07.12.22.13:36: BEYOND THE LCD APPROACH: CYCLES WITHIN CYCLES

In "Tangut Rhymes: An LCD Approach", I lumped rhymes apparently sharing the same vowel into large groups: e.g., R1-R7 = *-u, etc. The vowel *u is the lowest common denominator (LCD) of this group. Nobody thinks that these rhymes contained, say, *a or *i. However, there is no agreement on what distinguished the members of this group.

Gong's solution involves what I view as cycles within cycles. Almost any given rhyme belongs to

- one of the three cycles (lax, tense, retroflex - R104 and R105 are outside the system)

- one of the LCD groups

- is short or long

- is preceded by zero, -i-, or -j-

This template could be written in chart form:

I have been using Gong's reconstruction because this template is very elegant. But elegance is not necessarily correct.

First, I know of no spoken language that distinguishes CiV syllables from CjV syllables. (Ci-V disyllabic sequences don't count.)

Second, the evidence for a medial -i- and -j- is weak. These medials often correspond to zero in the Tibetan and Sanskrit transcriptive data, even though both those languages have a medial -y-. There are no Tibetan transcriptions like CiHV which might indicate iV diphthongs.

Third, the evidence for vowel length is weak. None of the Tibetan transcriptions use the Tibetan convention for indicating Sanskrit long vowels. That by itself would not be sufficient counterevidence, since we cannot assume that the Tibetan scribes would write Tangut as if it were Sanskrit. However, Gong's reconstructed vowel length for Tangut transcriptions of Sanskrit does not correspond to Sanskrit vowel length: e.g.,

R1 -u : Skt -uu (not -u)

R24 -jaa : Skt -a (not -yaa)

It is curious that R12 -ee and R54 -oo were not used at all in transcriptions of Sanskrit, even though they would be the obvious counterparts of Sanskrit e [ee] and o [oo], which are always long. In fact, Sanskrit e [ee] generally corresponds to Gong's short-vowelled R36 -jij, which I suspect was *[jej].

But it is not clear whether these counterexamples are found in 'pure' transcriptions of Sanskrit as opposed to Tangutizations of Chinese transcriptions of Sanskrit that disregarded vowel length.

Despite my reservations, I cannot help but wonder if medials and vowel length were (among) the distinguishing factors within the LCD groups. How else could the seven different types of *-u rhymes have been differentiated? Height and frontness?

Yet other possibilities exist: e.g., *-ʊw, *-yw, *-ɨw, *-əw, *-ow, a plain/velarized distinction (as in some rGyalrongic languages) or a plain/pharyngealized distinction (as in Hongyan Qiang and, I suspect, some stages of Old Chinese), etc. None of these possibilities can be easily verified using the extant evidence because they were probably absent from Tangut period NW Chinese, are definitely absent from Sanskrit, and cannot be written in the Tibetan script.

For now, I will continue to cite Gong's reconstructions, even though I do not take them literally.

07.12.21.16:48: NONEXISTENT TENSE TWINS (PART 3)

I just remembered that I have Gong's books in Hawaii. His Collected Papers on Tangut Philology contains "Phonological Alternations in Tangut", the article I needed to test my hypotheses.

It turns out that R62 (-jụ) corresponds not only to R2 (-ju) but also to R3 (-ju) and even R7 (-juu):

TT1354 dju R3 1.3 'complain; accuse'

TT4564 djụ R62 1.59 < *C-dju 'complain; grieve'

TT0473 nju R3 2.3 < *nju-H 'suck the breast'

TT2040 njụ R62 2.52 < *C-new-H 'breastfeed'

TT2338 gju R3 2.3 'suffering, hardship'

TT2337 gjụ R62 2.52 < *C-gjuH 'suffer; work hard'

TT1667 phjuu R7 1.7 'cover over'

TT4074 pjụ R62 2.52 < *C-pju(u)-H 'cap, crown, cover over'

Unfortunately, Gong did not find any tense-vowelled counterparts of R5 (-uu) and R6 (-juu) words. If such words existed, I predict that they would end in R61 (-ụ) and R62 (-jụ).

R4 (-u) corresponds to R61 (-ụ) like R1:

TT4179 lwu R1 1.1 'get mixed'

TT2036 lwụ R61 1.58 < *C-lwu 'to mix'

TT1719 ku R41.4 'loose; loosen (v.i.)'

TT3580 kụ R61 1.58 < *C-ku 'loosen (v.t.), release'

This chart sums up the correspondence patterns so far:

The differences between R1 and R4 (both -u in Gong's reconstruction) and R6 and R7 (both -juu in Gong's reconstruction) are as obscure as those between R2 and R3.

In Jacques (2003), R4 only corresponds to gDong-brgyad rGyalrong -u < Proto-rGyalrong *o, whereas R1 corresponds to other vowels in addition to -u:

This might imply that R4 was lower than R1.

Jacques (2003: 24) listed only one example each for R6 and R7. Both correspond to gDong-brgyad -u < PGR *-o.

R6 is extremely rare. It occurs in only five level tone syllables: khjuu (x 2) and ʑjuu (x 3). There is no rising tone version of R6 (which is why rhyme 2.6 is the rising tone counterpart of R7 1.7). If it were not for the fact that khjuu R7 1.7 contrasts with khjuu R6 1.6, I would conflate R6 and R7. I don't see any reason why kh- and ʑ- would be associated with a special rhyme. The two consonants have almost nothing in common.

07.12.20.18:05: NONEXISTENT TENSE TWINS (PART 2)

Writing this series here in Hawaii was probably a mistake because I didn't take Gong's articles on phonological alternations with me. I don't have his knowledge of the Tangut lexicon, so it would be very hard for me to find the alternations that I'm looking for: e.g.,

rhyme 2 (-ju) ~ rhyme 62 (-jụ)

rhyme 3 (-ju) 2165~ rhyme 62 (-jụ)

which would imply that R62 is the tense counterpart of both R2 and R3.

What is the difference between R2 (1.2/2.2) and R3 (1.3/2.3)? There are minimal pairs with the same initial indicating that R2 and R3 were not homophonous to the compiler(s) of Tangraphic Sea: e.g.,

TT0202 bju 1.2 'brothers' (Grinstead); also 'close relatives' (Shi et al. [2000: 24])

cf. TT5135 mju 1.3 'maternal uncle' (Grinstead), 'brothers' (Shi et al. [2000: 27])

m- < *N-b-?

TT0190 bju 1.2 'ox' (Grinstead); also 'elephant'? (Shi et al. [2000: 24])

cf. TT2159 phju 1.3 'ox' (Shi et al. [2000: 24])

does aspiration reflect *s-b-?

and TT2165 bju 2.3 'ox'

TT1160 bju 2.2 'border'

TT5225 bju 2.3 'elephant'

cf. TT0190 bju 1.2 'elephant'

TT0702 lju 1.2 'flute' < ?*liw < ?*lik

cf. Old Chinese 笛 *likw 'flute'

TT5231 lju 1.3 'saddle blanket' (Grinstead), 'carpet' (Shi et al. [2000: 29])

cf. TT4877 ljuu 2.6 'mattress' and TT4880 ljuu 2.6 'bedding'

TT0598 lju 2.2 'ounce'

borrowed from Tangut period NW Chinese 兩 ?*ljõ 'ounce'? - but the vowels don't match!

cf. TT0563 lju 2.2, the Tangutization of the TPNWC surname 呂 ?*ly - both have high vowels (u and y).

TT2665 lju 2.3 略 (many possible meanings including 'plan' 'invade'; Shi et al. [2000: 207]) or 'enter' (Li Fanwen [1986: 473])

if Shi et al.'s gloss is correct, could be borrowed from TPNWC 略 ?*ljo - but the vowels don't match!

cf. TT4613 lju 2.2 'plan' (Shi et al. [2000: 205])

cf. TT4665 lju 2.2 'attack' (Grinstead); 'enter; scatter' (Shi et al. [2000: 205])

TT2665 lju 2.3 'invade' could be thought of as 'attacking and entering'

Potential alternations between R2 and R3 suggest that one could be derived from the other via an affix (an infix or suffix?).

I'm surprised that I can't find any R62 words that could be plausibly connected to R2 words other than

TT1803 tɕju 1.2 'can only; able'

TT4912 tɕjụ 2.52 < *C-tɕju-H 'able; dare'

TT5164 ?lhju 1.2 'safe' (Gong and Li Fanwen [1986: 465] reconstructed initial ʑ-, but Gong reconstructed the fanqie initial speller with lh-, whereas Li reconstructed it with z- [1986: 478])

TT4576 ljwụ 1.59 'harmonize' (not the best semantic match)

cf. TT2036 lwụ 1.58 'mix'

I can't find any R3 ~ R62 alternations, so I'm not sure whether R62 really is the tense counterpart of R3 as well as R2. There are more R3 words than R2 words, so surely there must have been R3 words with tenseness-triggering preinitials in pre-Tangut:

I did find, of all things, a possible Chinese loanword with a tense vowel:

TT0731 ljụ 1.59 'flow' (Grinstead; Shi et al. [2000: 139]), 'float' (Li Fanwen [1986: 457])

cf. Late Old Chinese / Early Middle Chinese 流 *lu, Late Middle Chinese *liw 'flow' - but why wouldn't this be borrowed as Tangut ljiw?

also cf. TT4464 wụ 1.58 'Khitan', possibly from TPNWC 胡 *xu 'northern barbarian' - but in addition to the unexpected tense vowel, the initials don't match (due to prefixes? lenition?)

Does the tenseness of ụ reflect a native Tangut prefix attached to a borrowed root?

R2, R3, and R62 correspond to four gDong-brgyad rGyalrong rhymes (Jacques 2003):

The fact that R3 and R62 have similar correspondences implies that R62 might partly originate from R3 syllables plus preinitials.

Since the sole R2 word in Jacques (2003) has the same correspondence pattern as R3 words, gDong-brgyad evidence cannot determine how R2 differed from R3.

07.12.19.17:00: TANGUT RHYMES: AN LCD APPROACH

Ever since I first looked at Tangut reconstructions back in 1996, I have always been skeptical of the degree of phonetic detail in the various proposals. Such detail is difficult to justify on the basis of the extant evidence.

The Chinese transcriptive material is difficult to interpret due to the lack of non-Tangut evidence for the pronunciation of the 12th century northwestern dialect. And there is no way to write all but the most basic vowels in the Tibetan script. Sanskrit transcrption only requires a small number of vowels (a, aa, i, ii, u, uu, e, o, ṛ for most practical purposes*), so many of the 105 Tangut rhymes do not appear in Tangut transcriptions of Sanskrit (e.g., rhymes 2, 3, 5, 6, 7, 8, 9, 10, etc.). Presumably this is because those rhymes did not sound like Sanskrit vowels.

11 years ago, I even doubted the existence of tense and retroflex vowels. At the time, I thought that only a semi-algebraic reconstruction would be possible: e.g., instead of Nishida's -u, -ju, -juɦ, I would reconstruct *-u1, *-u2, *-u3. That would have resulted in rhymes like *-u11! To avoid that, yesterday I thought of indicating the rhyme cycle number (1, 2, or 3) in the number following the rhyme (modification 1 below). But now that I'm fairly sure that the second and third rhyme cycles had tense and retroflex vowels, why not just directly indicate those qualities (modification 2 below)?

If I apply the modification 2 approach to the entire Tangut rhyme system and remove the numbers, the result is what I call the 'LCD' (lowest common denominator) system:

This LCD system is heavily watered down from Gong's. Gong paired 65 with 76, whereas Arakawa grouped 76 with 77-79. The cyclical membership of 77-79 is uncertain.

There is nothing in the Tangut texts themselves that state where one cycle ends and another begins. The cycles were inferrred in modern times from the partly repeating sequences of vowels.

Assigning 77-79 to cycle II allows cycle III to begin with u like the other two cycles. However, cycle II would then end in -e, unlike the other cycles that end in -o.

I don't know why the cycles don't have the same vowel sequence.

Sofronov (1968 I: 138) placed 99-103 in their own cycle. Gong reconstructed 99-103 as -VV^r, but it's not clear why they wouldn't be adjacent to their -V^r counterparts unlike -(j)aa^r (rhymes 88-89) which is adjacent to -a^r, -ia^r, -ja^r (rhymes 85-87).

104-105 may have been left for last because all but one syllable (TT2621) in these rhymes is of foreign origin.

*Sanskrit ṛṛ and ḷ are very rare, and ḷḷ does not occur in any actual words.

07.12.19.14:21: NONEXISTENT TENSE TWINS (PART 1A)

The "A" stands for Arakawa.

So far, I've been looking at the problem of 'missing' tense rhymes using only Gong's reconstruction. Would Arakawa's reconstruction reveal a more systematic pattern of absence?

Gong's reconstruction has no -Ṿw, -Ṿ̃, or -ṾV rhymes. (I interpret rhyme 104 [-ũ] as lax-vowelled.)

Arakawa's 1999 reconstruction does not have any -Ṿu rhymes, but it does have -Ṿ̃ and ṾV rhymes (assuming his colon represents vowel length). Nasalization and length are in complementary distribution:

(I am ignoring medial glides and the -', -'', -2 symbols which seem to be arbitrary means of distinguishing between rhymes whose precise phonetic differentiation was unclear to Arakawa.)

It seems strange that the only long tense vowels are -ịi and -ạa (corresponding to short tense vowels in Gong's reconstruction!*) until one takes into consideration the fact that rhyme 70 (Arakawa's -ịi) corresponds to gDong-brgyad rGyalrong -a (Jacques 2003:7). This indicates that -ịi is from *-aa preceded by clusters. But it's still not clear why other long vowels do not have tense counterparts.

Moreover, why do only mid nasalized vowels have tense counterparts? What happened to the tense counterparts of high -ĩ and low -ã? (Arakawa's rhyme 104 [-ũ] only occurs in Chinese loanwords, so it has no tense counterpart -ụ̃**.)

Why don't any -I rhymes have tense counterparts?

If -ji, -je, -ju can have tense counterparts, why aren't there any such counterparts for -ja and -jo?

Why is -jẹ̃ unique? Didn't pre-Tangut have syllables like *CCjVN whose vowel was not *e?

Why are -ọ and -wọ (and similarly, -o^r and -wo^r) separate rhymes? Arakawa did not reconstruct lax -wo as a separate rhyme, and I don't know if Arakawa's rhymes 50 and 51 (both -o - a typo?) include -wo or not.

I also don't know if -ọ for both rhymes 73 and 75 is a typo or not. In Gong's reconstructions, 73 and 75 are distinguished by the presence or absence of -j-: -ọ and -jọ.

I am not confident about Gong's -j-, or about many other details in any reconstruction system I've seen.

Next: What I am sure about - more or less.

*Gong has a -j- absent from Arakawa's reconstruction, and he and Arakawa disagree about the vowel of rhyme 72. (One would expect Gong to have i or Arakawa to have I, but Gong's ɨ and Arakawa's I do not always correspond to each other.)

Arakawa's reconstruction also has a -jị (rhyme 69), but this corresponds to Gong's -iẹ!

**-ũ was a Tangut approximation of a Chinese rhyme. There is no evidence for tense vowels in the dialect of Chinese known to the Tangut, so I would not expect any Tangut loanwords of the shape Cụ̃. Moreover, Chinese had lost its obstruent-initial clusters long ago by the time the -ũ words were borrowed, so no words would have undergone the change

*CCṼ > *C̣Ṽ > *C̣Ṿ̣̃̃ > *CṾ̃

07.12.19.11:19: NONEXISTENT TENSE TWINS (PART 1)

In recent posts, I've been trying to explain why Tangut has no -Ṿw rhymes. If "A Louse-y Etymology" is correct, *-ịw merged with -jụ.

This triplet suggests that *-ẹw merged with -jọ and/or -jụ:

TT3649 new 1.43 'breast'

cf. Proto-Tibeto-Burman *nəw 'breast/milk' (Matisoff 2003: 604)

TT0219 njọ 2.64 < ?*C-new-H 'breastfeed'

TT2040 njụ 2.52 < ?*C-new-H 'breastfeed'

I still haven't found any instances of -ow alternating with a -Ṿ rhyme that could have come from an earlier *-ọw.

So far, I have only mentioned the absence of-Ṿw or -Ṿ̃ rhymes in Tangut. Six other types of lax vowel rhymes also lacked tense counterparts:

1. -jV(j) rhyme pairs corresponding to a single tense rhyme

1a. Rhymes 2 and 3 (-ju) have only one tense counterpart: 62 (-jụ)

1b. Rhymes 10 and 11 (-ji) have only one tense counterpart: 70 (-jị)

1c. Rhymes 19 and 20 (-ja) have only one tense counterpart: 67 (-jạ)

1d. Rhymes 30 and 31 (-jɨ) have only one tense counterpart: 72 (-jɨ̣)

1e. Rhymes 36 and 37 (-jij [jej]) have only one tense counterpart: 64 (-jịj [jej])

2. Rhyme 29 (-jə)

unlike rhymes 28 (-ə) and 30-31 (-jɨ)

3. Rhyme 50 (-jwo) and rhyme 53 (-jwo) with medial -w-

Rhyme 53 (-jo) without medial -w- corresponds to rhyme 75 (-jọ)

Medial -w- is an integral part of rhyme 53 (why?)

4. Rhyme 18 (-ia)

unlike rhymes 17 (-a) and 19-20 (-ja)

5. -Vj rhymes: rhyme 34 (-ej) and rhyme 41 (-əj)

unlike all other -Vj rhymes (-iej, -jij, -iəj, -jɨj) which have tense counterparts

6. Rhymes 1 and 4 (-u) have only one tense counterpart: 61 (-ụ)

When investigating the above gaps, I will assume the following:

1. Tense vowels originate from earlier clusters with obstruent preinitials:

*CCV > *C̣V > *C̣Ṿ > *CṾ

2. Any pre-Tangut rhyme could, in theory, be preceded by a cluster.

I can't think of any 'anticluster' vowels, codas, or rhymes in other languages.

3. If a pre-Tangut rhyme was not preceded by a cluster, this would be due to chance.

Next: Why do -jV(j) rhyme pairs only correspond to a single tense rhyme?

07.12.17.23:51: A LOUSE-Y ETYMOLOGY

Last night, I wrote

I will be back ... I can't flea my destiny (hint!).

I confused fleas with lice. That's a mistake no Tangut would make. Our ancestors were so much more perceptive about these things. On the other hand, they couldn't tell the difference between a PC and a Mac.

Why lice? Because TT2276 ɕjiw 2.40 'louse' might disprove my hypothesis that pre-Tangut *-q disappeared without a trace.

Its gDong-brgyad rGyalrong cognate is zrɯɣ (Jacques 2003: 19). zr- comes from an earlier *sr- (Jacques 2004: 319). -ɯɣ could come from three different Proto-rGyalrong rhymes (Jacques 2004: 261):

*-ɯk

*-ok

*-iq

The external evidence points toward a high front vowel rhyme:

Old Chinese 虱 *srit < *srik

Written Tibetan shig

Without other rGyalrongic evidence, I would guess that the PGR word for 'louse' was *sriq. I assume that a PGR *-q corresponded to a pre-Tangut *-q. If so, and if *-q was lost in Tangut, then the Tangut word for 'louse' should be ɕji. However, it is actually ɕjiw with a final *-w. Did bPT *-q

I suspect that this was indeed the case:

PT *-iq > *-ik (merging with original PT *-ik) > *-ix > *-iɣ > *-iɰ > *-iw

The change of *-iq to *-ik occurred independently in eastern rGyalrong (Jacques 2004: 261).

Note that even the gDong-brgyad rGyalrong reflex of PGR *-iq reflects an intermediate *-ik-stage:

PGR *-iq > *-ik > *-ɯk > *-ɯx > -ɯɣ

(the shift of *-q > *-k must have occurred after PGR *-ik was reduced to -i; otherwise PGR *-ik would have merged with *-iq)

gDong-brgyad -ɣ otherwise reflects PGR *-k. The normal gDong-brgyad reflex of PGR *-q is *-ʁ which corresponds to zero in Tangut.

Two days ago, I hypothesized that the missing -ịw might have become -jụ. It turns out that there is a word TT3588 ɕjụ 1.59, glossed by Jin et al. (2000: 138) as 蟣 'louse egg, nit' and by Grinstead (1972: 113) simply as 'louse'. I reconstruct their PT forms as

*sriq-H > ɕjiw 2.40

*C-sriq > *ɕjịw > ɕjụ 1.59

(The shift of *sr- to ɕ- is similar to the shift of *sr- to sh- in Written Tibetan. This shift must have occurred before autoretroflexion [the development of retroflex vowels in all syllables containing PT *r].)

Now I need to find examples of -ew ~ -jọ (< *-ẹw) and -ow ~ -ọ (< *-ọw) pairs as well as more examples of the alternation of -iw and -jụ.

07.12.16.23:46: RHINOPHOBIA

Old Chinese, Written Tibetan, and Written Burmese all had three final nasals in common:

-ŋ, -n, -m

Presumably Proto-Sino-Tibetan also had these three nasals. (Written Burmese palatal -ɲ is a secondary development from earlier *-ŋ and *-n, and I will not address the issue of whether Old Chinese or Proto-Sino-Tibetan also had a labiovelar *-ŋ^w here.)

rGyalrong is so conservative-looking that one might expect rGyalrongic languages to preserve those three final nasals. Yet Jacques' (2004: ) proto-rGyalrong reconstuction only had *-ŋ and *-m, and only the latter occurred before more than one vowel:

Why is Proto-rGyalrong so 'rhinophobic' toward nasal codas? What happened to all the PST *-n words and most of the PST *-ŋ words? Why would *-m outlast the other two nasals?

Gong's reconstruction of Tangut goes even further: it has no final nasals at all, and its nasalized vowels "occur mostly in Chinese loanwords" (Gong 2003: 602). (I should look into the presumably native exceptions later.) Presumably nearly all pre-Tangut *-VN rhymes have become *-V(G) rhymes in Tangut: e.g.,

TT4264 no 2.42 'old' < PT *naŋH

cf. OC 曩 *naŋʔ 'in the past'

TT2308 sji 2.10 'liver' < PT *sjinH

cf. gDong-brgyad rGyalrong tɯ mtshi, WT mchin < *m-syin 'id.'

TT3341 djɨj 1.42 'shallow' < PT ?*djɨm

cf. WB tim 'id.'

In "Ọw-mission", I looked at the derivation of CVw-CVj words from CVj roots. Although the verifiable roots were borrowings from Chinese *-j words, Tangut *-ow is partly derived from earlier *-VN, and I wonder if this pattern originated as *CVN-CVN. By the time the Chinese words ending in *-j were borrowed, there was no nasality in native CVw-CVj words or their CVj roots, so newer coinages could be based on CVj roots which never had any final nasals.

The two -j rhymes that appear in CVw-CVj words are 34 (-ej 1.33/2.30) and 42 (-iəj 1.41/2.36). These rhymes do not correspond to nasal-final rhymes in other languages:

Proto-rGyalrong *-ej corresponds to

Written Burmese -i, -e, -ai (ignoring tone/phonation)

Written Tibetan -i, -e

Old Chinese *-is, *-əj(ʔ)

(Jacques 2004: 213-215)

Gong (1995: 75) listed the only example that came even close:

TT1856 dze 1.8 'quarrel' : TT5509 dzeej 1.37 'to quarrel'

cf. Old Chinese 爭 *rtseŋ

Written Tibetan Hdzing 'to quarrel', zing-cha 'quarrel'

Written Burmese cac < *tsik 'war, battle'

Note that neither Tangut word ends in -ej, and there is no guarantee that the origin of -eej is similar to the origin of -ej. Moreover, one of the two Tangut 'quarrel' words might be derived from the other, so it's hard to tell whether it is -e or -eej which really derived directly from a Proto-Sino-Tibetan *-Iŋ (I = an unknown front vowel).