Andrew West pointed out that 5952 'ore, mine' from my last entry does occur outside dictionaries: e.g., in the 'ritual language' line 4A of the section on the tenth month in The Ode on Monthly Pleasures:

10.4A 1 2 3 4 5 6 7

Li Fanwen number 2992 0026 5429 2431 5952 5072 1420
Reading 2bia 2ŋwʊ 1po 2khwa 2nɤa' 1ɣiəʳ 2tʂɨụ
Gloss the Ba clan territory the Po clan Chinese (? - see below) make wing

The corresponding 'common language' line does not seem to match it at first glance:

10.4B 1 2 3 4 5 6 7

Li Fanwen number 1518 3437 2344 5882 0795 3497 4999
Reading 1tʂɨẹ 2ʔõ 2mi 1ɮaʳ 2riəʳ 2lʊ̣ 1giẹ
Gloss the Che clan the On clan Tangut Chinese PERF obstruct scissors

Nishida (1986: 62) translated those lines as topic-comment sequences:

4A. 'As for the Po and Chinese of the Pa [= my Ba*] territory - wings made of iron'

4B. 'As for the Tangut and Chinese of the Che Hon [= my Che On**] - scissors that obstruct'

He proposed four parallel pairs:

4A1-2 'Pa [= Ba] territory' : 4B1-2 'Che Hon [= Che On]'

4A3 'Po' : 4B3 'Tangut'

4A4 'Chinese' : 4B4 'Chinese'

4A5-7 'wings made of iron' : 4B5-7: 'scissors that obstruct'

Only the third pair is obvious.

I have suspected that the 'ritual language' is a non-Sino-Tibetan substratum language glossing the superstratum Sino-Tibetan 'common language'. For other interpretations of these 'languages', see Andrew's "The Myth of the Tangut Ritual Language".

But my hypothesis is problematic even for the third pair. If 4A4 2431 2khwa is a substratum word for 'Chinese' glossing the superstratum word 4B4 5882 1ɮaʳ 'Chinese', why do both words appear in the foreword to the Timely Pearl in the Palm, a text otherwise in the 'common language'? Is that a case of a substratum word borrowed by the superstratum language? Neither word has any connection to Chinese autonyms. (The Chinese autonym 漢 *xã 'Han' was borrowed as

5916 1xã

which like 2431 and 5882 is written with the unflattering components 'little' and 'insect'.)

The fourth pair only makes sense if 4A5-7 2nɤa' 1ɣiəʳ 2tʂɨụ 'wings made out of 5592' was the substratum term for 4B7 4999 1giẹ  'scissors'. 4999 can also be a verb 'cut' - is 'scissors' a derived noun 'cutter'? - but Kychanov and Arakawa 2006: 327 do not list any compound verb 3497 4999 'obstruct-cut', and a verb sequence 'obstructed and cut' is even harder to relate to the noun phrase 'wings made of 5592'.

Li (2008: 938) and Kychanov and Arakawa (2006: 296) agree that 5592 means 'ore'. (As Andrew pointed out, none of the textual examples in Li 2008 support 'mine'. Li's English gloss for 5592 seems to be a translation of his Chinese gloss 礦 which means both 'mine' and 'ore', but the Tangut word may have had a narrower meaning.) However, 'ore' might be odd in this context. Nishida translated 5592 as 'iron', even though 'iron' is a distinct word that combines with 5592 to form the phrase

5592 4995 2nɤa' 1ʂɨõ 'iron ore'.

in Homophones 14B33. Could 5592 also refer to some other metal? English ore is "partly from Old English ār brass". Also cf. the ambiguity of Japanese kane 'metal, gold'.

4B5-6 0795 3497 2riəʳ 2lʊ̣̣ 'obstructed' (0795 3497) has no substratum equivalent in line 4A. Was the superstratum verb understood even by speakers of the substratum language, or was it just omitted to make line 4A the same length as 4B?

Why would glosses have to match the lengths of the lines they glossed? Were the glosses meant to be poetry in their own right? And why place the glosses before the lines they gloss?

The '-ed' of my translation 'obstructed' corresponds to the perfective prefix 0795. Nishida (1966: 579) regarded 0795 as the Tangut equivalent of Classical Chinese 所, so I would have expected his translation to be 妨げるところ  'that which is obstructed' (cf. his translations of 0795 in 1966: 279), but that would make no sense before 'scissors'. His actual translation 妨げる 'obstructs' only corresponds to 3497.

4A6-7 5072 1420 1ɣiəʳ 2tʂɨụ 'wing made of ...' consists of words also in 'common language' texts. I don't know of any Sino-Tibetan cognates for those words. Could they be substratum loans in the superstratum language?

The first two pairs are the most baffling. When I see Tangut names, I feel as if I'm reading about characters in a TV show I've never seen. Who were the Ba, Po, Che, and On? I don't know, but I'm certain that Po is not a synonym of 2mi 'Tangut'. The names Che and On are together in Homophones 35B37, so they may have been a common collocation ('Che [and] On') or a disyllabic name ('Che'on'). Kychanov and Arakawa (2006: 315) favor the latter interpretation. Was the territory of the Ba also the land of Che and On (or Che'on)? If so, then the Po were a Tangut clan in that land, and 'the Po of the Ba territory' and 'the Tangut of Che (and?) On' refer to the same group of people. This hypothesis is impossible to explore further without learning more about these clan names.

4A2 0026 2ŋwʊ 'territory' in the 'ritual language' line is also in 'common language' texts. Is it a superstratum loan in the substratum, or vice versa? Like 'make' and 'wing' later in that line, it too has no known cognates.

*8.10.3:23: I follow Gong (1997) and Li Fanwen (1986: 218) who reconstructed the initial of 2992 as b-. Nishida (1986: 62) reconstructed it as p-; in 1966 he reconstructed it as m- (p. 396). Sofronov (1968 II: 307) reconstructed it as mb-. The character is in the labial chapter of Homophones, so there is no doubt that its reading had a labial initial of some kind. The only transcription for a member of its homophone group that I know of is *mba for

2bia 'belly'

in Timely Pearl in the Palm 19.1. I think the diacritic might indicate that the initial was b- (absent from Tangut period northwestern Chinese) rather than mb-. In any case, the Chinese transcription rules out Nishida's later reconstruction p-; it tells us that the initial was voiced (though whether it was nasal or prenasalized may be debated).

**8.10.4:04: I follow Gong (1997), Li Fanwen (1986: 425), and Sofronov (1968 II: ) who reconstructed the initial of 3437 as ʔ-. Nishida (1986: 62) reconstructed it as x-; in 1964 he reconstructed it as ɣ- (p. 134). (Nishida 1964 does not explicitly mention 3437, but it does list the reconstruction ɣõ for a rising tone rhyme 47 syllable without any homophones in chapter VIII of Homophones. 3437 is the only syllable that matches that description.) I do not know how those scholars reconstructed the initial of 3437, as its character has no transcriptions or homophones. Given that there are three basic chapter VIII initials (x-, ɣ-, ʔ-) and that the following rising tone rhyme 47 syllables can be reconstructed in Homophones chapter VIII -

xõ, ɣõ

- 3437 is likely to be ʔõ by a process of elimination. Tangut did not seem to permit xw- and ɣw- before o in native words, so xwõ and ɣwõ are unlikely. However, ʔw- was possible before o: e.g., in

4053 1ʔwọ 'ice'

which led to my interest in

2040 2nɤa' 'ice' (with 'water' on the left instead of the mystery element ヒ on the right)

and its homophones like 5952 'ore'. So perhaps 3437 was ʔwõ. Maybe it would be safest to write its reconstruction as Xõ with X representing an unknown back consonant. AN OCTET OF ICY HOMOPHONES

The last of the Tangut words for 'ice' that I have been writing about belongs to a set of eight characters in Homophones:

Homophones location Li 2008 number Tangraph Reading Gloss
14B24 2189 2nɤa' < *nraXH second half of 1937 2189 2kha 2nɤa' 'to stutter; sad' (both only in dictionaries)
14B25 2249 wrist (only in Homophones; synonym of the more common word 0682 1khwiə?; not sure if it can stand alone)
14B26 3556 to apply, smear
14B27 2726 colored glaze
14B28 2040

ice (only in dictionaries; not sure if it can stand alone)
14B31 2582 mud (only in Homophones; not sure if it can stand alone)
14B32 4765 yarn (only in Homophones and Miscellaneous Characters; not sure if it can stand alone)
14B33 5952

ore; mine (only in dictionaries)

Out of these eight characters,

- six may only be attested in dictionaries, judging from the absence of nondictionary examples in Li (2008)

- at least three are freestanding words; the others are only found in dictionaries next to other characters

I am still surprised there can be so many 2nɤa'. 2726 'colored glaze' may be an extended usage of 2040 'ice', and 3556 'smear' and 2582 'mud' may be the same root, judging from the analysis of the former which implies 'muddy' semantics:


3556 2nɤa' 'to apply, smear' =

'water' and 'earth' = left and center of 2005 1tʂɤoʳ 'mud' +

bottom left of 4737 2ma 'apply'

But the others do not appear to be related to each other. I doubt pre-Tangut had six homophonous roots. Were those roots nonhomophonous in pre-Tangut: i.e., is 2nɤa' a merger of *nrakH, *nratH, *nrapH, etc. (if *X was a consonant)? DISTRIBUTIO-NA-L ODDITIES IN TANGUT

I was surprised to find that

2040 2nɤa' < *nraXH

had seven homophones in Homophones. If my pre-Tangut reconstruction is correct, I would expect simple syllables to be more common than complex syllables. *na was more common than other *na-type syllables, but it's surprising that there were no *naH, *nra, *nraH, *naX, or *nraX while there were multiple *naxH and *nraXH. Nonexistent syllables are in gray.

Rhyme Grade Tangut syllable Pre-Tangut Number of characters per syllable
17.1.17 I 1na *na 12
17.2.14 (*2na) (*naH) 0*
18.1.18 II (1nɤa) (*nra) 0
18.2.15 (2nɤa) (*nraH) 0
(Grade III rhymes like 19.1.9/2.16 do not normally occur after dentals like n-. But see rhyme 21 below.)
20.1.20 IV 1nia *Cɯ-na 1
20.2.17 2nia *Cɯ-naH 3
22.1.22 I (1na') (*naX)o 0**
22.2.19 2na' *naXH 5
(23/1.X) II (1nɤa') (*nraX) 0
23.2.20 2nɤa' *nraXH 8
21.1.21 III 1nɨa' *Cə-naX? 2
21.2.18 2nɨa' *Cə-naXH? 3***
24.1.23 IV (1nia') (*Cɯ-naX) 0
24.2.21 (2nia') (*Cɯ-naXH) 0

Although it's not impossible for a language to have a complex syllable while lacking simpler, similar syllables (e.g., English has strength but not streng, trength, treng, rength, etc.), I wouldn't have predicted multiple instances of a complex syllable. Having eight 2nɤa' < *nraxH is like having eight unrelated strengths in English. I'll look at the eight 2nɤa' next time.

For now, I'll close by noting two peculiarities involving Grade III rhyme 21. First, it was placed before the Grade I and II rhymes in the Precious Rhymes of the Tangraphic Sea, disrupting the usual I-II-III-IV pattern. Second, normally Grade III rhymes do not combine with dental initials, yet there are five Grade III nɨa' (but no nɨa!). I am not happy with my pre-Tangut reconstructions for their sources; they are placeholders. See how *Cə-naXH became 2nɨa' here.

*5 according to Arakawa (1997: 30). Gong reconstructed those five as 2da.

**1 according to Arakawa (1997: 30). Gong reconstructed that syllable as 1daa (= my 1da').

***0 according to Arakawa (1997: 30), who used Nishida's reconstruction in which these three syllables had t- instead of n-. PRE-TANGUT *NRAXH 'ICE'

The last of the three words for 'ice' in Li (2008) is

2040 2nɤa' < *nraXH

In the past I would have reconstructed it as 2nææ with a low front long vowel. It is not clear how its Grade II (i.e., -ɤ-medial < *-r- + low vowel) a-type rhyme differs from that of the more common Grade II a-type rhyme -ɤa. Since I no longer think Tangut had long vowels, I write the less common rhyme with a ' reminiscent of a prime symbol to represent its unknown distinctive feature(s). Arakawa also uses ' for this rhyme (-ya' in his system), but I do not know if he regards it as a phonetic symbol or as a notational device.

I used to think that '-vowels (my former long vowels) came from vowel-consonant sequences, as the Tangut autonym

3752 3296 2miə 2nɨa' < *mə-naXH

corresponds to Tibetan mi-nyag which must have been borrowed into Tibetan before the loss of a final obstruent *X (probably *k; more details on the development of this word here*).

If that was the case, then ideally all Tangut -V' should correspond to -VC in related languages that preserve obstruent codas. But that is not the case: e.g.,

5700 2ni' < *Ci-naXH 'nose' (not *2ni!)

corresponds to Japhug rGyalrong tɯ-ɕna and Tibetan sna 'id.' which lack final obstruent codas. Could *-X in such cases be a pre-Tangut suffix absent from other languages?

Conversely, there are cases in which non-Tangut obstruent codas correspond to zero instead of -': e.g.,

5700 1sia < *Cɯ-sa 'to kill' (not *1sia'!)

corresponds to Japhug rGyalrong kɤ-sat, Tibetan gsod-pa, and Old Chinese 殺 *ksat 'id.' Were some codas lost in pre-Tangut under certain conditions before they could condition -' in Tangut?

Both types of cases require explanation.

The problems with 2040 2nɤa' go beyond the mystery of its rhyme. I'm not even completely sure it means 'ice'. I'm surprised Kychanov and Arakawa (2006: 296) also glossed it as 'ice'; they often disagree with Li (2008). I have not seen any attestations of 2040 outside dictionaries. That is a lexicographical red flag; it means any glosses cannot be confirmed in context. Moreover, its Tangraphic Sea definition is presumably in the lost second volume. Here are the only two instances of 2040 known to me:

Homophones: 2040 4053 2nɤa' 1ʔwọ

Homophones text D note: 2040 - 3058 2765 2ɮiəʳ' 1nwie

Li (2008: 339) regarded 2040 4053 as a pair of synonymous nouns: 'ice ice', whereas Kychanov and Arakawa (2006: 296) translated it as a disyllabic verb 'turn to ice'. Given that 4053 can also mean 'frozen', another possibility is a noun-adjective phrase 'ice frozen'; each entry in Homophones has one or two clarifier characters, and the clarifier 'frozen' would distinguish 2nɤa' from its homophones (more on them later). Have Kychanov and/or Arakawa seen 2040 4053 as a verb in a text?

3058 is 'water' without a doubt, but 2765 only occurs in dictionaries. Its Tangraphic Sea entry says,

'[The character 2765 is from] the left of earth (2627) and all of blood (2734). 2765 is 0975. It is what blood uniting (2734 3591) is called.'

Unfortunately 0975 is only known from dictionaries, and the Tangraphic Sea defines it as ... 2765 and 'blood gathering (2734 0269)'.

Li (2008: 454) translated 2765 as a verb 'to swell, coagulate', but Kychanov and Arakawa (2006: 296) translated it as a noun 'coagulated blood'. I favor 'coagulate' as 3058 2765 would make no sense as a note for 'ice' if it meant 'water [and] coagulated blood'.

The verbs

3591 2ni' 'to unite' and 0269 1khiə' 'to gather'

may only have those meanings in dictionaries; their characters are attested with different meanings in nondictionary texts:

3591: 'to attack; a shield; to cover; to die'

0269: 'second half of 4059 0269 investigate; hide; rigid'

I presume these sets are unrelated homonyms apart from the noun 'shield' and the verb 'to cover'.

I could try to force three of the above words into a single word family:

2040 2nɤa' < *nra-X-H 'ice'

2765 1nwie < *Pe-nra 'to coagulate'

3591 2ni' < *Ci-nra-X-H 'to unite'

However, I would then need to account for the functions of the various affixes. Moreover, it is not possible to determine for sure whether grade IV words like 2765 and 3591 originally had *-r-. If I did not link them to 2040, I would have reconstructed them in pre-Tangut without *-r- or front-vowel prefixes to condition *a-raising: *Pɯ-ne and *niXH.

*3752 3296 2miə 2nɨa' has a strange second syllable. Normally n- does not combine with Grade III (-ɨ-medial) rhymes. Perhaps the word developed like this:

Pre-Tangut *mə-naXH

Breaking of *a before nonlow vowel *ə: *mə-nɨaXH

Breaking of *ə: *mɨə-nɨaXH

Tonogenesis: *mɨə-2nɨaX

Tone spread: *2mɨə-2nɨaX

The timing of tonogenesis relative to the vocalic changes is unknown.

The first syllable may be an unstressed form of pre-Tangut *mi 'Tangut' which became

2344 1mi (cf. Tibetan mi 'person')

The second syllable may be cognate to

0176 1nɨa' < *Cə-naX (cf. Tibetan nag-po 'black')

which also has an anomalous n-Grade III rhyme combination. Was *mə-naXH originally *mi-naX-H 'black people'? 'Black' brings to mind the term

2750 0176 1ɣɤu 1nɨa'


which is a term for one subgroup of the Tangut people. PRE-TANGUT *TɅ-KU-H 'ICE'

The second of the three words for 'ice' in Li (2008) is

3177 2kʊʳ (also 'frozen')

which may be cognate to

4034 1kiụ < *S-ku 'cold' (adj.?; see below)

if it is from *Tʌ-ku-H with a root *ku instead of *Cʌ-kur-H with a root *kur.

The prefix *Tʌ- conditioned the lowering and the retroflexion of the root vowel:

*Tʌ-ku > *Tʌ-kʊ > *T-kʊ > *r-kʊ > *r-kʊʳ > kʊʳ (ignoring *-H; see below)

The suffix *-H conditioned the second ('rising') tone.

The semantic difference, if any, between 3177 2kʊʳ 'ice' and

4053 1ʔwọ 'ice'
is unknown (apart from the fact that 3177 can also mean 'frozen').

Unfortunately I do not know of any other pairs of the type

*Tʌ-√-H (noun) : *S-√ (adjective)

*S- is normally a verbalizing prefix but not in the adjective 4034 *S-ku 'cold' or in the pre-Tangut source of the noun 4053 1ʔwọ 'ice' which could have been *S-ʔʌ-pam (as reconstructed last week) or *S-P-ʔo (as reconstructed last night).

Perhaps I should not call 4034 *S-ku an adjective, as Li (2008: 652) does not list any instances of it as an independent word (or in any text outside a dictionary). It may occur only in the disyllabic words

4034 4051 1kiụ 1kiʳw < *S-ku T-kuk 'cold' (?)

4034 4077 1kiụ 1miẹ < *S-ku Sɯ-me 'cold' (?)

which might have originated as synonym compounds.

I am not certain about these glosses. Li (2008) does not list any attestations of 4034 4051 and 4034 4077 outside dictionaries. Both 4051 and 4077 appear as independent words for 'cold', but the latter is only in the Tangraphic Sea. I have no doubt that the meanings of 4034 4051 and 4034 4077 have something to do with cold, but without textual examples, I cannot be sure of their parts of speech.

Unlike Li (2008: 652) who regarded 4034 as an adjective 'cold', Kychanov and Arakawa (2006: 489) defined 4034 as nouns 'frost' and 'cold' (i.e., 'coldness'?). Perhaps they have seen it in contexts where those glosses are appropriate.

The Tangraphic Sea equates 4034 with

1. 4034 4051 (see above)

2. 4089 0143, lit. 'cold (?) cold (adj.)' (compound attested only in dictionaries; only second half confirmed by nondictionary textual examples)

3. 2720 'cold' (adj.; confirmed in nondictionary textual example)

4. 4077 (see above)

5. 1918 0115 'not hot' (adj.; confirmed in nondictionary textual example)

On the other hand, the meanings of 3177 can be confirmed in the nondictionary textual examples in Li (2008: 518). AN ICY *P-REFIX?

While writing about a possible *p-prefix in the Chinese word for 'ice' in my last entry, I forgot to mention that *P- could also be a source of the -w- in Tangut 

4053 1ʔwọ < *S-P-ʔo? 'ice'

I reconstructed *P- to account for pairs of semantically and phoneticaly similar words such as

1829 1tsha < *Kɯ-tsa 'hot' : 1825 1tshwia < *P-Kɯ-tsa 'to heat'

in which one member has -w- and the other does not.

Ideally I would like to pair 1ʔwọ 'ice' with a word like ʔo 'ice, freeze, cold, etc.' But none of the ten words glossed as 'cold' in Li (2008) sound anything like 1ʔwọ 'ice' or ʔo. Nor do words with similar meanings like 'frigid' or 'snow'. I am hesitant to reconstruct *P- if I cannot find a -w-less potential relative, though there is no guarantee that a language must have a bare form alongside each affixed form.

Moreover, Gong (2002: 46) found that most zero ~ -w-pairs "clearly show a morphological process of forming verbs from adjectives or nouns". Hence *P-, the source of -w-, was often a verbalizing prefix. Obviously 1ʔwọ 'ice' was not a verb. Gong found only one zero ~ -w-pair whose -w-member was a noun:

3354 1ɣɤi < *Cʌ-Kri 'power' : 5307 1ɣwɤi < *Pʌ-Kri 'power'

I would add

3596 1ɣwɤi < *Pʌ-Kri 'power' (homophonous with 5307!)

to this set.

By analogy with this pair, a hypothetical ʔo that was the root of 1ʔwọ 'ice' would also mean 'ice'.

I wonder if the 'power' set actually consists of two reflexes of *Pʌ-Kri rather than *Kri with two different prefixes. I reconstruct prefixes with low vowels to account for ɣ- which is (sometimes? often? always?) from a lenited velar obstruent and -ɤ-, the reflex of *-r- after a low presyllabic vowel (see the second table in "G-*r-adation in Tangut (Part 2)"). Incidentally there is a tangraph

5309 1ʔo

that Li (2008) glossed as ... 'power'! Alas, not the 'ice' I was hoping for.

If the -w- in 1ʔwọ 'ice' is not a lenition of a prefix *P- or a medial (root-initial?) *-P-, then I wonder if Cw-clusters such as ʔw- come from original clusters or unit phonemes such as the Old Chinese *ʔʷ- reconstructed by Baxter and Sagart (2011). S-PRƏNG FROM SOME COMMON SOURCE? (PART 2)

In my last entry, I forgot to address the fact that Old Chinese 冰 *prəŋ 'ice' had an *-r- absent from pam-words for 'ice' or 'snow' in non-Chinese Sino-Tibetan languages. I could try to explain away the *-r- as an infix or as a prefix that metathesized: *T-p- > *pr-. (Medial *-r- is so common in Old Chinese that I suspect it came from a variety of sources - *t- and *l- as well as *r- - that I symbolize as *T-.) The dissimilation of *-m to *-ŋ after *p- could have occurred before *T- moved into medial position as an *-r- that would have blocked the shift. However, this scenario would still require a pre-Shijing dissimilation, long before the dissimilation is evident in poetry.

Here's a very different scenario. In Guangyun, 冰 'ice' has two Middle Chinese readings, *pɨŋ and *ŋɨŋ. I know of no other case in which a character has both *p- and *ŋ-readings. The *ŋ-reading is homophonous with 凝 *ŋɨŋ 'to freeze'.

Was 凝 used to write two unrelated words for 'ice' which happened to have identical rhymes?

Or were the two words related? Zhengzhang reconstructed them as *pŋrɯŋ and *ŋrɯŋ. (His *ɯ is equivalent to in other reconstructions.)

This internal etymology has no phonological problems if one accepts the simplification of *pŋ- to *p-, but it does raise the question of what *p- was. In this case it could be a nominalizer or even a participial prefix (a fossil of an earlier system of conjugation?): 'ice' < 'frozen'. Are there other pairs of the type

*X 'verb' : *p-X 'nominalized verb' / 'verb-ed'?

Next: More Tangut words for 'ice'.

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