How can one reconstruct different rhymes for rhyme categories that have similar transcriptions? Do we have to settle for reconstructing those rhymes identically and indicating their Tangraphic Sea rhyme numbers: e.g.,

rhyme group 17: -a 1.17, -a 2.14

rhyme group 18: -a 1.18, -a 2.15

rhyme group 19: -a 1.19, -a 2.16

rhyme group 20: -a 1.20, -a 2.17

rhyme group 21: -a 1.21, -a 2.18

(cf. their Tibetan transcriptions -a, -aH)

This is what Gong (1997) has done in some cases: e.g., he reconstructed both 1.19/2.16 and 1.20/2.17 as -ya:

rhyme group 19: -ya 1.19, -ya 2.16

rhyme group 20: -ya 1.20, -ya 2.17

Is that the best we can do? For years, I wondered if it was.

Clauson (1964: 72) proposed that some of the 105 rhyme categories may have ended in "final plosives". Since the Tangut autonym

TT3536 5745

myï 2.28 nyaa 2.18

was Tibetanized as mi-ñag (rather than as mi-ña), he wrote that

[i]t seems more reasonable to reconstruct the rhyme [2.18] as -ag and the character [TT5745] as nyag.

Sofronov (1968 I.122), however, noted that if Tibetan ñag really was based on the Tangut reading of TT5745, it

could date back to a far older period of the history of the Tangut language than that reflected in the language of the Tangraphic Sea (tr. by AMR).

Thus what may have been a final -g during this period could have become something else by the time of the Tangraphic Sea. Sofronov mentioned three possibilities besides nyag for the TS period:

-k, -x, -'

I would add a fifth possibility: a final voiced velar fricative -G also found in Japhug rGyalrong. Voiceless -h and voiced -H may also be possible.

Since Sofronov did not believe that any extant transcriptional material dated from the Tangraphic Sea period*, he did not think he could reconstruct specific final consonants. Hence he reconstructed an unspecified -C (for согласный 'consonant') to differentiate 2.18 (the rhyme of TT5745) from 1.19/2.16 in the Tangraphic Sea period:

rhyme group 19: -ya 1.19, -ya 2.16

rhyme group 21: -yaC 1.21, -yaC 2.18

The final consonant of rhyme group 21 had become the high back unrounded vowel -U in his reconstruction of late 12th century Tangut:

rhyme group 21: -yaC > -yaU

But in general, Sofronov's Tangraphic Sea period -C disappeared without a trace in his late 12th century Tangut reconstruction: e.g.,

rhyme group 4: -uC > -u

merging with rhyme groups 1, 5

rhyme group 12: -eC > -e

merging with rhyme group 8

rhyme group 32: -əC >

merging with rhyme group 28

rhyme group 38: -aiC > -ai

perhaps the final -C blocked -ai from shifting to -ei unlike rhyme groups 34 and 35 which had no final -C and shifted from -ai to -ei; the lack of this shift might be considered a trace of -C

rhyme group 60: -yuoC > -yuo

merging with rhyme group 59

-C became -U only after -a and -o (but not after -uo).

There is no direct transcriptive evidence for -U. Tibetan had nothing like -U (except perhaps ï, whose phonetic value is uncertain) and that the Tangut period northwestern Chinese dialect probably did not have diphthongs ending in -ï. (Although some Chinese reconstructions have such diphthongs, I know of no attested form of Chinese with -Vï rhymes.) Sofronov's -U may correspond to Chinese -w or Chinese -Ø. This might indicate that there was no good way to write Tangut -U in Chinese.

Sofronov's reconstruction of the 105 rhymes could be summarized as follows:

Rhyme cycleRhyme groupTangraphic Sea periodLate 12th century
I: nontense cycle1-3-u type-u type
4-uC type
5-7-ũ type
8-11-e/i type-e/i type
12-14-eC type
15-16-ẽ type
17-19-a type
20-24-aC type-aU type
25-27-ã type
28-31-ə/I type-ə/I type
32-33-əC type
34-36-ai type-ei type
38-43-aiC type-ai type
44-49-eU, -əU type
50-53-o type
54-55-oC type-oU type
56-58-õ type
59-yuo-yuo type
II: tense cycle I61-73tense vowels and diphthongs; no final consonants or nasalization
74-75-о̣̃ type
III: tense cycle IĨ̃̃̃̃̃77-83tense vowels and diphthongs; no final consonants or nasalization
85-87-ạ type
́88-89-ạC type-ạU type
90-96tense vowels and diphthongs; no final consonants or nasalization
97-98-ụoC type-ụo type
IV: tense cycle IIỊ̣99-103tense vowels and diphthongs; no final consonants or nasalization
104-105-о̣̃ type (not clear how this is different from 74-75)

In short, he thought that the Tangraphic Sea was organized into

four rhyme cycles consisting of

rhyme groups sharing a common vowel, sorted by

a. absence of -C or nasalization

b. presence of -C

c. presence of nasalization

This view is not shared by others. Most Tangutologists did not reconstruct any final consonants other than -N (Hashimoto 1965) or -H (Nishida 1964). There is, however, one major exception.

Next: Arakawa's codas.

*I don't know why he was so certain about this. As far as I know, the Tibetan transcriptions of the Tangut are undated. They might be as old as the Tangraphic Sea. RIDDLE OF THE RHYMES (PART 5)

The sight of 97 level tone rhymes and 86 rising tone rhymes that are barely distinguished in the Tibetan transcriptions may make one skeptical of 'Dictionary Tangut'. Was it all a bluff to make the language look more complicated than it actually was? Was 'Dictionary Tangut' a composite language that was more complex than its sources, as I proposed in part 4?

Nishida (1966: 534) initially

doubted whether this large number of rhyme groups reflected the actual phonetic system of the Hsi-hsia language ... It is impossible to imagine that the 97 level tone rhymes and the 86 rising tone rhymes all represent single vowels ...

for no language has 97 different vowels.

In the words of Clauson (1964: 66), Sofronov (1963)

made a brave attempt, based on the Chinese and Tibetan transcriptions, to reconstruct these final sounds [the 105 atonal rhymes], but it is pretty clear that his reconstruction cannot really represent the final sounds of Tangut as they were in the 11th century, and if they represent the final sounds as they were when the language had been severely abraded it is probably rather too elaborate ... Sofronov's list contains sixty-five open vowels, thirty-one open vowels preceded by -j- [= y], three open vowels preceded by -w- and six final sounds of vowel plus -n, two of them preceded by -j- ... It does seem impossible that a Tangut phonetician [or any Tangut native speaker!], however acute his hearing, could have distinguished sixty-five different open vowel sounds, even if some of these were in fact diphthongs.

Sofronov's 1968 reconstruction attempted to avoid what Clauson (1964: 72) described as an

unnatural appearance ... with [an] enormous preponderance of final open vowels.

Sofronov (1968 I: 122; my translation) noticed that

the transcriptions of groups 8-10 and 12-14, 17-19 and 20-21, 28-30 and 32-33* are essentially identical. This signifies that either the syllables of these rhymes were not distinguished by their phonemic structure at the end of the 12th century [i.e., after Tangraphic Sea was compiled in the 11th century] or that they were distinguished by some qualities which could not be accounted for in the transcriptions.

Sofronov proposed two reconstructions:

- one for the Tangraphic Sea language which distinguished those groups

- one for the late 12th century language in which

groups 8-10 and 12-14 were identical

groups 17 and 20 were identical

(group 18 had no 2X counterpart)

(and groups 19 and 21 were still distinguished, but in a new way)

groups 28 and 32 were identical

(group 29 had no 3X counterpart)

groups 30 and 33 were identical

In short, Sofronov proposed that the Tangraphic Sea language really did have 105 rhymes which were reduced to a smaller number by the end of the 12th century: e.g., in his 1968 reconstruction,

TS rhymes 1.17 and 1.20 merged into -a in late 12th c. Tangut

What did the earlier language have that the later language lacked?

Next: The exonymic key.

*These group numbers refer to Sofronov's pairing of rhymes: group 8 is 1.8 : 2.7, group 9 is 1.9 : 2.8, etc. I have outlined Sofronov's pairings in part 3.

Here are the Tibetan transcriptions of each group from Nishida (1964) (with -eH from Sofronov's transcription data for 2.25):

Sofronov's rhyme groupsRhyme numbers-a-aH-am-i-iH-u-uH-e-o-ïH
8-101.8-1.10 : 2.7-2.9-i
12-141.12 : 2.11, 1.13, 1.14 : 2.12-i-iH-u-e
17-191.17-1.19 : 2.14-2.16-a-aH
20-211.20-1.21 : 2.17-2.18-a-aH-am
28-301.27-1.29 : 2.25-2.27-a-i-iH-u-uH-eH-o-ïH
32-331.31, 1.32 : 2.29-iH

8-10 and 12-14 don't seem that similar to me, and 28-30 and 32-33 are even more different. But 17-19 and 20-21 are identical if one ignores the aberrant -am (the only -m transcription I have ever seen). RIDDLE OF THE RHYMES (PART 4)

In Part 3, we saw that Homophones group I.16 (using Nishida and Li Fanwen's numbering) was a mixture of level (1.27) and rising (2.25) tone tangraphs:

1.27 x 4

2.25 x 10

1.27 x 1

2.25 x 6 (or 7?)

I used to wonder if Homophones reflected a Tangut dialect in which tone was nonphonemic. That clearly cannot be the case here. If it were, I would expect level and rising tone tangraphs to be mixed at random within a group. But instead they are neatly subgrouped. Yet they are not placed into two subgroups: i.e.,

1.27 x 5

2.25 x 16 (or 17?)

Why is there a single 1.27 tangraph

3A74 TT2033 筵 BAMBOO-MAT 1.27

breaking up an otherwise solid set of 16 (or 17) 2.25 tangraphs? Without examining distribution patterns throughout Homophones as a whole, I can only point out that we cannot be entirely sure that a tangraph listed under rhyme 1.27 in Tangraphic Sea was a 1.27 tangraph in the Homophones dialect (or any other dialect). Perhaps it was pronounced with the rising tone in the Homophones dialect:

3A74 TT2033 筵 BAMBOO-MAT

1.27 (Tangraphic Sea dialect)

2.25? (Homophones dialect)

If the rising tone originated from an affix, the Homophones dialect could have reflected an affixed variant of BAMBOO-MAT while the Tangraphic Sea dialect reflected the unaffixed stem.

We should constantly question the assumption of Tangut uniformity. Does one tangraph necessarily equal one pronunciation*? There is no reason to assume that all Tangut data reflect the same dialect or even very similar dialects. Surely there was both regional and temporal variation in Tangut during the short life of the Tangut Empire.

However, I would like to assume that the dialect within a given source (e.g., Tangraphic Sea, Homophones, Pearl) is uniform**. It would make no sense to mix readings from different dialects in a single dictionary organized by phonological criteria, unless 'dictionary Tangut' was an artificial creation like Qieyun Middle Chinese or Hphags-pa Chinese*** which attempted to encompass all phonological distinctions in a set of privileged dialects. Perhaps no Tangut dialect actually had 97 level tone rhymes and/or 86 rising tone rhymes. Could the Tangraphic Sea rhyme categories in fact represent tonal correspondence sets across dialects: e.g.,

Tangraphic Sea rhymeDialect A rhymeDialect B rhyme
TS1.1 (Gong: -u)A1.1 -uB1.1 -u
TS1.2 (Gong: -yu)A1.2 -yuB1.2 -yu
TS1.3 (Gong: -yu; same as 1.2)B1.3
TS1.4 (Gong: -u; same as 1.1)A1.3 -UB1.1 -u

In this hypothetical example, there are four TS rhymes corresponding to only three rhymes in two dialects. Each dialect maintains a distinction that the other does not:

Dialect A distinguishes between TS1.1 and TS1.4

Dialect B distinguishes between TS1.2 and TS1.3

The four categories TS1.1-1.4 would still be valid for historical-comparative purposes. Presumably some earlier stage of Tangut (pre-Tangut) had those categories which had merged in different ways in different dialects. We cannot assume that the Tangraphic Sea rhymes are pre-Tangut rhymes, since it's possible that pre-Tangut had, say, 100 level tone rhymes, and that some of them merged in all Tangut dialects: e.g.,

Pre-Tangut rhymeTangraphic Sea rhymeDialect A rhymeDialect B rhyme
PT1.1a -ungTS1.1 -uA1.1 -uB1.1 -u
PT1.1b -u

In this hypothetical example, pre-Tangut was long dead by the time the Tangraphic Sea was compiled, so no one in the Tangut Empire could have known that TS/A/B1.1 really had two origins, PT1.1 and PT1.2. However, correspondences such as

A1.2 -yu : B1.2 -yu

A1.2 -yu : B1.3

were still evident from living dialects and therefore were treated as separate rhyme categories (TS1.2 and TS1.3) in the Tangraphic Sea system.

Perhaps Homophones was compiled in a similar manner: i.e., grouping syllables by correspondences across dialects: e.g.,

1.27 x 4: all dialects have 1.27

2.25 x 10: all dialects have 2.25

1.27 x 1: 1.27 in most dialects?; 2.25 in some dialects

2.25 x 6 (or 7?): 2.25 in most dialects?; 1.27 in some dialects

I find it hard to believe that the compiler wrote 4 1.27 tangraphs, 10 2.25 tangraphs, realized he left out a 1.27 tangraph, inserted it in the middle, and then wrote the remaining 2.25 tangraphs.

One strange thing about Tangraphic Sea is that not all tangraphs are in the level and rising tone volumes. Some have been placed in the Mixed Categories volume and organized by initial class rather than rhyme. What distinguished the Mixed Categories tangraphs from the rest? What if they were tonally ambiguous: i.e., pronounced with a level tone in some dialects but with a rising tone in others? Could Mixed refer to mixed tones? Could we then interpret the three-way division of Tangraphic Sea as:

Level tone volume - tangraphs pronounced with a level tone in all dialects

Rising tone volume - tangraphs pronounced with a rising tone in all dialects

Mixed Categories - tangraphs pronounced with either a level or a rising tone depending on dialect

There is one problem with this interpretation. If Mixed Categories tangraphs were 'ambitonal', their fanqie should not have tonally unambiguous tangraphs as final spellers. Yet they do: e.g.,

TT1718 THREE sọ 1.70

final speller is TT3228 PIECE/LUMP thọ 1.70 from the level tone volume

TT3344 and TT4105, two tangraphs for the same word: MAN dzywo 2.44

final speller is TT1110 窮盡 END? sywo 2.44 from the rising tone volume of Precious Rhymes of the Tangraphic Sea (the rising tone volume of Tangraphic Sea has not survived)

Nonetheless I cannot help but wonder if there was something about the Mixed Categories tangraphs that made them unsuitable for indexing by tone. Do Mixed Categories tangraphs have more varied Tibetan transcriptions than other tangraphs? Did the compilers of Tangraphic Sea place all syllables with complex rhyme correspondences into Mixed Categories? Are Mixed Categories tangraphs treated slightly differently from the others in Homophones? (Surely a very different treatment would have been noticed by now.)

But before I tackle the Mixed Categories, I should first figure out what to do with the rest of the Sea.

Next: Maybe there really was a Tangraphic Sea dialect.

*There are a few interesting cases in which the Tangut dictionary tradition explicitly lists two readings for a tangraph (Nishida 1966: 554). These unusual tangraphs have a Chinese loanword reading as well as a presumably native reading. It is possible that some Tangut dialects had both words (borrowed and native) and that some had one or the other.

**I don't consider the Tibetan transcriptions a single source, since van Driem and Kepping (1991: 127) have identified at least four individuals' handwriting styles, and there is no guarantee that each man was only transcribing a single variety of Tangut.

***Coblin (2006: 102-103) wrote,

The 'Phags-pa Chinese orthography that was meant to write this composite language may have been a "pan-koine" diasystem, in that it could be read and pronounced by speakers of the different koine sub-types included under the rubric "'Phags-pa Chinese" ...

[If this hypothesis is correct], it would force us to set aside efforts to restore a unitary 'Phags-pa Chinese sound system as such. Instead, we must begin thinking in terms of the different koine sub-types which are encompassed by the orthography. It is these and only these which will have had historical reality as varieties of early Chinese pronunciation.

This could be paraphrased:

The Tangraphic Sea rhyme and fanqie system that was meant to record this composite language may have been a 'pan-koine' diasystem, in that it could be read and pronounced by speakers of the different koine sub-types included under the rubric 'Dictionary Tangut' ...

If this hypothesis is correct, it would force us to set aside efforts to restore a unitary Dictionary Tangut sound system as such. Instead, we must begin thinking in terms of the different koine sub-types which are encompassed by the orthography. It is these and only these which will have had historical reality as varieties of Tangut pronunciation. RIDDLE OF THE RHYMES (PART 3)

In Part 2, we saw that according to Nishida (1964: 49), rhyme 1.27/2.25 was transcribed with five different Tibetan vowel letters: a, i, u, o, ï - every vowel letter but e! Did a single rhyme really have so much variation, or is there a single vowel underlying all this transcriptive variation? Before drawing the latter conclusion, we should question its premise: that 1.27 and 2.25 really were one atonal rhyme.

Not all researchers agree on the number of atonal rhymes in Tangut, though there is a general consensus on the pairing of most level and rising tone rhymes:

Hashimoto (1962 chart, published in 1963, 1965): 98 rhymesNishida (1964): 102 rhymesSofronov (1968), Li Fanwen (1986), Gong (1997): 105 rhymes
1.1-1.4 : 2.1-2.4
unclear how 1.5-1.7 map onto 2.5-2.6 1.5 : 2.5
1.6 : no 2.X counterpart
1.7 : 2.6
1.8-1.12 : 2.7-2.11
1.13 : no 2.X counterpart
1.14-1.15 : 2.12-2.13
1.16: no 2.X counterpart
1.17-1.24 : 2.14-2.211.17-1.22 : 2.14-2.19
2.20 : no 1.X counterpart
1.23 : 2.21
1.24 : 2.22
2.22 : no 1.X counterpart
1.25-1.30 : 2.23-2.28
1.31 : no 2.X counterpart
1.32-1.37 : 2.29-2.34
1.38 : no 2.X counterpart
1.39 : 2.35
1.40 : no 2.X counterpart
1.41-1.46 : 2.36-2.411.41-1.45 : 2.36-2.40
1.46 : no 2.X counterpart
2.41 : no 1.X counterpart
1.47, 1.48 : no 2.X counterparts
1.49-1.80 : 2.42-2.731.49-1.67 : 2.42-2.601.49-1.56 : 2.42-2.49
1.57 : no 2.X counterpart
2.50 : no 1.X counterpart
1.58-1.67 : 2.51-2.60
1.68 : no 2.X counterpart
1.69-1.72 : 2.61-2.64
1.73 : 2.66 (sic)
1.74 : 2.65 (sic)2.65 : no 1.X counterpart
1.74 : 2.68
2.67 : no 1.X counterpart
2.68 : no 1.X counterpartsee above
1.75-1.77 : 2.69-2.71
1.78 : no 2.X counterpart
1.79-1.80 : 2.72-2.73
unclear how 1.81-1.85 map onto 2.74-2.771.81 : no 2.X counterpart
1.82-1.84 : 2.74-2.761.82 : 2.74
1.83 : no 2.X counterpart
2.75 : no 1.X counterpart
1.84 : 2.76
1.85 : no 2.X counterpart
1.86 : 2.77
1.86 : 2.78
unclear how 1.87-1.97 map onto 2.79-2.861.87-1.91 : 2.78-2.82
2.83 : no 1.X counterpart
2.84 : no 1.X counterpart
1.92-1.93 : 2.85-2.86
1.94-1.97 : no 2.X counterparts

All agree that 1.27 and 2.25 formed a pair: i.e., if 1.27 were -V(C)1, then 2.25 was -V(C)2 (i.e., the same vowel [and coda?] with a rising instead of a level tone). Here are their reconstructions of that pair compared with the Tibetan transcriptions:

RhymeTibetan transcriptions from Sofronov (1968: 21)Hashimoto (1965)Nishida (1964)Sofronov (1968)Li Fanwen (1986)Gong (1997)
1.27-a, -aa*, -i, -iH, -u, -o-aay1-uH11-u11
2.25-iH, -u, -uH, -eH, -o-aay2-uH22-u22

Nishida (1964: 49)'s transcription data also includes the rhyme -ïH not in Sofronov's data. (Nishida did not specify whether -ïH was used to transcribe the rhyme with one or both tones.) Thus all six Tibetan vowel letters were used to write this rhyme.

If this was a single rhyme, this indicates that the rhyme contained a vowel without a Tibetan counterpart: e.g., u (high central rounded; Nishida and Li) or ə (Sofronov and Gong).

Hashimoto's -aay is doubtful because it could have been transcribed as -aHi in Tibetan** and would not have been transcribed as -u(H) or -o.

Guillaume Jacques (2003) found the following correspondences between Tangut and the rGyalrong of gDong-brgyad:

Tangut rhymerGyalrong of gDong-brgyad (GDB)

Although there is no overlap between the GDB counterparts of 1.27 and 2.25, this does not necessarily mean that 1.27 and 2.25 had different vowels. This could simply be a side effect of the small number of (known) cognates. Moreover, the comparative study of Tangut and rGyalrong has just begun. Some of those correspondences may not hold up, and others may be discovered in the future.

It is likely that 1.27/2.25 came from several different rhymes in Proto-Qiangic, the common ancestor of Tangut and rGyalrong. It is even possible that some Tangut dialects kept these rhymes distinct, and that these distinctions are reflected in the Tibetan transcriptions***. Nonetheless, the issue is whether 1.27 and 2.25 formed a single class or not, and this should ideally be determined on the basis of Tangut-internal data.

If we look on page 3A of Homophones, we find 1.27 and 2.25 tangraphs placed in the same homophone group (I.16):

3A55 TT3678 THROW-AWAY 1.27

3A56 TT2604 禮儀 ETIQUETTE 1.27

3A57 TT0165 貨 COMMODITY 1.27

3A61 TT2032 (part of the name of an ancestor of the Tangut) 1.27

3A62 TT2162 牛 OX 2.25

(followed by nine other 2.25 tangraphs: 3A63-3A73)

3A74 TT2033 筵 BAMBOO-MAT 1.27

3A75 TT2163 (part of the name of a vegetable) 2.25

(followed by five [or six?****] other 2.25 tangraphs: 3A76-3B12)

If 1.27 and 2.25 were segmentally (i.e., nontonally) identical in Homophones, then we could conclude that the Tangraphic Sea rhymes 1.27 and 2.25 were also segmentally identical. Should we?

Next: The limits of similarity.

*In the transcription glaa, written as ga + subscript la + subscript Ha.

**The -aHi in Tibetan transcriptions of pre-Tangut period northwestern Chinese presumably represents a rhyme like -ay. It corresponds to -ai-type rhymes in modern languages: e.g.,

塞 Tib tr. saHi : Md sai 'place of strategic importance'

***Imagine that a far-future scholar has Roman transcriptions of Cantonese and a Mandarin-based rhyme dictionary in sinography. If one assumed they reflected the same language, one would be very puzzled by correspondences like these:

SinographTranscription of Cantonese rhymeMandarin rhyme dicitonary class
-ap-X (actually -ù, unbeknownst to the hypothetical researcher)

Should someone reconstruct a single 'Chinese' rhyme represented by all this data and write off the Cantonese final consonants as 'silent' (or 'ignored' in cases like 務父茹)? In reality, Cantonese has maintained rhyme distinctions that have been lost in Mandarin.

Similarly, the Tangraphic Sea rhyme classes 1.27 and 2.25 could correspond to multiple rhymes in the Tangut dialect(s?) that were transcribed into Tibetan. Hence it might be a mistake to assume that all the vowel variation in the transcriptions represents a single 1.27/2.25 rhyme class.

The many-to-one correspondences between Cantonese and Mandarin may also be analogous to the situation between rGyalrong and Tangut.

****In Li Fanwen's (1986) version of Homophones, 3B13 TT5409 2.25 is not in the 1.27/2.25 group, but is in the r 1.84 / myï 1.30/2.28 group. But TT5409 is in the expected group (22 by Sofronov's count) according to Sofronov's (1968 II: 395) index.

TT5409 has been glossed as 青 BLUE-GREEN (Li 1986: 209) and as 黃色 YELLOW (Shi et al. 2000: 240). Maybe a study of Tangut color terms would prove both interpretations to be correct. RIDDLE OF THE RHYMES (PART 2)

If Tangraphic Sea were a rhyme dictionary, the Tibetan transcriptions for tangraphs in each of its sections should ideally have the same rhyme: e.g.,

Rhyme 1: pu, tu, ku, tsu, chu, lu ...

Rhyme 2: puH, tuH, kuH, tsuH, chuH, luH ...

Rhyme 3: pi, ti, ki, tsi, chi, li ...


Of course, the reality is far messier. Here are all of the transcriptions that Sofronov (1968 II.6) listed for rhyme 1.1:

-u: pu, bu, gu, shu
-i: ñi, gni

-iH: gniH

-u, -i, -iH appear in transcriptions of tangraphs from other level tone rhymes as well (Sofronov 1968 II.6-54). (3.13.2:09: I've added -uH to the table to see if the distribution of -u and -uH is parallel to the distribution of -i and -iH.)


16 rhymes all seem to have contained an u-like vowel. But how did those 16 rhymes differ from each other? Did Tangut have 16 kinds of u? 23 kinds of i?

The problem is hardly confined to two vowels. Here are all the Tibetan transcription rhymes found in Nishida (1964) organized by rhymes disregarding tone. Note that Nishida's transcription data differs slightly from Sofronov's: e.g., he has no -i in his transcriptions for 1.1/2.1. Some Tibetan transcription rhymes (e.g., -ui, -eng) are absent since they represent unknown Tangut rhymes (Nishida 1964: 111-112).

(Set your browser font to the smallest size for optimum viewing.)

1.5/2.5no Tibetan transcription data
1.8/2.7no Tibetan transcription data
1.15/2.13no Tibetan transcription data
1.18/2.15no Tibetan transcription data
1.22/2.19no Tibetan transcription data
1.23/2.21no Tibetan transcription data
1.25/2.23no Tibetan transcription data
1.31no Tibetan transcription data
1.38no Tibetan transcription data
1.44/2.39no Tibetan transcription data
1.47no Tibetan transcription data
1.52/2.45no Tibetan transcription data
1.60/2.53no Tibetan transcription data
1.62/2.55no Tibetan transcription data
1.65/2.58no Tibetan transcription data
1.70/2.62no Tibetan transcription data
1.73/2.66 (sic)-e
1.83/2.75no Tibetan transcription data
1.85no Tibetan transcription data
1.88/2.79no Tibetan transcription data
2.83no Tibetan transcription data
1.94no Tibetan transcription data

-a appears in transcriptions for 12 rhymes, and the figures for e and o are also large: 22 rhymes and 17 rhymes. is 'only' in 4 rhymes. Are these really rhymes at all?

Next: How did Nishida pair level and rising tone rhymes? Would separating them help? We need all the help we can get! RIDDLE OF THE RHYMES (PART 1)

The table in "The Tibetan Test" gave me the impression that the Tibetan alphabet was more or less able to represent Tangut initials with a few exceptions: e.g., whatever was supposed to be represented by the un-Tibetan cluster ldth-.

However, the Tibetan alphabet was almost certainly inadequate to represent the rest of the Tangut syllable.

Tangut had at least 97 rhymes. The figure of 97 comes from the level tone volume of the Tangraphic Sea. The rising tone volume has 86 rhymes and the entering tone section in the Precious Rhymes of the Tangraphic Sea has 11 tangraphs with an unknown number of additional rhymes. If one counts each rhyme-tone combination as a separate rhyme, then Tangut had 97 + 86 + (X < 12) = 183 to 194 rhymes. Scholars ignoring tones have proposed figures ranging from 98 (Hashimoto) to 105 (Sofronov and Gong).

The Tibetan alphabet used in the transcriptions had only six vowel letters: a, i, u, e, o, and ï. The phonetic value of ï is uncertain.

Written Tibetan allowed 10 consonants in the coda position:

- the nasals m, n, ng (but not ñ)

- their stop counterparts b, d, g

- the sonorants l, r

- the fricative s

- which could combine with most nondentals for cluster codas:

nasals: ms, ngs

stops: bs, gs

(but not ns, ds, ls, rs, ss;

Hs is also not possible)

- H, whatever it may have been

The total number of codas is 15 (< 10 single consonants + 4 s-final clusters + zero consonant).

If a Tibetan adhered to Tibetan spelling conventions, he could theoretically write 6 x 15 = 90 different rhymes. I don't know if every single one of these rhymes existed in Tibetan, and in fact most of them do not appear in the Tibetan transcriptions of Tangut. The 21 rhymes attested in Nevsky's (1926) collection of 502 transcriptions for 334 different tangraphs and the transcriptions in Nishida (1964)* are in bold.


The rhymes -am, -an, -in are extremely rare. They each occur only once in Nishida (1964: 48, 61) and are therefore not a solid foundation for reconstructing the codas -m and -n. Tangut rhyme 1.21/2.18 was transcribed in Tibetan as -aH and -eH as well as -am (Nishida 1964: 48). -an is the only Tibetan transcription of Tangut rhyme 1.24/2.22.

Nishida's -in could be a typo for -ing which appears only once in Nevsky (#132) for TT3797 COLOR 1.68: rtsing ~ rtsiH ~ rtsi. Nishida (1964: 61) listed the transcriptions rtsin, rtsiH, rtsi for unspecified tangraph(s) in Homophones group VII.11: could this have been TT3797?

There are five rhymes not in the chart:

-aT: Nishida (1964: 65) has one instance of a transcription phaT with the Tibetan reversed t for Sanskrit retroflex T. Although Nishida does not cite the tangraph transcribed as phaT, I am sure it represented the Sanskrit magical syllable phaT. Thus it is not evidence for an indigenous Tangut rhyme -aT (or -at).

-ou?: Nevsky has one instance of a transcription combining the letters o and u (#114). Since other transcriptions for the same tangraph (TT3709 ABOVE 2.3) have o (pho and phoH), I presume that o + u represented a Tangut ou, though other values (uo or a vowel between u and o) cannot be ruled out. Gong reconstructed ABOVE as phyu.

-i with subscript -shi (Nevsky #254): Was thishi this an attempt to write thish? This tangraph (TT2214 THIS 2.28) also was transcribed as thi. The resemblance between thishi and this is, of course, totally coincidental. (Added 3.12.00:48.)

-ui: Nishida (1964: 111) listed Hdzwu and Hdzwui as transcriptions of unspecified tangraph(s) in Homophones group VII.41. (Added 3.12.00:52.)

-oong?: Nevsky (#269) romanized one Tibetan transcription for TT1718 THREE as oo even though his handwritten copy of the transcription looks like gswong to me. I suspect that either the romanization or his copy of the transcription is incorrect. The other transcriptions of THREE are gso, gsoH, gswong. (Added 3.12.1:03.)

The transcribed Tangut dialect(s) apparently only had up to five codas (-m, -n, -ng, -r, -H), assuming that they actually represent consonants rather than vowel features (nasality? retroflexion? breathiness or length?). Ignoring the five rhymes not in the chart, there could have been 36 Tangut rhymes (< 6 vowels x 6 codas, including zero; actually attested rhymes in bold)


but that figure is only one-third of the roughly 100 Tangut rhymes. So what were the other 60-plus rhymes? But before I tackle that question, I should ask a more fundamental question first:

Next: How do we know that Tangraphic Sea is a rhyme dictionary?

* transcription rhymes found only in Nishida (1964) are

(p. 44: 1.10/2.9, 50: 1.29/2.27, 52: 1.35/1.32)

-am (p. 48: 1.21/2.18)

-an (p. 48: 1.24/2.22)

-ïH (p. 49: 1.27/2.25)

-in (p. 61: 1.68)

-aT (p. 65: 1.82/2.74)

-ui (p. 111: rhyme class unknown to Nishida; 1.5/2.5 in Gong's system)

The notation 1.X/2.Y indicates a Tangut rhyme category with tonal variants 1.X (level tone) and 2.Y (rising tone). Some Tangut rhyme categories have only one tone: e.g., 1.68 has no 2.Y counterpart.

Tangut fonts by Mojikyo.org
All other content copyright © 2002-2007 Amritavision