The probable Tangut cognate of 殺 Old Chinese *r-sat 'kill' has a tangraph with a bizarre analysis:


TT0981 KILL sja 1.20 =

WOOD = (top of) TT4375 債 DEBT wee 1.12 +

? = (left of) TT5387 KILL ʔjirj 2.68 +

(tr. by Li Fanwen [1986: 279] as 伐 'cut down')

(tr. by Shi et al. [2000: 303] as 斬 'cut off', 去 ?'remove'.)

? =right of TT0381 CRUSH-LICE (i.e., killing lice?) dju 1.3 +

(tr. by Li Fanwen [1986: 279] and Shi et al. [2000: 29] as 擠 ?'squeeze')

I don't know what the second and third components mean, and I can't figure what WOOD is doing over them. DEBT has no semantic or phonetic function, not even if its Tangut period NW Chinese reading ?*tʃaj is taken into consideration.

There are no tangraphs pronounced s(j)a(a)(r) with WOOD on top which could have been potential sources of WOOD. (There are no tangraphs pronounced sia[a] at all.)

The problems with TT0981 KILL are not limited to its tangraphic structure. Its Chinese transcription in the Pearl was 薩 TPNWC ?*sa which had no medial *-j-. In the pre-Tangut period, 薩 transcribed

Skt saṃj-, sañ-

Pali sañ-, san-, sad-

Indic (unspecified) śaat-, sat-, sad-, (-)san-, sar-, -sal-, -sas-, -suṭ-, -st-

and was transcribed in Tibetan as sar. If TT0981 KILL were sja, it could have been transcribed with TPNWC ?*sja characters such as 些邪寫謝. (The last three were transcribed in Tibetan as sya in the pre-Tangut period.)

Although I cannot find any Tibetan transcription for TT0981 KILL, all Tibetan transcriptions for rhyme 1.20 and its rising tone counterpart 2.17 ended in -a, not -ya. Sofronov (1968 II: 17) listed the -y-less Tibetan transcriptions bsaɦ for sja 1.20 and saa 2.17 (tangraphs unspecified; the latter may be TT5705).

However, the Chinese and Sanskrit evidence sometimes points toward a -j- in a homophone of KILL and its rising tone counterpart:

TT4739 做 DO sja 1.20 (gloss from Li Fanwen 1986: 362)

transcribed (as?) TPNWC 散 *sã,*sa (in texts unspecified by Nevsky)

transcribed Skt saa, śaa (Nevsky 1960 II: 233)


(what is WAIST/BIRD doing on its left?)

transcribed as TPNWC 薛 *sja,*sjã

transcribed TPNWC 寫 *sja,*sja

transcribed Skt śa, ṣa, saa (Nevsky 1960 II: 312)

The use of sja for Skt śa, ṣa implies sj-.

So did sja really have a -j- or not? I think it did and it didn't. How do I resolve that contradiction?

There is no guarantee that Tangut phonology was stable across space and time. The transcription evidence may reflect different dialects in different periods.

Perhaps sja is the most conservative form, and other and/or later dialects

- lost the -j-: sja > sa

cf. Sino-Korean: earlier 샤 sja > modern 사 sa

- or fused sj- into a palatal fricative ɕ: sja > ɕa

If transcriptive practice within a text is consistent, but transcriptive practices among texts are not, then that may indicate different dialects. Of course, different dialects may also be reflected in a single text with multiple authors: e.g., TT5705 transcribed Skt śa, ṣa (reflecting a sja-dialect?) and Skt sasa-dialect?) in the Tangut Buddhist text 佛說守護大千國上經. Another possibility is that the author(s) of that text was/were sa-dialect speakers who had to use their single s-initial to represent Skt śa, ṣa as well as sa.

However, I can't help but wonder why Tangut ɕja-tangraphs were not used to represent Skt śa, ṣa. Had some dialect(s) merged s(j)- and ɕ- into a single initial?

APPENDIX: How did others reconstruct TT0981 KILL sja 1.20?

Nishida (1964: 193): sɑɦ

Sofronov (1968 II: 300) sa*saC

(11.4.00:31: The lost consonant was probably *-t, though it may have shifted to something else before disappearing.)

Shi et al. (my guess using their 1983 system): sæ̃

Li Fanwen (1986: 362): sa

Arakawa (my guess using his 1999 system): saa

Only Gong's reconstruction contains a medial -j-. Does any Tibeto-Burman language have a cognate with a medial -j-? None of the TB forms at Starostin's site have a medial -j-. A 'KILLER' PROBLEM (PART 3)

Laurent Sagart (1999: 69) reconstructed 319d 殺 Middle Chinese *ʂɛt 'kill' in Old Chinese as *s-rat, without specifying the syllable type (emphatic / nonemphatic in my reconstruction).

According to Sagart, 319d 殺 was the older spelling for the state whose name was later written with 337i 蔡 MC *tshayh. (Karlgren 1957, on the other hand, identified the earlier spelling as 319a-c 杀, the phonetic of 319d 殺.) Sagart reconstructed the earlier name of that state as OC *s-hrat-s and thought that the change of spelling 蔡 "presumably marks the acquisition of an affricate initial": e.g., it postdated his proposed shift of *s-hr- > *tsh-.

There are four problems with this proposal.

First, OC *(-)r- usually fused with an adjacent consonant to become an MC retroflex initial. OC *a after emphatic retroflex initials fronted to MC *æ. (The phonetic reason for this has long eluded me. Since retroflexion is antipalatal, was fronting dissimilation?) Yet Sagart's OC *s-hra- became MC *tsha-, not MC *ʂæ- or MC *tʂhæ-.

Was *hr phonetically *[χ]? (Cf. my proposal of *[ʁ] for *r.) That might explain the *r/r-phonetic 虍 in 虎 MC xoʔ 'tiger' (< OC *hraʔ; see Sagart 1999: 41).

Second, Sagart proposed on the previous page that OC *sr- became MC nonretroflex *dz-. I know of no evidence for reconstructing 319d 殺 'kill' with a nonemphatic initial, so I will rewrite his unspecified OC *s-rat as *s-rat. His rule on p. 68 would predict that OC *s-rat would become MC *dzat, not MC *ʂɛt. (It's not clear why MC [implying OC *r(C)e] sometimes appears instead of the expected MC *æ.)

Third, Sagart's OC root *rat for 'kill' doesn't match Proto-Tibeto-Burman *g-sat 'kill' (Matisoff 2003). There is no reason that OC and PTB have to have the same word for 'kill', so this is the weakest of the arguments against Sagart's proposal.

Fourth, I know of no derivatives of Sagart's OC root *rat that do not have sibilant initials. This implies that a sibilant was part of the root.

Here is how I would reconstruct phonetic series 319:

319a-c 杀 OC *C-sats > MC *tshayh 'name of a state' (*C = a consonant other than *k)

319d 殺

OC *r-sat ~ *r-set > MC *ʂɛt (reflecting the OC *e-variant) 'kill'

OC *r-sat-s ~ *r-set-s > MC *ʂɛyh (reflecting the OC *e-variant) 'diminish'

319e 閷 variant of 319d 'kill' with 閃 MC *ɕiem 'flash' (OC reading unknown; presumably semantic?) on the right

319f 樧

OC *r-sat ~ *r-set > MC *ʂɛt 'kind of fragrant tree'

OC *r-sat > MC *ʂɨet 'id.'

319g 摋 OC *sat 'slap from the side'

cognate to 319d 殺 OC *r-sat ~ *r-set 'kill'?

cf. German schlagen 'hit', Dutch slagen 'hit': English slay

(no Karlgren number) 剎 OC *ksat ~ *kset > MC *tʂhet 'pillar' (why is this written with 刂 'knife' on the right instead of, say, 木 'wood'?)

used to transcribe Skt kʂat, kʂet, kʂas(a) [see Coblin 1983: 242])

11.3.14:30: Although I wrote all of the above on the early morning of 11.2 (with a few later edits), I held off posting it until I looked at the Eastern Han sound gloss evidence in Coblin (1983) for series 319d. But I didn't find anything exciting, though this equation by Xu Shen in Shuowen caught my eye:

read 帴 like 殺 in 末殺 (i.e., read 帴 and 殺 with *-t)

155q 帴 belongs to an affricate-initial series: e.g.,

155a-b 戔 OC *dzan, *dzen

155d 棧 OC *rdzan(s), *rdzen'

155h 醆 OC *rtsen', *rtsVn'

155i 剗 OC *rtshen'

155j 錢 OC *tsVn', *dzVn

155k 淺 OC (*k-ts- >) *tshVn', *tsen

(*V may be either *a or *e. Other examples of the same initials are omitted.)

Graphs in such series usually have affricate initials in MC. Yet 155q 帴 had unexpected fricative as well as affricate initials in MC. I think they may have originated from *s-ts-type clusters (using Sagart's [1999: 65] proposal):

MC*dzan < OC *dzan

MC *dzienʔ (Karlgren 1957: 62) < OC *dzVnʔ

MC *tsien(ʔ) < OC *tsVn(ʔ)

MC *sanh < OC *s-tsans

MC *sɨe (Yupian; not reliable) < OC ?*s-tsal (anomalous coda)

MC *ʂɛt < OC ?*r-set < *r-s-tset

MC *ʂɨeyh (Jiyun) < OC ?*r-set-s < *r-s-tset-s

(All MC readings are based on Guangyun unless otherwise indicated.)

I have already proposed an *s-ts- cluster in

155r 綫 MC *sienh < OC ?*stsVns

The equation of 帴 with 殺 tells us that both were pronounced with the same initial (i.e.,*ʂ-) in Xu Shen's language. Their original initial clusters (*r-s-ts- and *r-s-) had simplified and merged by the first century AD.

11.3.14:57: Although r-s-ts- may look unreasonable, Written Tibetan allows clusters of four initial consonants:

bsky-, bskr-


bsgy-, bsgr-


sts- and rts- are attested in WT, as are bsts- and brts- (but not rsts-). (ts is a single consonant in WT as well as in Chinese.) A 'KILLER' PROBLEM (PART 2)

Edwin G. Pulleyblank came up with two different solutions to the 'killer' problem:

I. Sibilant-velar / glottal clusters (1962: 128-12)

OC *sk- > MC *k-

OC *skh- > MC *Chh- [tʂh]

OC *sg- > MC *G- (*g- in nonemphatic syllables?)

OC *sng- > *zng- > MC *ng-

OC *s-h-, *sngh- > MC *Sh- [ʂ]

II. Velar-sibilant clusters (1965: 206)

OC *ks- > MC *Sh- [ʂ]

OC *khs- > MC *Chh- [tʂh]

(I don't know what happens to *s-velar clusters in Pulleyblank's later OC reconstructions which have never been fully described.)

The first solution is incompatible with most of Sagart's (1999: 69) treatment of s-velar clusters (converted into my notation):

OC*sk- > MC *ts-

OC *sk- > MC *ts-, *sh- [ɕ]

OC *skh- > MC *tsh-

OC *skh- > MC *s-

OC *sg- > MC *dz-

OC *sg- > MC *z-

OC *sng-, *sng- > MC *s-

OC *s-hng-, *s-hng- > MC *?

maybe Pulleyblank was right, and these became MC *Chh- [tʂh]

OC *sx-, *sx- > (equiv. to Pulleyblank 1962's *s-h-; not in Sagart's table)

maybe these became MC *s-?

The second solution may be typologically unusual. I don't know of any language with phonemic aspiration which also have a phonemic distinction between /Cs/ and /Chs/ in initial position. Languages seem to have a default aspiration setting for initial consonants before a sibilant:

Greek: ξ-, ψ- but no χσ-, φσ-

Sanskrit: kʂ- but no ks-

Khmer: khs-, phs- but no ks-, ps-

(Khmer orthography distinguishes between ks- [in loanwords from Sanskrit] and khs- [in native words], but phonetically they are both [khs].)

Written Tibetan: gs-, gsh- but no khs-, khsh-

Mawo Qiang (Sun Hongkai 1981: 30):

khs-, khʂ-, khɕ- but no ks-, kʂ-, kɕ-

qhs-, qhʂ- but no qs-, qʂ-

Pulleyblank's solution implies other *Chs- clusters in OC, and possibly even *ChC- clusters with consonants other than *-s-, but I've never seen him propose OC *phs-, *phk-, etc.

If OC really did have a distinction between *ks- and *khs-, I would predict that Sanskrit kʂ- would have been approximated with OC *khs-, not OC *ks-.

Many years later, Pulleyblank (1991: 54-55) proposed that OC aspirates "were of secondary origin"and that they may have partly "reflect[ed] a kind of reduplication: *tt- > *th-". (He also proposed an *x-prefix as another source of aspiration. Cf. my solution 3 from last night.) Combining his 1965 and 1991 proposals, I could reconstruct OC *ks- and *kks- as sources of MC *Sh- and *Chh-. However, I have never seen an language with initial [kks], though such a cluster is not impossible to pronounce.

Next: Sagart's solution. A 'KILLER' PROBLEM (PART 1)

In recent posts, I've been deriving the Middle Chinese retroflex affricate Sh- [ʂ] from three different Old Chinese clusters: *sr-, *rs-, *ks-.

I've cited the use of 193a 山 MC *ShEn 'mountain' to transcribe the -xan- of Alexandria as evidence for*ks- in its OC pronunciation (following Pulleyblank 1991: 67; also see Pulleyblank 1962: 116):


OC *'a lək ksan (tə)ray

(11.1.1:48: I'll explain why I think 離 had a presyllable *tə- or at least a *t- in another post.)

The retroflexion of *s after *k would be like the retroflexion of s after k in Sanskrit (as noted by Pulleyblank 1962: 128, though there he was positing *sk- > Sh-).

Another potential piece of evidence is 319d 殺 MC *ShEt 'kill' < OC *r-set or *k-set, corresponding to Proto-Tibeto-Burman *g-sat (Matisoff 2003: 136, 330, 335, 442; cf. Kachin [= Jingpo] gəsat 'kill' [source]). The paradigm of Written Tibetan 'kill' -

g-sod-pa (present)

b-sad-pa (past)

g-sad-pa (future)

sod (imperative)

- indicates that the root was *s-t (hence Matisoff's index lists PTB *sat but not the prefixed form that appears in the body of his text). These words are probably cognate to

TT0981 KILL sya 1.20 (analysis coming soon)

If OC *ks- became MC *Sh-, one might expect Chinese transcriptions of Sanskrit kSh- [kʂ] to have been pronounced with MC *Sh-. However, those transcriptions (or their fanqie) were actually read with the MC retroflex aspirated affricate *Chh- [tʂh] (Pulleyblank 1965: 204-208):

Skt kSha: 叉 MC *Chhæ, *ChhE

Skt kShaan(ti): 羼 MC *ChhEnh, *tshænh

Skt kShet-, kShat-: 剎 MC *ChhEt

(same phonetic 杀 MC *tshat 'a place name' as 殺 MC *ShEt 'kill')

Skt -kShobh(ya): 閦 MC *Chhuk (< ?*-p)

Does this imply that MC *Chh- is the real descendant of OC *ks-? Not necessarily. Four possible solutions follow:

1. What if OC *ks- had two MC outcomes: *Sh- as well as *Chh-, depending on dialect? The problem is that I don't know of any MC *Sh- ~ *Chh- doublets. Nor do I know of any modern Chinese languages with affricates (< MC *Chh-) instead of fricatives for 山 MC *ShEn 'mountain' and 殺 MC *ShEt 'kill'. Hence I rule out this solution as unlikely.

In the next four solutions, MC *Sh- and*Chh- have different OC sources.

2. Presyllable vs. monoconsonantal prefix:

Variant 2a: Presyllable fell off.

*kə-s- > *kə-ʂ- > *ʂ- (why would *-s- assimilate to a non-adjacent *k-?)

*ks- ?[ksh] > *kʂ- ?[kʂh] > *tʂh-

Variant 2b: Monoconsonantal prefix fused with root initial. Presyllable lost vowel.

*kə-s- > *k-s- ?[ksh] > *kʂ- ?[kʂh] > *tʂh-

*ks- > *kʂ- > *xʂ- > *ʂ-

If I had to choose between the two, I would prefer 2b because it doesn't require *s to somehow assimilate to a nonadjacent *k.

3. Velar fricative vs. velar stop:

*xs- > *xʂ- > *ʂ-

*ks- ?[ksh] > *kʂ- ?[kʂh] > *tʂh-

Two problems:

First, it's not clear to me why Alexandria would be transcribed with an 山 OC *xsan (or *an?) instead of a *ksan (or *an?). But maybe at that point (the Western Han) 山 already had initial*ʂ-, as the coda of the preceding transcription graph 弋 *lək could represent the [k] half of -x- [ks].

Second, I see no need to reconstruct any *x-prefixes, whereas *k-prefixes are a must (see Sagart 1999). I know of no modern Chinese languages with x- or h-initial prefixes. k-prefixes, on the other hand, still exist.

(11.1.2:14: Perhaps *xs- is from an even earlier *sks-. Cf. Sagart's OC *s-ts- > MC *s-. But *s-ts- is a two-phoneme cluster, whereas *sks- is an even unlikelier three-phoneme cluster.)

4. Uvular stop vs. velar stop:

*qs- > *χs- > *xʂ- > *ʂ-

*ks- ?[ksh] > *kʂ- ?[kʂh] > *tʂh-

If this is correct, then I cannot reconstruct 'emphatic' *k as *[q], and I would have to reconstruct a dubious three-way opposition between *q, a backed *k, and a more frequent normal *k. Does any language have such an opposition?

5. Rhotic* vs. velar:

*rs- > *sr- > *ʂr- > *ʂ-

*ks- ?[ksh] > *kʂ- ?[kʂh] > *tʂh-

I think this is the most likely solution. It entails the following:

- The transcription of Alexandria must postdate the shift of *rs- > *ʂ-. There would be no reason to approximate -xan- as OC *-k-rs-.

- The OC reading of 殺 'kill' turns out to be *r-set, whose prefix doesn't match PTB *g-sat.

Next: Two more solutions from Pulleyblank.

*11.1.1:31: I use the vague term 'rhotic' because I'm not sure that both nonemphatic *r and emphatic *r were both liquids. Emphatic *r may have been a uvular fricative *[ʁ].

A fricative pronunciation of *r might account for the correspondences between (P)TB (*)r-C- and Late OC aspirates noted by Schuessler (2007: 58-59): e.g.,

PTB *r-ka(a)m 'cliff': 磡 OC ?*r-kəms [ʁqʌms] > LOC *khəmh 'id.'

PTB *r-kot 'dig': 窟 OC ?*r-kut [ʁqʊt] > LOC *khut 'cave'

PTB *r-kuw 'steal' : 寇 OC ?*r-kos [ʁqos] > LOC *khoh 'robber'

Written Tibetan rked-pa 'waist' : 广+絜 (not in my font) OC ?*(N)-r-ket [(ɴ)ʁqet] > LOC *get, *khet 'girdle'

(Schuessler has only one potential example of nonemphatic OC ?*r-k- becoming an LOC aspirate. I will deal with this in another post.)

However, I would expect OC *r-kə- and OC *r-ke- to merge into LOC *kE-. Perhaps TB had an r-prefix absent in the OC words for 'cliff' and 'dig', whose aspiration may have had a nonrhotic source (more on this in another post):

PTB *r-ka(a)m: 磡 OC ?*C-kəms

WT rked-pa: 广+絜 (not in my font) OC ?*(N)-C-ket

(OC *C*r)

Since the OC *r-prefix leaves no trace in LOC if followed by *-u (with or without a coda) or *-o (when not followed by a velar coda), it's hard to say whether the aspirated initials of 窟 LOC *khut 'cave' and 寇 LOC *khoh 'robber' came from OC *r-k- or OC *C-k- (*C*r). *PS-: DID CHINESE HAVE AN ASPIRATED SIBILANT?

In "Western Waters", I proposed that Middle Chinese *s ~ *tsh alternations within phonetic series could be explained by reconstructing Old Chinese *C-s- clusters as sources of *tsh: e.g., series 206:

僊 MC *sien < OC *san

遷 MC *tshien < OC *C-san

I proposed that the *C might have been a glottal stop. At that time, I didn't think it could have been another oral stop, since I assumed that

OC *p-s- > MC *ts-

(cf. OC *s-p- > MC *ts- [Sagart 1999: 65])

in both cases, *p assimilates to an adjacent *s

OC *t-s- > MC *ts-

OC *k-s- > MC *Sh- (IPA [ʂ])

(Like Sagart [1999: 110], I don't think OC had voiced stop prefixes.)

The first two developments are purely theoretical. Since OC had prefixes *p- and *t- (Sagart 1999: 85-97), they must have attached to some *s-initial roots. However, I don't know of any *s-roots with MC *ts- (< OC ?*p-s-, ?*t-s-) derivatives. The alternation of *s- and *tsh- seems to be more common in both word families and phonetic series: e.g.,

5k 傞 MC *sa ~ *tsha 'onomatope depicting a tipsy and continued dancing'

cf. 16e 娑 MC *sa 'dance'

355a 衰

MC*swa 'raincoat'

MC*Shwi (< OC *r-s-) 'decline, wane'

MC*tshwəy 'mourning dress' (related to 'decline'?)

MC *Chhwie 'reduce' (with an aspirated retroflex affricate [tʂh]; related to 'decline')

In most cases, MC *s- ~ *tsh- alternations go back to *s- plus OC nasal or liquid roots (Sagart 1999: 69):

MC *s- < OC *s-n-, *s-(h)l- (among other sources)

MC *tsh- < OC *s-hn-, *s-hl- (among other sources)

However, the roots in series 5 and 355 probably had initial *s-. Perhaps OC *s-s- became MC *tsh-. Or any non-velar stop and *s- plus *s- became MC *tsh-. So 5k and 355a could be reconstructed as

5k 傞 OC *say ~ *C-say

355a 衰

OC *soy

OC *r-suy

OC *C-suy


with *C- being *p-, *t-, *'-, or *s-.

None of the above addresses why OC *C-s- became an aspirated *tsh- rather than *ts-. Could OC *s- have been an aspirated *[sh] like Korean s? The aspiration would have been nonphonemic when OC *s- was a simple initial, but once a preceding consonant fused with *[sh], the aspiration became significant:

*/p-s/, */t-s/, */'-s/, */s-s/ > /tsh/ *[tsh], contrasting with unaspirated */ts/ [ts]

What about phonetic series in which MC *s- alternates with unaspirated MC *ts- as well as aspirated MC *tsh-? Sagart (1999: 65) proposed that OC *s-ts- became MC *s-, citing the word family

膝 MC *sit < OC *s-tsik 'knee'

節 MC *tset < OC *tsik 'joint', cognate to Tangut

TT1879 JOINT tserw 1.87 < pre-Tangut *r-tsek

Perhaps his derivation could account for MC *s-readings in affricate-initial phonetic series: e.g.,

131o 棷 MC *səw' < OC *s-tso' 'grassy marshland'

155r 綫 MC *sienh < OC *s-tsans 'thread'

592l 棲 MC *sey < OC*s-tsəy 'bird's nest; to roost'

but this is surely the same word as 594f 栖 'to roost' MC *sey, whose phonetic, 594a-e 西 'west' MC *sey, originated as a drawing of a bird's nest, and the evidence suggests a nasal, liquid, or fricative root initial, not an affricate!

611b 摲 MC *Sham < OC *r-s-tsam 'cut off'

cognate to its phonetic 611a 斬 OC *tsam' 'cut off'

1169d 繅 MC *saw < OC *s-tsaw 'reel off a thread'

(I assume that their roots had initial *ts-, as I think a root-initial cluster *sts- is improbable, unless their monosyllabic roots were compressed from earlier *sVtsV... disyllables.)

I wonder if my initial formulation (OC *p-s-, *t-s-, maybe even *s-s- > unaspirated MC *ts-) could account for instances where *s-initial phonetics are used to write *ts-initial words: e.g.,

174a 算 MC *swan'/h < OC *son'(s) 'count'

phonetic in 174d 纂 MC *tswanh < OC *C-sons 'plaited or woven silk band; continue'

433a 巽 MC *sonh < OC *suns 'humble'

phonetic in

433c-d 僎 MC *tswin < OC *C-sun 'ceremonial assistant'; loan for 433g (see below)

433f 選

MC *swien' < OC *son' 'select'

MC *swien'/h < OC *son's 'promote'

MC *swien' ~ *swan' < OC *son' ~ *son' 'count' (same word as 174a above)

433g 撰 MC *JwEn < OC *N-r-C-son or *N-k-son 'create'; loan for 433f above

812a-d 生 MC *Shïang < OC *r-seng or *k-seng 'live; fresh (as greens)'

Starostin's entry for 生 lists Min forms with affricates: Xiamen tshĩ, Chaozhou tshẽ, Fuzhou tshiang. (Strangely, his dialect database only lists Xiamen tsĩ without aspiration.) Do these reflect a prefix other than *r- or *k-?

phonetic in / cognate to 青

MC *tsheng < OC *C1-seng 'green'

(*C1 = *'-, *s-?)

MC *tseng < OC *C2-seng 'luxuriant (vegetation)'

(*C1 = *p-, *t-?)

did different prefixes lead to the difference in aspiration?

1134a-b 喿 MC *sawh < OC *saws 'crowd of birds chirping'

phonetic in 1134g 澡 MC *tsaw' < OC *C-saw' 'wash'


In "Western Waters", I struggled to reconcile the sibilant-initial 西 series with nasal-initial 迺.

Earlier forms of 西 'west' are 卤 and 卥. These look similar to 囟 'fontanelle', though the two words are too semantically different to possibly be related. Yet according to Shuowen, 囟 coincidentally has also derivatives with sibilant and nasal initials in later Chinese, though this is obscured by the replacement of 囟 with 田 / 曲 / 西 in the modern script (Pulleyblank 1962: 132):

囟 MC *sinh, *sin', *sih 'fontanelle' (latter two readings from Jiyun)

氵+ 囟 MC *sinh, *seyh

(Neither have Karlgren phonetic series numbers, implying that neither is attested in Old Chinese texts known to Karlgren.)

(Pulleyblank lists his equivalent of my MC *Næw as the reading of 氵+囟, but that is actually the reading of 石+囟 'sal ammoniac' which is not in Shuowen or, as far as I know, in OC texts.)

1241l 細 MC *seyh 'small, minute'

973a 思 MC *sï 'think'

1005a 農 MC *nowng 'agriculture; farmer'

206a MC *tshien 'rise high' (not in my font; see it here)

206b 僊 MC *sien 'caper about; dance'

206c 遷 MC *tshien 'remove; alter'

I would add

1244f 腦 MC *naw' 'brain'

It's possible to derive their initials from a common Old Chinese nasal prototype, though as Pulleyblank pointed out, their "finals are very divergent" - far more than expected within a phonetic series, as their Karlgren numbers above indicate (similar numbers are generally correlated with similar OC rhymes):

MC *sinh < OC *snir's

MC *sin' < OC *snir'

MC *sih < OC *snils

氵+ 囟

MC *sinh < OC *snirs

MC*seyh < OC *snils

石+ 囟 MC *Næw < OC *rnu or *rnaw

細 MC *seyh < OC *snes

思 MC *sï < OC *snə

農 MC *nowng < OC *nung

206a MC *tshien < OC *C-snan 'rise high'

僊 MC *sien < OC *snan

遷 MC *tshien < OC *C-snan

腦 MC *naw' < OC *nu(ng)' or *naw' (see Schuessler 2007: 397 on the ambiguity of the rhyme)

(MC *-owng', the expected reflex of OC *-ung', is strangely missing, implying that *-ung' lost its nasal and became something else:

OC *nung' > *nu'* > MC *naw'

A final -ng' in 'brain' may correspond to the *-k in Tibeto-Burman words for 'brain', though a final glottal stop : *-k correspondence is also possible.)

If one insists that all the graphs contained a sibilant in OC, then the *n- readings could have had a *sə- presyllable -

石+ 囟 MC *Næw < OC *sə-rnu or *sə-rnaw

農 MC *nowng < OC *sə-nung

腦 MC *naw' < OC *sə-nu(ng)' or *sə-naw'

- though this would conflict with Sagart's [1995: 215] proposal of Proto-Austronesian *punuk 'brain' as an external cognate of 腦 'brain', unless *sə- were added after *pu- was lost).

If 囟 is not phonetic in all of the above graphs, it is difficult to assign a semantic function to it: e.g., what does 'fontanelle' have to do with 細 'small' (yes, a fontanelle is part of a small, young skull, but still ...) or 農 'agriculture'? Have drawings of different, nonhomophonous objects been conflated as 囟? I suppose one could regard 囟 as a element representing a skull (a whole rather than a part) in 思 'think', 'rise high', and 腦 'brain'.

思 can also represent words for 'bearded' which were MC *sï and *səy. If these words are reconstructed with a nasal in OC as *snə and *snə, they may be cognate to 982a-b 而 OC *nə 'whiskers' and 133a-d 須 / 鬚 MC *suo 'beard' which may go back to OC *sno (although there is no evidence for a nasal within phonetic series 133).

(10.30.00:18: Schuessler reconstructed a Proto-Sino-Tibetan *sno 'mouth' on the basis of 須 / 鬚 OC *sno and Tibeto-Burman words meaning 'mouth').

(I assume that OC *sn- became MC *s- just as OC *sng- and *sm- became MC *s- [Sagart 1999: 69]. Schuessler [2007: 124] proposed this sound change but gave no example. Perhaps 思 OC *snə > MC *səy could fill the blank in his table.)

Yesterday, I mentioned that 僊 could also represent 仙 OC *san 'an immortal', which in turn has the phonetic 山 OC *ksan, representing the -xan- in a transcription of Alexandria. This implies that series 206 could not have had an initial nasal, unless *sn- had simplified to *s- by the time 僊 was used to write 仙 OC *san.

(10.30.00:12: According to Schuessler [2007: 527], 僊 is the original graph for 仙. If so, then did the graph 仙 postdate the shift of *sn- to *s-? Schuessler proposed Written Tibetan gshen 'shaman' [note the g- reminiscent of the *k- in 山 OC *ksan] as a possible cognate of [or loanword from] 僊. If 僊 were OC *snan, this would have been borrowed into Tibetan as snan, not gshen.)

When did the *sn- to *s- shift happen? Pulleyblank (1962: 132) cites the word 葫荽 OC *ga snoy 'coriander' as evidence for an OC cluster *sn- because it might be a loan from an Iranian form like Middle Persian goshniz. One problem is that this word may postdate the shift of *sn- to *s- in mainstream late OC, as Pulleyblank gives a citation from the third century AD. Could this have been a loan into a conservative OC dialect retaining clusters (and with *-[w]i < *-oy, since the Iranian forms had an i rather than an o in the second syllable?). What bothers me most about the finals of the 囟 series is the lack of a common coda. *-r ~ *-l (and maybe ~ *-n) alternation is fine, but final liquids shouldn't alternate with zero (ignoring final *-'/s), *-ng, and *-w.

The fact that four or five (*i *u *e *ə; possibly also *a) out of the six OC vowels appear in the 囟 series also doesn't help. This can't be explained away as an extreme case of root vowel ablaut, since derivatives of several unrelated roots were written with 囟.

Were OC scribes so desperate for an *s(ə)n- phonetic element that they used 囟 to write almost any syllable with *s(ə)n-? I doubt that, as nasal-initial phonetics were used to write other *sn-syllables: e.g.,

354a 妥 OC *hnoy' 'tranquil' in 荽 OC *snoy, the second half of 'coriander'

388a 人 OC *nin 'person' in 信 OC *snins 'believe' (00:49: if this was not *sngins and 言 *ngan 'words' is not phonetic) WESTERN WATERS

In part 7 of "Did Proto-Koreanic Have Two Liquids?", I mentioned Old Chinese *sri' as one source of Mandarin sa. This sa is 洒 'sprinkle', written as 氵 'water' (semantic) plus 西 'west' (phonetic). The 'western' phonetic series was pronounced with three types of initials in Middle Chinese: alveolar s-, retroflex Sh-, and palatal sh- (the codes are from Karlgren 1957):

594a-e 西 'west' MC *sey

594f 栖 'roost' MC *sey

594g-h 洒 'wash' MC *sey', *sen'; 'sprinkle' MC *Shæyh (from Guangyun; Karlgren [1957: 159] lists a reading that I would reconstruct as MC *Shæy')

594i 哂 'smile' MC *shin'

A possible fifth member of the series is

946a-d 迺 'then; thereupon' MC *nəy'

also written 乃

In my previous post, I reconstructed 洒 as OC *sri' following Schuessler (2007: 452) but I am not sure how to reconstruct this series. Here are two possible solutions and their problems:

1. The nasal solution

西栖 MC *sey < OC *snər


MC *sey' < OC *snəl' (variant of *sər')

MC*sen' < OC *snər'

'sprinkle': MC *Shæy'/h < OC *r-snər'(s) or *k-snər'(s)

哂 MC *shin' < OC*hnər'

迺 MC *nəy' < OC *nə(ng)'


洒 MC *Shæy'/h < OC *rsnər'(s) or *ksnər'(s) 'sprinkle' may be related to

878i 灑 'sprinkle'

MC*Shiə'/h < OC *sre'

MC *Shæy' < OC *sre'

MC *Shæ' < OC *sra(y)'

(is the root *sr-'?)

which cannot be reconstructed with a nasal. However, there is no evidence for a final *-r in 灑, and there is no known suffix *-r. There is a remote chance that all the readings of 灑 go back to OC *-y' < *-l' and that the root was *sr-l', but I would not count on this.

10.29.00:39: According to Pulleyblank (1962: 84), the MC rhyme *-æy "sporadically" lost its final glide: MC *Shæy' > MC *Shæ'. Hence there is no need to posit an OC source *sra(y)' for MC *Shæ'. If 灑 never had any OC reading ending in *-y' < *-l', then I cannot claim that 洒 and 灑 were *-r' ~ *-l' variants of the same root.

*-r phonetics are not used as phonetics for syllables like 迺 ending in ə(ng)'.

2. The liquid solution

西栖 MC *sey < OC *s-hlər


MC *sey' < OC *s-hləl' (variant of *sər')

MC*sen' < OC *s-hlər'

'sprinkle': MC *Shæy'/h < OC *r-s-hlər'(s) or *k-s-hlər'(s)

哂 MC *shin' < OC*hlər'


There are still two obstacles to linking the two words for 'sprinkle' (洒 and 灑):

The initials don't match: *(r/k-)s-hl- and *sr-/sr-. I know of no other case of *hl and *r in the same word family.

As already noted, 灑 cannot be reconstructed with a final *-r, and there is no known suffix *-r that could account for the final *-r of 洒.

*hl-phonetics are not used as phonetics for syllables like 迺 with initial *n-.

*-r phonetics are not used as phonetics for syllables like 迺 ending in ə(ng)'.

3. The sibilant solution:

西栖 MC *sey < OC *sər


MC *sey' < OC *səl' (variant of *sər')

MC*sen' < OC *sər'

'sprinkle': MC *Shæy'/h < OC *r-sər'(s) or *k-sər'(s)

哂 MC *shin' < OC*sCər' (C = voiceless stop)

迺 MC *nəy' < OC *sə-nə(ng)'


哂 and 迺 are the odd men out, because I don't know of any *sC- and *sə-syllables written with *s-phonetics: e.g., 午 *sə-nga' 'horse' (in the Chinese zodiac; cf. Tai forms with s-: Ahom shi-nga,sa-nga, Dioi sa [Li 1945 quoted in Pulleyblank 1991: 59]) is a velar phonetic in sinographs for velar-initial syllables like 許 *hnga' 'allow', not an *s-phonetic.

There's no way to relate 洒 OC *rsər'(s) or *ksər'(s) to 灑 unless the initial of 洒 was *sr-. Even if it were, 洒 still has a final *-r absent in 灑 that cannot be regarded as a suffix.

*-r phonetics are not used as phonetics for syllables like 迺 ending in ə(ng)'.

Conclusion: Without appealing to external evidence, I'm not certain which solution is correct. For now, I would try to reconstruct the 西 series without any reference to 迺 or 灑.

Although 洒 and 灑 could both be read as *Shæy' as MC, perhaps they originated as two non-homophonous OC words which later converged.

10.29.1:23: 哂 'smile' can also be written as 弞 (with 引 OC *lin' 'pull' as abbreviated phonetic according to Shuowen) and 吲 (again with 引 as phonetic). These alternate forms indicate that 西 was a lateral series. Perhaps 西 OC *s-hlər 'west' is cognate to Tangut

TT1523 WEST lyi (tone unknown)

No, I haven't forgotten about Tangut!

10.29.1:57: In Baihu tongyi (late 1st century AD), 西 has the paranomastic gloss 206c 遷 (Coblin 1983: 155). This implies the Late Old Chinese readings *sen and *tshien. The *tsh- of 遷 comes from some sort of *s-cluster. I would reconstruct phonetic series 206 with a common core *san:

206a (not in my font) OC *C-san 'rise high'

206b 僊 OC *san 'caper about, dance'; loan for 仙 OC *san 'an immortal'

206c 遷 OC *C-san 'remove; alter'

I think the initial *C- of 206a and 206c may have been a glottal stop: OC *'s- > MC tsh-. This is similar to Schuessler's (2007: 62) proposal of OC *s'- > MC tsh- (but in those cases, the OC root had an initial glottal stop, whereas I assume the OC roots of 206a and 206c had initial *s-.)

Other possiblities are *ks- (proposed by Schuessler [2007: 61-62]) and *sh- (not a palatal fricative, but an aspirated s, proposed by Schuessler 1987 and Starostin 1989). Like Pulleyblank, I prefer to view *ks- as a source of MC retroflex *Sh-, not MC alveolar *tsh-: e.g.,

山 OC *ksan > MC ShEn (irregular vowel) 'mountain', used to transcribe the -xan- of Alexandria

Tangut fonts by Mojikyo.org
All other content copyright © 2002-2007 Amritavision