I've been trying to come up with a cute name for David Boxenhorn's Tangut Search program. Last night, I realized that Tan sounds like Mandarin 探 tan 'search', so I was stuck at 探gut. (Gut searching?) The only gut thing I could think of was 글 gŭl 'writing' as in hangul. But the final consonant doesn't match, and Korean has nothing to do with Tangut. This afternoon, I came up with

探古特

Md Tangute

'search ancient special'

唐古特

Md Tanggute

'Tangut'

with 唐 Tang 'Tang Dynasty' instead of 探 tan 'search'.

David's program has three-letter alphacodes for tangraphic elements. One can try to form existing words out of these letter sequences, though I haven't succeded yet. (Element names that match existing words: e.g., fan, don't count.) I thought it'd be neat to give Tangut Search an alphacode-based name, but there is no tangraph with the alphacode tangut.

There are no alphacodes ending in -t. I wonder why, because -s and -u (but not -v, -w, -y, or -z) appear as third letters in alphacodes. (Finally looked at the source code. There are no alphacodes ending in stops other than -k.)

However, there are four tangraphs with an element tan:

Tangraph	Alphacode	Reading	Rhyme	Rhyme.tone	Gloss	Notes
	diotan	dʒiaa	24	2.21	(a surname)	dʒ- is normally not in a grade IV rhyme like R24*. Is this name of foreign origin? The Tangraphic Sea analysis derives tan from ソ丨 plus 丰 bor from cuobordex thiõʳ R98 2.83 'to wave; to shake' (describing the Jaa family?). ソ 丨 is from the 隹-like component of diotan's homophone diobishon 'male', based on Chn 雄 'male'. Hence dioソ丨 is phonetic in diotan.
	tangex	bi	11	1.11	grow; develop	tan is from tantua gex is from the right of foudex dʒu R2 2.2 'person' atop the bottom right of basciadim mieʳ R79 2.68 'person'
	tandim	ŋeʳw	93	1.87	to fear; to dread	tan is from tangex dim is from baepikdim ziẹ R64 1.61 'to hit; to touch; to fit exactly'
	tantua	bi	11	1.11	first half of bi-vəiʳ 'abundant'	tan is from tandim tua is from the left side of tuadex vəiʳ R82 1.77, the second half of bi-vəiʳ 'abundant'.

tan in diotan may have nothing to do with the tan in the last three tangraphs.

The sources of tan in the last three tangraphs form a closed circle:

tangex < tantua < tandim < tangex

The semantic functions of dim in tandim and the two abbreviations of 'person' in tandim are also unclear.

*5.2.16:00: Li Fanwen's (1986: 196) table of possible Tangut syllables excludes dʑǐâ (equivalent to my dʒiaa), even though he reconstructed dʑǐâ for diotan on p. 391. But he also reconstructed ȵǐâ (which is in his table) for diotan and

cirpancok 'glutinous rice'

but not for

cirgemcok 'marsh; bog'

Their fanqie is

hingom 'clear away' + halbaequn (transcription tangraph)

Sofronov 1968: ni̭e + ndɪ̭a

LFW 1986: ȵɩe + dǐa (sic; should be dǐâ)

Gong: nji + djaa

(the 1997 edition of Li Fanwen's dictionary has dʑji as well as nji; the 2008 edition only has nji)

This site: ni or dʒi + diaa

hingom is in the fourth chapter of Homophones for 'retroflex'-initial syllables, but the three tangraphs whose initial is represented by hingom are in the seventh chapter for palato-alveolar-initial syllables. Why wouldn't a fanqie initial speller and the tangraphs with that initial be in the same chapter of Homophones?

Li Fanwen (1986: 294-295) reconstructed palatal (not retroflex!) ȵ- for most syllables in chapter four and has no reconstructions for the remaining syllables.

Homophones contains no fanqie. The fanqie for diotan and cirgemcok are in the Mixed Categories volume of Tangraphic Sea (18B22-31). There is no fanqie for cirpancok, but its fanqie should be the same as diotan and cirgemcok since they are in the same homophone group in Homophones.

Perhaps Tangraphic Sea and Homophones represent slightly different dialects of standard Tangut:

It's hard to believe that n- would alternate with dʒ-. Li Fanwen's reconstruction of palatals for all four tangraphs seems more plausible, but how can it be reconciled with the fact that hingom's initial fanqie speller

tolgaidil: nɨə R30 1.29 'skirt'

has the initial fanqie speller

dexbiobaebescin: ni R11 2.10 'to listen; to hear'

(cognate to niu 'ear', Old Chinese 耳 *nəŋʔ 'ear')

which is in chapter three (dentals) of Homophones, not four ('retroflexes')?

Could Homophones preserve a three-way contrast lost in the dialect of Tangraphic Sea?

Tangraph
Alphacode	dexbiobaebescin	hingom	diotan, cirpancok, cirgemcok
Tangraphic Sea	n- (all members of the same fanqie chain)
Homophones	n- (chapter III)	ɲ- (chapter IV)	dʒ- (chapter VII)

I reconstruct a palatal ɲ- for hingom following Li Fanwen (1986) but using an IPA symbol instead of ȵ-. hingom belongs to R10 which is a grade III rhyme associated with palatals rather than dentals.

10.4.30.23:59: WHICH WAY WATER?

When I looked up

jux: reʳw R93 2.78 'ladder; stairs; steps'

in Li Fanwen's 2008 Tangut dictionary on Thursday night, I saw

biuduudexcok: nəu R1 1.1

across it in the adjacent column. By sheer coincidence, it containst the graph for 'black' that I've been blogging about lately. Its Tangraphic Sea analysis is

=+

biuduudexcok = cirzaa + duudexcok

nəu = top (sic!) of zɨəʳ 'water' + all of nɨaa 'black'

(in Tangut noun + adjective order!)

What does biuduudexcok mean? Select the blank space below for the answer.

'mud; mire; sludge'

Notice that cir is replaced by biu in biuduudexcok. I suspect "top" should be "left", since cir is equated with biu in

=+

biuhanfoi = cirgar + biohanfoi

ləu = left of niọ 'lubrication; oiliness' + bottom of ləu 'storehouse; warehouse'

What does biuhanfoi mean?

'moist; smooth; juice'

and the top of 'water' is fam (see below), not biu.

Similarly, cir is replaced by cen in wincen (and fam + dos = win):

=+

wincen = famdosdux + cirzaa

pa = diu 'have' + zɨəʳ 'water'

(in un-Tangut verb + object order instead of object + verb)

What does wincen mean? Select the blank space below for the answer.

'wave'; loan from Middle Chinese 波 *pa 'wave (since the Tangut were inland)

There are at least five different abbreviations of zɨəʳ 'water' in Tangraphic Sea:

	1. biu	2. fam
3. cir	0. cirzaa	4. cen
		5. bil

The only part of cirzaa which might not be used as an abbreviation is 干 bel, the bottom left of zaa. (Perhaps one or more of the lost analyses in the second volume of Tangraphic Sea contain bel < 'water'.)

Only one instance of fam < 'water' is known:

=+

famhordex = cirzaa + hordex

mia = top of zɨəʳ 'water' + all of vəi 'ripe; cooked'

What does famhordex mean? Select the blank space below for the answer.

'river' which has nothing to do with 'ripe'!

fam appears in 145 other tangraphs, not including components which incorporate fam like zaa on the right side of 'water'.

Three instances of bil < 'water' are known:

=++

famdiibil = famdaiher + foudex + cirzaa

lã = top of ləụ 'stone; rock' + right of dʒu 'person' + bottom right of zɨəʳ 'water'

What does famdiibil mean? Select the blank space below for the answer.

'aerolite' which has nothing to do with water!

=+++

balcuubaebilcin = cirvaigii + giifexbaebul + cirzaa + bumdexcin

xe = top left of niooʳ 'water, dew' + bottom center of miuu 'deep pool' + bottom right of zɨəʳ 'water' + right of ʃwa 'river'

What does balcuubaebilcin mean? Select the blank space below for the answer.

'sea'; borrowed from Tangut period northwestern Chinese 海 *xe (since the Tangut were inland; the Sea of Tangraphic Sea is a native word cuojoi = ŋiõ)

Note how cir (not from 'water'!) is abbreviated as

bal

and how giifexbaebul is reduced to a single vertical line!

bae

=+

biubilceubil = biubilceugar + cirzaa

riʳ = frame of mɛ 'cheese; fermented milk' + bottom right of zɨəʳ 'water' (合編甲 23.134)

What does biubilceubil mean? Select the blank space below for the answer.

'juice; soup'

bil appears in 261 other tangraphs, not including components which incorporate bil like zaa on the right side of 'water'.

In Tangraphic Sea equations

cir, biu, and cen

are equivalent. They may be considered to be allographs of a single element whose shape is determined by position:

cir: left or center (260 tangraphs; abbreviation of 'water' in at least 19)

biu: top or top right (9 tangraphs; abbreviation of 'water' in at least 1)

cen: right (47 tangraphs; abbreviation of 'water' in at least 3)

The 'water' figures are based on analyses in Tangraphic Sea. The actual figures may be doubled if the lost second volume of Tangraphic Sea is ever found.

cir is probably based on 氵, one of three allographs of Chinese 水 'water' in combinations:

氵: left (default)

氺: bottom

水: bottom and elsewhere

There are minimal pairs in sinography differing only in the placement of 'water':

江 Md jiang 'river' : 汞 Md gong 'mercury'

泊 Md bo 'anchor a boat', po 'lake' : 泉 Md quan 'spring; source'

One might think that X + 氺/水 graphs were created after the combination 氵 + X was already used, but that is not usually the case; e.g, 氵 + 將 does not exist, so I don't know why 水 is at the bottom of 漿 'thick liquid'.

Similarly, one might expect minimal pairs in tangraphy differing only in the placement of 'water' abbreviations. Was biuduudexcok created because

cirduudexcok

with cir in default left-hand position already existed? cirduudexcok even has the same two source tangraphs as biuduudexcok:

=+

cirduudexcok = cirzaa + duudexcok

nɨaa = left of zɨəʳ 'water' + all of nɨaa 'black'

What does cirduudexcok mean? Select the blank space below for the answer.

'deep, black'; cognate to its homophone nɨaa 'black'

On the other hand, an abbreviation of 'water' was placed on the right side of

wincen

even though there is no

cirwin

with an abbreviation of 'water' on the left.

'Water' itself contains cir on the left, but where does its cir - or any of its other parts - come from? Unfortunately, the analysis for it and

cirzuo: ʒɨu R2 2.2 'fish'

which contains the same components in a slightly different arrangment are lost.

What determines which part of a tangraph is abbreviated? Why wasn't 'water' just abbreviated in one or two ways to make it recognizable? What determines where that part goes in a new compound tangraph?

5.1.14:51: Maybe biu is for liquids which aren't water. Here are all nine tangraphs with biu:

Tangraph	Alphacode	Reading	Rhyme	Rhyme.tone	Gloss	Notes
	vaobiufil	jooʳ	103	1.94	to weave; to knit	Only tangraph with biufil Phonetic vaobiu from vaobiuceubil (see below)
	vaobiuceubil	jooʳ	103	1.94	cyst	biu from biubilceubil 'juice; soup'
	biuduudexcok	nəu	1	1	mud; mire; sludge	biu from cirzaa 'water'
	biuhangol	ləu	1	2.1	juice	No analysis known Cognate to biuhanfoi via *-H suffixation > rising tone?
	biuhanfoi	ləu	1	1.1	moist; smooth; juice	biu from cirgar 'lubrication; oiliness'
	biubilceubil	riʳ	84	2.72	juice; soup	biu from biubilceugar 'cheese; fermented milk'
	biubilceugar	mɛ	35	1.34	cheese; fermented milk	biu from biubilceubil 'juice; soup'
	biufal	tsõ	56	1.54	soup; broth	No analysis known
	biuguxdiadea	ʃɔ̃	57	2.48	otter	No analysis known guxdiadea without biu is a homophone ʃɔ̃ 'gray, ashy color' (phonetic?) Why 'liquid' instead of 'water'?

I got a new version of it from David Boxenhorn yesterday, but I fell asleep before I could try it out!

It enables me to set the size of tangraphs in the results for my searches. Larger is better. Pairs like

dexjux and gesjux

are difficult to distinguish at smaller sizes

though they're still easier to distinguish than this trio. (One is easy to tell apart, but I needed Photoshop to figure out the difference between the remaining two.)

The absence or presence of a dot is all that distinguishes dexjux and gesjux which have completely different readings:

dexjux: bi R11 2.10* 'below; down'

gesjux: lyu R3 1.3** 'low'

dexjux may have dex ('person' as an independent tangraph) on the left because its Chinese translation equivalent 低 has 亻 'person' on the left.

I thought ges in gesjux might be dex plus a diacritic dot rather than an element commonly identified as 'earth'. However, the Tangraphic Sea analysis of gesjux identifies gesgir 'earth' as the source of its left side.

=+

gesjux = gesgir + dexjux

lu 'low' = left side of lɨə̣ 'earth' + right side of bi 'below, down'

'Earth' is semantically appropriate for 'low'.

The Tangraphic Sea analysis of dexjux is unknown, but I would not be surprised if it included gesjux as the source of its right side:

=(...)(...)+?

dexjux = (...)dex(...) + gesjux?

One might think that the common right side of dexjux and gesjux

jux: reʳw R93 2.78

by itself would mean 'low' but it means 'ladder; stairs; steps' as an independent tangraph and may be based on 弟, the phonetic of Chn 梯 'ladder'. Why write 'down'-graphs with 'ladder'? Conversely, is 'ladder' an abbreviation of a 'low' graph, and if so, why? (The Tangraphic Sea analysis of jux is unknown.)

4.30.00:25: Although 'ladder' makes no semantic sense as an element in 'low' graphs, perhaps the logic for its choice went something like this:

1. The Chinese graph for 'ladder' is 梯 'ladder'.

2. Our graph jux for reʳw 'ladder' is based on its right side: 弟.

3. 弟 is nearly homophonous with 氐 'root, foundation, base'.

4. The Chinese graph for 'low' combines 氐 with 亻 'person': 低.

5. Therefore our graph dexjux for bi 'below; down' combines jux, based on a near-homophone of 氐, with dex, our equivalent/derivative of 亻.

Tangraphs are based on abbreviations of sinographs (弟 < 梯) as well as other tangraphs.

*4.30.0:50: Li Fanwen (2008: 612) lists the reading of dexjux as bji 2.10 in Gong Hwang-cherng's reconstruction. The underlining indicates a long vowel. However, 2.10 is R11 which has a short vowel, so I reconstruct dexjux as bi, equivalent to GHC's bji sans underlining.

**4.30.2:10: Li Fanwen (2008: 612) lists the reading of gesjux as ljwu 1.3 in Gong Hwang-cherng's reconstruction. GHC reconstructs a four-way contrast between

Although this looks neat, does any attested language have such a contrast? Also, it does not differentiate between R1 and R4 or R2 and R3.

I reconstruct a seven-way distinction:

I think all seven rhymes came from a pre-Tangut *-u and were conditioned by different preceding consonants and/or presyllables:

Note that R2 and R3 initials are not in completely complementary distribution. Exceptions to the basic pattern (e.g., bɨu R2) may be loanwords or conditioned by preinitials.

Tangut Grade II rhymes may have had pharyngealized vowels like Hongyan Qiang (whose /iˁ/ resembles [ɛ] but approaches the centrality of [ə]) or velarized vowels in Puxi rGyalrong. Just as earlier *-w- "may have given rise to a backing of the nucleus in Hongyan" (Evans 2006: 114), a pre-Tangut back consonant may have conditioned the lowering of *u.

David Boxenhorn's tangraphic element alphacodes are a dream come true. Four years ago, I experimented with naming tangraphic elements using letter sequences, but I never got around to creating a full system.

I can use such sequences as pronounceable substitutes for unknown Tangut B sound values. For example,

dexduudexcok and duudexcok

'night' and 'black'

would have Tangut B readings that were homophonous except for (a prefix?) 'dex-':

dex-duudexcok 'night' (black thing?) = dex- (nominalizer?) + duudexcok 'black'

Are there any other examples of dex- as a nominalizing prefix? Using David's search engine, I could look up "^dex" and see if there are any other examples of nouns written as 'dex' + adjective. Unfortunately, I don't have time to look at all 555 dex-graphs. I tried looking up a few adjectives to see if they had dex-prefixed versions, but couldn't find any noun-adjective pairs.

I did, however, find this unusual pair:

pə R28 1.27 'big; thick; head' (dexdiodal)

pə R28 1.27 'little livestock' (diodal)

They are homophonous yet have somewhat opposite meanings and cannot be cognate.

They remind me of the pair

bɨə R31 1.30 'tall, high' (bukgak; gak is also an independent tangraph for the cognate word bie 'high')

bɨə R31 1.30, first half of bɨə-bi 'below' (bukjux)

that Gong Hwang-cherng found.

4.28.1:14: The Tangraphic Sea analyses of the pə tangraphs above are circular:

=+

dexdiodal = dexwun + diodal

pə 'big; thick; head' = left of pe 'firstborn' + all of pə 'little livestock'

pə 'head' and pe 'firstborn' are probably cognate (cf. Chinese 頭生 'headborn' for 'firstborn').

pə 'litte livestock' is presumably phonetic.

=+

diodal = dexdiodal + dexgoldal

pə 'little livestock' = center of pə 'big; thick; head' + right of jiw 'achievement' (why?)

Which came first, dexdiodal or diodal? Similar questions could be asked about the A and B tangraphs whose derivations follow this pattern:

A = B + C

B = A + D

=+

bukgak = bukjux + feigak

bɨə 'high' = left of bɨə (first half of 'below') + right of so 'tall; high'

The analysis of feigak is unknown, but it wouldn't surprise me if it were

feigak = (...)fei(...) + bukgak

=+

bukjux = bukgak + dexjux

bɨə (first half of 'below') = left of bɨə 'high' + right of bi (second half of 'below')

The analysis of dexjux is unknown, but it wouldn't surprise me if it were

dexjux = bukgak + bukjux

10.4.26.23:51: THE BEST ARTICLE ON TANGUT WRITING EVER

... is Andrew West's, not mine, and certainly not this post.

I confess that I was one of the "others" Andrew is referring to below (emphasis mine):

[...] unlike most Chinese radicals, Tangut radicals do not have a single fixed meaning, and so giving names to them (as Nishida and others have done) is at best not very useful, and at worst misleading.

I use names because it is inconvenient to constantly insert graphics. At the moment I think using David Boxenhorn's alphacodes avoids the problem of erroneous semantic assocations. Thus I'd say that

na 'night' (dexbeldex)

at first glance consists of three elements arbitrarily named dex, bel, and dex instead of PERSON, SURROUND, and PERSON.

However, I think names like PERSON and SURROUND have some value as they represent hypotheses about their referents' origins (as opposed to their referents' functions within a given tangraph).

The element dex (PERSON) represents the Tangut word dzwio 'person' as an independent tangraph. It may be derived from a four-stroke variant of Chinese 人 'person' that I can't find in Unicode or the Taiwanese variant dictionary, though it is in Grinstead (1972: 56). I have seen that variant in modern times but don't know if it was in use in the northwest of China during the Tangut period.

The element bel (SURROUND) may be derived from the cursive form of the phonetic 韋 of Chinese 圍 'surround'. The modern simplification 韦 of 韋 is based on that cursive form.

=+

dexbeldex = dexbelpax + dexduudexcok

left of na (first half of na-raʳ 'tomorrow') + left of kõ 'night'

dividing it into two

instead of three parts:

kõ 'night' in turn was analyzed as

=+

dexduudexcok = dexbeldex + duudexcok

left of na 'night' + all of nɨaa 'black'

i.e., as

In short, the two tangraphs for 'night' are supposedly derived from each other.

And 'black' was derived from ... can you guess?

=(?)+

duudexcok = ? + dexduudexcok or dexduudexcok

Unfortunately, part of the equation is missing from Tangraphic Sea, so it's not clear whether 'black' should be split into

or into

(There is no independent tangraph duu.)

or even

My guess is the second or third analysis since liọ 'how' has no obvious semantic or phonetic relationship with nɨaa 'black'.

I assume that dex was created for 'person' before any compounds with dex were created. But what is dex doing in the tangraphs for 'night' and 'black' in the first place? In more abstract terms, how do we get from X for A to combinations of X for B which has no obvious semantic or phonetic similarity to A? What determines the method of abbreviation? Why aren't characters abbreviated in a consistent manner? dexduudexcok can be abbreviated either as dex or as (dex)cok or even as dexduu, the left and center of

swie 'clear, obvious' (dexduupak)

And what determines the placement of abbreviations? Why place the abbrevaition of dexduudexcok on the left of dexduupak but on the right of dexbeldex?

I think I already blogged about this Tangut word

na-raʳ 'tomorrow'

that came up in last night's post, but I don't mind covering the same topics again because I might come up with something new, especially now that I've got a second, even more powerful version of David Boxenhorn's tangraphic search engine. Thanks, David!

The meaning of 'tomorrow' can't be easily guessed from the dictionary analyses of its components:

=+

na- = left and center of na 'night' + left of dzị 'to cross' (海 22.231)

(Li Fanwen 2008 has a mistranslation of the analysis as 夜左過全 instead of 夜左過左.)

=+

-raʳ '= right of na(-raʳ) + center of raʳ 'to flow, leak, pass' (合編甲 23.021)

I thought 'tomorrow' might be from 'night (na) flow (raʳ)': i.e., what follows after the night flows away. But I feel that's contrived. Is there any language with a similar compound word for 'tomorrow'?

The element

'surpass' (alphacode pax; Nishida radical 241; < 戉, phonetic of Chn 越)

is shared by the tangraphs for both halves of na-raʳ 'tomorrow' and is presumably semantic if my 'night flow' etymology is correct.

The right half of the second half of na-raʳ 'tomorrow'

(no known meaning; alphacode dai; Nishida radical 104)

is phonetic and appears in other tangraphs pronounced raʳ. Could it be derived from the Khitan small script character ra? The strokes (丿一 丨) can be rearranged into a ㄇ shape and the horizontal line at the top could be doubled. Even if that hypothesis is wrong, the possibility of Khitan influence on tangraphy should be investigated.

Is there a more plausible Chinese source for it? Did 罗 (a simplification of 羅 Late Middle Chinese *la) already exist by the 11th century when tangraphy was created? The top element 罒 could have been reduced to 二. The bottom element 夕 could have been rotated 45 degrees counterclockwise minus one stroke to form ㄇ.

'majestic, glorious' (alphacode: dexbex)

is a shared element in the homophones

na(-raʳ) 'tomorrow' (alphacode: dexbelpax)

and

na 'night' (alphacode: dexbeldex)

which is read as pi by itself and means 'majestic, glorious'. I think dexbex is a graphic calque of

Chn 偉 'great' =

人 'person' (cf. PERSON dex) +

韋 'leather' which is also in 圍 'surround' (cf. 干 SURROUND bel, probably a simplification of 韋 - did the Chinese simplication 韦 already exist in the 11th century?)

'Majestic' has nothing to do with 'tomorrow', so I would expect dexbel to be phonetic in dexbelpax and dexbeldex. Yet those two tangraphs are pronounced na, not pi. Was dexbel chosen because pi or even something like *wi (cf. 偉 Late Middle Chinese *wi) was a Tangut B word for 'night' (scenario A)? Or was na a Tangut B word for 'majestic' - the translation equivalent of Tangut A pi (scenario B)?

(Could the Tangut B word for 'tomorrow' also be 'night-flow': *pi ... or *wi ... with a Tangut B word for 'flow'?)

Scenario B

In either scenario, dexbel is a double-function phonetic. Japanese writing conains a few double-function phonetics.Japanese 國字 kokuji (made-in-Japan characters) on occasion have phonetics based on native Japanese readings (訓 kun) as well as Sino-Japanese readings (音 on): e.g., 升 is a phonetic for both the on reading shou and the kun reading masu:

枡 'measuring box' has no on reading because it is a kokuji. Similarly, some tangraphs may not have had Tangut B readings because they were devised to write Tangut A words without Tangut B equivalents. I assume all tangraphs had Tangut A readings which are recorded in Tangut dictionaries.

Although there is no good evidence for Tangut B, is there any other way to explain why 'night' and the first half of 'tomorrow' incorporate 'majestic', a semantically and phonetically different tangraph?