-I-w

-A ~ -wA < *-A-w

but the actual pattern is

-I ~ -wI (2/3 of cases)

-A ~ -wA (1/3 of cases)

and there are no -I ~ -Iw-type alternations at all.

There are, however, alternations of the type -I ~ -jo implying a suffix *-o:

R10/11 -ji, R36 -jij ~ R53 -jo

R14 -jii ~ R55 -joo

R70 -jị ~ R75 -jọ

R85 -ier ~ R96a -ior

R86 -jir ~ R96b -jor

but note that

R34 -ej ~ R56 -ow

R38 -eej ~ R54 -oo < ?*-oow

R42 -iəj ~ R56 -iow

don't quite fit the pattern and look more like ablaut than suffixation

could -j ~ -w alternation reflect *-CI ~ *-CU alternation?

If a suffix *-o ever followed nonfront vowels, it has disappeared without a trace except after R7 -juu:

R7 -juu ~ R55 -joo < ?*juu-o

why doesn't R7 alternate with R50 -jwo, which looks like a likely fusion of *-juu with *-o?

why are there no cases of short -u alternating with -o rhymes?

I have been assuming that affixes could have combined with roots of all phonological types in the past. This need not have been the case: e.g., the Korean nominative postposition is -ka after vowels but -i after consonants. Pre-Tangut *-o might have occurred only after some vowels and a different pre-Tangut suffix might have disappeared after roots with other vowels.

Tangut fonts by Mojikyo.org
All other content copyright © 2002-2008 Amritavision