Peter Boodberg proposed that some Chinese disyllabic words were the products of dimidiation (halving) of earlier monosyllabic words with initial clusters: e.g.,

痀僂 *koroʔ < ?*kroʔ 'hunchbacked'

(example from Boltz [2003: 171] with my reconstructions)

The 历口 Likou variety of Hui has a very different kind of dimidiation which I would describe as nasogenesis:

*bans > *bɨanh > *bua̤n > *vyæn > *fuːɛn > *fũːɛ̃n > *fũːɛ̃ > fũːmɛ̃ 'rice'

(The relative chronology of some of the segmental changes is questionable, but -m- must have developed after the vowels became nasalized.)

I wonder if *ia before *nasal codas has similarly become ĩːnɛ̃.

Although Likou has nasalized vowels as vestiges of earlier nasal codas, the rhymes of other dialects have undergone nasoexodus:

屯溪 Tunxi: 线 siːɛ < *sienh 'line', 张 tɕiau < *ʈɨaŋ 'sheet'

the correspondence of u : is also found in Sino-Japanese: e.g., 張 chou < tiyau.

休宁 Xiuning: 园 yːɛ < *wuan 'yard' FRICATIVE ONSET VOICING

In my last post, I mentioned that z is rare compared to other voiced obstruents. Moreover, s occurs in three times as many languages as z in the UPSID sample. Nonetheless, *s has become [z] in Dutch and German: e.g.,

'six': Dutch zes, German sechs [zɛks]

As a result of this change, [z] may be more common in Dutch and German than [s].

Moreover, *f has become v in Dutch and its German reflex [f] is spelled as v (implying an earlier *v < *f?) as well as f: e.g.,

'four': Dutch vier, German vier [fiːɐ]

'five': Dutch vijf, German fünf

What accounts for the v ~ f spelling variation in German?

Initial *s- and *f- also voiced in some dialects of English. Vixen, cognate to fox, is a borrowing from an *f-to-v dialect.

Why did initial *s- and *f- voice before vowels? Were they assimilating to a following segment? Why didn't other voiceless obstruent onsets voice before vowels?

Are there any other languages in which voiceless obstruent onsets have voiced before vowels?

Lastly, does the German spelling z for [ts], the reflex of earlier *t, imply an obsolete voiced pronunciation *[dz], the affricate counterpart of [z] < *s? I doubt that *[dz] existed since [ts] < *t is the acute counterpart of pf < *p, and I know of no evidence for a *bv stage between pf and *p. Hence I think the letter z was chosen semiarbitrarily for [ts]. (z and ts are both sibilants, though they differ in terms of voicing and degree of obstruction.) INTERVOCALIC VOICING IN NORTHEASTERN JAPANESE DIALECTS: ARCHAISM OR INNOVATION?

The Touhoku (Northeastern) dialects of Japanese have intervocalic voicing in various degrees: e.g.,

Akita: /k s t/ > [g z d]

but I can't find any cases of /s/ > [z] at Wikipedia or at akitafan.com (not the best sources, I know)

Kesen, Shinjou, Sendai:: /k t/ > [g d] (but not /s/)

(Akita, Shinjou, and Sendai have intervocalic [tɕ], but I don't know if Kesen has [tɕ] or [dʑ].)

Shimokita: /k c t/ > [g dʑ d] (but not /s/)

If intervocalic voicing is an archaism, sibilants were the first to devoice.

Conversely, if intervocalic voicing is an innovation,

- it might have begun with *p which has presumably lenited to [w] or zero as in 'mainstream' Japanese:

Touhoku *p > *b > > *w

'mainstream': *p > > > *w

but intervocalic voicing could have occurred after *p > > > *w in Touhoku as well as in the 'mainstream'

- it has not yet spread to sibilants in some dialects (are there any clear-cut cases of /s/ as [z]?)

Maybe early Japanese /s/ was like Korean /s/ which is aspirated. Korean aspirates /kh ch th ph s/ never voice intervocalically unlike Korean nonaspirated /k c t p/. Similarly, an early Japanese aspirated */s/ would not have voiced intervocalically unlike */k t p/. Zero reflexes of */s/ in intervocalic position could be from *[sʰ] via *[h]. Avoidance of intervocalic [z] in early Japanese would not be surprising, since z is not in nearby languages (Korean and Manchu), is not reconstructible in the ancestors of 'Altaic' languages*, and is much rarer than g, d, or b in UPSID:

g z d b
% of the 451 languages in the UPSID database 56.1% 13.7% 26.6% 63.6%

I'm surprised that g is almost as common as b, given that g seems more likely to become something else (e.g., [ɦ] in Czech, Slovak, and Ukrainian, [ɣ] in Belarusian, [ɣ] ~ [x] in Dutch, and [dʒ] in Arabic). The low frequency of d may be due to mergers with other acute voiced consonants like l or n. (For example, Taiwanese l is a merger of *d < *nd and *l.)

*I use the term 'Altaic' purely in an areal sense. PRENASALIZATION IN YUSUHARA

The Yusuhara dialect of Japanese as spoken by a 73-year-old man in 2003 had at least eight voiced obstruents:

ʑ dz d b
ŋg nd mb

The three prenasalized voiced obstruents were in common words including Chinese and recent Western loans, whereas their nonprenasalized counterparts tended to be in less common and literary words as well as the helping verbs /ga/ and /da/. (It's not clear whether the nonprenasalized counterpart of ŋg could be g as well as ŋ.)

Prenasalized obstruents contrasted with nasal-obstruent sequences in medial position: e.g.,

/hida/ [hinda] 'a fold' : /hiNda/ [hinnda] 'it's the dignity'

Yusuhara also had a segment that Nagano-Madsen transcribed as [nŋ]. I have never heard of a prenasalized nasal and I wonder if it is simply a long nasal ŋŋ (even in initial position!?).

This chart of velar voiced obstruent distribution is presumably incomplete.

initial medial after vowel medial after /N/
ŋg + ?
ŋ ? + +
nŋ +

I don't know if initial /g/ in uncommon words was realized as ŋ and/or g or if medial /g/ in uncommon words was realized as ŋg. I really wish Nagano-Madsen's paper had more example words.

The distinction between initial ŋg and nŋ has no parallel in modern standard Japanese. The only example of initial nŋ is onomatopoetic, so I wonder if that unusual sound is limited to onomatopoeia.

Voiced sibilants were never prenasalized. I presume all three voiced sibilants were in both common and uncommmon words.

This is reminiscent of how nasal onsets in northwestern Late Middle Chinese became prenasalized stops:

*ŋ- > *ŋg-

*n- > *nd-

*m- > *mb-

with the exception of *ɲ- which became a nonprenasalized sibilant *ʑ- (transcribed in Tibetan as zh-), presumably via a shortlived stage *ɲʑ-.

Prenasalized sibilants are rare in the UPSID database:

Prenasalized voiced obstruent ŋg ɲʑ ɲdʑ nz ndz nd mb
Number of languages in UPSID 45 3 10 1 5 26 48

I can understand why prenasalized sibilant fricatives are rare since it is difficult to pronounce ns and nz without an intrusive transitional stop. However, I don't understand why prenasalized sibilant affricates are almost as rare.

Tangut may have had such an affricate (*ndz-) which was often transcribed in Tibetan as Hdz-. (H- can represent prenasalization in Tibetan.) mdz- and Hj- are rarer transcriptions. However, nonnasal transcriptions also exist (Tai 2008: 190):

ts-, gts-

dz-, gdz-, bdz-



Yusuhara lacked intervocalic voicing of voiceless obstruent phonemes: e.g., /gaikoku/ 'foreign country' was [ŋgaikɔku], not [ŋgaigɔgu]. Thus it resembles the Middle Japanese dialect recorded by missionaries and the reconstruction of Western Old Japanese in my book, though it is not a descendant of those dialects.

Yusuhara is a Kouchi dialect, and Polivanov called Kouchi "Sanskrit Japanese" (in Murayama's 1976 translation) because of its archaism. What other aspects of Yusuhara are archaic? ARTE DA LINGOA DE IAPAM: EVIDENCE AGAINST A PUSH CHAIN?

In general, I am skeptical about push chains in historical phonology such as the one I proposed here for Japanese: e.g.,

*nd began to lose its nasality and become more like *d (e.g., *Ṽd with nasalization on a preceding vowel rather than a preceding nasal consonant)

*d (an intervocalic allophone of /t/) 'fled' to *t

If a sound A became more like B, wouldn't it merge with B unless B 'fled in time' and became C?

I've thought that modern Japanese words like

ada 'enemy' < earlier *ata (not *anda!)

contain medial obstruents that didn't 'get away in time' (i.e., that didn't devoice intervocalically*):

*/ata/ [ada] > ada (instead of the expected [ata])

If the push chain hypothesis is correct, prenasalized obstruents began to denasalize before intervocalic voiced obstruents 'fled' from them (i.e., devoiced).

However, according to João Rodrigues' Arte da lingoa de Iapam (1604–1608),

... vowels preceding [d, dz, g]. were always nasalized while vowels preceding [b] were sometimes nasalized** (Hashimoto, 1950; Toyama, 1972). (Hamano 2000)

and the 17th century missionary romanization of Japanese used voiceless symbols for intervocalic obstruents corresponding to modern voiceless intervocalic obstruents: e.g., Feiqe *[ɸeike] rather than Feigue *[ɸeige] corresponding to modern Heike. Thus denasalization was in progress when intervocalic devoicing was already complete as predicted by the pull chain hypothesis: e.g.,

*/-t-/ [-d-] > *t

the 'd-space' once occupied by */t/ is now empty and ready to be filled by

*/Nd/ [nd] > *~d > *d

One problem with the pull chain hypothesis is that it provides no motivation for intervocalic devoicing. In the push chain scenario, devoicing is motivated by a need to further distinguish the nonnasal obstruents from their denasalizing counterparts. In the pull chain scenario, on the other hand, the unusual process of devoicing happens for no known reason and random words like */ata/ [ada] don't participate. My anti-devoicing view by definition avoids devoicing, yet it too cannot explain why random intervocalic obstruents voiced in words like ada.

*Google has only two hits for "devoice intervocalically". One refers to a constructed language and the other refers to a possible case of intervocalic devoicing (perhaps via geminates) in nonstandard Welsh. Since intervocalic devoicing is rare, I still doubt that it occurred in Japanese. (And did it occur independently in Ryukyuan languages, or was intervocalic voicing purely a mainland innovation?) Nonetheless, I assume that devoicing occurred for most of this post.

**Among the prenasalized voiced obstruents of Middle Japanese, */b/ (or */Np/) had the least need for nasalization. Other intervocalic obstruent pairs differed only in terms of nasalization, whereas /Np/ was a stop unlike its normal*** nonnasal counterpart:

[-nasal] */k/ *-g- */s/ *-z- */t/ *-d- */ɸ/ *-β-
[+nasal] */Nk/ *-ŋg- */Ns/ *-nz- */Nt/ *-nd- */Np/ *-mb-

My preferred anti-intervocalic reconstructions are:

[-nasal] */k/ *-k- */s/ *-s- */t/ *-t- */ɸ/ *-ɸ-
[+nasal] */Nk/ *-ŋg- */Ns/ *-nz- */Nt/ *-nd- */Np/ *-mb-

***I am ignoring the less common phoneme /p/ which did not contrast with */b/ = */Np/ intervocalically. JAPANESE CONSONANTS THROUGH TIME: A SPECULATIVE OVERVIEW

What follows is not a complete argument. These are my guesses at the moment which require testing.

Stage 1: Prehistoric

Voiceless consonants all had voiced counterparts, and vice versa:

*k *s *t *p
*g *z *d *b
*Nk *Ns *Nt *Np
*Ng *Nz *Nd *Nb
*n *m
*hn *hm
*y *r *w
*hy *hr *hw

y = IPA [j].

The voiceless sonorants were unit phonemes which could be written as n̥, etc.

The choice of ɦ as the voiceless counterpart of ʔ was influenced by Sanskrit which has [ɦ] (romanized as h) but no [h] in noncoda position. ([h] or visarga, romanized as ḥ, occurs only as a coda.) Czech and Slovak are modern languages with /ɦ/ but no /h/*.

A handful of Wei zhi transcriptions have sinographs implying voiceless (but not voiced!) velar or glottal fricatives in some language(s) spoken in 3rd century Japan:

*xɔ or *hɔ

*xɑwʔ/h or *hɑwʔ/h

The transcription graph

*ɣwɛk or *ɣwæk or *ɦwɛk or *ɦwæk

has a cluster similar to my early Japonic *hw- but with voicing.

Stage 2: Old Japanese

Voiced-voiceless pairs merged, and ʔ and ɦ merged into Ø (which is absent from my table):

*k *s *t *p
*Ng *Nz *Nd *Nb
*n *m
*y *r *w

The loss of the original voicing contrasts resulted in the phonemicization of once-predictable vowel pitches: e.g.,

*ʔá > but *ɦà > (if early Japonic was voiced-low like Cantonese)

*ʔà > but *ɦá > (if early Japonic was voiced-high like Siamese)

I am not yet sure whether early Japonic was VL or VH.

Stage 3: Middle Japanese
*k *s *t (*p)
*x > > *y (*ç > *y)
*Ng *Nz *Nd *Nb
*n *m
*y *r *w

(I regard *kw and *gw as clusters like *Cy and not as unit phonemes.)

With marginal exceptions (onomatopoeia and clusters in Chinese loanwords), *p lenited to *ɸ. *k lenited to *x and then before i in endings. *s (phonetically *[ɕ] before i?) also lenited to in some dialect(s). These fricatives voiced and lenited further into glides *w and *y. The latter disappeared completely before *i.

Stage 4: Modern Standard Japanese (allophones in parentheses):

h k ts t p
g dz or z d b
(ç) ɕ s (ɸ)
ɴ (ŋ) n m
y r w

I have nothing new to add to the story of the expansion of the Japanese consonant inventory, so I will only point out two things:

- the nasalized series of stage 3 has generally lost its nasality (with the exception of *Ng whose modern reflex is often pronounced as [ŋ])

- modern ç is an allophone of /h/ and is not a descendant of the short-lived *ç in stage 3

*In Slovak, /ɦ/ becomes [h] before a voiceless consonant, whereas Czech /ɦ/ becomes [x] in the same environment (Short [1993: 536, 458]). INTERVOCALIC DISSIMILATION IN JAPANESE?

Some believe that early Japanese voiceless obstruents had voiced allophones in intervocalic position:

/k s t p/ > [g z d b] / V_V

This is similar to modern standard Korean (except for the voicing of /s/).

I have long been skeptical about this idea because it is not strongly supported by transcriptive evidence: e.g., early Japanese intervocalic /t/ was not normally transcribed with Middle Chinese *d, etc. Although one could argue that the early Japanese script was phonemic and not phonetic, people can transcribe subphonemic distinctions in their own language using a script designed for a language in which those distinctions are phonemic. For example, there are Koreans who transcribe /k t p c/ differently according to position. Most Koreans who transcribe /k t p c/ as g d b j in intervocalic or even initial position do not transcribe final [k t p] with voiced symbols. Phonemic romanizations like Yale which are phonemic have never caught on. Thus 바둑 /patuk/ [paduk] tends to be romanized as paduk or baduk but not patuk or badug, even though the latter two are more phonemic (i.e., they consistently use one type of symbol [voiceless or voiced] for Korean plain obstruents).

However, it is possible that intervocalic voicing occurred after Chinese influence waned and the Japanese script evolved independently to become more phonemic. This would account for the near-total absence of intervocalic voicing in early texts on the one hand and later intervocalic lenitions on the other. Intervocalic voicing is a kind of assimilation that enables a speaker to keep his voice setting set to 'on' (indicated in bold) beginning with the first vowel:

*taka (with voicing turned off for -k-) > *taga (with voicing from the first -a- onward)

One might expect this -g- to lenite to -ɣ- before disappearing entirely. Yet the general reflex of early Japanese intervocalic *-k- is -k- in modern standard Japanese:

Early Jpn Early Jpn after voicing Normal MSJ reflex Abnormal MSJ reflex
*-k- *-g- -k- -Ø-
*-s- *-z- -s- none (but -Ø- in western Japan)
*-t- *-d- -t- (none?)
*-p- *-b- -w- -u- (phonetically lengthening of /u/ and /o/), -t- (before -t-)

The 'abnormal' cases generally involve grammatical endings.

There are doublets like hayaku and hayou (< *-yɔɔ < *-yau < *-yaɣu < *-yagu) from earlier *payaku; -ku is an adjectival suffix.

On Friday, I realized that regular intervocalic voicing would have to be followed by regular intervocalic dissimilation:

*taga (with voicing from the first -a- onward) > taka (with on-off-on voicing)

Is there any language in which intervocalic voiced obstruents almost always devoiced*? Isn't it simpler to assume that most intervocalic obstruents with the exception of *-p- never voiced except in or near endings (which are high-frequency and hence more susceptible to phonetic erosion) and a handful of nongrammatical items? (*-t- doesn't seem to have any lenited reflexes in MSJ.)

*p has an history unlike its brethren, as it lenited to in initial and intervocalic positions whereas initial *k, *s, and *t remain unlenited to this day in MSJ. Later, initial *ɸ- became MSJ h- before /a e o/ and MSJ ç- before /i/ and medial *-ɸ- normally became MSJ -w- (presumably via an intermediate stage *-β).

*I wondered if devoicing was part of a push or pull chain:

Push chain: *-mb-, *-nz-, *-nd-, *-ŋg- were losing their nasality, forcing *-b-, *-z-, *-d-, *-g- to become something else.

Pull chain: *-b-, *-z-, *-d-, *-g- devoiced, leaving a gap to be filled by *-mb-, *-nz-, *-nd-, *-ŋg- which no longer needed nasality to distinguish them from other obstruents. DID GSR 727A'-E' HAVE A LATERAL OR DENTAL ROOT INITIAL?

I found 17 variants of GSR 727a'-e' 䵼 MC *ɕɨaŋ 'cook' at dict.variants.moe.edu.tw. They have four types of phonetics:

1. 727f 將 OC *Cɯ-tsaŋ

2. 720a-d 昜 OC *Cɯ-laŋ

3. 732a 羊 OC *Cɯ-laŋ (or *ɢaŋ in Sagart's 2007 reconstruction)

4. 734a-e 商 OC *Cɯ-staŋ or *sɯ-taŋ

used as a loan graph for 賞 OC *Cɯ-staŋ or *sɯ-taŋ

昜 and 羊 imply that 'cook' was OC *Cɯ-hlaŋ and 商 implies that 'cook ' was OC *Cɯ-staŋ or *sɯ-taŋ. OC *Cɯ-hl- and *st- (or *sɯ-t-) are common sources of MC *ɕ-. (I don't know what Sagart's uvular equivalent of OC *Cɯ-hl- is.)

Karlgren doesn't list any of the variants in GSR. The oldest of the variants seems to be 𩰱 in Shuowen which lists 羊 as phonetic. This may mean that the *ts-like OC initial of 䵼 (which I reconstruct as *sɯ-tɯ-ts-) and OC *Cɯ-hl- had merged by the 1st century AD.

Tangut fonts by Mojikyo.org
Tangut radical font by Andrew West
All other content copyright © 2002-2008 Amritavision