14.10.23.23:57: AURAL DOUBLES (PART 2)

A recap of part 1: Tangut had two syllables with similar fanqie

=+

3369 1mia 'transcription character for Skt ma, mā' and sixteen homophones = 5026 1mi 'to hear' + 3853 1tia 'topic marker'

=+

5025 2mia 'transcription character for Skt mya' = top and bottom left of 5026 1mi 'to hear' + left of 5314 2ʔia 'transcription character for Sanskrit ya'

If both syllables were mia (disregarding tones), why was 3369 1mia used to transcribe Sanskrit ma and mā without -y-? And why create 5025 2mia as an 'aural double' of 3369 1mia etc. if 1mia was already a good match for Sanskrit mya?

The answer to both questions is the same: 3369 etc. were actually 1ma, not 1mia, so a special character had to be created to transcribe Sanskrit mya.

Why did I reconstruct -i- in rhyme 20? Can this -i- be salvaged?

- I need to write about 'grades'. I've already covered the topic in "G-*r-adation in Chinese" (part 1 / part 2) and "G-*r-adation in Tangut" (part 1 / part 2), but I've changed my mind about a few things over the past day.

In the Yunjing rhyme tables for some unknown variety of Late Middle Chinese, a-type syllables were placed in four tables:

Vietnamese, Korean, and Japanese loans from Late Middle Chinese have /-(w)a/ for all of those rhymes, so their vowels must have been a-like. One could reconstruct a single Yunjing phoneme */a/ and compress the four tables into two:

But why didn't the author of the Yunjing do that? I think it's because */a/ had two allophones, back *[ɑ] and central and/or front *[a]. The *[ɑ] rhymes were placed in tables 27 and 28, while the *[a] rhymes were placed in tables 29 and 30. I reconstruct these allophones on the basis of correspondences with standard Mandarin and Cantonese. (The latter two languages are probably not descendants of the Yunjing language, but their ancestors were probably similar to it.)

Standard Mandarin

Standard Cantonese

The Cantonese pattern is quite clear:

- Grade I: back vowel

- Grade II: central vowel

- Grades III and IV: front vowels

The Mandarin pattern is complicated by these shifts:

*[ɔ] > [ɤ] after velars, [wɔ] elsewhere

*[wɨɑ] > *[wja] > *[ɥa] > *[ɥɛ]

*[ɛ] > [ɤ] after retroflexes

Sino-Vietnamese, Sino-Korean, and Sino-Japanese data for some non-a rhymes indicate that Grade IV was more palatal than Grade III (which may have been entirely nonpalatal in the source dialects of SK and the Go-on layer of SJ): e.g.,

Similarly, Mandarin Grade IV [jɛ] is more palatal than Grade III [ɤ].

- I was backer than the others

- IV was more palatal than III

We do not know for sure that Tangut also had grades. I do not know of any Tangut term for 'grade'. However, patterns of correlation between Tangut rhymes and Chinese grades in transcriptions have been known for over half a century. Moreover, those patterns also correlate with Tangut initials.

Here is a new Tangut-internal definition of 'grades'. One could identify the grade of a Tangut rhyme by looking at which initials may precede it:

The table above is only a first approximation.

I classify rhymes which can be preceded by any initial as Grade III/IV. One could also consider such rhymes Grade V, though such a term would have no parallel in the Chinese tradition.

Compare that distribution of initials with the distribution of Chinese initials in Yunjing:

The two patterns are not identical, but there are similarities:

- Labiodentals and r- never appeared in Grade IV.

- Dentals and sibilants were in near-complementary distribution with shibilants.

- l- was infrequent in Grade IV.

I think these similarities were due to Chinese influence on Tangut. Of course, Tangut had its own history, which is why the parallels are not absolute: e.g.,

- Tangut had Grade I and II v- unlike Chinese (in which *w- became *ɣw- in Grades I and II - a change absent from Tangut).

- Tangut had Grade I r- unlike Chinese (in which *r- became *l-; Yunjing *r- is from *n-)

The distribution of initials in each grade tells us whether certain grades were 'friendly' or 'hostile' toward certain initials. Such 'attitudes' give us clues about the phonetic characteristics of both grades and initials. For example, the fact that shibilants never occur in Grade IV, the most palatal of the grades, tells us that they were not palatal in either the Yunjing language or Tangut. That is why I reconstruct retroflex shibilants. One can also make historical inferences from (near-)complementary distribution: e.g., Chinese shibilants derived from dentals and sibilants, and Tangut shibilants may have partly derived from dentals and/or sibilants.

Having established strong parallels between grades in the two languages, I used to think that Grade IV a was the same in Yunjing Chinese and Tangut: i.e.., -ia. But I could not explain why Tangut rhyme 20 -ia

- transcribed Sanskrit -a and -ā (and nearly all rhyme 20 characters for Sanskrit -ya syllables were fanqie tangraphs combining part of an initial speller with the left side of the character transcribing Snaskrit ya: e.g., 5025)

- was transcribed as -a(H) in Tibetan

Now I think I have a solution:

If the Chinese dialect known to the Tangut was similar to the Yunjing language, it had four kinds of a-rhymes which were similar to Tangut rhymes 17-20.

The Grade IV a-type rhyme of late Tang Dynasty northwestern Chinese was transcribed in Tibetan as -ya, matching the *-ia I reconstructed for the Yunjing language. Maybe that rhyme was still *-ia in the eleventh century, and the Tangut thought its front (?) *a was like the front vowel of their rhyme 20.

The Tangut transcribed Sanskrit central a and ā - vowels absent from their language - with both back ɑ (rhyme 17) and front a (rhymes 18-20).

I suspect that rhyme 20 was once an *-ia that simplified to -a after all initials except glottal stop. Hence the rhyme 20 tangraphs

1767 1ʔia and 5314 2ʔia)

transcribed Sanskrit ya and yā. There was no rhyme 20 *ʔa. The *i of pre-Tangut *-ia may have been conditioned by a preceding presyllable with a high vowel, as Japhug cognates identified by Guillaume Jacques (2006) lack i:

0335 1pha < *Cɯ-pha : J ɯ-phaʁ 'side'

1530 1ma < *Cɯ-ma : J smar 'river'

2098 2ŋa < *Cɯ-ŋa-H : J aʑo < *ŋa-jaŋ 'I' (also cf. Old Chinese 吾 *ŋa 'I')

4225 1sa < *Cɯ-sa : J kɤ-sat 'to kill' (also cf. Old Chinese 殺 *ksat 'to kill')

4459 2ba < *Nɯ-ba-H 'to cut': J kɤ-mbaʁ 'to be cut'

3926 and 4601 2na < *Cɯ-naH 'thou' and second person singular verb suffix

correspond to Old Chinese 汝 *Cɯ-naʔ 'thou'.

If Grade IV rhyme 20 lacked -i-, and Tangut Grade IV was characterized by frontness contrasting with the backness of Grade I, I can revise my vowel reconstructions as follows:

That table is not as simple as its predecessor from four months ago, but it fits the Tibetan and Sanskrit transcription evidence better.

14.10.22.23:59: AURAL DOUBLES (PART 1)

I remain troubled by my reconstructionse of Tangut rhyme 20 (1.20/2.17) as -ia. Let's look at the transcription evidence for (or should I say against?) the syllable 1mia from my last post:

1. In Pearl in the Palm,

0092 1mia 'mother'

was transcribed in 12th century northwestern Chinese as 麻 *mbɤa. Granted, there was no Chinese *mia, so this does not necessarily mean 1mia is wrong.

2. On the other hand, it is possible to write mya in the Tibetan script, and yet 0092 was transcribed eight times as ma. Moreover, all rhyme 20 syllables were consistently transcribed without -y-. The Tibetan evidence favors reconstructions of rhyme 20 like Arakawa's -a: and Sofronov's -a (see this table).

3. Moreover, rhyme 20 was often used to transcribe Sanskrit -a and -ā. That is another point in favor of Sofronov's -a. Sofronov did not reconstruct a length distinction in Tangut, whereas Arakawa did. I would expect Arakawa's length distinction to correspond to the length distinction of Sanskrit, but it doesn't: e.g., Arakawa's long -a: may corresponds to Sanskrit short -a as well as long -ā, and vice versa. (10.23.1:09: Gong's length distinction that I used to carry over into my reconstruction also did not correspond to Sanskrit length:

Colors indicate length: pink = short, green = mixed, blue = long.

Rhyme 22 could not have been a simple -a or -aa, as it was never used to write Sanskrit. Rhymes 18 and 23 were also un-Sanskrit.)

If rhyme 20 were -ia, there would be no reason to create a special fanqie character

=+

5025 2mia = top and bottom left of 5026 1mi 'to hear' + left of 5314 2ʔia 'transcription character for Sanskrit ya'

to transcribe Sanskrit mya, since one of the seventeen 1mia characters with the fanqie

+

5026 1mi 'to hear' + 3853 1tia 'topic marker'

would have been sufficient. However,

3369 1mia

actually transcribed Sanskrit ma and mā without -y-!

(10.23.0:33: One might think that 5025 was created for Sanskrit mya because the second tone was favored for Sanskrit words. But tones in Sanskrit transcription seem to be random: e.g.,

- ma was transcribed with both 3369 1mia and 4737 2ma

- mi was transcribed with 5026 1mi, the initial fanqie speller for 5025 and 3369

- Cya syllables were transcribed with both first and second tone tangraphs

I doubt tones in Tangut transcriptions of Sanskrit had anything to do with Vedic pitch accent which was absent from Buddhist Sanskrit.)

In Arakawa's (1997) Nishida-style reconstruction, the reason for 'aural doubles' - tangraphs with slightly different fanqie containing 5026 1mi 'to hear' - is clear: 5025 was 2myaɦ, whereas 3369 and its sixteen homophones were 1maɦ without -y-.

In Arakawa's own reconstruction, 5025 might be 2mya: contrasting with 3369 and sixteen other 1ma:. (Yet there is no 2mya: or 2ma: on pp.128-129 of Arakawa's 1997 syllabary, though there are seventeen 1ma:.)

(10.23.1:46: I don't know how Sofronov would reconstruct 5025 and 3369 today. In 1968, he reconstructed them as 2ma and 1ma.)

At least everyone agrees that rhyme 20 was a-like, which is why I render it as -a in my lay transcription of Tangut.

Next: Why did I reconstruct -i- in rhyme 20? Can this -i- be salvaged?

14.10.21.23:59: WHY SO MIA-NY?

I have been writing about names of Kumārajīva lately (part 1 / part 2) such as Tangut

3948 3369 3284 2152 3284 (again!) 1kɨa' 1mia 2lɨa 1ʂɨi 2lɨa

The tangraph transcribing mā was one of the rhyme 1.20 syllables in the Tangraphic Sea that I listed last week.  Most were written with one or two tangraphs, but 1mia was written with seventeen! (For comparison I have also included the corresponding rising tone syllable 2mia with rhyme 2.17.)

Tangraph	Li Fanwen number	Reading	Li Fanwen gloss	Type (* = only in dictionaries)
	0092	1mia	mother (cf. 3334)	free morpheme 1
	0409		former times (only in dictionaries?; combines with regular word for 'day')	bound morpheme 1*
	1178		first half of 1mia 2nie 'end' (only in dictionaries; cf. 3369)	free morpheme 1 in a compound 'end-tail'*
	1215		first half of 1mia 2mɤe' 'to think of, to long for' (only in dictionaries)	morpheme half 1*
	1216		ten thousand (loan from Late Old Chinese 萬 *mɨanh 'id.'?)	free morpheme 2
	1458		second half of 2ni' 1mia 'salamander' (only in dictionaries)	bound morpheme 2* after a Chinese loanword 鯢 'salamander'
	1530		river	free morpheme 3
	1721		stirrup	free morpheme 4
	1803		first half of 1mia 1ɬiu' 'gray', name of an ancestor (only in dictionaries)	morpheme half 2*, free morpheme 2*
	2270		last syllable of (2mɪ) 2mɪ 1mia 'a kind of bird' (only in dictionaries)	morpheme part 3*
	2648		first half of 1mia 1khiu 'underground' (1khiu is 'under')	bound morpheme 1
	3334		female, woman (cf. 0092)	free morpheme 1
	3369		end, tail, east (only in dictionaries; cf. 1178); first syllable of 1mia 2ɬiụ 'plantain' and 1mia ?xa 'water buffalo'; transcription of Sanskrit ma, mā	free morpheme 1*, morpheme half 1, morpheme half 2, (not in Tangut words)
	3527		analogy; generally; doubt, fear (i.e., uncertain); and; few; should (i.e., to be time for), time; clothes	free morphemes 5-11
	3569		fishing hook	free morpheme 12
	3718		second half of 1ɣa 1mia 'doorframe' (1ɣa is 'door')	bound morpheme 2
	5118		second half of 1niu 1mia 'earring' (1niu is 'ear')	bound morpheme 3
	5025	2mia	transcription of Sanskrit mya	(not in Tangut words)

Why are there so many 1mia - and no native 2mia? The lower frequency of second tone syllables indicates that the source of the second tone must have been something extra which I reconstruct as a final glottal *-H by analogy with Chinese.

I reconstruct *Cɯ-ma(C) as the pre-Tangut source of 1mia. The high presyllabic vowel conditioned the breaking of the main vowel:

*C₁ɯ-ma(C₂) > *C₁ɯ-mɨa > *mɨa > 1mia

I don't know when the final consonant was lost relative to vowel breaking.

The various 1mia may have had different presyllabic and/or final consonants in pre-Tangut: e.g.,

*kɯ-map, *tɯ-mak, *pɯ-ma, etc.

I count 24 types of 1mia:

17 in texts (not just dictionaries; pink):

12 free morphemes (0092 = 3334, 1216, 1530, 1721, 3527 [seven homophones!?], 3569)

3 bound morphemes (2648, 3718, 5118)

2 parts of polysyllabic morphemes (3369 [two homophones])

7 only in dictionaries (blue; possible 'ritual language' words and/or words that didn't happen to appear in Buddhist, Confucian, military, etc. texts: e.g., 'salamander'):

2 free morphemes (1178 = 3369, 1803)

2 bound morphemes (0409, 1458)

3 parts of polysyllabic morphemes (1215, 1803, 2270)

Green indicates a tangraph (3369) that represents one morpheme only in dictionaries and parts of words in texts.

Further analysis may be able to reduce the number of types of 1mia: e.g., the 1mia in 1458 2ni' 1mia 'salamander' may be 'river' and the 1mia in 4681 5118 1niu 1mia 'earring' may be 'hook'.

Although one could describe tangraphy as 'logography' (i.e., as a word-per-character writing system), 3527 might have represented up to seven unrelated words! Conversely, the word 1mia 'female' was written with two tangraphs (0092 and 3334) depending on whether it referred to mothers or females in general. And 1mia 'end' was written differently depending on whether it was an independent word (3369) or in the compound 1178 5734 1mia 2nie 'end-tail'.

10.22.1:54: A high degree of homophony is tolerable: e.g., English can can mean

1. to be able

2. a container

3. to place in a container

4. prison (if preceded by the?)

5. toilet (if preceded by the?)

6. to be ready for release (in the can)

7. to be released from employment (mostly passive: was/got canned?)

8. Canada (e.g., in Canwest)

and various other meanings I have never encountered. Context is sufficient to disambiguate these many uses.

None of those meanings are opposites. One might look up

1530 1mia and 2648 1mia

in Li Fanwen (2008) and think they are near-opposites ('river' and 'land'), but in fact the latter apparently only occurs in the disyllabic expression

2648 5399 1mia 1khiu 'underground'

and I suppose that is much more common than

1530 5399 1mia 1khiu 'under a river'

so there is little risk of ambiguity. (In Google, under a river has 8.74 million hits, which sounds like a lot, but underground has 335 million hits! And many references to under a river involve underwater construction that would have been unimaginable to the Tangut nearly a thousand years ago.)

14.10.20.23:21: 'ZEN': A REMNANT OF TANGUT EMPIRE CHINESE?

KJ Solonin's article made me think about the Tangut name for Zen

=+

3504 1ʂɨã =

all of 2833 2diẽ 'calm, quiet' (probably 'not' + top and bottom right of 'to move')

left of 5593 1bɤo' 'to look, watch, observe'

as well as the Tangut names of Kumārajīva (part 1 / part 2). 1ʂɨã is a borrowing from Tangut period northwestern Chinese 禪 *ʂɨã which in turn is from Late Old Chinese (LOC) *dʑian, a Sinified form of Pali jhāna- (< Sanskrit dhyāna 'meditation'). (Japanese Zen is from Middle Chinese 禪 *dʑien.) Coblin (1994: 323) reconstructed 禪 as *śan ~ *źan in the 9th and 10th centuries AD on the basis of these Tibetan transcriptions:

大乘中宗見解: shan, zhan

南天竺國菩提達摩禪師觀門: zhan, Hzhan

LOC *dʑ developed differently in premodern northwestern Chinese and in Mandarin in 'level' tone syllables:

I don't understand the phonetic motivation for the split. Why were 'nonlevel' tones incompatible with a voiced affricate? (Voiceless affricates were possible before 'nonlevel' tones.)

Although modern northwestern Chinese generally has Mandarin-style reflexes of *dʑ, 禪 'Zen' still has a fricative initial in some varieties (Coblin 1994: 323):

Xining ʂã⁴⁴

Dunhuang ʂæ̃²⁴

Early 20th century Xi'an (as recorded by Karlgren): ʂæ̃ (tone unknown)

I thought these fricatives might be substratum retentions. I had either forgotten or overlooked this passage earlier in Coblin (1994: 101):

Occasional exceptions are found [to the Mandarin pattern of reflexes of *dʑ ...], e.g. 禪 [0678] (QYS źi̯än) "Zen Buddhism": [mid-Tang Chang'an] *dźan > *źan; CSZ [colloquial Suzhou] *śan (~ *źan?); XN [Xining]: ʂã⁴⁴; DH [Dunhuang]: ʂæ̃²⁴. These exceptional modern reflexes appear to derive directly from forms like those found in CSZ.

I looked for those "occasional exceptions" and found

蟬 LOC *dʑian 'cicada' is ʂæ̃²⁴as well as tʂʰæ̃²⁴(cf. standard Mandarin chan) in Xiaoxuetang's Xi'an data

辰 LOC *dʑin 'fifth Earthly Branch' is ʂɛ̃ (tone unknown) in Karlgren's Xi'an data (Coblin 1994: 361) and ʂẽ²⁴as well as tʂʰẽ²⁴ (cf. standard Mandarin chen) in Xiaoxuetang's Xi'an data

This last graph has two Sino-Korean readings, chin (without aspiration!) and shin. The first reading may be an old borrowing from Early Middle Chinese *dʑin; the second is from Late Middle Chinese *ɕin.

The multiple Sino-Korean readings of 什 in 鳩摩羅什 'Kumārajīva') may also be from different strata of borrowing: 집 chip from Early Middle Chinese *dʑip and 십 ship from Late Middle Chinese *ɕip. (집 chip becomes -jip with secondary voicing after a sonorant. That voicing is due to a Korean phonological rule and does not preserve the voicing of Early Middle Chinese *dʑip.)

A third Sino-Korean reading 습 sŭp is difficult to explain; it may be from a different Late Middle Chinese dialect in which *-ip became *-ɨp rather than vice versa.

The Xining reading of 禪 'Zen' also has an irregular 'yin level' tone (which would normally reflect an earlier *voiceless initial) instead of the expected 'yang level' tone (reflecting an earlier *voiced initial). I don't think the tone of 禪 'Zen' indicates that it had a voiceless initial in pre-Xining. I hypothesize that the original dialect of the region had a 'yang level' tone that sounded like the 'yin level' tone of the Mandarin dialect that displaced it.

If I am correct, then a study of irregular tones in Xining may reveal something about the substratal tone system. Unfortunately, it may not reveal the exact values of the tones at the time of borrowing because all tones - substratal and superstratal may have changed since then. So I don't know if 44 was the 'yang level' tone contour in the substratum dialect.

It would be interesting if other modern northwestern dialects also have a seemingly 'yin level' tone for 禪 'Zen'.

Dunhuang only has one 'level' tone which may be a merger of earlier 'yin level' and 'yang level' tones.

I don't know the modern Xi'an reading of 禪 'Zen', but I do know that both the substratal fricative-initial and superstratal affricate-initial readings of 蟬 'cicada' and 辰 'fifth Earthly Branch' have 'yang level' tones in modern Xi'an. Were the tones of the substratal readings shifted to match the superstratal tones?

One last question: Why would northwestern Chinese retain an old word for 'Zen'? The answer probably has something to do with the religious history of the region.

I am reminded of how Japanese Buddhist terminology consists of Early Middle Chinese-based borrowings (呉音 Go-on) that were not displaced by Late Middle Chinese borrowings (漢音 Kan-on) during the Tang Dynasty: e.g., 禪 Zen was not replaced by a newer borrowing *Sen. (One might think that Zen Buddhism was practiced in Japan before the Tang Dynasty, but in fact it took root in the 12th century when 1ʂɨã 'Zen' was practiced in the Tangut Empire. An old reading Zen was used for a new school because of the strong association between Go-on and Buddhism in Japan.)

On the other hand, Korean Buddhist terminology generally consists of Late Middle Chinese borrowings: e.g., 禪/선 Sŏn 'Zen' probably replaced an earlier borrowing that would have become modern 전 *Chŏn. A rare exception is the 什 -jip in 鳩摩羅什/구마라집 Kumarajip. But that is not the most common reading of 鳩摩羅什. Here are Google frequencies for the three readings of the name:

구마라십 Kumaraship: 215,000

구마라집 Kumarajip: 21,900

구마라습 Kumarasŭp: 19,300

The newer reading 십 ship outnumbers the older reading 집 jip by nearly ten to one.

The older voiced affricate reading of 禪 'Zen' has left no trace in Sino-Vietnamese. The only Sino-Vietnamese reading of 禪 is Thiền from southern Late Middle Chinese *ʑien; there is no *Chiền from southern Early Middle Chinese *dʑien.

14.10.19.23:54: THE TANGUT NAMES OF KUMĀRAJĪVA (PART 2)

The third Tangut name of Kumārajīva shares no characters with the other two:

1429 4575 4710 4867 1kiew 2mo 1lo 1ʂɨəʳ

It is obviously based on Tangut period northwestern Chinese 鳩摩羅什 *kɨwmbɔlɔʂɨi from a 4th century *kumaladʑip.

As I mentioned yesterday, 1429 is also the transcription character for 鳩 in the Tangut translation of the Forest of Categories (Gong 2002: 438).

4575 and 4710 are also transcription characters for Sanskrit mo and lo (Arakawa 1997: 111).

4867 was also used to transcribe other Chinese characters pronounced *ʂɨi (十實失室) and 涉 *ʂɨa (Li 2008: 770). The retroflexion in Tangut may have reflected subphonemic vowel retroflexion in Chinese after retroflex affricates: /ʂi/ = [ʂɨiʳ] and /ʂia/ = [ʂɨaʳ].

In theory the name could have been borrowed in a more Sanskrit-like form as *kʊ ma raʳ dzi va via Tibetan kumaradziba [kumaradziwa] or directly from the variety of Sanskrit known to the Tangut which had [dz] for j. (My Tangut reconstruction has no rhyme -u. Retroflexion was almost always obligatory after r- in Tangut.)

I was curious to see how Kumārajīva was rendered in other languages. Judging from Wikipedia entry titles:

Czech Kumáradžíva preserves the long vowels.

Polish Kumaradżiwa [kumaradʐiva] has retroflex dż for Sanskrit palatal j [dʑ]. I would have expected *Kumaradziwa [kumaradʑiva] with palatal dz (pronounced like dź [dʑ] before i). The combination of retroflex dż and palatal i is unusual in Polish. I wonder if that i is pronounced [ɨ] as in the normal Polish combination ży [ʐɨ].

Ukrainian Кумараджива [kumaradʐɪva] has [ɪ] instead of [i]. I presume the spelling was taken from Russian Кумараджива [kumaradʐɨva].

Korean 쿠마라지바 [kʰumaradʑiba] has an un-Sanskrit (and English-influenced?) initial aspirate. I presume it is a modern term. Older names are 鳩摩羅什 Kumarasŭp/Kumaraship/Kumarajip (the last character is read three different ways) and 羅什 Nasŭp (with initial r- becoming n- before a-).