Home WERE R77-79 AND R102-103 RETROFLEX?

According to Gong they were, but Arakawa reconstructed R77-79 and R102 as tense and R103 as plain. If they were retroflex, they should have some Tibetan transcriptions with preinitial r- and/or syllables with initial r-. r- can only precede retroflex rhymes with some apparent exceptions that I will look into next time. R77-79 fit the criteria for retroflex rhymes, but R102-103 do not:

Rhyme AMR Gong Arakawa Tibetan transcriptions with preinitial r-? Syllables with initial r-?
R77 -eʳ -ejr -yeq2 yes (see Nishida 1964: 64) yes
R78 -ɛʳ -iejr -eq'2 no; only initials are ʔ- and ʒ-
R79 -ɨeʳ/-ieʳ -jijr -yeq'2 yes
R102 -ooʳ -oor -woq2 no (only one Tibetan transcription exists: lha) no; only initials are k-,g-, ʃ-, l-, lh-
R103 -iooʳ -joor -ya:n (no known Tibetan transcriptions) no; only iniitals are n- and kh-

However, the absence of r- in R102-103 does not necessarily entail the absence of a retroflex rhyme. If there were more Tibetan transcriptions for those rare rhymes, some might have had preinitial r-.

I don't know why Gong reconstructed R102-103 as retroflex, but here's my guess:

The Chinese transcriptions for R102-103 syllables represent Tangut period NW Chinese syllables with -o-like rhymes:

R102: 果 ?*kwɔ, 我餓 ?*ŋgɔ (all Chinese Grade I)

R103: 娘 ?*njo (Chinese Grade III)

(assuming a chain shift of > > *o)

All the possible o-slots in Gong's reconstruction are used except for three:

Tangut Grade I -oor

Tangut Grade II -ioor

Tangut Grade III -joor

Gong may have chosen -oor for R102 and -joor for R103 since he assumed that Chinese grades correspond to Tangut grades.

I have no idea why Arakawa reconstructed R102 as -woq2 with a tense vowel indicated by -q. (I don't know what the function of the -2 is.)  Why is this tense rhyme following his retroflex rhymes (R80-R101) instead of being next to his R73 -oq?

I wonder what Arakawa's equivalent of Gong's TT4828 ʔwọ R73 is.  I presume it would be a 'woq contrasting with R102 -woq2. (12.31.14:18: There is no -oq2 or even an -o2 in Arakawa's reconstruction.)

Arakawa's R103 -ya:n ?[jãã] looks like -jaŋ, the late Middle Chinese rhyme of the R103 syllable transcription 娘. But the rhyme of 娘 had shifted to *-ɔ even before the Tangut period.

The only Tibetan transcription of these last two rhymes in Tai (2008: 229) is lha for TT5171 R103. Since -a is the default vowel of a Tibetan consonant letter without any vowel diacritic added, perhaps lha is an error for lho or the section of the page with the letter o atop la has been damaged or lost. 105 RHYMES FOR CHRISTMAS

I finally got around to writing an Excel file comparing my reconstruction to Gong's and Arakawa's. Eventually I will add more reconstructions.

The table excludes medial -w-variants of rhymes: e.g., my -wəu is not listed for R1.

The table makes it obvious that my reconstruction is derived from Gong's. His first eleven rhyme groups are identical to mine. I consider his twelfth rhyme group consisting of R104 and R105 to be a couple of leftovers

Rhyme AMR 081225 Gong 1997
R104 -əũ -ũ
R105 -ya -jwa

that should not be grouped together since other rhyme groups do not mix oral and nasal vowels or -u and -a rhymes.

His Grade I and II are identical to mine with two exceptions (R4 and R6). I have split his Grade III into Grade III and IV. My grades indicate the quality of the first vowel after bending, if any:

Rhyme groups Base vowel Grade I: mid Grade II: low Grade III: high nonpalatal Grade IV: high palatal
I u əu ʊ ɨu (or u?) iu
II-III, IX i əi ɪ ɨi i
IV-V a a æ ɨa ia
VI, IX ə ə ʌ ɨə
VII-VIII e e ɛ ɨe ie
X-XI o o ɔ ɨo io

Nasal, long, tense, and retroflex vowels have been excluded from the above table. The anomalous diphthong ya in R105 has also been left out.

The colors in the Excel file indicate three rhyme cycles:

1. rhymes with nontense, nonretroflex vowels: V

2. rhymes with tense vowels:

3. rhymes with retroflex vowels:

The first 60 and the last two rhymes belong to the first cycle in all three reconstructions, but the classification of several rhymes is disputed:

Rhyme AMR and Gong Arakawa
R77-R79, R102 retroflex tense
R103 plain

I will examine these five rhymes in my next post. MY LONGEST ERROR

has a short explanation. I thought that the unknown reading of

TT2080 ? R? ?.? 'stretch; widen; make longer'

might be dʒɨọ < ?*s-dʒɨo, derived from

TT5109 dʒɨo R53 1.51 'long'

with a causative prefix *s-. But this is probably wrong because Homophones, Tangraphic Sea and Precious Rhymes of the Tangraphic Sea all list a tangraph pronounced dʒɨọ R75 2.64 without any homophones:


TT0783 'ink' (from 'wood' from 'charcoal' plus an abbreviated phonetic from the bottom of dʒɨo R53 2.44 'to have'; this bottom element is not attested as an independent tangraph)

Perhaps 'stretch' was dʒɨọ R75 1.72 with a level tone instead of a rising tone. But if it differed only in tone from TT0783, the two might have been grouped together since tonal distinctions were often ignored.

Another possibility is that 'stretch' was dʒɨoʳ R96 < ?*r-dʒɨo which has not yet been reconstructed for any tangraph. However, the only other similar pair that I know of has a derived causative without retroflexion:

TT1275 kii R14 1.14< *ki-C 'sharpen'

TT1267 kiʳ R84 1.79 < *r-ki 'sharp'

Lastly, 'stretch' may have had a different lax vowel. The rhyme of 'long' (R53) alternates with the following lax-vowelled rhymes:

R3 -iu (Grade IV)

R10 -ɨi (Grade III) / R11 -i (Grade IV)

R36 -ɨe (Grade III) / R37 -ie (Grade IV)

However, alternations with R10/R11/R36/R37 generally involve verb inflection, not derivation*. There are no known R3/R53 alternations involving adjective-to-verb derivation and R3 is a Grade IV rhyme that cannot combine with the Grade II and III initial dʒ-. One could reconstruct 'stretch' with the Grade III rhyme R2 -u, but I know of no other cases of R2/R53 alternation. The syllables

dʒu R2 1.2 and 2.2

dʒɨi R10 1.10 (but not 2.9)

dʒɨe R36 1.35 and 2.32

are all attested and distinct from 'stretch' in Homophones, and it is doubtful that 'stretch' was dʒɨi R10 2.9 since R10/R53 alternations always involve verb inflection.

There is a short/long alternation between R51 -o and R54 -oo (< ?*-o-C) so perhaps R53 -ɨo/-io alternated with R55 -ɨoo/-ioo. Reconstructing 'stretch' as dʒɨoo R55 would not conflict with any other known syllables. But there are no known cases of R53/R55 alternation.

At present, dʒɨoʳ R96 may be the most probable out of a set of unlikely possibilities, assuming that 'stretch' was derived from 'long' and not an unrelated word which happened to have the same initial.

There is a word

TT5003 sio R53 1.51 'long and slender'

which happens to have the same rhyme as 'long' (albeit in a Grade IV variant required after s-) but it cannot be related since there are no other cases of dʒ- ~ s-alternation.

*An exception is the pair

TT0441 nie R37 2.33 'words'

TT4630 nio R53 2.44 'words'

(why is 'hand' on the right of 'language'?) THE LENGTH OF A HORNED HAT

I left one derivative of

TT5109 dʒɨo R53 1.51 'long'

out of my last post:

TT2080 ? R? ?.? 'stretch; widen; make longer'

Only one fact is known about its pronunciation: it had a palatal initial since it was listed in the palatal sections of Homophones and the Precious Rhymes of the Tangraphic Sea. I suspect that it is a causative derivative of dʒɨo 'long'. Although some causative derivatives have voiceless aspirated initials corresponding to voiced initials of their bases, I doubt that TT2080 had initial tʃh- because it was placed in the Mixed Categories volume of PRTS whose palatal section only contains tangraphs for -syllables. I suspect that TT2080 had a tense vowel that was a trace of a causative prefix *s-:

?dʒɨọ < ?*s-dʒɨo

The pair dʒɨo 'long' : dʒɨọ 'lengthen' would be parallel to adjective-verb pairs like

TT3591 bi R11 1.11 'low'

TT5019 bị < *s-bi R70 1.67 'to lower' (i.e., to make low)

TT1309 bie R37 2.33 'high'

TT1232 biẹ < *s-bie R64 1.61 'to heighten' (i.e., to make high)

The 'horned hat'

may signify a verbal derivative in TT2080, but it does not have that function in other tangraphs: e.g.,

TT1839 liẽ R43 2.37 'ball'

has no semantic or phonetic relationship to hatless

TT1309 bie R37 2.33 'high'

According to the Tangraphic Sea, 'long' consists of the bottom right of TT2080 plus the left of riʳ 'branch' (something that is long?):


The right side of 'branch' is

TT1620 tʃhɨə R30 1.29 'narrow'

which looks like Li Fanwen radical 024 (meaning unknown) plus 'person' (why?)

and its left side is


which may be an inversion and distortion of Chinese 長 'long'. THE LENGTH OF DEATH METAL

How do I know that the readings of these tangraphs began with g-? All three are in chapter V (velars) of Homophones, so they must have begun with a velar.


TT2592 gwii R14 1.14 'banner'

analyzed as 'metal' (from top of bə̣i 'spear') atop dʒɨo 'long'

belongs to Li Fanwen's (1986: 50) fanqie initial speller chain 20 equivalent to Tai's (2008: 185) chain 15 which was generally transcribed with Tibetan Hg- ?[ŋg]. (I will discuss the sole exception later.) It is probably a loanword from NW Late Middle Chinese 旗 *gɨi 'banner' with a medial -w- that may be a trace of an earlier Tangut labial prefix *P-. The vowel length may imply an earlier Tangut suffix.


TT4053 gie R37 1.36 'valley; mountain stream gorge; ravine'

analyzed as 'earth' (from left of mee 'cave; rock; river; aperture') and dʒɨo 'long'

belongs to Li Fanwen's (1986: 50) fanqie initial speller chain 21 equivalent to Tai's (2008: 185) chain 16 which was most frequently transcribed with Tibetan Hg- ?[ŋg]. Other transcriptions are bg-, b- plus an illegible letter, and rg- (before nonretroflex R3! - more on this later).

There is no known fanqie for the third tangraph


TT3137 gi R11 2.10 'banner'

which looks like a combination of 'death' (why? - it does look flaglike, but that can't be the reason) plus gwii 'banner'. Like gwii, it is probably also a borrowing from NW Late Middle Chinese 旗 *gɨi 'banner' but without any Tangut affixes added.

TT3137 is listed as a solitary tangraph without homophones in Homophones (27B73) but in Precious Rhymes of the Tangraphic Sea, it is listed among the members of Homophones group 23A11-23A16 reconstructed by Li Fanwen (1986: 308-309) as ŋi and by Gong as gji (= my gi). The Tibetan transcription Hgyi ?*[ŋgi] (NIshida [1964: 99], Tai [2008: 207]) for this group indicates a voiced stop initial. (The Tangut period NW Chinese transcription 宜 ?*ŋgi is ambiguous since TPNWC had no syllable *ŋi, whereas the Tibetan script could distinguish between ŋyi, gyi, and Hgyi.) Since the second volume of Tangraphic Sea has been lost, it is impossible to determine with certainty whether TT3137 had homophones or not in the TS dialect, as opposed to the dialect of the unknown copyist of PRTS.


TT5109 dʒɨo R53 1.51 'long'

(a loanword from NW Late Middle Chinese 長 ?*dʒɨo 'id.'?)

does not sound like its g-derivatives

perhaps it would be better to treat it as an abbreviation of gwii 'banner' in which it is presumably semantic ('banner' < [something hung from a] 'long' + 'spear').

A fifth derivative of 'long' has a non-g reading


TT2786 jị R70 2.60 'stretch'

because it is presumably a semantic compound of 'hand' + 'long'. HOMOPHONES 21B11-21B17

TT3777 giaa R24 2.21 'to rejoice; to enjoy'

belongs to a homophone group with seven members:

Homophones number tangraph Tangut telecode reconstruction gloss

3777 giaa R24 2.21 to rejoice; to enjoy
21B12 1844 able; skillful; first syllable of a Tangut surname
21B13 1404 swallow; eat

1401 pair; also read twəəi R12 2.11 'two; opposition' (loan from Tangut period NW Chinese 對 ?*twej 'pair; opposition')
21B15 5416 (toponym transcription)
21B16 4476 disease of horse hoof
21B17 0959 name of a tree; fence

Since none have known fanqie, how can one reconstruct their readings?

All but TT1401 appear in the section for rhyme 2.21 in the Precious Rhymes of the Tangraphic Sea, so the rhyme of all seven must be 2.21. The rising tone volume of PRTS only has seven entries for 2.21: six of the seven tangraphs above plus

TT0968 miaa R24 2.21 'great; big; important; (great) business'

1.23, the level tone counterpart of 2.21, has eleven tangraphs in the level tone volume of PRTS with four more initials:

initial ph- b- g- m- d- l-
number of tangraphs 1 1 1 5 1 2

(The insertion of velar-initial giaa 1.23 between labial-initial biaa 1.23 and miaa 1.23 is strange. It is not due to a copyist's error in the handwritten PRTS because Tangraphic Sea has the same peculiarity.)

Most R24 (1.24/2.21) syllables have grave initials with three exceptions. 

Written Tibetan does have -ya, so I can't explain why R24 was transcribed as -a(H) without -y-. In any case, there is no doubt that R24 had some sort of a-like rhyme.

There is, however, disagreement on the initial of the homophone group 21B11-21B17. The initial has to be velar since that group is in Homophones chapter V for velars. Thus it has only four possible initials:

k- kh- g- ŋ-

Gong and Li Fanwen reconstructed g- whereas Sofronov reconstructed ŋ- (since his system has no g-). Why did they choose a voiced initial? I don't know.

None of the tangraphs in this group have Tibetan or Chinese transcriptions that could indicate the initial (Tai 2008: 211 and Nishida 1964: 99). Hence Nishida (1964: 104) reconstructed the reading of the group as with G representing an unknown velar. (12.21.21:25: Shi et al. [2000: 235-236] also regard the initial of teh group as an unknown velar.) Gong's (1991) list of Forest of Categories transcriptions does not include any tangraphs from this group. How many other reconstructions are as seemingly shaky? Perhaps there is transcription evidence I am not aware of pointing to g-.

I favor Nishida's agnostic stance, but if I had to defend Gong's g-, I would point out that

TT0959 giaa R24 2.21 name of a tree; fence

may share a phonetic element

with the following tangraphs with initial g-:

tangraph Tangut telecode reconstruction rhyme tone.rhyme gloss
2592 gwii R14 1.14 'banner'
3137 gi R11 2.10
4053 gie R37 1.36 'valley; mountain stream gorge; ravine'

Now I'm afraid to find out whether their initial g- is supported by external evidence.

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