To Chinese eyes, the Khitan large script at first appears to be a random mix of Chinese characters and alien shapes.

Given that the Khitan large script is said to have been 'invented' c. 920 using the Chinese script as a model, one might expect it to be something like the modern Japanese script in which Chinese loans are generally written with Chinese characters and kana almost always represent non-Chinese words*:

Khitan large script characters resembling Chinese characters : Chinese loanwords

Khitan large script characters not resembling Chinese characters : native Khitan words

However, the reality is more complex:

Khitan large script characters resembling Chinese characters :

Chinese loanwords

e.g., 皇帝 (looks like Liao Chinese *hongdi 'emperor') for Khitan hongdi 'id.'

and native Khitan words

e.g., 五 (looks like Liao Chinese *ngu 'five') for Khitan tau 'id.'

Khitan large script characters not resembling Chinese characters :

native Khitan (or at least non-Chinese**) words

e.g.,  doro (?) 'seal'

and Chinese loanwords

e.g., gün 'army' for Liao Chinese 軍 *gün 'id.'

One could also hypothesize that Chinese character lookalikes were used to write Khitan syllables that had (near-)homophones in Chinese, whereas nonlookalikes were used to write non-Chinese Khitan syllables and words with un-Chinese segments and phonotactics: e.g., Khitan iri 'name' with an un-Liao Chinese -r-.

But in fact, syllables shared by Khitan and Chinese were sometimes written with nonlookalikes:

e.g., for ai (why not write it with a lookalike of Liao Chinese *ai-graphs like 愛?)

And syllables and words with un-Chinese elements were sometimes written with lookalikes:

e.g., 午 (looks like Liao Chinese *ngu 'horse (calendrical)') for Khitan iri 'name'

Did the creator(s) of the Khitan large script take the Chinese script as used in the early 10th century, keep random characters, change the sound values of some of them, and then make up new characters?

One might come up with such an explanation for Cyrillic: its inventors took the Latin alphabet, kept some letters (e.g., А), changed the sound values of some of them (e.g., В for [v] instead of [b]), and then made up new characters (e.g., Б for [b] and Г for [g]). However, that is not what what happened. Both the Cyrillic and Latin alphabets are derived from the Greek alphabet. They are sisters, not daughter and mother.

If Janhunen (1994, 1996) is correct, the Khitan large script is to the Chinese script what Cyrillic is to Latin. Like Cyrillic, the Khitan large script was not invented on the spot; it was an adaptation of an existing script: the Parhae script, a Manchurian offshoot of the early Chinese script. The following seven Khitan large script characters might then be inherited from the Parhae script rather than taken from the 10th century Chinese script:

Sinograph Liao/Jin Chinese Khitan large script Khitan Jurchen large script Jurchen
*ho (< Middle Chinese *ɣɑ) ha ha
*she (< Middle Chinese *ɕjæˀ) ? sha
*sien (< Old Chinese *sˁir < *sˁər) ? shira or shïra
*gung ? (no similar Jurchen character) (*gung***)
gung gung
*ong (< Old Chinese *ɢʷaŋ) ong ong

Janhunen then proposed that the Jurchen large script was another derivative of the Parhae script rather than a direct successor of the Khitan large script.

Let's suppose the conventional wisdom is correct and that the Jurchen large script was invented c. 1120 with the then-current Chinese script as a model. Why was Jin Chinese 公 *gung 'duke' written with Jurchen 王, a lookalike of the characters for Jin Chinese *ong 'prince' and Khitan ong 'prince'?

Jin Guangping and Jin Qizong (1980: 56) proposed that Jurchen 王 gung was derived from Jin Chinese 工 *gung 'work' with an added stroke. Why not just copy 公 or 工?

Here is a wild speculation. In Old Chinese, 王 was pronounced *ɢʷaŋ. In mainstream Chinese *ɢʷ- weakened to *w-, and later, *waŋ became -ong in the northeast. What if a now long-extinct Manchurian Chinese dialect retained a stop initial for 王? Then perhaps 王 had two readings in Parhae, *gung based on the colloquial stratum of Manchurian Chinese, and *ong based on a literary stratum borrowed from mainstream Chinese. The first reading is the source of the Jurchen reading and the second is the source of the Khitan reading.

3.29.0:34: I am skeptical of the stop-retention scenario because there is no other evidence for *ɢʷ- surviving as a stop at such a late date in the northeast or anywhere else. Nor is there any evidence for *-ʷaŋ becoming *-ung in the northeast.

3.29.0:46: The Jurchen characters

for ong resemble those for ja (see my previous entry)

with two extra strokes on top.

However, Jin Qizong (1984: 236) regarded the ong-graphs as derivatives of the Khitan small script character

071 <ong>.

How would Janhunen explain that resemblance? Do the Jurchen large script and Khitan small script characters both go back to a Parhae prototype? Could the Jurchen character retain a 'roof' lost in the Khitan small script character?

*3.29.0:57: Although there is a strong tendency to write Chinese loans with Chinese characters in Japanese, some Chinese loans are in kana: e.g., サンゴ sango 'coral' (instead of 珊瑚).

Furthermore, Chinese characters do not always represent Chinese loans. In many cases they represent native Japanese words: e.g., 薔薇 for bara 'rose' as well as the much rarer borrowings shōbi and sōbi.

**3.29.1:01: Not all non-Chinese words in Khitan are native: e.g.,

053-051 <qa.gha> 'qaghan'.

may ultimately be of Xiongnu origin. (Has this word been identified in the large script?)

***3.29.1:14: Jin Qizong read two different Jurchen characters

as gung (in my notation), so in theory either could have transcribed Jin Chinese 工 *gung 'work'.

However, the second is only attested as a transcription of 宮 'palace' which was transcribed as

334-019-345 <g.iu.ung>

in Khitan.

So I suspect that the two Jurchen characters originally represented two different syllables, gung and giung, that merged into gung in the Yuan Dynasty Old Mandarin dialect of the Zhongyuan yinyun but not Phags-pa Chinese where they are still distinct as ꡂꡟꡃ <g.u.ng> and ꡂꡦꡟꡃ <g.ee.u.ng>.


When I first became interested in Jurchen, I assumed that its (large) script was "obviously derived from the Chinese script and the Khitan large script, with many innovations of its own" (Kane 1989: 21).

Then I discovered Janhunen's (1994: 114) hypothesis which I still regard as plausible after almost twenty years:

It was the other Sinitic script [of Parhae] that, due to its firm local [i.e., Manchurian] roots, was later transmitted first to the Khitan, and then to the Jurchen. All of this means that the conventional view, according to which the Jurchen script was successive to the Khitan «large» script, cannot be correct. As graphic systems, and heirs of the Bohai [= Parhae] script, the Khitan and Jurchen «large» scripts should be viewed as parallel, rather than successive developments.

There is much more to Janhunen's argument than that, but for now I want to focus on one of its implications. If the Khitan and Jurchen large scripts are offshoots of the Parhae script developed at some point prior to the end of the Parhae state in 926, then the readings of their Chinese-based elements are likely to reflect pre-10th century Chinese phonology to some extent. Such a scenario has a precedent in Old Japanese man'yōgana whose readings contain archaisms from the Chinese learned by the Paekche centuries earlier: e.g.

支 for Old Japanese ki < *ki and *ke is closer to Late Old Chinese *kie than Middle Chinese *tɕie

止 for Old Japanese is closer to Old Chinese *təʔ than Middle Chinese *tɕɨəˀ

(But Gerald Mathias views 止 as a kungana whose reading is based on Old Japanese töma- 'stop' [my təma-]; if so, then the resemblance to Old Chinese is coincidental.)

富 for Old Japanese is closer to Late Old Chinese *puəh than Middle Chinese *puʰ

Conversely, if the Khitan and Jurchen large scripts had no deeper roots, the readings of their Chinese-based elements should be derivable purely from Liao and Jin Chinese, as there would be no way for their creators to know about earlier readings.

Jin Guangping and Jin Qizong (1980: 56-57), Kane (1989: 23), and Kiyose (2004: 93) list Jurchen characters* with readings as well as shapes of Chinese origin**:

Jurchen Jurchen reading Sinograph Liao/Jin Chinese*** Middle Chinese Old Chinese
aci *ci *tɕʰiek *tɯ-qʰjak
ging *ging *kɨeŋ *Cɯ-qraŋ or *qɯ-raŋ
gung *gung *koŋ *koŋ
hi *si *sej *sʌ-ləj
i *u < *wuo *Cɯ-ɢʷa
i *u < *wuoˀ *Cɯ-waʔ
ja *jr *tɕi < *tɕɨʰ *təs
ki *ki *gɨ *gə
ngu *ngu *ŋo *ŋʷa
sa *cha *ɖæ *rla
u *ngu *ŋoˀ *ŋaʔ
dai *da(i) *dɑjʰ *lats
fu < pu *fu *fu < *puoˀ *poʔ
jul *ju *tɕu < *tɕuo *Cɯ-to
shang *shang *ɕɨaŋ < *dʑɨaŋˀ *Cɯ-daŋʔ or *Nɯ-taŋʔ
tai *tai *tʰɑjʰ *l̥ats
ha *ho *ɣɑ *ɢaj
sha *she *ɕjæˀ *l̥jaʔ
shira (Kiyose) or shïra (Jin and Jin) *sien *sen *sˁir < *sˁər < *Cʌ-sər

Out of that incomplete sample of nineteen characters,

- eleven have readings based on Liao/Jin Chinese (green)

- five have readings that could be based on either Liao/Jin Chinese or Middle Chinese (bluish green)

- two have readings that resemble Middle Chinese (blue)

- at least one has a reading that resembles Old Chinese (yellow)

I'll discuss a less likely instance in my next entry.

The last three characters (which all have have Khitan large script predecessors that look exactly like Chinese 何舍先) are hardly solid proof for Janhunen's hypothesis.

The Khitan and Jurchen may have used Liao/Jin Chinese 何 *ho for ha in their languages because there may not have been a character for *ha in Liao/Jin Chinese. (The only character read ha in the Phags-pa Chinese of the Yuan Dynasty is rare: 閜.)

Nonetheless the other two are difficult to explain if they were devised c. 1120 or perhaps even c. 920. Why write Jurchen sha with a derivative of Jin Chinese 舍 *she when Jin Chinese 沙 *sha was a closer phonetic match? And is the close match of Jurchen shira ~ shïra and Old Chinese *sˁir < *sˁər just a coincidence?

*3.28.2:50: Since this post does not deal with the Jurchen small script, I will refer to Jurchen large script characters simply as Jurchen characters.

**3.28.2:58: There are Jurchen characters with shapes of Chinese origin and native readings that are translations of Chinese: e.g.,


looks like Jin Chinese 一 *i 'one' but represented the native Jurchen word emu 'one'.

***3.28.3:15: I wrote Liao/Jin Chinese forms in an orthography resembling my transcriptions of Khitan and Jurchen to facilitate comparison. Khitan and Jurchen voiced obstruents may have been unaspirated and voiceless: e.g., Jurchen jul may have been [tɕul], a close match for Middle Chinese 朱 *tɕu(o).

15.3.26:23:49: QUINTUP-<UL> TROUBLE (PART 3)

In part 1, I proposed that Khitan small script character


might have represented <ül> because

131-366 <u.?> 'winter'
corresponds to Written Mongolian ebül 'id.'

In a generic 'Altaic' language, harmonic rules prevent the mixture of segments from two classes which I will call A and B*: e.g.,

a, u, ł, ɣ ... e, ü, l, g ...

'Neutral' segments can occur with segments of either class A or B.

Hence <ül> should be a class B character that should only co-occur with class B and/or neutral characters within a Khitan small script word block.

I used to think that

098 and 261

represented class A <ał> and class B <(e)l>, but in fact they not only coexist with each other but even with 366 in

340-098-366-261-349-021 <x.ał.ül.el.ge.mó>** (興宗 26.6)

(021 <mó> looks like an error for the dotless verb ending 020 <ei>)

which is unexpected from an 'Altaic' perspective. I would have expected


class A *130-098-206-098-051-122 <x.ał.uł.ał.ɣa.ai>*** or class B *340-261-366-261-349-020 <x.el.ül.el.ge.ei>.

366 can also coexist with both class A 051 <ɣa> and class B 349 <ge> in the same text (道宗):


161-366-261-051-189-123 <aú.ül.el.ɣa.a.ar> (道宗 12.30)

(instead of 161-206-261-051-189-123 *<aú.uł.ał.ɣa.a.ar>)

and 131-097-372-366-334-140 <u.úr.û.ül.g.en> (道宗 18.6)

That would also be unusual for an 'Altaic' language.

I am conflicted.

On the one hand, Khitan has sets of suffixes implying the presence of an 'Altaic'-style harmonic system: e.g., the causative-passive suffixes (class A?) and (class B?) in the above pair of words.

On the other hand, there seem to be harmonic violations. Are those violations artifacts of incorrect class assignments (e.g., is 366 a neutral character?), or are they real and perhaps even predictable?

The earliest known small script text is dated 1053, over a century after the invention of the small script c. 925. Do all small scripts discovered so far reflect Khitan after its harmonic system began to break down? Would the very first texts in the small script have more harmonic spellings?

*3:27.2:13: I got the A/B terminology from EG Pulleyblank who used it to describe Old Chinese syllable types. Norman (1994) was the first to draw parallels between Old Chinese and Altaic syllable types. I have gone even further and proposed harmony rules for Old Chinese.

I use the terms A and B to avoid specifying the nature of the classes: e.g., front vs. back, ±RTR, etc. As Khitan is in the Manchurian linguistic area, I suspect it had RTR harmony like its neighbor Jurchen.

**3.27.2:17: This is Andrew West's reading. Qidan xiaozi yanjiu has

340-067-366-261-349-020 <x.eü.ül.el.ge.ei>

which is not only harmonic but also has the dotless verb ending 020 <ei> instead of dotted 021 <mó> which is not a verb ending. I have not seen the handwritten copy of 興宗, and the original stele is inaccessible, so I do not know who is correct.

***3.27.2:24: I assume 206 is a type A character since it is flanked by a-characters in

029-206-189 <tau.uł.a> 'hare'.

15.3.25:23:59: QUINTUP-<UL> TROUBLE (PART 2)

In part 1, I built upon Aisin Gioro's work by equating the following five Khitan small script characters and regarding the first three as variants of each other:

  013 <ul> = 050 <ul> = 206 <ul> = 228 <ul> = 366 <ul>

The second and third appear in the same word:

050-131-206 <ul.u.ul> (道宗 16.21, 20.13 [1101 AD], 蕭仲恭 33.33 [1150 AD])

Did scribes of two different inscriptions nearly fifty years ago apart really use two variants so close together in three instances, or did  050 and 206 have two different readings?

3.26.1:10: Was 050-131-206 for ulul (?) above related to (or at least partly homophonous with)

050-131-366-311-162 <ul.u.ul.b.c> (宣懿 18.2 [also 1101 AD])

050-131-366-311-222 <ul.u.ul.b.ń> (道宗10.25, 15.19, 28.24, 宣懿 17.11)

which have 366 instead of 206 for their second <ul>? Or did 206 and 366 have different readings?

15.3.24:23:24: QUINTUP-<UL> TROUBLE (PART 1)

In my last entry, I proposed that the rare Khitan small script character 013 might be a variant of 050 which Aisin Gioro (2008) read as <ul>. Both in turn resemble 206 which Aisin Gioro (2003) also read as <ul>. Could 206 be yet another variant of 050?


050 <ul> = 013 <ul> = 206 <ul>

If Aisin Gioro is correct, then

029-206-189 'hare'

was <tau-ul-a> = taula, and Khitan may have lost a final -i retained in Written Mongolian taulai.

How can taula be reconciled with the History of the Liao Dynasty transcription 陶里 *tauli for 'hare'? There is no guarantee that Chinese transcriptions and the Khitan small script represent the same variety of Khitan. Perhaps *ai simplified differently in different dialects of Khitan:

Proto-Khitan-Mongolic *taulai
Proto-Khitan *taulai Proto-Mongolic *taulai
Standard taula Nonstandard tauli Written Mongolian taulai

Aisin Gioro (1999, 2004) identified two more small script characters for <ul>. Why did the Khitan have five characters for the same VC sequence?

050 <ul> = 013 <ul> = 206 <ul> = 228 <ul> = 366 <ul>?

The first three may be allographs, but the last two do not resemble them. Did 050/013/206, 228, and 366 originally represent three different sequences? If Khitan were like Mongolic, an obvious two-way distinction would be between <ul> and <ül>. 366 might have been <ül> since

131-366 <u.ul> 'winter'

corresponds to Written Mongolian ebül 'id.' But what would have been a third value contrasting with <ul> and <ül>? Kane (2009: 29) wrote that Khitan "was exceptionally rich in rounded vowels." Was there a three-way contrast between front [y], back [u], and near-high [ʊ] (like Manchu ū)? Did these three characters

131 <u>, 245 <ú>, 372 <û>

represent those vowels without a following lateral? (I almost wrote [l], but /l/ may have had different allophones depending on the adjacent vowel.)

At first, one might identify 131 as ü since it preceded 366 which might have been ül. However,

226 <ü>

transcribed Liao Chinese ü, whereas the three other <u>-type characters were used to transcribe Liao Chinese *u. Were they always interchangeable, or was that interchangeability due to later mergers?

Has anyone looked at Khitan spelling over time? Spelling variation may give us clues to changes in Khitan over a two or even a three-century period. If Nova N 176 is from, say, 1200 - the eve of the fall of the Qara Khitan - its large script spelling could differ from the norm established c. 920. Moreover, some variation may be due to Jurchen speakers' perceptions of Khitan phonology: e.g., Jurchen speakers may have heard only one or two kinds of /u/ in Khitan which might have had three. (3.25.0:16: First-language influence in Khitan texts written by Jurchen speakers has yet to be explored.)


Last night, I accidentally miswrote

070-050 <w.?>

in the Qidan xiaozi yanjiu transcription of 興宗 15.19 as

070-013 <w.?>

with a slightly different and much rarer character that only appears twice in the texts in Qidan xiaozi yanjiu:

028-067-013 <sh.eu.?> (道宗 27.9) and 013-224-327 <?.mu.ie> (耶律撻不也 12.1)

Is 013 in any of the texts that have been found in the three decades since the publication of Qidan xiaozi yanjiu? Could 013 be a variant of 050? Is that why Aisin Gioro did not include 013 in 契丹小字の音価推定および相関問題?

028-067 <sh.eu> (a transcription of Liao Chinese 守 *sheu; could it also be a native word?)

occurs by itself. Does that imply 028-067-013 <ś.eu.?> is a suffixed form, or are they unrelated partial homophones? There are eight forms beginning with 028-067; some have known suffixes (e.g.,

028-067-273 <sh.eu.un> ending in what may be genitive <-un> in 蕭令公 25.16 and 許王 50.5)

and others do not (e.g.,

028-067-041 <sh.eu.?> 'dew' in 宣懿 25.14 and 許王 cover 1.5).

Aisin Gioro read 041 as <us>* and 050 as <ul> ~ <l-> for reasons unknown to me. The Khitan small script has many sequences of the same vowel in two adjacent characters, so sequences such as

028-067-041 <sh.eu.us> and 028-067-013 <sh.eu.ul> (if 013 = 050)

look plausible. Moreover, 028-067-041 <sh.eu.us> may be a variant spelling of

028-067-244 <sh.eu.s> (巴拉哈達洞壁墨書 I.2.4; <s> may be a plural ending)

Unfortunately, I know of no

*028-067-261 <sh.eu.l>

corresponding to 028-067-013 <sh.eu.ul>.

*3.24.1:56: If Aisin Gioro's reading of 041 is correct, then

028-067-041 <sh.eu.?> 'dew'

woulld be less of a match for Written Mongolian sigüder(i) 'dew'. If Mongolian -der(i) is not a suffix, then perhaps the Khitan form is a reduction of an earlier sigüder(i)-like form to sheu with a plural suffix -s: i.e., drop of dew. The two words may also be unrelated.


Two nights ago, I wrote,

It seems that Khitan VC characters can also double as CV characters. I've guessed that they are CV before consonants and VC before vowels, but that does not always seem to be the case.

Offhand certain types of VC characters are less likely to have reversible readings than others: e.g., VN characters seem to have nonreversible readings with the sole definite exception of

222 <ń> for ~ ńi (see Kane 2009: 61 for the Chinese transcription evidence)

There are dedicated characters for some NV sequences other than ńi and ngV*: e.g.,

139 <na> and 191 <mú>

Conversely, vowel-liquid characters may have had reversible readings:

084 <ar> ~ <ra>, 098 <al> ~ <la>, 261 <el> ~ <le>?

See "<Ra>-Construction 5" and "Did Khitan Have Two Laterals?".

The VG sequence character

020 <ey> (Kane's <ei>)

represented <y> in word-initial position: e.g.,

020-084-131-344 <y.ar.u.ud> '耶律 Yelü'

Could <w> also represent a VG sequence: i.e., Vw? That is unlikely because such sequences already have characters:

019 <iu>, 023 <iu> (?), 067 <eu>, 138 <iû>, 161 <au>, 164 <au> (?), 210 <aú>, 289 <iú>

(I could also transliterate them as <iw>, etc. to match <ey> instead of <ei>.)

Might some of those characters be read as wV in word-initial position? Could 019, for instance, have been read wi 'to not exist, die'? I doubt it because Chinese w- was always written with initial

070 <w>

instead of any of the above <Vu> (= <Vw>) characters. That character never appears in medial position. If Khitan had medial -w-, it must have been written in some other way: e.g., could 019 <iu> stand for -wi- after vowels? Were

262 and its variant 263

ever read as wi instead of [uj]? Was

210-262-140 <aú.ui.en> 'woman of noble rank-GEN' (耶律撻不也 16.14)

from my last entry pronounced something like awiən?

The only non-Chinese Khitan word with an initial <w> is

070-050 <w.?> (興宗 15.19)

according to Qidan xiaozi yanjiu. Its reading is open to question, as Andrew West read it as

073-? <ên.?>

Alas, I can't consult the original because only a handwritten copy remains, and I have not been able to examine a good reproduction of it.

3.23.1:35: That mystery word occurs after a space that may indicate respect. Does it have an aristocratic referent?

Could it be an error for

072-050 <?.?>

from 道宗 25.17 and 耶律撻不也 16.4? Neither of those instances was preceded by a space.

*3.23.1:27: ng was only in Chinese loanwords. Initial and occasionally final ng were written with

264 <ng>.

The absence of ng in native Khitan words is not surprising since Janhunen (2003: 6) did not reconstruct it for Proto-Mongolic which may be the descendant of a 'sister' of Khitan. However, there is no guarantee that Khitan and Proto-Mongolic lacked the same consonants: e.g., Khitan had p, but Proto-Mongolic did not. Similarly, the absence of Proto-Mongolic *w does not guarantee the absence of w in native Khitan words.

Tangut fonts by Mojikyo.org
Tangut radical and Khitan fonts by Andrew West
Jurchen font by Jason Glavy
All other content copyright © 2002-2014 Amritavision