I usually say that two out of three volumes of the Tangraphic Sea have survived, but for brevity I don't note that the early parts of the level and rising tone sections of the Mixed Categories (MC) volume are missing. This is why Andrew West's electronic version of MC begins with class IV of the 'level' tone and class V of the 'rising' tone.

The Precious Rhymes of the Tangraphic Sea (PRTS) manuscript has bits of those sections. Here is the number of tangraphs per class in each section of Mixed Categories (MC) in PRTS:

'level'/1 3 2 ? 5 6 106 82 22* 74
'rising'/2 1 1 7 2 6 82 72 13 74

The figures (based on Shi et al. 2000: 319-343) are not complete, but what remains leads me to doubt that they could be much higher: e.g., the list of 'level' tone class I tangraphs begins and ends on the same page.

The majority of tangraphs in MC in PRTS are in the class VI, VII, and IX sections largely containing tangraphs for syllables with dz- (VI), j- (VII), and lh- (IX). I don't know why those syllables were not listed in the 'level' or 'rising' tone volumes. dz- and j- are both voiced obstruents, but they do not form a natural class with the voiceless sonorant lh-.

Conversely, why are scattered non-dz-/j-/lh-tangraphs in MC in PRTS: e.g., why was

5405 2ma1 'the Tangut surname syllable Ma'

the sole tangraph in the 'rising' tone class I section of MC instead of in the 'rising' tone volume of PRTS?

*3.1.0:06: This figure includes the class VIII tangraph

0222 1horn1 'to roar, howl'

which was actually listed toward the end of class VII.


I don't want this blog to become a survey of Tangraphic Sea homophone groups. I want to only look at a few cases that differ enough from each other to warrant posts.

This trio in Tangraphic Sea rhyme 1.2 caught my eye because Arakawa reconstructed different initials (š-, š²-) in his Nishida-style reconstruction while reconstructing a three-way merger in his own reconstruction:

Tangraphic Sea 1.2 homophone group Tangraphic Sea circle Example Homophones A Homophones B Nishida-style reconstruction in Arakawa (1997) Arakawa (1997) Gong This site Tangraphs
8 36A38-36A51 41B42 1šĭu 1shyu 1ɕju 1shu3 A 1
9 36B67 1š²ĭu 1shu3 B 3
12 37A54-37A63 37B75-38A11 1šĭu 1ɕjwu 1shwu3 5

Each homophone group has a distinct fanqie. Their initial spellers are in three nonoverlapping chains including the example tangraphs above:

8. <> (initial transcribed in Chinese as 室 *sh- and Tibetan as sh- and (g)j-; transcribed Sanskrit ś- and possibly c-; Sofronov 1968 II: 22 and Tai 2008: 192)

9. <> (initial transcribed in Tibetan as (b)sh-, gs-, and zh(w)-; transcribed Sanskrit ś-, ṣ-, s-; Tai 2008: 192)

12. <> (initial transcribed in Chinese as 說 *shw- or 姪 *chh-; Sofronov 1968 II: 17)

The external evidence generally points to sh- for all three. Group 7 has initial chh-, so group 12 might have initial shw- by process of elimination.

If Tangut had two kinds of sh-, I would expect them to correspond to Sanskrit palatal ś- and retroflex ṣ-, but both were written with tangraphs whose initial spellers were in the fanqie chain for the initial of homophone group 9.

The fanqie final spellers for 8 and 9 are in the same chain:


group 8 < 1013 < 3003 > group 9

implying that the distinction between the two groups is in their initials. Yet Homophones A regards the two groups as homophones unlike Homophones B.

The placement of circles in the Tangraphic Sea also implies that the two groups are somehow united (they are undoubtedly similar), though circles are also missing between groups that are undeniably very different: e.g., 1.1.7 1nu1 B and 1.1.8 1ku1.

The Tangraphic Sea and both editions of Homophones agree that group 12 is distinct, and the final speller 0622 is in a separate chain from that of groups 8 and 9:


0622 1wu3 <> 2842 1shwu3

I have not found any strong transcription evidence for the -w- that Gong reconstructed for 0622 and its fanqie initial speller

1795 2wi4

However, if 2842 had shw-, then its final speller probably had -w- too.


In my last entry, I found that the 'A' and 'B' homophone groups in Tangraphic Sea rhyme 1.1 had identical rhymes because their fanqie final spellers were in the same chain. That may seem like a tautological statement, but there is no guarantee that two or more rhymes (1.1a, 1.1b ...) were not conflated under a single heading. Arakawa reconstructed two types of 1.1 rhymes, 1-u and 1-u2. (All of the 'A' and 'B' pairs have 1-u in his system.)

Groups 6 (1nu1 A) and 7 (1nu1 B) also had fanqie initial spellers in the same chain, so their fanqie seem to indicate they were homophonous. Why would identical syllables be arbitrarily split into two groups?

What about the fanqie initial spellers of the other two 'A/B' groups?

Group 9 (1khu1 A) has a 'rising' tone fanqie initial speller without any known fanqie:

2782 2khi4 < Chinese 氣 *2khi3 'gas'

If the missing volume of the Tangraphic Sea is ever rediscovered, we could see if 2782 is in the same fanqie chain as 4807, the fanqie initial speller of group 10 (1khu1 B):


4807 1khi4 <> 5399 1khu4

2782 and 4807 are in the same homophone group in Homophones A, implying that they had the same initial, but they are in different homophone groups in Homophones B, implying that they had different initials as well as finals.

I thought group 10 (1khu1 B) might have had secondary kh- from an earlier g-. If 4807 'to lose' was a speller for *g-, it should have cognates with voiced initials. However, Jacques (2014: 81, 94, 250) identified its Japhug cognate as kra 'to drop' with a voiceless initial. (The mismatch in aspiration requires explanation.) On the other hand, I think 4807 is a loan from Chinese 棄 *3khi4 'to discard'. Either external connection points to kh- and not to g-.

Lastly, groups 13 (1tshu1 A) and 14 (1tshu1 B) have initial fanqie spellers in the same chain:

group 13 < 3291 < 4996 > 1319 > group 14

so they had the same initial (tsh-) as well as the same rhyme (1-u1) ... and yet groups 13 and 14 are separate in both the A and B versions of Homophones!

13: Homophones A 33B37Homophones B 34A47 (isolated characters without homophones at the end of chapter VI)

14: Homophones A 32B15Homophones B 33A31

I give up ... for now. I doubt this is the last time I'll try to wrestle with this problem. AN UN-NU-N DISTINCTION

Gong reconstructed Tangraphic Sea rhyme 1.1 homophone groups 6 and 7 identically even though they (1) have different fanqie, (2) are separated by circles, and (3) correspond to different (though adjacent) homophone groups in Homophones:

Tangraphic Sea 1.1 homophone group Homophones Tangraphs Gong
6 17B44-17B45 1nu1
7 17B41-17B43
(The third tangraph in parentheses is in Homophones but not the Tangraphic Sea.)


Yesterday I hypothesized that the two groups might have had different initials: voiceless hn- and voiced n-. If the distinction were in the initials, then their fanqie initial spellers 3226 and 4027 should not be in the same fanqie chain. But in fact they had the same fanqie initial speller (0616):


group 6 < 3226 < 0616 > 4027 > group 7

Their fanqie final spellers are also in the same chain:

'A' groups 6, 9, 13 < < < > > 'B' groups 7, 10, 14

So how could groups 6 and 7 be distinct if they had the same initials and finals? A U-NIQUE TANGRAPH

I added a note to my previous entry about

1520 0wu1

which is the only possible rhyme 1 syllable in the Class VIII section of the 'level' tone section of the Mixed Categories volume of the Tangraphic Sea. I wrote that its "reconstruction is problematic." Here's why.

1. Tone

I transcribed the tone of 1520 as 0 (= unknown) following Gong. I should have followed Arakawa and regarded its tone as 'level' because it is in the 'level' (= first) tone section of the Mixed Categories and its fanqie contains a final speller with the 'level' tone:


1520 0wu1 = 0434 1i3 'alas' (with a glottal stop initial) + 3044 1u1 'grave, death'

1520 is a fanqie character, so that formula doubles as its graphic analysis: 'speech' (the left side of 0434) plus all of 3044 (phonetic).

2. Medial

I don't know why Sofronov (1968 II: 379) and Gong reconstructed a medial -w- and Li Fanwen (1986: 429) reconstructed an initial w-, as neither fanqie speller contained -w-. 0434 transcribed Sanskrit i, not vi (Arakawa 1997: 110). (Sanskrit has no w.) 3044 belongs to homophone group 16 which was transcribed in Tangut period northwestern Chinese as 烏 *1u1 without w (Sofronov 1968 II: 6).

My guess is that w was reconstructed to account for the fact that Homophones lists no homophones for 1520: i.e., 1520 could not be homophonous with 3044 1u1.

However, such a w is difficult to reconcile with the fact that 1520 transcribed Sanskrit u, not vu (Arakawa 1997: 110). Moreover, 1520 combined with 'long' to form 1540 which transcribed Sanskrit ū, not (Grinstead 1972: 87, 184):


1540 'Sanskrit ū' = 1520 'Sanskrit u' + 0443 1jo3 'long'

(The placement of 0434 and 1520 in Grinstead's table on p. 184 imply they were for Sanskrit long ī and ū, but I think they should be in the previous row for short vowels.)

3. Final

Arakawa (1997) reconstructed 1520 with what appears to be rhyme 1.4* (1-uɦ) using Nishida's system on page 110 but with rhyme 1.1 (1-u) using Nishida's system and his own system on pages 98 and 125. Yet the final fanqie speller of 1520 has rhyme 1.1. I cannot explain this inconsistency. (Nishida 1966: 420 reconstructed 1520 with rhyme 1.1 1-u.)

4. Placement

If 1520 was 1u1 as its fanqie indicates, why was it in Mixed Categories instead of the 'level' tone volume with 3044 1u1? Is it because 1520 is a Sanskrit transcription character like others in the Class VIII section of Mixed Categories? But not all Class VIII characters in Mixed Categories transcribe Sanskrit, and there are Sanskrit transcription characters in the other two volumes of Tangraphic Sea.

The bigger question is why the Mixed Categories volume exists at all. Was it a compilation of characters that were accidentally left out of the other two volumes, or do its characters have something else in common?

*Arakawa wrote "1.? øuɦ" without specifying a rhyme, though -uɦ can only be rhyme 4 in Nishida's system. The question of whether Sanskrit vowel transcription characters were pronounced with zero initials and/or glottal stops merits investigation. TANGRAPHIC SEA RHYME 1.1: 'PREFACE'

It's neat that the name of the first Tangut rhyme

5085 1bu1

means 'preface' rather than, say, 'epilogue'.

It is the first of six tangraphs in homophone group 3 in the 'level' (first) tone volume of the Tangraphic Sea:

Page.position Homophone group Circle Class Transcription Tangraphs
5.131-5.142 1   I 1pu1 4
5.143-5.161 2 1phu1 4
5.162-5.212 3 1bu1 6
5.213-5.231 4 III 1tu1 4
5.232-5.252 5 1thu1 6
5.261-5.262 6 1nu1 A 2
5.271-5.272 7 1nu1 B 2
6.111 8 V 1ku1 1
6.112-6.122 9 1khu1 A 4
6.131-6.132 10 1khu1 B 2
6.141-6.153 11 1gu1 5
6.161 12 VI 1tsu1 1
6.162 13 1tshu1 A 1
6.171-6.211 14 1tshu1 B 3
6.212-6.213 15 1su1 2
6.221 16 VIII 1u1 1
6.222-6.231 17 1hu1 2
6.232-6.242 18 IX 1lu1 4
6.243 19 1zu1 1
6.251 20 III 1nwu1 1
6.252 21 V 1ngwu1 1
6.261 22 VI 1tshwu1 1
6.262 23 1swu1 1
6.271 24 IX 1lwu1 1

You can see the tangraphs and fanqie spellings for all of those groups on Andrew West's site.

I discussed the tshu-groups (13, 14, and 22) in my last post, so I created that color-coded table to see them in context. I'd like to go through both surviving volumes of the Tangraphic Sea to study their organization.

Groups 1-19 follow the order of consonant classes also in Homophones. The order of consonants within each class in turn follows the Chinese tradition which in turn is based on the Indian tradition: e.g.,

k-, kh-, kh- (< *g-?), g- [ŋg] (groups 8-11)

The consonant class cycle starts again in group 20, implying that groups 20-24 constitute a set distinct from the first 19 groups. Gong reconstructed -w- in groups 20-24, though there is no Tibetan, Chinese, or Sanskrit transcription evidence for a medial. (Sofronov 1968 II: 6 collected all the Tibetan and Chinese evidence for the first rhyme of the Tangraphic Sea.) Arakawa (1997: 17, 125) wrote the rhyme of groups 20-24 as -u² in Nishida's system and as -u2 in his own system.

The absence of Class II is due to chance. Rhyme 1.1 is a Grade I rhyme that cannot have Class IV and VII initials.

Circles divide homophone groups 1-6, but the correlation between circles and groups breaks down when circles do not separate groups 7-9 even though 7 is Class III and 8-9 are Class V.

Gong reconstructed groups 9 and 10 identically even though they are separated by a circle, have different fanqie, and are in different columns in a Tangut phonetic table. 10 should have had *kɦ- or even *g- if the Tangut were following Chinese consonant order. Gong pointed out that groups 9 and 10 were transcribed with the same Chinese initial *kh- in the Pearl in the Palm (1190) but that does not necessarily mean they had the same initial when the Tangraphic Sea was written. Late 12th century northwestern Chinese *kh- is from both *kh- and *g-, and late 12th century Tangut kh- may also have had two sources in earlier Tangut:

Tangraphic Sea Tangut k- kh- g- ŋg-
Pearl in the Palm Tangut k- kh- (ŋ)g-

This hypothesis predicts that kh- and old g- (not to be confused with the prenasalized stop transcribed as g-) should have different fanqie spellers and different correspondences with consonants in other languages. I should test both predictions later.

Groups 13 and 14 may also have had different initials:

Tangraphic Sea Tangut ts- tsh- dz- ndz-
Pearl in the Palm Tangut ts- tsh- (n)dz-

I have arbitrarily used the letters A and B to distinguish between groups with uncertain differences in my transcription here. I do not intend to imply that 1khu1 A (group 9) was to 1khu1 B (group 10) what 1nu1 A (group 6) was to 1nu1 B (group 7), though perhaps there are parallels: e.g., if group 6 n- was voiceless [n̥] and group 7 n- was voiced [n], then A syllables had voiceless (aspirated) initials and B syllables originally had voiced initials that later merged.

I am baffled by the absence of a circle separating group 19 from groups 20-24. Were the latter considered to be self-evidently different (e.g., with Gong's medial -w- or with a different rhyme)?

I also do not understand why there are no m- or d- syllables with rhyme 1.1 (though rhyme 1.4 has one m- syllable and six d-syllables - the only rhyme 1.4 syllables with nonback initials).

2.24.11:18: Nor do I understand why dzu-syllables with rhyme 1.1 - and all other syllables with dz- and either tone - were placed in the Mixed Categories volume of the Tangraphic Sea along with

1520 0wu1

whose reconstruction is problematic.

2.24.23:21: One other rhyme 1.1 syllable in the Mixed Categories volume of the Tangraphic Sea is

5621 1lhu1 'to increase'.

Why are most lh-syllables in the Mixed Categories volume? ADMONISHING POTATOES

The first character in the fragment I discussed last week has this fanqie in the Mixed Categories volume of the Tangraphic Sea:


3543 1dzwy1 'chapter' = 0524 1dzu3 'to admonish, instruct' + 3605 1tshy1 (first syllable of 1tshy1 2on4 'potato')

Why does 1dzwy1 contain a -w- absent from its fanqie spellers?

The short answer is that my transcriptions are based on Gong's 1dzwə, 1dzju, and 1tshə.

At times like this I want to reconstruct from scratch. Let's look at the evidence for 3543.

It is a fact that 3543 is in the tone 1 / Class VI section of Mixed Categories. So the tone is certain, and the initial must be ts-, tsh-, dz-, or s-.

The initial speller of 3543 has this fanqie in Mixed Categories:


0524 1dzu2 = 1092 1dzy4 +  3003 1u3


- was transcribed in Tibetan as HdziH (whose initial was [ndz])

- was transcribed in Chinese as 尼則 *nji tsy = *ndzy

- transcribed Sanskrit j (which was [dz] in the variety of Sanskrit known to the Tangut)

so its initial - and the initials of 0524 and 3543 - must have been dz- [ndz].

The final speller of 3543 has this fanqie in the 'level' tone (= tone 1) volume of the Tangraphic Sea:


3605 1tshy1 = 0984 1tshwu1 + 3732 1ky1

I think 3605 should be 1tshwy1 instead of 1tshy1. Then 3543 would also have -w-:

3543 1dzwy1 = 0524 1dzu3 + 3605 1tshwy1

Yet both Gong and Sofronov reconstructed 3605 without -w- and 0984 with -w-.

There is no transcriptional support for -w- in 0984 which belongs to the last of three tshu-type homophone groups in rhyme 1 of the 'level' tone (= tone 1) volume of the Tangraphic Sea:

1tshu1 group 1:

1tshu1 group 2:

1tshu1 group 3:

The third group is isolated toward the end of rhyme 1, far from the other two which are next to each other. Gong and Sofronov reconstructed a -w- in the third group and regarded the first two groups as homophonous despite their different fanqie. I suspect there was a three-way distinction that no one has reconstructed yet.

2.23.4:03: I forgot to justify reconstructing -y for 3543. Unlike the other two volumes of the Tangraphic Sea, Mixed Categories is organized by tone and initial consonant class, not tone and rhyme. So the rhymes of Mixed Categories entries have to be determined from their fanqie final spellers. The final speller of 3543 is 3605 which is listed under rhyme 27 of the 'level' tone (= tone 1) volume of the Tangraphic Sea. That rhyme was transcribed in Tibetan as -a, -i, -e, and -o. The corresponding tone 2 rhyme was transcribed with -u as well as the other four Tibetan vowels. This diversity of transcriptions indicates a Tangut vowel that was unlike anything in Tibetan: e.g., a schwa. (5:03: I transcribe all nonlow central and back unrounded vowels as -y.)

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