08.4.26.23:59: THE ORIGINS OF POLISH Ę AND Ą

The previous post may have given the impression that Polish ę and ą are directly descended from earlier Slavic *ę [ẽ] and *ǫ [õ]. But I've known for some time that this isn't the case. Stone (in Comrie [1987: 357]) gave the following examples of nasal vowel correspondences:

'Matches':

'son-in-law': Polish zięć : Common Slavonic *zętь

'tooth': Polish ząb : Common Slavonic *zǫbъ

'Mismatches':

'hand': Polish ręka : Common Slavonic *rǫka

'row': Polish rząd : Common Slavonic *rędъ

(In current Polish, ę and ą are pronounced as oral vowels followed by homorganic nasal consonants in the above four words: e.g., zięć [ʑeɲtɕ]. However, these four words did originally have nasal vowels.)

I didn't understand how this happened until tonight when I found Rothstein's explanation in Comrie (1993: 691). An earlier length distinction was reinterpreted as a front-back distinction:

(Stone in [1987: 357] represents the Old Polish nasal vowel as an ø-like letter without a / inside the o. Neither Rothstein nor Stone state what the phonetic value of ø might have been. I wonder if it was [ə̃].)

Thus it's not surprising that Polish has ę ~ ą (< ø ~ øø) alternations parallel to its o ~ ó [u] (< *o ~ *oo) alternations.

The frontness of PS *ę left a trace in the preceding initial. Hence *zętь became zięć (not zęć) and ?*ręędъ became rząd (not rąd). Conversely, the initial of ręka (not rzęka) indicates a PS nonpalatal *ǫ. These initial qualities must have been established before the merger of front and back nasal vowels in Old Polish: e.g.,

(I presume that rz was once a palatalized r, though it is now homophonous with modern Polish ż [ʐ].)

4.27.00:54: Other Slavic languages added a height opposition to the original front-back opposition whereas Bulgarian fronted and delabialized *ǫ (adapted from Schenker in Comrie [1993: 80]):

Schenker did not list any reflexes for long PS nasal vowels, but I assume they are identical or similar on the basis of these cognates:

'row': PS ?*ręędъ > Bulgarian and SC red, Czech řad, Russian rjad

(but Slovak has riadok instead of rädok - did *ęę become ia after r?)

'tooth': PS ?*zǫǫbъ > Bulgarian zъb, SC, Czech, Slovak, and Russian zub

(I am guessing PS vowel length on the basis of ą < PS *ǫǫ in Polish rząd and ząb.)

08.4.25.23:59: TANGUT Ẹ̃ AND Ọ̃: FURTHER PARALLELS?

I accidentally reversed my reconstructed values for R65 and R76 in the previous post. I've fixed that and a number of other typos.

Although I currently reconstruct only front nasal tense vowels, I'd also like to reconstruct at least one back nasal tense vowel as well. Then my reconstruction would have a mid front vs. back nasal tense vowel opposition similar to the lower mid front vs. back nasal nontense vowel opposition of Polish, Old Church Slavonic, and Late Proto-Slavonic (as reconstructed by Schenker in Comrie [1993: 82]):

Tangut: ẹ̃ : ọ̃

Polish: ę [ɛ̃] : ą [ɔ̃]

OCS ѧ [ẽ] : ѫ [õ]

Late Proto-Slavonic *ẽ(ẽ) : *õ(õ)

LPS had both long and short nasal vowels, whereas Gong's reconstruction and my revision of it only have short nasal vowels.

How typical is this Slavic pattern of mid nasal vowels? The top ten nasal vowels in UPSID are

I have conflated higher mid and mid vowels. If lower noncentral mid vowel figures are added to the (higher) mid vowel figures, õ and ẽ would both outrank a:

One might expect a language that has only two nasal vowels to have õ and ẽ.

4.26.1:46: I looked at the vowel inventory of every language with ẽ, ɛ̃, õ, and ɔ̃ in UPSID. Nearly all of those languages had two or more other nasal vowels: e.g., ã, ĩ, ũ, ẽ. Only three had two nasal vowels with a front-back opposition:

Changzhou: æ̃ and õ

Seneca: ɛ̃ and ɔ̃

Yuchi: ẽ and õ

None had a single nasal vowel. I am now even more reluctant to reconstruct ẹ̃-type vowels in Tangut without back counterpart(s).

Neither Polish nor Old Church Slavonic are in UPSID.

According to Stone and Polański in Comrie (1993), Cassubian and the extinct Polabian language have a somewhat different two-way opposition of nasal vowels: ã (low nonlabial) vs. õ (mid labial). This is disguised by the use of Polish vowel symbols in Cassubian spelling: ę [ã] : ą [ɔ̃].

UPSID also has examples of laryngealized and pharyngealized nasal mid vowels:

Southern Nambiquara: ẹ̃, ọ̃ (also has ị̃, ạ̃)

!Xu: õ^ʕ, ɔ̃ɔ̃^ʕ (also has ã^ʕ, ãã^ʕ, ãẽ^ʕ, ãõ^ʕ, õĩ^ʕ, õã^ʕ)

Although it is dangerous to draw conclusions from a sample of two, perhaps exotic nasal vowels also tend to appear in groups.

08.4.24.23:59: TANGUT Ẹ̃ AND Ọ̃: POLISH PARALLELS?

In part 4 of "Pine Hair and Hard Thread", I wrote,

It's not clear what happened to syllables of the type *S-N(H) since Gong's reconstruction has no tense nasal rhymes. I presume *S-N(H) syllables merged with *-N(H) or *S-(H) syllables.

I forgot an obvious example of a pre-Tangut *SCVN(H) (= *S-N(H) in last night's notation) syllable:

TT1183 THOUSAND tụ R61 1.58 < pre-Tangut ?*stuŋ

cf. Written Tibetan stong

Perhaps pre-Tangut *SCuŋ became Tangut Cụ, but what about SC- combined with other vowels and nasal codas? Did all *SCVN(H) become CṾ(H)̣, merging with original *SCV(H) syllables?

Recently I've come to suspect that Tangut might have had tense nasal rhymes. As far as I know, Nishida was the first to reconstruct them, and they exist in other reconstructions as well, though their distributions are not identical:

With the exception of Gong, who did not reconstruct any tense nasal vowels, all others only reconstruct one or two mid tense nasal vowels:

Nishida: ɛ̣̃, ọ̃

Sofronov: ọ̃

Arakawa: ẹ̃, ọ̃

This site: ɛ̣̃, iẹ̃

Are high and low tense nasal vowels impossible? No, for Jianchuan Bai has high and low as well as mid tense nasal vowels. Four tense nasal vowels (in bold) appear in Lee and Sagart (1998) and three more appear in Xu and Zhou (1984), including a nasalized tense syllabic v (!).

All seven of these nasal vowels have nonnasal tense counterparts.

Are there languages with only mid tense nasal vowels? I know of no living language other than Bai with tense nasal vowels of any height. However, I do know of a living language which only has mid nasal vowels: Polish (ę [ɛ̃] and ą [ɔ̃]). If a language has only one or two nasal vowels, would those vowels tend to be mid?

Could Tangut mid vowels like ẹ̃ and ọ̃ be partly descended from nonmid pre-Tangut tense nasal vowels?

ẹ̃ < ?*ị̃, ?*ẹ̃, ?*ə̣̃

ọ̃ < ?*ạ̃, ?*ọ̃

but ?*ụ̃ became ụ (as in THOUSAND above) instead of merging with ?*ọ̃

Such mergers should be reflected in morphological alternations: e.g.,

ẹ̃ ~ i and ə as well as e

ọ̃ ~ a as well as o

4.25.00:05: It seems more likely that *SCVN(H) merged with *SCV(H) if *V was nonmid:

*SCuN(H) > Cụ (e.g., THOUSAND)

*SCiN(H) > Cị

*SCəN(H) > Cə̣

*SCaN(H) > Cạ

08.4.23.23:59: PINE HAIR AND HARD THREAD (PART 4)

The lack of correlations between Tangut and Chinese tones in transcriptions led me to propose that Tangut 'tones' were actually phonations that were absent from Tangut period northwestern Chinese: e.g.,

'level tone' = modal voice?

'rising tone' = creaky or breathy voice?

Even now, I suspect that the 'rising tone' originated from pre-Tangut final glottal(s) that I write as *-H.

However, tones and voice quality need not have been mutually exclusive in Tangut. Perhaps they were correlated as in Jianchuan Bai. The following table is based on Edmondson, et al. (2001: 88; color PDF here). Numbers represent pitch height on a six-point scale from one to six:

A similar Tangut table might look like this (with x and y standing for unknown tonal values and the prime symbol standing for 'similar tonal value': e.g., x' was similar to x, and [x] and [x'] were allotones of a single toneme /x/):

(Gong's reconstruction does not allow tense vowels to be nasal.)

If Tangut were like Jianchuan Bai, tense vowels would have higher pitches than lax vowels unless they were combined with harsh voice. But no known evidence indicates that Tangut tense and lax vowels had different pitches.

(If Tibetan transcriptions of Tangut have tonal spelling as proposed by Arakawa [1999], the tones of tense and lax vowels may have been transcribed differently. Arakawa found a distinction between transcriptions for 'level tone' and 'rising tone' syllables and other distinctions may yet be discovered.)

The Tangut table could be rewritten in terms of the pre-Tangut segments that have may conditioned nasality and voice qualities:

It's not clear what happened to syllables of the type *S-N(H) since Gong's reconstruction has no tense nasal rhymes. I presume *S-N(H) syllables merged with *-N(H) or *S-(H) syllables.

What segmental distinctions led to the fifteent-way opposition in Bai? Here are my guesses on the basis of the main correspondences between Chinese and Bai in the earliest layers of Chinese loans (A1 and A2) as presented by Lee and Sagart (1998). *C- represents a voiceless initial and *C- represents a voiced initial. *-T represents voiceless stop codas other than glottal stop.

(4.24.00:21: Unlike Edmondson et al., Xu and Zhao [1984: 12] and Lee and Sagart describe 35 as lax, not tense. Perhaps there is variation among Jianchuan dialect speakers.)

Unlike Tangut tenseness which originated from a preinitial *S-, Bai tenseness may have originated from

- voiced initials not followed by any glottal codas (42)

- stop codas (44, 21 [and 35?])

Two tones with tense vowels (66 and 35) seem to occur only in Chinese loans, even though I know of no Chinese languages with a tense-lax vowel distinction or a harsh-modal voice sequence. The Bai may have recycled tones 44 and 21 to imitate the pitches of tones which were not associated with vowel tenseness in Chinese.

08.4.22.23:55: WRESTLING WITH 'RANDOM' REFLEXES (PART 2)

'Random' reflexes aren't just a problem with Jingpho. In Matisoff (2004), Proto-Tibeto-Burman *a corresponds to twenty-one different Tangut rhyme categories:

The situation is messy, but not as bad as it seems. If tense and retroflex rhymes were conditioned by lost segments and length is ignored, the number of correspondences can be reduced to six:

(4.23.0:36: All but one instance of PTB *-a corresponding to Tangut -o can be ignored since Tangut -o is a suffix.

I think the sole example of a PTB *-a : Tangut root vowel -o correspondence is dubious:

PTB *na 'ill' : Tangut ŋo

The initials do not match and I know of no other case of *n- > ŋ- in Tangut. But that still leaves us with *a becoming five different vowels.)

The conditioning factors that led PTB *a to develop into (almost) any of the six vowels of pre-Tangut and their twenty-one Tangut reflexes are still not completely understood, though Matisoff has identified some tendencies (e.g., raising of *a after sibilants). I suspect that much of the variation is due to *a raising (and rounding?) to match the qualities of a vowel in a lost presyllable. A lot must have happened between PTB and Tangut, and we may never be able to reconstruct all the intermediate steps.

Although Zbu rGyalrong must surely be closer to Proto-rGyalrong than Tangut is to PTB, it too seems to have 'random' reflex syndrome. Here are some of the correspondences between rhymes in Zbu and other rGyalrong varieties from Jacques (2004). Zbu reflexes present in Jacques' detailed tables but absent from his summary tables are in bold.

I have no idea what conditioned the different Zbu reflexes. If one learned Zbu first, one would have a hard time trying to guess the vowels of cognates in other varieties of rGyalrong and vice versa.

Moreover, a single etymon can have multiple vowels in Zbu: e.g.,

'bake': kɐ-sqʌʔ, sqʰɐʔ, sqeʔ < PGR ?*sqa

cognate to Tangut ɣjii 1.14 < ?*Cɯ-qi (with ?*qi < ??*Ci-qa)

(Pre-Tangut *sqa would have become kạ.)

In such cases, the root initial couldn't have been the sole conditioning factor. Perhaps the answer lies partly outside phonology - e.g., in morphology*. (Note that 'bake' is a verb.)

One could reconstruct a unique vowel or diphthong in Proto-rGyalrong corresponding to each and every correspondence between the modern languages that wasn't conditioned by root initials, but this would result in an improbably gigantic vowel system. Although Tangut did have 105 rhymes ending in vowels or glides, I would rather not assume that Proto-rGyalrong also had a huge number of vowel-final rhymes in addition to the consonant-final rhymes that must also be reconstructed.

The difficulty of reconstructing low-level proto-languages or even pre-Tangut makes me very skeptical about high-level proto-languages like PTB. I don't literally believe in Matisoff's PTB reconstructions. I regard them as vague prototypes containing possible characteristics of earlier forms.

*4.23.0:30: I assume that Tangut doublets with different rhymes reflect different affixes: e.g.,

'bitter lettuce' kha 2.14 < ?*qha-H

'bitter' khie 1.9 < ?*Cʌ-qhi (with ?*qhi < ??*Ci-qha)

Cf. various Qiangic qha-words for 'bitter' and Old Chinese *khaʔ [qhɑʔ] 'bitter'.

08.4.21.23:59: WRESTLING WITH 'RANDOM' REFLEXES (PART 1)

I was quite frustrated last night trying - and failing - to find a general pattern of correspondence between Jingpho initials (in Matisoff's voicing-based notation) and Matisoff's Proto-Tibeto-Burman reconstruction:

Only reflexes that I found last night are listed. There may be others: e.g., I expect some b from *p.

Like Matisoff (2003: 15-16), I presume that the variation in voicing and aspiration was due to prefixes which were (mostly?) lost:

More crucial for the complication of TB [Tibeto-Burman consonant] manner developments are the intricate patterns of interaction between prefix and root initial. A *voiceless [root-initial] C could easily assimilate in voicing to a voiced prefix (e.g. *m-), while a voiceless prefix (e.g. *s-) could devoice or aspirate an originally *voiced [root-initial] C. The prefix might then drop, leaving only the change in voicing of the [root-initial] C as a trace of its former presence. Nothing in fact is more unstable in diachronic TB phonology than the voicing or aspiration of initial obstruents; there are innumerable TB word families with both voiced and voiceless allofams.

Matisoff (2003: 129) assumed that at least one PTB prefix *r- was preserved in Jingpho as lə-. However, his PTB initials did not behave consistently after this prefix: e.g.,

*r-kəw > lə-gu 'steal'

*r-ko-t > lə-khot 'scoop'

Why did *k voice in the first instance but aspirate in the second? One possibility is that the different reflexes of *k resulted from sound changes applying to multiple layers of *r-prefixes. Bold type indicates newly added prefixes.

In this scenario, early layer *r- became a fricative (stage 2) that merged with a following root initial as an aspirate (stage 3) whereas later layer *r- voiced root initials (stage 4). Then *r- became lə- and was attached to khot (stage 5), resulting in a word that appears to have only one prefix but actually also bears the trace of an earlier prefix. Synchronically, there is no difference between the two lə- of 'steal' and 'scoop', but the first lə- is older than the second.

I am extremely uncomfortable with scenarios like this. Although they are phonetically plausible, they are also difficult to test. Presyllables* and affixes that just so happen to vanish after conditioning the phenomena I want to explain are convenient and hence dangerous. Nonetheless, I do believe that presyllables and affixes are the key to variation within Tangut and Chinese and perhaps within Jingpho as well. (I know almost nothing about Jingpho, so I am not very confident about the solution presented above.)

What are the alternatives? Lexical diffusion? Borrowing from dialects which had undergone different sound changes? Both may account for some of the variation in these languages, but can they account for all of it?

*Does any language have postsyllables? I could only Google one instance of this term. It was synonymous with 'suffix' and its antonym was foresyllables (prefixes).

4.22.1:11: Matisoff (2003: 461) traced both Jp lə-khot 'scoop' and got 'be scooped out' back to PTB *r-ko-t, but I wonder if 'be scooped out' once had a voicing prefix: *C-k(h)ot > *C-got > got.

got is the base for a number of other Jingpho words: lə-got 'scoop' (again!), sə-got 'scoop up', and (Hkauri dialect) də-got 'scoop, ladle'.

Although Matisoff (2003: 380) reconstructed variation between *ko-t and *go-t all the way back to the PTB level, I suspect that the Jingpho got-forms all historically contain one or more prefixes. Alternately, *k- ~ *g- variation in PTB was due to prefixation (*k- ~ *C-k-) at the pre-PTB level.

I couldn't claim that *k regularly became kh or g before lə- because Jingpho also has lə-k-words: e.g., lə-ka 'ladder' (lă-kạ in Liu [1984: 113]), lə-kuŋ 'limb, branch' and lə-kuŋ 'dexterity' (both from Matisoff 2003: 130). Matisoff derived the lə- prefix in the last two from PTB *lak 'hand/arm' rather than PTB *r-. Maybe the final *-k of *lak prevented the root initial from voicing:

*r-k- > *r-g- > lə-g-

but

*lak-k- > lə-k-

If there are instances of lə-kh- and lə-g- from *lak- plus a *k-root, I would have to regard them as doubly prefixed:

*(lak-)r-k- > lə-kh-

*(lak-)C-k- > lə-g-

The *l-prefix could have been added before or after the innermost prefix conditioned aspiration or voicing.

08.4.20.23:59: PINE HAIR AND HARD THREAD (PART 3)

In Hani, tense vowels could not follow aspirated consonants.

At first glance, Jingpho appears to have a similar distribution of tense and lax vowels:

Perhaps Jingpho also has a constraint against 'double laryngeal' syllables.

But notice that the above table is missing a row for voicing. According to Liu Lu (1984), Jingpho only has one voiced obstruent* (ʒ; cf. the similar status of standard Mandarin r if pronounced as [ʐ]). I cannot find any examples of voiced initials followed by tense vowels in Liu's Jingpo jianzhi. The tense-lax distinction seems to only occur after voiceless obstruents**. And unlike Hani, Jingpho still has final consonants which can be preceded by both tense and lax vowels.

Here's what I think might have happened:

Pre-Jingpho had a distinction between voiceless and voiced obstruents. The voiceless obstruents became tense and this tenseness spread into the following vowel. (Tenseness also spread from onset to nucleus in Tangut, though voicelessness was not the conditioning factor.) Then tense and (lax) voiced consonants merged into a single voiceless series, and tenseness after this series became phonemic:

*kV > *ḳV > *ḳṾ > kṾ

cf. Tangut *sCV > *sC̣V > *C̣V > *C̣Ṿ > CṾ

*gV > kV

Vowels after aspirated and voiced initials remained lax.

The distribution of tense vowels in Jingpho is reminiscent of the distribution of tense consonants in Korean which cannot be aspirated or voiced:

Matisoff (2003: 15) regarded tense-vowelled syllables like kṾ as [kV] and lax-vowelled syllables like kV as [gV]. Perhaps his view and the official Jingpho romanization in the PRC*** reflect Jingpho pronunciation in Burma.

4.21.2:34: Matisoff's book contains examples of earlier voiceless initials in kṾ-type syllables and earlier voiced initials in kV-type syllables as well as cases which match my expectations. 'Mismatches' are in red. Only one example per pattern is given with one exception ('lick' and 'arm' which share an unusual course of development).

Initially, I thought that Jingpho added a prefix that voiced the original root initial of 'thorn': *C-ts- > *dz-. That might explain this case, but can prefixes account for all the other cases of voicing 'mismatches': e.g., the various reflexes of PTB alveolar affricates (Matisoff ̣2003: 34)?

PTB *ts- > Jingpho ts- ~ dz- (Matisoff did not mention tʃ- and dʒ-; see above)

PTB *dz- > Jingpho ts- ~ dz- ~ ʃ-

Moreover, Jingpho aspirates are not necessarily original: e.g., lə-khot 'scoop' < *r-kot (Matisoff 2003: 129). Why did *k aspirate in 'scoop' but voice in 'steal'? On p. 130, Matisoff reconstructed PTB *l-paŋ and *l-baŋ as possible sources of Jingpho lə-phaŋ 'deaf' and traced Jingpho dʒə-khu 'nine' back to PTB *dgəw (cf. Old Chinese kuʔ 'nine' with a voiceless unaspirated initial!). Jingpho ə-kha 'bitter' does have an aspirate like Old Chinese *khaʔ, but I wouldn't count on the Jingpho aspirate being a retention. Matisoff (2003: 419) even linked both Jingpho ʃə-tin 'buttock' and ləthin 'heel' to a single PTB root *til.

I don't understand what's going on with Jingpho initials. I'm only clear about one thing: Jingpho tenseness has little in common with Tangut.

*4.21.2:43: Jingpho ʒ was a sonorant that became a voiced obstruent (presumably after all other voiced obstruents had devoiced). Matisoff (2003) wrote this sound as r:

But here's one instance of ʒ which seems to be from an earlier lenited affricate (cf. my hypothesis of Tangut z- < *C-ts-):

Maybe ʒep is a typo for dʒep, though the word isn't in Matisoff's corrigenda.

Conversely, Matisoff (2003: 340) derived Jingpho kə-tsap from PTB *jaap 'fan/paddle', implying fortition of *j as another source of Jingpho ts-.

Matisoff (2003: 502) also derived medial -ts- from -s-: kə-tsut < PTB *sut.

Untangling the history of Jingpho initials might be a worthwhile topic for an MA thesis or even a PhD dissertation..

**4.21.2:55: Tense vowels cannot occur after s and ʃ even though they are also voiceless obstruents. Perhaps they are slightly aspirated like Korean lenis /s/ (phonetically [s] and [ɕ]).

Tenseness may be more widely distributed in other varieties of Jingpho. Dai et al. (1983) has examples of tense vowels after sonorants corresponding to preglottalized sonorants in Matisoff's (1963) fieldwork.

***4.21.0:12: In the official Jingpho romanization, voiced and voiceless obstruent symbols symbolize vowel tenseness:

ga [ka] : ka [kạ]

(is this romanization historically and/or dialectally based? would ga be [ga] in Burma?)

Similarly, some romanizations of Korean use voiced obstruent symbols for lax obstruents but use doubled voiceless obstruent symbols for tense obstruents:

ga

[ka] : kka [ḳạ]