In "Buffers Behind You and Me", I quoted from the Old Korean poem 兜率歌 (도솔가) Tosolka 'Tuṣita Song'. The Korean and Sanskrit don't match very well partly because Korean Tosol was borrowed from Middle Chinese *təwʂwit rather than directly from Sanskrit Tuṣita-:

Sanskrit t u
i t a
Middle Chinese *t *əw *w *i *t
Sino-Korean t o s o l


Sanskrit and MC ʂ are just two different ways of representing the same consonant (a voiceless retroflex fricative).

Inevitable mismatch

There was no syllable *tu in MC or even late Old Chinese, so MC 兜 *təw < late OC *tou 'helmet' was the closest available match for Skt tu.

Sino-Korean -l is from an earlier SK *-rʔ which in turn is from a late Middle Chinese dialect with *-r < *-t. So it is not as if Koreans heard Skt t and borrowed it as l. Modern Koreans borrow foreign -t as -thŭ rather than -l: e.g., Eng set > Kor sethŭ. (An unaspirated -t- is not possible between vowels in modern Korean.)

Unsurprising mismatch

Sanskrit final -a often corresponds to zero in Chinese. Perhaps this reflects the pronunciation of those words by clergy from the west during the period of borrowing. The Chinese tendency toward shorter words may also have been a factor: e.g., the reduction of disyllabic 佛陀 *butda 'Buddha' to monosyllabic 佛 *but. (6.8.2:14: There is a trisyllabic Chinese form  兜率哆 with 哆 for Skt -ta. I'll mention other trisyllabic forms next time.)

Odd mismatches

MC *-əw normally corresponds to SK -u, not SK -o. SK dictionaries list the expected tu as the reading of 兜 'helmet'. Yet the word for 'Tuṣita' is Tosol, not Tusol, even though the latter is closer to Sanskrit. The word must have been borrowed from a Chinese dialect which had not shifted 'helmet' from *tou to *təw. (Later, the word became dou [tow] again in Mandarin.)

I would expect the correspondence

Skt -i- : MC *-i- : SK -i-

rather than

Skt -i- : MC *-wi- : SK -o-.

in the second syllable. MC 瑟 *ʂit 'lute' would be a perfect phonetic match for Skt -ṣit-. Why wasn't it used? (6.8.2:15: It was used in another transcription I'll analyze next time.)

率 MC *ʂwit 'lead' (and its antonym 'follow'!) should correspond to SK sul and Mandarin shui, but its actual readings are SK sol and Md shuai. I suspect that they reflect 'emphatic' variants of the 'nonemphatic' OC word that developed into MC *ʂwit*ʂujh. MC forms in parentheses are not attested in the MC dictionary tradition.

emphasis early Old Chinese middle Old Chinese late Old Chinese Middle Chinese modern languages
nonemphatic *srut *srut *ʂuit *ʂwit Cantonese søt, Sino-Vietnamese suất
*srut-s *sruts *ʂuis *ʂujh (no descendants?)
emphatic *Cʌ-srut or *sʌ-rut *srut > *srout > *srot *ʂɔt (*ʂwæt) SK sol, Sino-Japanese sotsu
*Cʌ-srut-s or *sʌ-rut-s *sruts > *srouts > *srots *ʂɔs (*ʂwæjh) Md shuai

Next: Urban Heaven

6.8.1:51: Did 'lead' and 'follow' originally differ in terms of affixation (and later emphasis) before merging?

6.8.2:19: Skt Tuṣita- looks like it should be the past participle of √tuṣ 'be content'. But in fact the past participle is tuṣṭa- < tuṣ-ta- without -i-. Perhaps √tuṣ once had two past participles, and the one without -i- came to be used as the name of a heaven. Another Chinese name for the Tuṣita heaven is 妙喜足天 'heaven full of wonderful joy' - i.e., contented heaven. This translation suggests that there was a perceived (if not etymological) connection between Tuṣita and √tuṣ 'be content'. Not all nontranscriptive equivalents are necessarily good semantic guides: e.g., another name for Tuṣita is 知足天 'heaven full of knowledge' even though √tuṣ does not mean 'know'. BUFFERS BEHIND YOU AND ME?

In modern Korean, many postpositions and endings have two forms: one after consonant-final stems and another after vowel-final stems:

type subtype gloss after consonants after vowels
invariable initial vowel dative e
initial consonant comitative ('and') hago
variable 'buffer' consonant (in bold) topic marker ŭn nŭn
accusative ŭl rŭl
'buffer' vowel ŭ- (in bold) instrumental ŭro ro
two consonants after a consonant!* comitative ('and') kwa wa
completely different** nominative i ga

It is not entirely clear why there are so many different subtypes instead of a simple pattern like

- V-initial postpositions and endings after consonants: e.g., ŭn

- C-initial postpositions and endings after vowels: e.g., nŭn

I have long found (n)ŭn and (r)ŭl to be intriguing because their 'buffer' consonants are copies of their final consonants. (Korean /r/ is [l] in final position, so rŭl is /rŭr/.) Was there a rule to copy the final consonants and insert them as buffers? I find that rather implausible, and I'd like to extend Johanson's (2011) claims about Turkic to Korean, another 'Altaic'-type language.

The earliest attested forms of those Korean particles (隱 and 肹) may not have had n- and r-:

Old Korean 汝隱 (兜率歌 2.8-9)

semantogram <you> + phonogram (early Sino-Korean *ŭn)

'you' (topic)

cf. modern nŭn 'you TOP'

Old Korean 吾肹 (獻花歌 3.1-2)

semantogram <I> + phonogram (early Sino-Korean *hŭrʔ)

'me' (accusative)

cf. modern na rŭl 'I ACC'

How are 汝隱 and 吾肹 to be read? There is no agreement. (I have transliterated the hangul reconstructions of the various scholars.)

Scholar 汝隱 'you' (topic) 吾肹 'me' (accusative)
Kim Wan-jin nŭn na răr
Ogura Shinpei na r
Yang Chu-dong na hăr
Chi Hyŏn-yŏng năn
Kim Sŏn-gi na nan uri kkar 'us' (acc.)
Sŏ Chae-gŭk nŏ n na hăr
Kim Chun-yŏng nŏ ŭn
Yu Chhang-gyun nŏhŭi n 'you (pl.)' (topic)
Modern translation nŭn na rŭl

(6.7.00:52: Note the slight variations in the readings for the semantograms 汝 'you' and 吾 'I'. Most assume that OK and modern Korean had identical pronouns, but in theory the OK pronouns could have been noncognates.)

I don't think the topic and accusative postpostitions had OK initial *n- and *r- because they were not written with *n- and *r-initial phonograms. 隱 and 肹 were also used to write the OK particles corresponding to modern ŭn and ŭl: e.g.,

Old Korean 君隱 (安民歌 1.1-2)

semantogram <sovereign> + phonogram (early Sino-Korean *ɯn)

'sovereign' (topic)

cf. modern kun ŭn

Old Korean 花肹 (獻花歌 4.1-2)

semantogram <flower> + phonogram (early Sino-Korean *hɯrʔ)

'flower' (accusative)

cf. modern kochh ŭl

On the other hand, one could hypothesize that the particles had fixed spellings and that readers knew when to insert buffer consonants: e.g., they would see 汝隱 <nŏ ŭn> and read it as [nŏ nŭn]. This would be like writing the English past tense ending as -d and reading headd as [hɛdəd] with a vowel [ə] not indicated in the spelling.

However, I prefer WYSIWYG (what you see is what you get) reconstructions that more directly reflect the hard data. If the OK topic and accusative postpositions originally had zero and h- initials after vowels, how did they develop n- and r-?

There is no need to posit such a change if OK was not the direct ancestor of modern standard Korean. Perhaps Proto-Korean *r- in the accusative became OK *h-.*** But did Proto-Korean *n- in the topic marker disappear between vowels in OK?

*6.7.1:11: This highly anomalous pattern is the result of intervocalic lenition:

*-C kwa > -C kwa (no change)

*-V kwa > -V gwa > -V ɣwa > -V wa

See Martin (1992: 663) on late attestations of unlenited kwa after vowels.

Martin (1992: 662) derived kwa in turn from the verb kărp-a 'lining up':

*kărp-a > *kărβ-a > kăβa (attested!) > kwa

Modern colloquial hago 'and' is also from a verb: hago 'doing'.

**6.7.1:33: The original nominative was i which was reduced to y after vowels. ga was later added to y, and y was then lost:

*V i > V y > V y ga > V ga

Traces of y remain in the front vowels of these standard Korean nominatives:

*na i > *na y > nae [nɛ] ga 'I' (nom.; cf. na 'I' in isolation)

*nŏ i > y > ne ga 'you' (nom.; cf. 'I' in isolation)

*tsŏ i > tsŏ y > che ga 'I' (humble, nom.; cf. chŏ 'I' in isolation)

***6.7.1:50: In Lao,  initial *r- became h everywhere: e.g.,

*rɔɔy > hɔɔy 'hundred' (cf. Thai rɔɔy 'id.')

However, medial *-r- apparently did not become -h- everywhere in Old Korean or in any later attested variety of Korean. The *r- > *h- shift, if it occurred at all, may have been unique to the accusative. BOGUS BUFFERS?

Johanson (2011: 34) drew the following conclusions about Turkic:

The popular myth about the insertion of "connective" vowels after consonant stems is highly implausible. The alleged insertion of "bridging" or "buffer" consonants in order to prevent vowel hiatus is unproven.

I haven't read the article yet, but even before finding it, I was reluctant to posit something out of nothing as opposed to the reverse. In other words, I prefer loss to growth, though the latter certainly does happen: e.g., Bhat (2001: 68)* gave the following example of insertion (a.k.a. epenthesis):

Sanskrit sn- > Pali sin-

I expect epenthetic segments to be conditioned by their neighbors (or by analogy) rather than be random.

If I am reading Masica (1991: 189-190)'s :#NIA (New Indo-Aryan)-6 rule correctly, he seems to be saying that epenthetic segments were inserted to break up vowel sequences and prevent them from fusing:

[...] the separate vowels could be preserved by the use of śruti [lit. 'hearing']-glides (ẏ, v̇ = [w], and in some languages, particularly Sinhalese, also h) [...] Marathi tended to prefer v where other languages had y.

"[U]se" could imply a conscious effort to insert 'buffers' whereas "could be preserved by the presence of ..." could imply that the 'buffers' just happened to be there.

Masica's rule NIA-6a(2)
Glides after ā sometimes remained to give rise to diphthongs of a new type [āi, āu], often written āy, āv

applied to the ancestor of Skt rājā 'king', but I still don't understand how a word without any labials developed a labial segment. The velar glide ɰ is to a what y and v/w are to i and u. Did medial -y- assimilate to the a surrounding it?

Skt jā > Middle Indo-Aryan ya > *rāɰa > *rāwa > *rāv > Hindi o?

If this hypothesis is correct, j only became *v > o when next to a. But are there any other Hindi nouns ending in -o? rāo seems to be obsolete except as a surname, judging from the fact I can't find it in any of the Hindi or Urdu dictionaries at DSAL. Did other -āo nouns also become obsolete, or did rāo undergo a sui generis change?**

Turner #10674 lists another Hindi rāo in rāo-cāo 'love, amusement' which is from Skt rāga- 'passion'. The development of this o is much more straightforward. A velar stop became a velar glide***:

Skt ga- > *rāɣa > *rāɰa > *rāwa > *rāv > Hindi o?

Perhaps the prothetic j- of Slavic (Schenker 1993: 68) originated as *ɰ- before a****:

'apple': Proto-Indo-European *ʕebōl- ~ *ʕbōl- > *ʕebl- (with *e generalized) > *abl- > *ɰabl- > Russian jabloko (but English apple without initial y-)

*Bhat's book Sound Change is actually a selective catalogue of sound changes in languages on the Indian subcontinent in spite of its general title.

**6.6.1:22: One might expect Skt jāyā 'wife' to become H jāo just as Skt rājā became H rāo, but it became H 'mother' (Turner #5205; now apparently obsolete like rāo). (Is the fact that Skt jāyā and rājā have different genders and belong to different declensions relevant?)

***6.6.00:56: A similar shift occurred in Tangut:

*-k > *-ɣ > *-ɰ > -w (e.g., in *kʌ-tek > 1lew 'one').

****6.6.1:18: According to Schenker, *j- developed before long but not short *a. However, 'apple' has a short first vowel in Beekes' 1995 reconstruction, so the *j- in Slavic words for 'apple' is unexpected. Moreover, Czech has few native words with initial a-. Are all native Czech ja- from *ā? RĀY-DUCTION

In "Like a Dragon's Eye", I mentioned that Sanskrit initial y- hardened to Hindi j-. Conversely, Sanskrit medial -j- softened to Hindi -y- or even  zero:

Skt rājan- : Hindi राय(ा) rāy(ā), राओ rāo 'king' (Turner #10679)

Hindi rāyā looks like a descendant of the Skt nominative singular rājā. Is this too good to be true?

What Skt form is H rāy from?

And is H rao from the Skt ablative/genitive singular rājño < rājñ-as?

6.4.2:08: Platts (1884: 584 and 585) listed several more Hindi words for 'king':

राए rāe

"old" राइ rāi

राव <rava> rao

"dial." राऊ rāū

"dial." राउ rāu

(I am excluding obvious loans from Sanskrit like राजा rājā.)

Platts derived the -o, -ū, -u forms  from a suffixed Skt rāja-kas. If this is correct, were the intermediate stages like

Skt -akas > *-ak > *-aɣ > *-aw > -o, -ū, -u?


Skt -akas > -ako > *-aɣo > *-ao > -o, -ū, -u?

But I doubt that Skt -o or even -ako became H -o because masculine -(ak)o was very common in Sanskrit, whereas the most common masculine H final vowel is -ā.

Shakespear (1834: 957) derived rāe and rāī (with a long final -ī) from Sanskrit rai 'wealth'. Presumably they would specifically come from the oblique stem rāy-, as the nominative singular of rai is rās. But a semantic shift of 'wealth' to 'king' is doubtful.

Masica (1991: 190) derived both -y and -o-type forms from Skt rājā:

Skt rājā > Middle Indo-Aryan rāā, rāya > New Indo-Aryan rā, rāy, rāv > rāo

However, I don't understand how NIA v developed if there were no labial segments in the MIA stage.

Tangut fonts by Mojikyo.org
Tangut radical and Khitan fonts by Andrew West
Jurchen font by Jason Glavy
All other content copyright © 2002-2012 Amritavision