15.10.24.23:59: WHY RECONSTRUCT BOTH *-J- AND *-I̯- IN PROTO-HMONG-MIEN?

After a two-day detour into Korean ... back to my favorite topic!

I've written a series of posts starting here about how I've been troubled by Ratliff's (2010) Proto-Hmong-Mien (PHM) distinction between *j *ʷ in onsets and i̯ u̯ at the beginning of rhymes. Why not simply abandon the distinction, reduce the four to two (*j *w), and rewrite *-ji̯- as *-j-? The short answer is because the distinctions indicated by her notation are real, even if that notation may not be phonetically precise. Let's look at *tshji̯əŋ 'new' (last seen here) again and break it up into an initial and a rhyme.

I have no doubt that PHM had a distinction between the initials that Ratliff reconstructed as *tsh- (3.2) and *tshj- (3.17) which have different reflexes. Initial 3.17 clearly has a palatal quality absent from initial 3.2. Ratliff already reconstructed an aspirated palatal stop *ch- (4.2).

(I have excluded the initials of 'thousand' from 3.2 since some forms look like they might be post-PHM borrowings from Chinese. I excluded the initials of 'new' from 3.17 [even though I chose 3.17 because 'new' was reconstructed with it!] for reasons I will discuss next time.)

Those PHM initials remained unchanged in Ratliff's reconstructions of Proto-Hmongic and Proto-Mienic.

If 3.17 were something other than *tshj-, it might have been *tʃh- or *tʂh- (as in White Hmong tsh [tʂ]) which are absent from Ratliff's reconstruction. A retroflex value may fit Pulleyblank's hypothesis of *ks-retroflexion.

I am hesitant to reconstruct a palatal affricate *tɕh- since I know of no language contrasting palatal affricates with stops.

Similarly, rhyme 18g (*-i̯əŋ) is clearly more palatal than rhyme 21d (*-əŋ):

It does not help that Ratliff (2010: 161) gave only a single example of 21d, though it is a basic word (*hməŋH 'night'). I am surprised that *-əŋ is rarer than *-i̯əŋ; that may imply that 21d was more marked than 18g rather than the other way around. Or is the rarity of 21d simply random?

10.25.16:57: Are there reflexes of 18g and 21d with schwa? None of the sample of eleven languages have schwa. Of course there is no need for a proto-phoneme to be preserved intact in at least one descendant language, and a mid central vowel could easily lower to a or back to o.

Ratliff reconstructed the Proto-Hmongic reflex of 18g as *-in. Other PHM rhymes that merged with 18g in Proto-Hmongic are 18b *-im, 18c *-in, 18d *-iŋ, 18e *-i̯əm, and 18f *-i̯ən. I wonder if Proto-Hmongic *-in was phonetically something like *[iən] (cf. Southern American English diphthongization). White Hmong -ia more or less retained the original diphthong but lost the *-n, whereas Zongdi reversed the diphthong but kept the nasal: *-iən > -æin. Yanghao lost the schwa while Jiwei, Fuyuan, and Jiongnai fused *iə into a mid front vowel. Pa-Hng generally lost the schwa like Yanghao, but its word le³ for 'plum' has a front vowel. (I expected Jiwei, Fuyuan, and Jiongnai to also have front-vowel words for 'plum', but Pa-Hng is the only Hmongic language with 'plum' in Ratliff's language sample.)

15.10.23.23:33: WHY DOES KOREAN'S 'LARGE' CONSONANT HAVE A LOW FREQUENCY?

I love phonetic statistics. The first data set I ever saw was in Whitney's Sanskrit Grammar. The symmetry of Sanskrit phonology contrasted with the skewed distributions of phonemes in actual usage: e.g., this 5 x 5 grid of stops and nasals in which t is 665 times more common than jh!

The numbers are tiny and are hard to read. I apologize for any transcription errors. They

give the average percentage of frequency of each sound, found by counting the number of times in which it occurred in an aggregate of 10,000 sounds of continuous text, in ten different passages, of 1,000 sounds each, selected from different epochs of the literature: namely, two from the Rig-Veda, one from the Atharva-Veda, two from different Brāhmaṇas, and one each from the Manu, Bhagavad-Gītā, Çakuntalā, Hitopadeça, and Vāsa-vadattā (J.A.O.S., vol. X. p. c1). (p. 26)

Last night I found Pak Chae-yŏn's 고어사전 Koŏ sajŏn (Dictionary of Old Words) which listed the number of entries beginning with each premodern Korean consonant. I have rearranged his table to facilitate discussion:

The aspirates are much less common than their unaspirated counterparts, and aspirated kh-, the initial consonant of khɯ- 'large', is the least common - even though its unaspirated counterpart k is the second most common! Why?

I think those figures support the hypothesis of secondary aspiration in Middle Korean. If aspiration is from a preceding *h- or *k-, and if *kk- simplified to k-, then kh- had only one source whereas the other aspirates could have at least two - or perhaps even up to five in the case of tsh- which is twice as common as ph- and th-:

I propose *Cs-clusters as additional sources of tsh-, just as I proposed *ks- as a source of *tsh- in Old Chinese and in Proto-Hmong-Mien.

15.10.22.23:59: WHY DO KOREAN 'SEVEN' THROUGH 'NINE' END IN THE SAME CONSONANT? THE SEVEN-*UP HYPOTHESES

Last night I was thinking about Korean numerals, and it occurred to me that if ire 'seven days' was from ilgop 'seven' minus -op plus -e, then the root for 'seven' was il.

And the root for ahop 'nine', ah-, can be extracted from ahŭre 'nine days', whose nonharmonic -ŭre- might be by analogy with yŏdŭre.

Did early Korean have a suffix *-Up (whose vowel depended on vowel harmony) for 'seven' through 'nine'?

*nirk-up > Middle Korean nirkup > modern Korean ilgop (the shift of -u- to -o- is irregular)

*yətɯr-up > Middle Korean yətɯrp > modern Korean yŏdŏl (the final -p remains in some dialects: e.g., yadap, yədap, yədəp (see Chhoe 1978: 1047-1048 for details on locations) and yɐdɐp as well as yɐdɐl throughout Cheju

*ah-op > Middle Korean ahop > modern Korean ahop (*o harmonized with the low vowel *a)

That 'seven-*up hypothesis cannot be correct because the gamma of Middle Korean nirɣwəy 'seven days' is a trace of a root-final *-p:

*nirkup-əy > *nirgubəy > *nirɣuβəy > *nirɣβəy > Middle Korean nirɣwəy > modern Korean ire

I forgot that I had more or less proposed that shift last year. (In that earlier version, I wrote *-k- instead of *-g-. I don't know why I did that. I assume the lenition of *k after *r is a case of irregular compression. If such lenition were regular, *nirkup should have become Middle Korean *nirɣup.)

In that earlier post - or this even earlier post about counting days with an erroneous explanation of the words for 'seven days', 'eight days', and 'nine days' - I did not address why 'seven' through 'nine' all end in -p. I didn't even notice that until last night, almost thirty years after I first learned to count in Korean. I think it's because the -p of 'eight' is silent in modern standard Korean. I did, however, notice the -t of the four previous numerals right away: set 'three', net 'four', tasŏt 'five', and yŏsŏt 'six'.

Here's a different seven-*up hypothesis - one in which 'seven' is not a root 'seven' followed by a suffix *-up. What if root-final *-up spread by analogy to the following two numerals which were originally *yətɯr and *ah (cf. Middle Korean ah-ʌn 'ninety')?

*yətɯr > *yətɯr-up > *yətɯr-ɯp > yətɯrp

*ah > *ah-up > *ah-op (with suffix harmony) > ahop

(10.23.14:03: Chhoe 1978: 1046 lists ahup in Kyŏngnam, Phyŏngnam, and Phyŏngbuk. Could this be a retention of *ah-up without suffix harmony? Another Kyŏngnam form also found in Chŏnnam, agop, may retain a medial velar that weakened to -h- elsewhere; if so, then the root was *ak-.)

But if that were the case, why isn't the Middle Korean word for 'nine days' *ah-ay with the lower vowel variant of the suffix *-Ay? Perhaps the actual word ahʌray 'nine days' has -ʌray by analogy with yətʌray 'eight days'.

The latter word has an unexpected mix of higher and lower class vowels unlike yətɯrp 'eight' which only has higher class vowels. How can I account for that mismatch? I think Cheju yɐdɐp 'eight' reflects an earlier *yʌtɯr 'eight'. *yʌ then monophongized to *e outside Cheju. When Korean (i.e., non-Cheju Koreanic) developed vowel harmony, *etɯr became *etʌr and 'eight days' became *etʌr-ay. *-ʌray then spread to 'nine days'. Then *e broke to yə (via *ye?) with the higher vowel ə, and *yətʌrp (which may have 'grown' a -p by analogy with 'seven' by this point) was harmonized again as yətɯrp (regaining the ɯ lost during the first harmonization!) while yətʌray remained nonharmonic (perhaps to maintain parallelism with ahʌray 'nine days').

10.23.13:09: Perhaps Korean vocalic history can be described as a tug-of-war between harmony and analogy followed by the breakdown of harmony. Middle Korean yətɯrp 'eight' has harmony, whereas Middle Korean yətʌray maintain its original vocalism because of analogy. But neither harmony nor analogy can account for the lowering of u in modern Korean ilgop from Middle Korean nilkup. I would expect modern Korean *ilgup.

APPENDIX: The 二中歷 Nichūreki transcriptions of c. early 12th century Korean numerals

In Nichūreki, the Korean words for 'seven' through 'nine' were transcribed as <tarikuni>, <tirikuni>, and <etari>.

The first two look like attempts to write 'seven'. Their *n- might have been *[ⁿd] or *[nᵈ] or even *[d] and was hence transcribed with <t>-initial kana. Yoshida (2008: 2) cited Takeyasu (2004) on how modern Korean denasalized onsets can be perceived as voiced stops by non-Koreans. Could denasalization have been an old phenomenon in Korean?

The third looks like an attempt to write 'eight' (even though it was glossed as 'nine'). Perhaps the Korean informant said something like, "ndirkun, ndirkun, (y)...tʌr", repeating 'seven' twice, and the Japanese scribe missed the word for 'nine' and misinterpreted 'seven, seven, eight' as 'seven, eight, nine'.

I can't explain why 'seven' was transcribed with <ni> instead of an <f>-kana. No Korean dialect in Chhoe (1978: 1050-1051) has an -n-final word for 'seven'. The only nasal-final form, Chhŏngju ilgom, has m, not n. The Chinese transcription in 雞林類事 Jilin leishi (c. 1103) is 一急 *iʔ kiʔ without a final nasal. (My reconstruction may be too innovative. A more conservative reconstruction *il kip would still lack a final nasal. In any case, the transcription also lacks an initial nasal. Perhaps Korean */ni/ was *[ɲi], and this was misheard as *[i] since I doubt 孫穆 Sun Mu's Chinese dialect had *ɲ. By coincidence, Vietnamese has the readings nhất and nhứt for 一; that nh- [ɲ] is from *ʔj-.)

The Japanese transcription of 'eight' as <etari> was probably [jetari] in Japanese. <e>/[je] could have approximated *yʌ, *yə, *ye, or *e. If my theory of harmony above is correct, the target word might have been *(y)etʌr or *yətʌr prior to the second wave of harmonization. I assume that <ta> is for *tʌ with a nonhigh vowel, though I cannot rule out the possibility that it represented *tɯ with a high vowel. (Note how the high vowel *i of 'seven' was transcribed in Japanese with a low vowel kana <ta> as well as a high vowel kana <ti>.)

15.10.21.23:59: NO 'CLEAR' EVIDENCE FOR MY 'NEW' PROPOSAL ABOUT HMONG-MIEN MEDIALS

The quotes around 'new' aren't sarcastic; my proposal really is new. But my attempt to apply it to the Proto-Hmong-Mien word for 'new' was not convincing.

Unfortunately, nor is this attempt to apply it to Proto-Hmong-Mien (PHM) *ntshji̯əŋ 'fresh' (Ratliff 2010: 75) which is a Chinese loan with a 'double medial'* like *tshji̯əŋ 'new'. (The scope of "like" in the preceding sentence is conveniently ambiguous, as it is not clear whether 'new' is also a Chinese loan.)

In Baxter & Sagart's (2014: 356) Old Chinese (OC) reconstruction, 清 'clear' is *tsheŋ without any medial *-j- or *-i-. Why would this have been borrowed into Proto-Hmong-Mien as *ntshji̯əŋ?

The initial nasal (which is still in Hmongic languages today) is not an insurmountable problem, as it may have been from a prefix *N- in an OC dialect that fused with *tsh-, resulting in the voiced dz- or dʑ(ʰ)- in a few Gan forms. (10.22.10:22: Sagart 2002 suggested a possible Hmong-Mien substratum in Gan, so perhaps the OC dialect with *N- was ancestral to Gan.)

However, I would have expected OC *-eŋ to have been borrowed as PHM *-eŋ, not *-ji̯əŋ. Of course OC could not have been uniform, so perhaps there was a dialect in which *e broke to *jə or *iə before *-ŋ. It would be nice if I had a list of PHM words with *-jəŋ-type sequences corresponding to Baxter and Sagart's OC *-eŋ, but I don't. Yet. Going back to the single word I have on hand, dialectal OC *-jəŋ or *-iəŋ could have been borrowed as PHM -jəŋ or -i̯əŋ with a single medial. Why a double medial?

1. No preinitial

In Late OC, *e broke to *ie after nonemphatic initials like *tsh-. Then the initial *tsh- could have palatalized before *i, and PHM *-ji̯əŋ would reflect both the palatalization of the Chinese initial and the diphthong of the Chinese rhyme. (There is no PHM *-i̯eŋ, though there is an *-i̯en.) The trouble is that the palatalization of *tsh- is quite recent in modern Chinese languages, and there is no independent evidence for it at the Old Chinese level. And I am still reluctant to accept reconstructions with a distinction between *-j-, *-i̯-, and *-ji̯-.

2. PHM prefix

3. OC prefix *k-

Mainstream OC: *kɯ-seŋ > *kɯ-sieng > *ksieng > *tshieng

In my OC reconstruction, consonants before low class vowels like *e became 'emphatic' unless preceded by presyllables with high vowels: e.g., *kɯ-. The high vowel conditioned the partial raising of the following mid vowel: *i ... e > *i ... ie > *ie. The *k- fused with *ts- to become *tsh- (cf. *k- as a source of aspiration in Korean and perhaps also Tangut).

OC dialect source of PHM loan: *N-kɯ-sieng > PHM *N-ki-siəŋ > *N-ki-sjiəŋ (presyllabic *i conditioned secondary *-j-) > *N-ksjiəŋ > *ntshjiəŋ

Here I have rewritten Ratliff's glide-vowel sequence *i̯ə as a diphthong *iə (like modern Vietnamese ia/iê [iə]) to avoid a double glide sequence *-ji̯-.

The trouble is that Pulleyblank proposed that OC *ks- fused into *kʂ- rather than *tsh-. Then again, offhand I recall *kʂ- only being in formerly emphatic syllables, so perhaps *kˁsˁ- (phonetically [qˁsˁ]?) became *kʂ- whereas *ks- became *tsh-.

4. OC prefix *ni- (added 10.22.10:44)

1f one objects to *k-, I could rewrite scenario 3 above by placing *i before *N-:

Mainstream OC: *Nɯ-tsheŋ > *Nɯ-tshieng > *tshieng

OC dialect source of PHM loan: *Nɯ-tshieng > PHM *Ni-tshiəŋ > *Ni-tshjiəŋ (presyllabic *i conditioned secondary *-j-) > *ntshjiəŋ

In this scenario, a nasal is not a Gan innovation but was once present in the ancestor of all (?) Chinese languages.

*What I call a 'double medial' is really part of the onset followed by part of the rhyme in Ratliff's phonemic analysis. *-ji̯- looks as if it should be pronounced [jj], though I suggested other possibilities.

15.10.20.23:50: HOW WOULD RATLIFF'S PROTO-HMONG-MIEN *-JI̯- HAVE BEEN PRONOUNCED?

In my last post, I mentioned a hypothetical Proto-Hmong-Mien medial cluster *-jw. The closest actual cluster in Ratliff's (2010) reconstruction is *-ju̯- which she analyzed as the end of an onset plus the beginning of a rhyme. She also reconstructed *-ʷi̯- which has a similar analysis: labialization of the onset plus the beginning of a rhyme. These sequences could have been pronounced [jw] and [wj], and perhaps one was [ɥ]. (Does any attested language have a phonemic distinction between /jw/ and /wj/?) But how could *-ji̯- in her *tshji̯əŋ 'new' (p. 74) have been pronounced?

At first I wanted to rewrite one of the palatal medials as a preinitial or presyllable with a high front vowel: e.g., *T-tshjəŋ or *Ci-tshjəŋ. The word vaguely resembles Old Chinese 新 *sin 'new' whose *-n might be an *-ŋ that fronted after a front vowel. What if the Proto-Hmong-Mien word were borrowed as *Ki-sjəŋ from an Old Chinese dialect with a *k-preinitial or presyllable and schwa that blocked the fronting of the; nasal coda? In later Proto-Hmong-Mien, *-jəŋ could have become something palatal but without *-j-, and *Ki-s- could have fused into *tshj- with a new *-j-. The trouble is that there is no Chinese-internal evidence for a velar preinitial. Although Matisoff did reconstruct Proto-Tibeto-Burman *g-sik 'new' with a *g- matching my *K-, that word ends in *-k, not *-ŋ which may be a Chinese innovation.

(I don't believe Sino-Tibetan has a Tibeto-Burman subgroup, but I provide Matisoff's form anyway as a rough composite of non-Chinese Sino-Tibetan words for 'new' such as the kə-ʃək-type forms in rGyalrongic languages [see #1700 in Nagano and Prins' list].)

15.10.19.23:59: ARE PREINITIALS SOURCES OF RATLIFF'S PROTO-HMONG-MIEN MEDIALS?

I've continued to ponder the problem of /j ʷ/ and /i̯ u̯/ coexisting in Ratliff's Hmong-Mien reconstructions. Could they be reinterpreted as four different glides: e.g., /j w ɥ ɰ/? But I would rather not posit exotic glides when the data does not support them.

What if Hmong-Mien has a mixture of primary and secondary /j w/?\

- syllables with original /j w/ developed in certain ways

- preinitials *T- and *P- conditioned new /j w/ which developed in different ways

- the idea of a preinitial coronal stop conditioning medial /j/ is from Jaxontov's (1965) Old Chinese reconstruction:

- I reconstruct *P- as a pre-Tangut source of Tangut medial -w-

Let's suppose that Ratliff's Proto-Hmong-Mien /jæn/ (p. 157) and /i̯æn/ (p. 113) could be rewritten as *T-æn and *jæn:

- Early Proto-Hmong-Mien *jæn > Late Proto-Hmong-Mien *iæn > Proto-Hmongic *i and Proto-Mien *æn (e.g., 'he/she/it')

- Early Proto-Hmong-Mien *T-æn > Late Proto-Hmong-Mien *jæn > Proto-Hmongic *jæn and Proto-Mien *jæn (e.g., 'footprint/track')

10.20.13:13: The above scenario involves a chain shift. Original *j strengthened to *i, forming a diphthong with the following vowel. (Proto-Hmongic and Proto-Mienic preserve opposite halves of *iæ). Then a new *j developed from *T-, resulting in *jV sequences filling the gaps left by the earlier *jV which had become *iV.

*K- is grave like *P- and could be another source of secondary *-w-:

*K-C- > *ɣ-C- > *ɰ-C- > *Cw-

I didn't think of it at first because I reconstruct pre-Tangut *K- as a source of aspiration in Tangut rather than a source of Tangut -w-.

*Ci.CV > *Ci.CjV > *CjV

*Cu.CV > *Cu.CwV > *CwV

*Tu.CV > *Tu.CwV > *T.CwV > *CjwV

In any case, I remain skeptical that there were two kinds of /j w/-type sounds in early Hmong-Mien, and I continue to think that the rich variety of correspondences was in part conditioned by factors other than original *j and *w. Although I lack hard evidence for presyllables, I think Proto-Hmongic */P.ɢa/ (or *Cu.ɢa?) 'to escape' and /ɢwa/ (or /ɢʷa/?) 'duck' are more likely than a proto-phonemic distinction between /ɢʷ + a/ (labiouvular + monophthong) and /ɢ + u̯a/. (uvular + monophthong). (I am ignoring tones.) Is there any attested language with a phonetic distinction between a 'weak w' and a 'strong w'? If Ratliff's Proto-Hmongic was not such a language, then I would assume /ɢʷa/ 'to escape/ and /ɢu̯a/ 'duck' were both phonetically [ɢwa] - but they developed differently, so they could not have been homophonous.

15.10.18.23:59: RATLIFF'S PROTO-HMONG-MIEN MEDIALS

The problems with rhyme-final /j i/ and /w u/ in Ratliff's (2010) Proto-Hmong-Mien (PHM) notation that I discussed in my previous post have parallels earlier in the syllable. Ratliff reconstructed onsets with /j ʷ/ and rhymes beginning with /i̯ u̯/. How did /j/ in Proto-Hmongic (PH)* /mphje^D/ 'daughter' (p. 44) differ from /i̯/ in PHM /mphi̯ɛk/ 'chaff/husk' (p. 114)? Similarly, how did /ʷ/ in Proto-Hmongic** /ɢʷa^D/ 'to escape/ (p. 99) differ from /u̯/ in PH /ɢu̯a^C/ 'duck' (p. 103)?

Could /i̯ u̯/ be rewritten as halves of diphthongs? What if, for instance, PH 'escape' was [ɢwa] with a single vowel whereas PH 'duck' was [ɢua̯] with a diphthong [ua̯] with a stressed [u]? The trouble with that scenario is that I'd expect a stressed [u] to more or less survive into modern Hmongic languages, but neither of the modern reflexes of 'duck' in Ratliff's data have [u]:

Yanghao ka⁶ (cf fa⁸ 'to escape')

Fuyuan ʁwa^C (cf ʁwa^D 'to escape')

The reinterpretation of /i̯/ as a stresssed [i] in [i]-diphthongs leads to similar problems; 'chaff/husk' has no reflex with [i] in Ratliff's data:

Pa-Hng (the only Hmongic language in this set; the rest are Mienic) m̥e⁵

Mien bwa⁷

Mun va⁷

Biao Min bja⁷

Zao Min bjɛ⁷

10.19.0:49: I suppose one could reinterpret /j i̯/ as the palatalization of a consonant /ʲ/ opposed to a true glide, but I don't know of any language that has a phonemic distinction between consonants with labialization /ʷ/ and consonant-/w/ sequences. As far as I know, labialization and /w/ are phonetically identical; the difference is phonemic: e.g., both Cantonese and Mandarin have the syllable [kwaŋ], but in Cantonese, /kʷ/ is a unit phoneme whereas Mandarin /kw/ is a phoneme sequence.

*Ratliff (2010: 31) listed /mphj-/ as a PHM onset, but I can't find an example of it in a reconstructed PHM word in her book.

**10.19.0:57: I used Proto-Hmongic examples because I wanted a clear minimal pair. Here's a slightly less parallel PHM pair I just found:

PHM *qʷuw 'far' : *qu̯oH 'old' (< Late Old Chinese 故 *koh, phonetically [qoʰ]?)

Although some Hmongic and Mien forms for 'old' end in [u], that is not necessarily evidence for a diphthong [uo̯] with a stressed [u] since those forms could be borrowings from (or influenced by?) later ku-like forms in Chinese.

10.19.13:30: Another solution is to reinterpret weak vowel-strong vowel /V̯V/ sequences as balanced /VV/ diphthongs like [uo] which would phonetically contrast with glide-vowel sequences like [wo]. I reconstruct Late Old Chinese and Middle Chinese with /VV/ diphthongs.