After this I think I'll be really be done with the topic of Tangut rhyme 101 (1.93/2.86) for some time.

I have updated my database of Tangut readings (download version 1.2 here) to incorporate the changes I have proposed in this series:

- the reinterpretation of rhyme 101 as -er' instead of -ir' (see part 1)

- the reassignment of

2705 'to help; right side of character (i.e., assistant)' and 2928 'to explain, note' (with 'speech' on its right side; probably a different spelling of a specialized usage of 'to help')

from rhyme 2.54 to 2.86 (see "Explaining the 'Right' Reading")

I didn't upload version 1.1 in which I replaced a lot of symbols for nasalization with glides after mid vowels, bringing my transcription closer to Gong's reconstruction: e.g.,

-en > -ey (corresponding to Gong's -əj, -iəj, -jɨj)

-on > -ow (corresponding to Gong's -ow, -iow, -jow)

I have retained the nasalization from 1.0 in 1.2. THE 'RIGHT' RHYME (PART 5)

I didn't expect to write a five-part series on this topic, but I forgot to mention the Tangut-internal evidence on the rhyme in part 4, so it's getting its own part.

"It" consists of alternations between rhyme 101 and other rhymes (70 -iq3/4 and 84 -ir3/4) in what superficially appear to be synonym pairs (Gong 2002: 103):


1. 5683 2er'4 ~ 5209 2iq4 'to stretch, lengthen'


2. 1928 1ler'3 ~ 5850 1liq3 'to rub'


3. 5742 1tser'4 ~ 3641 1tsir4 'to choose'

Gong reconstructed rhyme 101 as -iir which matched the i of the non-101 members of those pairs. However, the placement of rhyme 101 in the Tangraphic Sea (see part 1) and the Tibetan and Chinese transcription evidence (see parts 2 and 3) point to a nonhigh vowel. The merger of *-ir' with *-er' that I first proposed in part 1 can account for the vowel mismatch.

Those three pairs could be reconstructed in accordance with the proposals in part 4 as

1. 'to stretch, lengthen'

*r((ɯ)-s)ɯ-ʔa-X/*r((ɯ)-s)ɯ-ʔe(n/ŋ)-X ~


2. 'to rub'

*r((ɯ)-s)ɯ-la-X/*r((ɯ)-s)ɯ-le(n/ŋ)-X ~


3. 'to choose'

*rɯ-tsa-X/*rɯ-tse(n/ŋ)-X ~



*Cɯ-tsar-X/*Cɯ-tser-X ~


All three pairs require a -vowel presyllable to condition the raising of *a and/or *e to their Grade III/IV reflexes.

The rhyme 101 members of pairs 1 and 2 must have had a prefix *r(ɯ)- to condition retroflexion absent in their rhyme 70 counterparts ending in -iq. If their bases had *sɯ-prefixes, then there is no need to reconstruct  after *r- since the  of *sɯ- would be sufficient to condition Grades III/IV. But the possibility of *rɯ-sɯ- cannot be ruled out.

All three pairs involve the presence or absence of the mysterious factor *X that I have arbitrarily written at the ends of syllables.

Without external evidence, there is no way to narrow down the possibilities.

And without narrower definitions, there is no way to be sure about the functions of the various affixes. (*X may not have been an affix, though for convenience I write it as if it were a suffix.) THE 'RIGHT' RHYME (PART 4)

I originally wanted to end this series with comparative evidence for Tangut rhyme 101 (1.93/2.86) words, but I don't know of any. Which is not surprising as there are only thirteen characters with a total of five different readings ending in rhyme 101 (Arakawa 1997: 91):

Homophones initial class Tone 1 Tone 2
I   2ber'4
V 2ker'4
VI 1tser'4 2tser'4
IX 1ler'3  

Grades III and IV are normally in complementary distribution. If there are no Grade III/IV minimal pairs in a rhyme, I assign Grade III to Class II and VII consonants and the Class IX consonant l-. The default grade for all other initials including the Class IX consonant lh- is IV. This assignment parallels the general pattern of distribution of initials in rhymes that have Grade III and IV minimal pairs. The different distribution of l- and lh- suggests that they did not simply differ in terms of voicing. I think l- was velar [ɫ] (as in Nishida's 1ɫĭə̣r corresponding to my 1ler'3) whereas lh- was a fricative [ɬ]. (Note that Nishida reconstructed l- and ɫ- as distinct initials, whereas I think /l/ was always [ɫ] except before Grade IV rhymes where it was [l].)

Nishida (1964: 67) was the first to identify rhyme 101 as retroflex, and that classification has been carried over into the reconstructions of Arakawa, Gong, and this site. (Sofronov does not reconstruct retroflexion in Tangut vowels in any of his three reconstructions.)

Vowel retroflexion has two sources in Tangut: (pre)initial *r- and final *-r. The r- of Tibetan transcriptions of Tangut rhyme 101 syllables may directly reflect a preinitial r- preserved in a nonstandard Tangut dialect (see part 2). Preinitial *r- may have been *rɯ- with a high vowel conditioning Grade III/IV.

Nishida (1964: 67) also reconstructed tension in rhyme 101. Gong reconstructed preinitial *s- as the source of tension in Tangut vowels, though he reconstructed length rather than tension as an extra nonretroflex feature in rhyme 101: -iir. If Nishida and Gong are both right, the sources of rhyme 101 syllables would have preinitial *s(-r)- or *(r-)s-.

I can't rule out Nishida's tension, but I do not think Gong's length was present in this rhyme or any other, at least not during the Tangut imperial period. Sanskrit has phonemic vowel length that does not correlate with Gong's reconstructed vowel length in the readings of Tangut transcription characters. For now I simply acknowledge that this rhyme was somehow different from regular -er3/4, and indicate that difference with an apostrophe that I call 'prime'. I arbitrarily indicate the unknown source of 'prime' as a final *-X in pre-Tangut. The position of *-X is simply carried over from *-'; the actual conditioning factor of -' could have been anywhere in the syllable. I have never seen -'/*-X correlate to anything in any other Sino-Tibetan language. It is remotely possible that Tangut preserves something lost in the rest of its gigantic family, but I am hesitant to make such an extreme claim until I look harder. Which won't be tonight. All I can say for now is that *-X seems to have blocked raising in rhyme 40, the nonretroflex counterpart of 101:

R10 -i3 < *Cɯ...-en, *Cɯ...-eŋ

R11 -i4 < *Cɯ...-en, *Cɯ...-eŋ

R40 -e'3/4 < *Cɯ...-enX, *Cɯ...-eŋX

and perhaps *Cɯ...-anX, *Cɯ...-aŋX?

Integrating the above proposals (other than *s-) with Guillaume Jacques' sources for -i/e3/4 (in my notation; -ji(j) in his) and my hypotheses of mergers from part 1 and part 3, I have come up with up to twelve possible sources of -er'3/4:

Early pre-Tangut Late pre-Tangut Standard Tangut
*rɯ-CaX *Cir'3/4 Cer'3/4
*r(ɯ)-CukX *Ciwr'3/4
*rɯ-CekX *Cewr'3/4
*rɯ-CanX *Cer'3/4

Obviously not all twelve had to exist in pre-Tangut. THE 'RIGHT' RHYME (PART 3)

Nishida (1964: 67) listed four Chinese characters that transcribed Tangut rhyme 101 (1.93/2.86) syllables or vice versa (the directionality is not clear):

精 *1tse4

賊 *4tshy1

造 *3tshaw1

草 *2tshaw1

I would add a fifth: *3me3 (Timely Pearl 32.2.8), a transcription of

2705 'to help; right side of character (i.e., assistant)'.

The first and fifth end in an -e that matches the -e- of -er', my proposed transcription of rhyme 101 in part 1. But how I do explain the other three? Where are they from? Kwanten (1982) and Li Fanwen's (1994) indexes to the Chinese character transcriptions of the Golden Guide do not list them as corresponding to rhyme 101 characters. Here are the rhymes of Tangut characters transcribed by 賊造草 on pp. 499, 511, and  of Li's index which is more complete than Kwanten's (which does not list 造草):

Li index 0991 賊 *4tshy1 < *-ək (phonetically [tsʰəɰ] or [tsʰeɰ]?; tsej or tsɛj in modern northwestern dialects which seem to have overrun the earlier dialects like the one known to the Tangut)

R 2.7 -i1 (12.6, 19.4; phonetically something like [əj] or [ej]?)

R 1.30 -y4 (5.4, 8.3, 16.4, 16.4; phonetically something like [jɨ]?)

R 1.32 -y'4 (21.1, 32.5; phonetically something like [jɨ] plus the mystery 'prime' quality indicated by -')

R44 2.38 -ew1 (33.6)

Li index 0441 造 *3tshaw1 (phonetically [tsʰɑw]?)

R44 1.43 -ew1 (21.6, 23.1)

Li index 0556 草 *2tshaw1 (phonetically [tsʰɑw]?)

R44 1.43 -ew1 (13.6, 21.4, 23.6)

Without knowing the exact nature of the correspondences between 賊造草 and Tangut characters that Nishida did not specify, I can only make a few speculations (which are partially mutually exclusive: i.e., they can't all be right):

- The retroflex quality of -er' may correspond to Chinese *[ɰ] and *[w].

- Chinese *-aw reflects a nonstandard Tangut rhyme *-ewr' that merged with *-er' in the standard dialect,

- The correspondences of -er' with [ə] and [ɑ] suggest that it may have been centralized and/or lower mid.

- Although I mechanically converted Gong's Grade III -j- into my -3/4, if rhyme 101 corresponds to Chinese Grade I (賊造草) as well as Grades III and IV (命精), perhaps it contains three subrhymes, -er'1, -er'3, and -er'4. Or rhyme 101 could be ... gradeless? (I just noticed Sofronov 2012 also lists this rhyme as gradeless; his reconstruction is -ɪ.)

Next: What are the pre-Tangut sources of -er'? THE 'RIGHT' RHYME (PART 2)

In part 1, I proposed reinterpreting Tangut rhyme 101 (1.93/2.86) as -er' instead of -ir'. A mid vowel e fit the Chinese transcription *3me3 (Timely Pearl 32.2.8) for

2705 'to help; right side of character (i.e., assistant)'

better. And if rhyme 101 had e, that would be the vowel expected after y in rhyme 100 following the usual order of Tangut vowels.

In this part, I will start to look at the transcriptional evidence for rhyme 101.

1. Sanskrit transcription evidence

There isn't any. This tells us that 101 probably didn't sound like anything in Sanskrit. V'-rhymes are rare in Tangut transcriptions of Sanskrit and Vr'-rhymes seem to be nonexistent. That tells me that the unknown quality that I write with a prime symbol was absent from Sanskrit.

2. Tibetan transcription evidence

Tai (2008: 229) lists 22 transcriptions of two tangraphs with rhyme 101:

0467 1tser' 'method, art, skill, dharma'

transcribed as rtsi (x 1), rtse (x 5), rdze (x 1), rc? (x 1)

2698 2tser' 'nature, character'

transcribed as rtse (x 12), ?e (x 1)

Out of 22 transcriptions, 20 end in -e, 1 ends in -i, and 1 ends in an unknown vowel. The obvious conclusion is that the vowel of rhyme 101 was something like Tibetan e.(Why did Gong reconstruct long i instead of long e for rhyme 101?)

The preinitial r- of the transcriptions may either indicate the retroflexion of the following vowel or reflect an actual preinitial r- in the transcribed Tangut dialect corresponding to retroflexion in the standard dialect described in the Tangut phonological tradition:

Tangut dialect transcribed in Tibetan Standard Tangut
CV (plain vowel) CV (plain vowel)
rCV (r- + plain vowel) CVr (retroflex vowel)

If the table above is correct, the dialect transcribed in Tibetan had fewer vowels than the standard dialect; the latter had retroflex vowels absent in the former.

I would expect the Chinese transcriptions of rhyme 101 characters other than 2705 to also contain *e, but we will see that is not the case in part 3. THE 'RIGHT' RHYME (PART 1)

I wouldn't have guessed that

2705 2bir'4 'to help; right side of character (i.e., assistant)'

was transcribed in the Timely Pearl as *3me3 (32.2.8) with -e rather than -i. Why not transcribe it as, say, 彌 *1mbi4 with *-i?

It is a fact that 2705 is listed under rhyme 2.86 in the Precious Rhymes of the Tangraphic Sea.

On the other hand, it is merely a hypothesis that rhyme 2.86 was 2-ir'4: i.e., second tone Grade IV i with retroflexion and some unknown quality marked as -'. Should -ir' be -er' with a mid vowel like the transcription *3me3 of 2705?

In my systerm, there is no -er' or -ur', though all other Tangut vowel types are represented in -Vr' rhymes: -ir', -ar', -yr', -or'. Did *-ir' and *-er' merge into one rhyme while *-ur' and *-or' merged into another? It would be neat if the merged rhymes were of the same height: e.g., mid -er' and -or' or high -ir' and -ur'.

The usual order of vowels in the Tangraphic Sea is u-i-a-y-e-o (with iw/ew between e and o). This order is not followed in the -Vr' rhymes. Reinterpreting rhyme 101 (1.93/2.86) as -er' (cf. Arakawa's -yer2) would make the -Vr' rhyme order closer to the norm:

88. 1.83 -ar'1

89. 2.75 -ar'3/4

99. 2.84 -ir'1 (a merger of *-ir'1 and *-er'1?)

100. 1.92/2.85 -yr'3/4

101. 1.93/2.86 -er'3/4 (formerly written as -ir'3/4; a merger of *-ir'3/4 and *-er'3/4?)

102. 1.94 -or'1 (a merger of *-ur'1 and *-or'1?)

103. 1.95 -or'3/4 (a merger of *-ur'3/4 and *-or'3/4?)

(The absence of -Vr'2 rhymes may tell us that the phonetic quality of -' was incompatible with Grade II which was at least partly from *-r-. Retroflex vowels were conditioned by [pre]initial *r- or final *-r but not *-r-.)

The only remaining oddity in the order of -Vr' rhymes is the placement of 88-89 -ar' not only before 99 -ir' but also in the middle of the -Vr rhyme sequence:

77. -er1

78. -er2

79. -er3/4

80. 1.75/2.69 -ur1

81. 1.76/2.70 -ur4

82. 1.77/2.71 -ir1

83. 1.78 -ir2

84. 1.79/2.72 -ir3/4

85. 1.80/2.73 -ar1

86. 1.81 -ar2

87. 1.82/2.74 -ar3/4

88. 1.83 -ar'1

89. 2.75 -ar'3/4

90. 1.84/2.76 -yr1

91. 1.85 -yr2

92. 1.86/2.77 -yr3/4

93. 1.87/2.78 -ewr1

94. 1.88/2.79 -iwr4

95. 1.89/2.80 -or1

96/97. 1.90/2.81 -or2/3/4

The placement of the -er rhymes (77-79) before the -ur rhymes (80-81) instead of after the -yr rhymes (90-92) defies explanation.

Next: The transcriptive evidence for the 'right' rhyme. EXPLANING THE RIGHT READING

Until now I've been reading the character

2705 'to help; right side of character (i.e., assistant)'

in Tangut character analyses as 2beq4 (a reading converted from Gong's reconstruction 2bjịj in Li Fanwen's 1997 dictionary). But last night I discovered that it should be 2bir'4 (a reading converted from Gong's reconstruction 2bjir in Li Fanwen's 1997 dictionary).

2705 and its homophone

2928 'to explain, note' (with 'speech' on its right side; probably a different spelling of a specialized usage of 'to help')

form a two-character homophone group in the first chapter of Homophones. All characters in that first chapter have readings with labial initials (p-, ph-, b-, m-). The Timely Pearl transcription of 2705 is *3me3 (32.2.8); the diacritic indicates b- rather than m-.

I have no idea how Nishida, Sofronov, and Gong determined the rhyme

N -ɛ̣ 2.54

S -ɪ̭e 2.? (no rhyme number listed on II: 307, but implictly 2.54 since its transcription 命 is listed under 2.54 on II: , presumably a typo for -ɪ̭ẹ since -ɪ̭e 2.54 should either be 2.35 or 2.37 according to I: 137)

G -jịj 2.54

= my 2-eq4

before the rediscovery of the Precious Rhymes of the Tangraphic Sea which listed 2705 under rhyme 2.86 (-ir'). 2928 is not in PRTS, but its rhyme must be identical to that of 2705 since the two are homophones.

(8.29.0:59: Perhaps 2.54 was determned by a process of elimination. The Chinese transcription *3me3 most likely reflected a Tangut reading like be4 (be3 would be anomalous in Tangut). 2705 could not have had the first tone because 2705 was not in the Tangraphic Sea's volume for first-tone characters. 2705 was thus thought to be in the lost second-tone volume [and later it indeed was found in the second-tone volume of the Precious Rhymes of the Tangraphic Sea]. But was it 2be2 [2.33], 2be'2 [2.35], 2beq4 [2.54], or 2ber4 [2.68]? 2705 was not in the homophone groups thought to be for 2be2 and 2be'2, so 2beq4 and 2ber4 were the remaining possibilities. I don't know why 2.54 was favored over 2.68. The actual rhyme turned out to be 2.86 - which was none of the above!)

All this makes me wonder how many other second tone readings have been revised in Li Fanwen's 2008 dictionary in accordance with the Precious Rhymes of the Tangraphic Sea.

Tangut fonts by Mojikyo.org
Tangut radical and Khitan fonts by Andrew West
Jurchen font by Jason Glavy
All other content copyright © 2002-2015 Amritavision