I have long been puzzled by the feminine stems of 'three' and 'four' in Sanskrit:

gloss masculine/neuter feminine
'three' tri- (weak), traya- (strong) tisra-
'four' catur- (weak), catvār- (strong) catasr-

I used to think that the -s- was inserted into the masculine/neuter stems:

tri- > ti-s-r-a- (with metathesis of i?)

catvār- > cata-s-r-

Years later, I read in Beekes (1995: 212) that

The [Proto-Indo-European] element -s(o)r- [in 'three' and four'] also perhaps appears in Hitt[ite] hassu-sara- 'queen' (from hassu- 'king')

but that didn't explain why tri- became ti- or catur- became cata-.

Today I realized something that eluded me for twenty years: the first -r- was dropped to avoid two -r-s in the same word:

Pre-Proto-Indo-European *tri-ser-* > Proto-Indo-European *tiser- > Skt tisras-

Proto-Indo-European *kʷetur-sr- > Skt catasr-

This cannot be the whole story, though. It still doesn't account for the -a- in catasr- instead of the expected *catusr-. The unusual -a- cannot be a Sanskrit innovation because it is also in Avestan cataŋr- 'four' (f.; -ŋ- is from *-s-). Was *ur irregularly reduced to a syllabic *ṛ that became a to avoid two rs in the same cluster?

*Beekes (1995: 212) reconstructed Proto-Indo-European *-ser- in 'three' but *-sor- as the suffix that became Hittite -sara-.

2.10.0:17: Skjærvø (2003: 207) wrote:

The [Avestan] element -šr-/-ŋr- is an ancient suffix found in the fem. forms of the numerals "3" and "4" in several Indo-European languages. It may be related to strī- (< *srī-) "woman" and -ŋhar- in xvaŋhar- "sister" (if originally *xva-har- "one's own woman"?)

Avestan xvaŋhar- goes back to Proto-Indo-European *swesor-, so 'one's own woman' is not an Avestan innovation.

Wikipedia's entry for Proto-Indo-European *swésōr (the nominative singular with a long vowel in the second syllable) has a different etymology (I have rewritten the laryngeals using Beekes' [1995] values):

Possibly a compound of reflexive pronoun *swé (“self”) and *ʔésʕr̥ (“blood”), so literally “woman of one's own kin group” in an exogamous society.

Is there any evidence for laryngeals in 'sister', or were they lost without a trace? UN-SHYR-TAIN EVIDENCE FOR RETROFLEXION IN TANGUT PERIOD NORTHWESTERN CHINESE

While checking the entry for

2gii '(to) hope'

in Li Fanwen (2008: 770), I noticed that the tangraph in the adjacent entry

1ʂɨəʳ 'the surname Shyr'

was used as a transcription character for Chinese syllables that are now pronounced shi [ʂr̩] in modern Mandarin: 十什失實室涉. For years I assumed that these syllables would have been pronounced *ʃi in the northwestern Chinese dialect known to the Tangut. I thought that did not become retroflex ʂ until the last few centuries due to a chain shift:

Stage 1: > ʂ

Stage 2: *s > ɕ (not quite ʃ, but close)

However, if the Tangut transcribed such syllables with a retroflex vowel according to most current reconstructions*, that might be evidence for reconstructing them with a retroflex initial and possibly even a retroflex syllabic *r̩. I need to dig deeper before drawing any firmer conclusions.

2.6.2:27: In Gong's 1991 study of transcriptions of Chinese in the Tangut translation of 類林 The Forest of Categories, tangraphs read with retroflex vowels (i.e., with rhymes 77-103) are completely avoided with four exceptions:

1ʂɨəʳ (rhyme 92): 什涉寔 ?*ʂr̩ (see above)

1thwəʳ (rhyme 90): 盾 *thwə̃ < Middle Chinese *don

1mɔʳ (rhyme 96): 邈 *mɔ < Old Chinese *mrakʷ

2lwəʳ (rhyme 90): 論 *lwə̃ < Middle Chinese *lon(h)

There are no Tangut rhymes ending in -ə̃. Was -əʳ an attempt to approximate Chinese *-ə̃?

Is the correspondence between -ɔʳ and Old Chinese medial *-r- coincidental? I doubt that *-r- had left a trace as retroflexion over a millennium later in the Chinese dialect known to the Tangut.

The general avoidance of retroflex vowel transcription tangraphs indicates that Chinese dialect lacked retroflex vowels and that Tangut had not merged retroflex and plain vowels. (If such a merger had occurred, tangraphs with retroflex as well as plain vowels would frequently transcribe Chinese syllables.)

*2.26.2:29: Gong (1997) and Arakawa (1999) also reconstructed a retroflex vowel in 'the surname Shyr'. However, Sofronov's new 2012 reconstruction still completely lacks retroflex vowels:

Tangraph Gong Arakawa Sofronov This site
1śjɨr 1shI:r 1-ə̂, 1-jə, 1-ɪ? 1ʂɨəʳ

I do not know how Sofronov would reconstruct the initial.

Sofronov reconstructed rhyme 92 three different ways. I do not know which reconstruction he would choose for this tangraph.

Tangut fonts by Mojikyo.org
Tangut radical and Khitan fonts by Andrew West
Jurchen font by Jason Glavy
All other content copyright © 2002-2013 Amritavision