I've been trying to come up with a term for the study of the shapes I was seeing in my secondary vowel diagrams: e.g., triangles for Belarusian and Pacoh, a square for Tangut, etc. Phonogeometry was long-winded and there is no geo 'earth' in these diagrams, so I considered phonometry, but that word already exists with a different meaning. Phonogeometry has zero Google results ... at least until I upload this.

Primary vowel systems also have shapes: e.g., the triangle or V of Classical Arabic.

i   u

(For simplicity I'm omitting vowel length here and generally throughout this post.)

That triangle can be considered a 'hollow triangle' to distinguish it from the 'filled triangle' (or Y) of Proto-Austronesian and older reconstructions of Proto-Japonic:

*i   *u

Nowadays Proto-Japonic is reconstructed with a 'thick V' or 'double V'-shaped system like those of Old Chinese and my pre-Tangut:

*i   *u
*e *o

I would like to think that Proto-Sino-Tibetan also had such a six-vowel system, but I am not sure.

Frellesvig and Whitman's Proto-Japonic reconstruction and Gong's Tangut reconstruction have seven (core) vowels:

*i  *ɨ *u
*e *o

I don't know what to call that shape. Suggestions?

Some vowel systems are square, and their subsystems (in yellow) only fill part of that square: e.g., Pacoh (excluding diphthongs and tense mid vowels):

i ɨ u
e ə o
ɛ a ɔ

Vowels which appear in presyllables as well as main syllables are in yellow.

Middle Korean had ... I don't know what shape this is:

i   ɯ u
  ə   o

Green, blue, and pink indicate neutral, yin, and yang vowels. The minimal vowels ɯ and ʌ are in different shades of yellow.

Maybe o was [ɔ] which would be as far from u as a and ʌ were from ə and ɯ.

If exact heights are ignored in favor of a yin-yang split, the Middle Korean system is almost square:

i ə ɯ u
  a ʌ o

I reconstruct Old Korean *e in the lower left gap.

Schenker's (1993: ) Late Proto-Slavonic has an inverted U-shaped system:

*i *y *u
*e *o

Minimal vowels (jers) are in yellow. Vowels that can also be nasal are in purple; in East Slavic they shifted in opposite directions: nasal became low ja whereas nasal became high u.

The study of these shapes (and shapes within shapes) can help us determine which shapes (and changes in shapes) are more likely than others. A knowledge of probability can increase the realism of our reconstructions.

Graphic models could also be developed to represent consonantal and even morphological and syntactic systems, but such models would unfortunately be increasingly arbitrary because there is no objectively obvious way to organize their elements in a two or even three-dimensional space.

ADDENDUM: 3.15.3:07: Here's an alternate T-shaped interpretation of the Middle Korean vowel system:

i ɨ u
  ə o

I am skeptical because ɐ (= ʌ in my interpretation) merges with a and ɨ (= ɯ in my interpretation). How could ɐ raise to ɨ without merging with ə? That's not an issue in my interpretation: ʌ lowers to a and rises to ɯ without any vowel in the way. SHAPES OF SECONDARY VOWEL SYSTEMS

Last night, I realized that unstressed Belarusian vowels formed a triangle:

i ɨ u
(*e > a) a (*o > a)

It is symmetrical in origin as well as shape: mid vowels lowered and centralized.

That got me thinking about the shapes of other secondary vowel systems: i.e., subsets of vowels that only occur in unstressed syllables (e.g., East Asian presyllables).

Russian also has a symmetrical triangular system, though its origins are not symmetrical.

 *e, *i > ɪ *e, *ɨ > ɪ̈, *u > ʊ̈  *u > ʊ
  *o, a > ə (elsewhere)   
  *o, a > ɐ (pretonic)  

The diagram above is based on Wikipedia.

I just noticed that the system in Comrie's (1993: 832) description is triangular, albeit not symmetrical:

*e, *i > ɪ *e, *ɨ > ɪ *u > ʊ
  *o, a > ə (elsewhere)  
    *o, a > ʌ (pretonic)

(Comrie's ʊ corresponds to Wikipedia's ʊ̈ as well as ʊ.)

Pacoh isn't related to Belarusian, but it too has a symmetrical triangle-shaped secondary vowel system with ə instead of ɨ (Watson 1964: 144):

i   u

It has a three-way distinction in open presyllables and a single vowel ə in closed presyllables: e.g.,

papiː 'to converse'

tinol 'a post'

kucet 'to die' (cognate to Vietnamese chết 'id.').

bəmbaːr 'to divide by two' (cognate to Vietnamese hai < Proto-Vietic *haːr 'two'? -3.14.1:14)

Pacoh caught my eye almost exactly six years ago because I wondered if Old Chinese presyllables had a similar three-vowel system (without *ə) instead of the vertical two-vowel system I had reconstructed for years:


could have resulted from a merger of an earlier *i and *u.

These vowels do not have to necessarily be back. My choice of symbols was influenced by the Middle Korean 'minimal' vowels ɯ and ʌ. I also wanted symbols that would not overlap with my symbols for vowels in main syllables (*a *i *u *ə *e *o).

Ramsey (1991: 237) noted that the 'minimal' vowels of Middle Korean

occur in non-initial position in Middle Korean verb stems almost to the complete exclusion of the other vowels (except /i/, which is the neutral vowel in the vowel harmony system).

Could those noninitial minimal vowels /i ɯ ʌ/ have been reduced from an earlier full set of vowels?

I used to reconstruct a similar two-vowel system for pre-Tangut presyllables, but now I reconstruct a square four-vowel system:


The capital letters distinguish presyllabic front vowels from the main syllable front vowels *i and *e. For consistency I could rewrite the presyllabic back vowels as *U and *O or *A. This square system is reminscient of *a-less reconstructions of Proto-Indo-European and Schenker's (1993: 66) Early Proto-Slavonic, though those inventories were for complete vowel systems.

Proto-Indo-European vowels

*i *u
*e *o

*i and *u are allophones of /y/ and /w/. Only *e and *o are true vowel phonemes.

*a developed from *ʕe (Beekes 1995: 138).

Schenker's Early Proto-Slavonic vowels

*i *u
*e *a

All four vowels could be long or short. Short *a became short *o. I guess the long *oː he reconstructed for Late Proto-Slavonic (p. 82) was the result of compensatory lengthening and/or vowel contraction (pp. 80-81). VOWEL CLASS CLASHES IN KOREAN: THE RESULTTS OF RAISING

Last night, I wrote,

I have never understood instances of vowel harmony violation in modern Korean like modu 'all' (cf. Middle Korean moto which Kim Wan-jin projected back into Old Korean) or abŏ-nim 'father' (cf. Middle Korean apa- in apa-nim).

I started thinking about other instances of vowel harmony violation: e.g.,

1. Modern Korean pad-a  > nonstandard pat-ŏ 'receive, and ...'

2. Middle Korean atăr  > modern adŭl 'son'

3. Middle Korean amă  > modern amu 'any'

4. Middle Korean namo  > modern namu 'tree' (cf. modu 'all')

I thought the first type resulted from analogy (since most Korean verbs take the yin suffix -ŏ) and knew the second and third types were the result of regular sound changes -

- ă > ŭ in syllables (hence the attested transitional form amŭ 'all' from 捷解新語 1676)

- ŭ > u after labials

- but couldn't explain the fourth.

Tonight, I found this passage in Lee and Ramsey (2011: 265; vowels rewritten in the romanization on this site):

Early Modern texts showed the change /o/ > /u/ in non-initial syllables with increasing frequency during the period [the 17th-19th centuries AD]. This development represented a serious erosion of vowel harmony between stems and endings.

So that takes care of the fourth type ('all' and 'tree').

I realized that all of the above instances involve the raising of noninitial yang vowels to yin vowels:

a > ŏ [ə] ('father', 'receive, and ...')

ă [ʌ] > ŭ [ɯ] ('son'; > u after labials: 'any')

o > u ('all', 'tree')

I can't think of any cases of yin vowels becoming yang in any position.

Conversely, the yang vowel ă lowered in initial syllables:

hă- > ha- 'to do'

Modern Korean completely lacks ă, though it survived in spelling into the twentieth century. Apparent instances of the raising of the initial syllable yang vowel a to ă and the lowering of the noninitial syllable yang vowel ă to a may reflect the merger of a with ă: e.g., these spellings for [adɯl] 'son' (whose correct spelling was adăl until ă was abolished):

ădăl (敬信錄諺解 1880)

ădal (華音啓蒙諺解 1883)

Why did vowels in different positions move in different directions? Were initial syllables stressed when the yang vowel ă lowered in initial syllables? Korean does not have stress now, but what if it had it in the past?

3.13.0:52: I also wonder why Russian and Belarusian unstressed vowels shifted in different directions (blue for raising and red for lowering):

Proto-East Slavic Russian Belarusian
*u [ʊ̈] after palatalized consonants
[ʊ] after nonpalatalized consonants
[u] (no change)
*i [ɪ] [i] (no change)
*e [ɛ] [ɪ] after palatalized consonants
[ɨ] after nonpalatalized consonants
initial [ja]
[a] with palatalization of the preceding consonant
*o [ɔ] pretonic [ɐ]
[ə] elsewhere
*a pretonic [ɐ]
[ə] elsewhere
[a] (no change)
[ɪ̈] no change

Unstressed Russian vowels in general shifted toward schwa (if not to schwa itself) except for

*e which raised to [ɪ] and [ɨ]

pretonic *o which centralized and lowered to [ɐ]

Unstressed Belarusian vowels lowered and bleached (lost their palatal and labial qualities).

Korean, on the other hand, mostly raised, and its lowering back vowel was the odd man out. VOWEL CLASS CLASHES IN MING JURCHEN?

Manchu has three types of vowels: 'yin' (blue), 'yang' (red), and neutral (green):

i e [ə] u
a o ū [ʊ]

Yin and yang vowels generally don't mix*, though both can coexist with the neutral vowel: e.g.,

gisurembi 'to speak'

hūlambi 'to read aloud, shout' < Chn 呼 *xu 'to call'

goidambi 'to last a long time'

How far back does this system go? Manchu is Jurchen with a new name dating from the 17th century, and a few of the words from the Sino-Jurchen Vocabulary of the Bureau of Interpreters (c. 1500) violate later Manchu vowel harmony rules:

'frost': semanggi (instead of *semenggi or *samanggi)

'rainbow': juwelemo (instead of *juwelemu or *juwalamo; see this note on u coexisting with a)

'rain': J agu (instead of *agū, *ago, or *egu)

Do these reflect

- retentions of an earlier stage of Jurchen without vowel harmony?

- limitations of Chinese transcription?

- errors in Chinese transcription?

- something else?

1. 'Frost'

The Jurchen spelling in the Sino-Jurchen Vocabulary of the Bureau of Translators (also from the Ming Dynasty) is harmonic:


Could the e of semanggi be a monophthongization of an earlier *ai in the Interpreters dialect? Perhaps it was [ɛ] or [æ]: i.e., a vowel lower than original e [ə]. This lower e could be romanized as ē by analogy with ū [ʊ] for the lower counterpart of u. I could use macrons and breves to indicate other vowels in an extended Manchu-style romanization of Jurchen**:

yin i e ă ŏ u
yang ī ē a o ū

However, I hesitate to reconstruct ē in 'frost' unless I can find more examples of e in the Translators dialect corresponding to ai elsewhere.

There is also a less exciting explanation. Jurchen se (or sē?) was transcribed in Interpreters as 塞 which could have been read as *sə or *saj. Perhaps the author intended the latter reading in 塞忙吉 *s-?-maŋ-ki, though he could have avoided ambiguity by using 賽 *saj instead. (The latter does appear for Jurchen sai elsewhere: e.g., 賽因 *saj-in for sain 'good'.)

2. 'Rainbow'

As far as I know, this word is unique to Interpreters. Maybe the final character of the transcription 拙勒莫 *tʂwə-lə-mo was an error for one of the various characters with 莫 as a phonetic pronounced as *mu: 幕墓慕募暮. However, Jurchen mu was normally transcribed as 木 in Interpreters (e.g., in 木都力 *mu-tu-li for muduri 'dragon'). (Oddly Kane 1989 left out 木 from his list of transcription characters in Interpreters.)

I doubt the word was *juwalamo since that would require two errors in transcription, as opposed to substituting one character for a similar-looking and sounding character.

Could the Interpreters dialect be like modern Korean whose vowel harmony is falling apart? I have never understood instances of vowel harmony violation in modern Korean like modu 'all' (cf. Middle Korean moto which Kim Wan-jin projected back into Old Korean) or abŏ-nim 'father' (cf. Middle Korean apa- in apa-nim).

3. 'Rain'

The Jurchen spelling in the Sino-Jurchen Vocabulary of the Bureau of Translators (c. 15xx) is harmonic and matches Manchu aga:


I wonder if the Interpreters transcription 阿古 *a-ku is either a mistake or an attempt to transcribe ag < *aga with apocope or an ag-like pronunciation of /aga/ like [ˈɑɢʌ] or [ˈɑɢɤ] whose second vowel is unstressed, short, and barely audible.

There was no way to indicate a distinction between u and ū [ʊ] in Chinese transcription, so the Interpreters transcripton  阿古 *a-ku might represent agū as well as agu. Although Kiyose (1977: 45-46) argued against a u : ū distinction in the Translators dialect on the basis of an absence of such a distinction in Ming Jurchen spelling, that distinction may have existed in Jin Jurchen and might have survived in the Interpreters dialect as well as the ancestor of standard Manchu. However, I am not yet aware of a regular shift of *a to u (or vice versa) without any nearby labials in Jurchen***.

*The yin vowel u and its yang counterpart ū are normally in complementary distribution: ū is after uvulars and u is elsewhere. Both uvulars and velars are romanized with the same letters (h, g, k), though they are written with different letters in the Manchu script.

u after nonvelars can coexist with yang vowels (e.g., juwan 'ten'), but u after velars generally cannot (e.g., (e.g., *guwan [guwan] is impossible except in Chinese loans). Strictly speaking, u is almost neutral; it is partway between wholly neutral i and wholly yin e.

ū cannot coexist with yin vowels: e.g., *gūwen [ɢuwən] is impossible.

** I intend this romanization to be as compatible with Manchu romanization as possible. Hence there are no macrons on the yang vowels a and o and no breves on the yin vowels i, e, and u.

I doubt I need ă since e is the yin counterpart of a.

ŏ also happens to be the McCune-Reischauer romanization of a Korean yin vowel, though Korean ŏ is [ə] like Manchu and Jurchen e rather than [o] which is the yin counterpart of o [ɔ].

A comparison of vowels in the extended Manchu-style romanization of Jurchen and the modified McCune-Reischauer romanization on this site

Question marks indicate vowels that may not have existed in Jurchen at any stage.

IPA Jurchen Korean
[i] i or i i
[ɪ] ī? n/a
[ɯ] ŭ
[e] n/a e
[ə] e ŏ
[ɛ] ē? ae
[ɐ] ă? n/a
[a] a a
[ʌ] n/a ă
[u] u u
[ʊ] ū n/a
[o] ŏ? o
[ɔ] o ŏ

Korean ㅓ ŏ  is listed twice since it has a range of pronunciations. See Martin (1992: 26) for [ɔ]. I recall that Leon Serafim suggested a similar range of pronunciation for Old Japanese ə.

I have considered romanizing Korean [ɛ] as ĕ, but the romanization ae is well-established.

***Kane (1989: 118) listed a couple of cases which may involve (unstressed?) *a assimilating to m in the Interpreters dialect:

'fish': nimuha (cf. Manchu nimaha)

'return': muri- (cf. Manchu mari-)

I am surprised that a did not simply round to o. Perhaps it raised before it rounded: > > > u.

Kane (1989: 117) also noted a single case of a in the Interpreters dialect corresponding to Manchu ū:

'palm of the hand': falangga (cf. Manchu falanggū) NO LONGER MID US: LEXICAL CHANGE IN ENGLISH, JAPANESE, JURCHEN, AND KOREAN

Much reconstruction of Old Korean is based on the implicit premise that Old Korean was basically Middle Korean written in Chinese characters instead of hangul. See my previous post for examples of several reconstructions of this type. But it is highly unlikely that Korean changed so little in the centuries before Middle Korean was first written alphabetically.

Today I found this list of extinct Old English words: e.g., mid which was replaced by with. How can we be sure that the Old Korean words for 'go' and 'spring' are not as extinct as mid 'with' - or, closer to Korea, Old Japanese in- 'go away'? (Old Japanese yuk- 'go' did survive in modern Japanese, but the general modern root for 'go' is its cognate ik-.)

The Sino-Jurchen Vocabulary of the Bureau of Interpreters (c. 1500) has Jurchen words without cognates in standard Manchu from the 17th century onward: e.g.,

'frost': J semanggi, M beiguwen, gecen, gecuhun

'rainbow': J juwelemo, M nioron

'bridge': J hufurun, M doohan, kiyoo (< Chn 橋 *kʰjaw)

Moreover, those words which do have cognates are not necessarily identical to them: e.g.,

'heaven': J agūwa* [ɑʁwɑ], M abka

'rain': J agu, M aga

'dew': J šilei, M silenggi

These differences are not dew, er, due to sound changes between Jurchen and Manchu; they reflect the fact that the Jurchen dialect of the Vocabulary was not the Jurchen dialect ancestral to Manchu. Similarly, the dialects of Old Korean poetry may not have been ancestral to standard Middle Korean.

*This is a Manchu-style transcription. LONGING FOR THE LOST SOUNDS OF SPRING

The speculative readings for the Khitan small script character


based on known Turkic, Mongolic, and Jurchen/Manchu words reminded me of the speculative Old Korean readings of hyangchhal semantograms. Here are several scholars' reconstructions of the first line of the Old Korean poem "Ode to Knight Chukchi" from c. 700 AD (translated by Peter H. Lee) with speculations in bold. Sinograph meanings and guesses based on them are in pink; sinograph readings and reconstructions based on them are in blue. All Old Korean reconstructions other than mine are in a McCune-Reischauer-like transliteration of the various scholars' hangul spellings. Middle Korean forms are in the same transliteration system for ease of comparison.

Middle Chinese *kʰɨəʰ *ʔɨnˀ *tɕʰwin *kɛj *lɨˀ *mejˀ
(southern *majˀ)
Chinese gloss to leave hidden spring all reason, to manage rice
Old Korean ? *-(ɯ)n ? ? *-Rɯ-? *-me or *-may
Middle Korean ka- sum- 'to be hidden'
-nă-n (processive modifier)
-(ă/ŭ)n (modifier)
pom ta, moto tasări- 'to manage' psăr 'rice'
-(ă/ŭ)m-ay 'because'
-(ă/ŭ)m-y-ŏ 'and'
Ogura Shinpei 1929* *ka *-năn *pom-i *ta *tasări- *myŏy
Yang Chu-dong 1942 *ka *-n *pom *kŭ *ri *may
Chi Hyŏn-yŏng 1948 *mŏy
Kim Sŏn-gi 1967-75 *kka *pam *ka *may
Sŏ Chae-gŭk 1974 *ka *pom *kă
Kim Chun-yŏng 1979 *kŭ
Kim Wan-jin 1980 *mot o- 'not come'
Yu Chhang-don 1994 *kŭ *myŏ
Peter H. Lee 1981 'All men sorrow and lament / Over the spring that is past'
cf. Middle Korean kŭri- 'long for'

Many Koreanologists seem to take these speculative readings for granted, even though they are based on the assumption that Old Korean is very much like Middle Korean. But there is no guarantee that Middle Korean preserved Old Korean words for 'go', 'spring', 'all', etc. without any changes for centuries. I consider *ka- for Old Korean 去 'go' to be as certain as au for Khitan 'heaven'; both are merely plausible guesses. We may never know the readings of and 去 unless phonetic spellings are found. So what I do think?

去: Semantogram. There is no way to know that its reading was the ancestor of Middle Korean ka-, but even if it was, there is no guarantee that it was ka- centuries earlier. Vovin (2010: 28) reconstructed 'go' in Proto-Korean as *kan- or *kal-. If 隱 represented a vowel-initial suffix, then 'go' may have been a consonant-final stem (*kan-, *kal-, or something else entirely).

隱: Phonogram for modifier suffix *-(ɯ)n for 'go'. 隱 was later abbreviated to 阝 and even ㄱ (no relation to the hangul letter ㄱ k) as a kugyŏl symbol for n.

春: Semantogram. Reading unknown. Even if its reading was ancestral to Middle Korean pom, it could have been something like *palum* if it was related to Old Japanese paru 'spring'**.

An analogy: Suppose English fall were written semantographically as 秋 . If future English retains autumn and loses fall, one might misread 秋 as *autumn. *pom (or even *palum) could be to some lost Koreanic word for 'spring' what autumn is to fall in this scenario.

皆: Phonogram for the first syllable of a polysyllabic word. The following phonogram 理 is for a second syllable since Proto-Koreanic probably did not have initial liquids. Oddly nothing in later Korean indicates an in earlier Korean, so I am puzzled by the choice of an *-ɛj sinograph. (Modern Korean [ɛ] is from *ay and is not inherited from Old Korean.)

Kim Wan-jin's reconstructed reading *mot o- 'not come' (based on an assumed reading *moto for 皆 'all') is ingenious, but I can't remember any other instance of a phonogram representing an entire word followed by the beginning of another word.

理: Phonogram for a liquid-initial syllable *-Rɯ'-. I cannot tell if the final vowel is part of a stem *kVRɯ- or a following suffix for a stem *kVR-.

米: Phonogram for *me or *may. I don't know which Chinese reading the writer had in mind: *mejˀ (cf. idealized Middle Sino-Korean myŏ̌y < *mey) or *majˀ (cf. early Sino-Japanese mai, probably borrowed from Sino-Paekche).

*me could not be the source of Middle Korean -m-y-ŏ [mjə] 'and', though it's possible that Proto-Koreanic *-y-ə fused into *-e in this Old Korean dialect which then could not be ancestral to Middle Korean.

*may may have survived intact in Middle Korean, though it might not have meant 'because' in Old Korean.

*I use *l to symbolize an early Koreanic liquid that was lost in medial position unlike *r. This liquid survives in early transcriptions: e.g., Jpn nirimu for Paekche *nilim 'master' cognate to Middle Korean nim. I wrote about the possibility of two Proto-Koreanic liquids in a four-part series: 1 2 3 4. This *l is not to be confused with the [l] allophone of the modern Korean liquid phoneme.

Sino-Korean has -o (idealized as -ow) for Middle Chinese *-aw. I have long assumed that an un-Korean *-aw was Koreanized as -o, but perhaps o is the result of monophthongization that followed medial *-l-loss: e.g.,

Early Koreanic *palum > *paum > Middle Korean pom 'spring'

寶 Late Middle Chinese *paw > Old Sino-Korean *pau > Middle Sino-Korean po(w) 'treasure'

Such a monophthongization occurred in Japanese: e.g.,

寶 Late Middle Chinese *paw > Old Sino-Japanese *pau > Modern Sino-Japanese hō 'treasure'

**Vovin (2010: 105) rejected a connection between Middle Korean pom and Old Japanese paru.

Tangut fonts by Mojikyo.org
Tangut radical and Khitan fonts by Andrew West
Jurchen font by Jason Glavy
All other content copyright © 2002-2013 Amritavision