I conclude my trilogy on

4517 1dzi3 'eat'

by looking into the origin of its reading.

Although it is tempting to assume that Tangut dz- is a direct retention of an earlier voiced obstruent, dz- may have been [ndz] with prenasalization as in Muya ndzɯ³⁵, Japhug ndza, and Naxi ndzɯ³³. That prenasalization in turn may have been from an *m-. (8.21.23:19: See Matisoff 2003: 119 for the reasoning behind his reconstruction of *m.)

As for the rhyme, there is no doubt that it has been 'brightened' to use Matisoff's (2004) term: i.e., *a has raised to i. There are two problems regarding brightening that remain to be solved.

First, why did *a raise to i1 in some cases and i3/i4 in others? According to Guillaume Jacques (2014), the presence or absence of *-j- determines the type of raising:


*Cja > Ci3/Ci4 (Guillaume does not make a distinction between Grades III and IV; he uses Gong's reconstruction in which both grades are characterized by a -j- that is not supported by Tibetan transcriptions.)

But for years I have proposed that presyllabic vowel harmony conditioned the changes leading to Tangut's rich inventory of open-syllable rhymes. I have yet to fully integrate my views with Guillaume's, but here's a first attempt. *Ca by itself or with a lower vowel presyllable became Ci1, whereas *Ca with a higher vowel presyllable became Ci3 or Ci4 depending on the initial:

*(Cʌ-)Ca > Ci1

*Cɯ-CaCi3/Ci4 (normally Pi4, Vi3, Ti4, Ki4, TSi4, CHi3, Qi4, li3, rir4, zi4, zhi3)

e.g., *Nɯ-dza > 1dzi3 (!; see below)

Unfortunately I have not found the Sino-Tibetan equivalent of Hittite that preserves the presyllables predicted by my hypothesis.

Second, why did *a become i3 after dz- in 'eat' instead of i4? There is a related word

4513 2dzi4*Nɯ-dza-H 'eat, drink, food'


The second half of the fanqie spellng of the Tangut cognate of Wobzi dzí 'eat' is also puzzling:


4517 1dzi3 'eat' = 5051 1naq4 + 0932 1i3 (Mixed Categories of the Tangraphic Sea 4.252)

(9.4.23:59: The first half really isn't puzzling at all. The correct initial speller is 4973 1dzu4 with the expected dz-, not 5010. See "Eating Begins with Love, Not Marriage".)

Normally Class VI initials like dz- only combine with Grade I and IV rhymes: e.g., 1dzi1 and 1dzi4. But 4517 and its homophones

0382 1dzi3 'equal' and 4912 1dzi3 'cut'

have a Grade III rhyme -i3!

There are only ten syllables combining Class VI initials with Grade III rhymes. The other seven are


0524 1dzu3 'admonish, instruct' 

4973 1dzu3 'love'

4977 1dzu3 'father-in-law, uncle' (only in dictionaries)

5121 1dzu3 'dream' (only in dictionaries)

3408 1tsa3 'broil, roast' (only in dictionaries)

3976 2tsha3 (transcription character; only in dictionaries)

3371 1dza'3 'hair worn in a bun or coil; peak'

(8.21.1:54: Is it significant that Class VI initials can only precede four different Grade III rhymes?)

Why is the syllable type TSV3 (Class VI initial + Grade III rhyme) so rare? Was there something 'antialveolar' about Grade III? There certainly was no such quality in Grades I and IV.

Gong demonstrated a strong (but not absolute) correlation between Tangut and Middle Chinese grades. In Middle Chinese, Grade III was less palatal than Grade IV: e.g., in Chinese borrowings in Korean, Grade III -i is sometimes -ŭi, but Grade IV -i is always -i. For years I reconstructed Tangut Grade III -i as [ɨi] and Grade IV -i as [i]. The velarity of the first half of [ɨi] was the 'antialveolar' quality that usually made it incompatible with Class VI initials. But now I write those rhymes algebraically as -i3 and -i4 because I am no longer certain about their phonetic values. If -i3 were really [ɨi], why was

0932 1i3 'many, more, much'

a transcription character for Sanskrit i? Sanskrit has no [ɨi] or even [ɨ]. I would have expected Sanskrit i to be transcribed as 1i4: i.e., as [i].

8.20.23:54: Why can't I just reverse my reconstructions of Tangut i3 and Tangut i4 on the basis of 0932 1i3 transcribing Sanskrit i? Because that goes against what is known about the Chinese grades and because i3 is less common than i4 (88 : 209). It would be strange if i3 [i] were less common than [ɨi].

8.21.1:59: Another solution would be to reconstruct i3 as simple [i] and i4 as [ji]. But i4 was transcribed in Tibetan as i and not as yi. And characters with i4 were used to transcribe Sanskrit syllables with i, not yi. THE PAST AND PRESENT SOUND OF EATING IN TANGUT (PART 1)

After mentioning the Tangut cognate of Wobzi dzí 'eat' in my last entry, I looked up its fanqie and was surprised:


4517 1dzi3 'eat' = 5051 1naq4 + 0932 1i3 (Mixed Categories of the Tangraphic Sea 4.252)

Shouldn't those initial and final spellers add up to 1ni3 instead of 1dzi3? I don't know of any Chinese or Tibetan transcriptions of 4517 or its homophones

0382 1dzi3 'equal' and 4912 1dzi3 'cut'

so its reading has to be determined using internal evidence. 4517 is in the section of Mixed Categories for class VI initials: i.e., dental sibilants (ts- tsh- dz- s-). n- is a class III initial. The closest class VI initial to n- is dz- which is voiced and possibly prenasalized: [ndz]. I'm not entirely satisfied with that resolution to the conflict.

9.4.23:57: The correct initial speller is 4973 1dzu4 with the expected dz-, not 5010. See "Eating Begins with Love, Not Marriage".

Next: The rhyme of 4517.

8.19.22:57: Here is some indirect evidence for reading 4517 with an affricate initial. 4517 is the initial speller for


4855 2dze4 'one of a pair' = 4517 1dzi3 + 4517 2tse4

which is a transcription character for Sanskrit je. Palatals in standard Sanskrit pronunciation correspond to dental affricates in Tangut (as in Tibetan and Chinese). So the initial of 4855 - and its initial speller 4517 - must have been dz- (or something close like [ndz]).

Could someone explain why Sofronov (1968 II: 85) lists the initial speller of 4517 (5051) as part of class VI fanqie chain 12? The fanqie of 5051 has an initial speller which doesn't even have a class VI initial:


5051 1naq4 = 5700 2ni'4 + 5371 1taq4

It has a class III initial n- that can be confirmed by its Tibetan transcription ni (Tai 2008: 208). RIGHTWARD REDUPLICATION IN WOBZI

In "Triply Awake in Sanskrit", I asked,

Does any language have rightward reduplication?

I should have been more specific and asked about rightward reduplication in verb inflection as opposed to other kinds of rightward reduplication that I'm used to like

Japanese hitobito 'people' (< hito < pitə 'person')

Burmese kàũgàũ 'well' (< kàu 'good')

Thanks to Guillaume Jacques who understood what I was really getting at, I now know that the answer is yes. Lai Yunfan wrote an entire paper about rightward reduplication (RDP) in Wobzi: e.g.,

14a. nú ǽ-ɕʰə-vɣí-n-vɣɑ? (< vɣí 'full')

'thou DIR*-Q-full-2SG-RDP'

'Are you full or not? (I don't think you are.)'

14b. ŋó zɑmɑ̀ dz-ɑ́ŋ-dzu gædì. (< dzí 'eat', cognate to Tangut 4517 1dzi3 'id.')

'I meal eat-1SG-RDP after'

'(I will be) after I have finished eating my meal (but I don't really know if I'll finish it).'

Note that the person suffixes go between the root and its reduplication. I never expected that because I am accustomed to Sanskrit which has the structure


e.g., ād-a (< a- + √ad 'eat')


'I ate' (perfect)

I also asked,

Is there a tendency not to have a reduplicated root right before an inflectional affix?

In other words, is the structure √-RDP-INFL rare or nonexistent? I wasn't even thinking of RDP-INFL-√ when I wrote that.

8.18.22:09: There are six possible sequences of root, reduplication, and a single inflectional affix:


2. √-INFL-RDP (Wobzi)

3. RDP-√-INFL (Sanskrit)



6. INFL-RDP-√ (Sanskrit reduplicated aorists with the augment a- in the INFL slot; does any language have an agreement prefix in that position?)

Are 1, 4, 5, and 6 with an agreement prefix attested? Is the distribution of these sequences due to chance? What are the historical implications, if any, of these different orders? For instance, the first person singular and second person suffixes of Wobzi seem to be derived from pronouns.* (You can see both those suffixes and their possible source pronouns in the sentences above.) Did the grammaticalization of pronouns predate reduplication?

*√ > *√-INFL > √-INFL-RDP?

*On the other hand, the first person plural suffix -j does not appear to be related to ŋgə́ɟi 'we'. There is no third person suffix. HOW FREQUENT ARE REINTERPRETED REDUPLICATIONS IN SANSKRIT? The short answer: Not very.

For perspective, here are the frequencies of a few basic verbs in the Digital Corpus of Sanskrit:

kṛ 'do': 28926

as 'be': 20762

gam 'go': 13531

'give': 8626

'go' (extended version of the next root): 4166

i 'go': 1014

ad 'eat': 237 (Sanskrit speakers didn't write much about eating, but they must have talked about it. Written language is not a mirror of spoken language in terms of content as well as style.)

Now here are the frequencies of all the reduplicated verbs reinterpreted as simple verbs that I know of and their source verbs:

√jāgṛ 'awake': 155 (see "Triply Awake in Sanskrit" and §641 and §1020a of Whitney's Grammar)

√gṛ: 81

√jakṣ 'eat' (less common synonym of √ad): 69 (see "Tell Me Why They Speak Differently")

√ghas: 0 (the corpus is not comprehensive; see Monier-Williams for citations)

√cakas 'shine': 20 (see §641 and §677 of Whitney's Grammar)

√cakṣ 'tell' (< 'verbally shed light on'?): 12 (see "Tell Me Why They Speak Differently")

both of the above are related to √kāś 'shine' (which only appears once in the DCS without reduplication): 38

√dīdī 'shine': 3 (0 for its later variant √dīdhī; see §641 and §676 of Whitney's Grammar)

√dī: 0

√daridrā 'run': 0 (see §641 and §1024a of Whitney's Grammar)

√drā: 0

√vevī 'go': 0 (see §641, §676, and §1024a of Whitney's Grammar)

√vī: 284

√pīpī 'swell': 0 (see §676 of Whitney's Grammar)

√pī: 0

I now don't feel so bad about having either forgotten or never having learned these verbs because they're so rare. I suppose that if a verb is rare but had reduplicated forms in use (e.g., √kāś above), one might encounter the nonreduplicated forms so infrequently that one might regard the reduplicated forms as basic and build upon them: e.g., with rereduplications like the perfect 3rd sg jajāgāra 'awoke' < √jāgṛ < √gṛ.

8.17.23:40: One problem with that hypothesis is that the reduplicated forms of two of those verbs are rare in the DCS:

√gṛ: only one out of 81 instances is reduplicated

√vī: none of 284 instances is reduplicated

How could jāgr- and vevī- be reanalyzed as simple stems if they were hardly ever used? Or were they common in speech but not in writing - at least until their reanalysis was complete? TRIPLY AWAKE IN SANSKRIT

I almost mentioned √jāgṛ 'awake' as an example of a reduplicated root reinterpreted as a simple root in "Tell Me Why They Speak Differently", but I decided not to because it has a complication absent from √cakṣ 'tell' and √jakṣ 'eat': a long first vowel in the reduplication. The original root is √gṛ, and if it conjugated like a regular class 3 -verb such as

√ṛ 'move' > i-yar-ti 3rd sg

√pṛ 'fill' (cognate to its translation) > 3rd sg pi-par-ti

√bhṛ 'bear' (cognate to its translation) > 3rd sg bi-bhar-ti

I would expect the reduplication in the present indicative to be ji-. But it is jā: e.g., 3rd sg jā-gar-ti (not *ji-gar-ti). Is the length of ā a remnant of a lost laryngeal *H: *gʷeH-gʷorH > jā-gar-? Or has the intensive stem been generalized? (8.16.19:57: The latter. See §676 of Whitney's Grammar. But why generalize the stem from a low-frequency form? I don't see any intensive forms of √kṛ 'do' in the Digital Corpus of Sanskrit. Bucknell's Sanskrit Manual does not even list forms of this type of intensive. [It does list footnotes of a second type in footnotes.])

In any case, it is strange how a CVCV-verb was reinterpreted as a simple verb when most simple verb roots have the shape (C)V(C)(C). (On the other hand, it is not surprising that √cakṣ 'tell' and √jakṣ 'eat' were reinterpreted as simple verbs since their reduplicated forms do fit the CVCC template for simple verb roots like rakṣ 'protect'.)

The title refers to how the already reduplicated verb can be subject to reduplication: e.g., the perfect 3rd sg jajāgāra with two copies of √gṛ (the earlier perfect was jāgāra with only one copy).

8.16.21:18: All this brought to mind old and new questions:

Old: Why do fifty verbs (i.e., class 3) reduplicate in the present indicative? They do not have any common semantic feature: e.g. (examples from Burrow 2001: 322): jigharti 'sprinkles', piparti 'fills', bibharti 'bears', jigāti 'goes', mimāti 'bellows', śiśāti 'sharpens', siṣakti 'cleaves', dadāti 'gives', dadhāti 'places', jahāti 'leaves', babhasti 'eats', vavartti 'turns', sasasti 'sleeps', saścati 'they accompany'. 'Sharpens' might be considered an inherently repetitive action, but 'gives' does not necessarily entail giving again and again.

On the other hand, the motivation for reduplication is transparent in the intensive "which expresses intensification or repetition (emphasis mine) of the sense expressed by the root (Burrow 2001: 355). And reduplication in past forms (the perfect and one type of aorist) makes me think the verb was partly repeated to confirm that something had happened. The perfect 3rd sg cakāra and the reduplicating aorist 3rd sg acīkarat 'did' remind me of English did do. (Of course, the weak parallel ends there, as did is used with all English verbs. Dadau 'gave' is not reminiscent of *gave give.)

New: Sanskrit verb reduplication is leftward. Does any language have rightward reduplication? Is there a tendency not to have a reduplicated root right before an inflectional affix? The only inflectional affix in Sanskrit that precedes reduplicated roots is the augment a- (e.g., in the reduplicating aorist 3rd sg acīkarat 'did' above); all other prefixes are derivational: e.g., in the reduplicating aorist 3rd pl ācakrire 'they drove near' < ā 'toward' + √kṛ 'make'. A FISHY VOWEL IN JAPHUG

I initially thought it was unusual that Japhug only has the vowel /y/ in /qaɟy/ 'fish' and its derivatives. I might expect such an anomalous vowel to be in a loanword, but I don't know of any external source of the word, and I wonder if it is cognate to Tangut

2zhu3 'fish' < *CV-CuH.

Then I realized that I knew of another case of a language with a vowel only in one root and its derivatives: Sanskrit syllabic is only in √kl̥p 'arrange'; its external cognate Latin corpus has -r-.

And I also realized that Japhug /y/ may be a fusion of /wi/ that became a new phoneme for some speakers.'Fish' for other speakers is /qaɟwi/ which I assume is the older form.

8.15.22:59: Unfortunately I do not know of any other cases of Japhug /wi/ corresponding to Tangut u. This is not surprising since /wi/ is an infrequent rhyme in Japhug.

Until now I would have been surprised if *wi became Tangut u, because

- Tangut wy is from *wi: e.g.,

1200 1khwy4 'dog', cognate to Japhug khɯna and Written Tibetan khyi < *kwi

- Tangut u is mostly from *o (with at least a couple of exceptions from *ə)

There is no Tangut syllable *zhwy3, so one could claim that *wi became u3 after zh, but that is unlikely since the syllable shwy3 does exist, and it is probable that vowels developed the same way after sh and zh which only differ in voicing.

Even if I don't understand the sound correspondences involved, I don't think it is a coincidence that Macro-rGyalrongic languages have z-words for 'fish'. The vowels of those words are all over the place. STEDT lists the Proto-Qiangic reconstruction *r-dzwa. I am skeptical because such a form might have become Tangut *tswar, not 2zhu3. But I am unable to propose an alternative.

Tangut fonts by Mojikyo.org
Tangut radical and Khitan fonts by Andrew West
Jurchen font by Jason Glavy
All other content copyright © 2002-2015 Amritavision