Retroflexion in IA (was Re: Munda in Early NW India)

Vidhyanath Rao rao.3 at
Sat Apr 21 11:15:41 UTC 2001

"petegray" <petegray at> wrote:

> The distribution of the retroflex series is severely limited in the older
> language [...] (list deleted)

I am not sure how to understand `retroflex': Isn't that supposed to include
the retroflex sibilant (.s) and the nasal (.n) in addition to the non-nasal
stops? If so, here are the regular processes that led to them in Sanskrit:

ruki-s -> .s
n that follows r or .s, even with vowels, labials, y, v intervening -> .n
t/th that immediately follows .s -> .t/.th
s' (from PIE *k') +t(h) -> .s.t(h)
h from PIE gh' + t/dh -> .dh (with h -> zero, preceding vowel lengthened.)
ruki s + d/dh -> .d/.dh (with s -> zero, preceding vowel lengthened if not
long by position)

Only the last two have retroflexing environments that go to zero in
Sanskrit. However, they are rare in practive: the penultimate can occur
only with ani.t roots ending in *gh' (only a handful in Sans); the standard
examples of the last are isolated words such as ni:.da (< *ni-sd-o, nest)
and aorist 2nd pl mid, an infrequent category.

Other processes mentioned by Peter are irregular (may be dialect mixture,
may be prakritisms, ...). An oddbal is Fortunatov's law: l+t -> .t
[but there is RV galda] Its status seems to have gone back and forth.
I don't know where it stands now.

In Prakrits, -rt- sometimes becomes -.t.t- (and vocalic r+t can become
a/i/u+.t). But this is dialect dependent and it seems to vary within other
families as well. Similar developments seem to have occurred in Norwegian,
and this is why Hock (loc. cited by Oberoi) objects to attributing this to

Turning now to substratum explanations:

See now T.A. Hall in Lingua 102 (1997) pp. 203-221: Contrast of
two laminal shibilants, one alveopalatal and one palato-alveolar (his terms),
is rare to non-existent and when sound changes lead to such a contrast,
a shift to a more stable system will follow. One common repair
mechanism is to make one sound apical. This is the explanation for IA
retroflex s.

It is interesting to see the evolution of the view on .s: Once upon a time,
it was considered to be limited to South Asia. Masica, "Genesis of a
linguistic area" lists only Chinese in addition. Hall makes gives a large
number of examples of .s vs 's like contrast, and deems it a `near-universal'.

Also, all the three sibilants fall together as s in MIA. If .s is due to
substratum influence, this is weird. The substratum induces a change, then
conveniently disappers allowing Prakrits (which one would to show more
substratum influence) to reverse the process.

Note also that -rt- becomes some sort of (apical) shibilant in (Late?)
Avestan: This suggests that -rt- was already retracted in PI-Ir and that the
development of -rt- to -.t.t- in MIA need not invoke the deux ex machina
of substratum influence.


Regarding phonemic status of retroflexes: First of all, note that s/'s/.s
contrast of Sanskrit is exactly parallel to h/s/s^ of Avestan. Nobody, to
my knowledge, worries whether Avestan has three phonemes here or
two. Why then should we worry about Sanskrit?

Secondly, I don't know how phonemes are defined in inflected languages:
Do we use inflected words or dictionary forms? Theory seems to dictate
the former, but people seem to use the latter to find contrasts. This is a
more serious problem in Sanskrit as the dictionary form is the stem, which
is, synchronically, a grammatical fiction, seldom uttered in actual speech.
An additional problem is sandhi: If two words are homophones in some
contexts due to sandhi, but not in all contexts, do they count for contrast?

Due to the fact that the conditioning environments for retroflexion do not
generally go to zero in Sanskrit, it is hard to find inherited contrasting
words.  If we add the accent, it gets even harder: For example, u:Dha's (masc.
nom sing of PPP of vah) and u:'dhas (udder, replacement for u:dhar/n, an r/n
heteroclitic) are not homophones if we count the accent.  But, if they are all
allophones, then .sa.tbhis (with six) and satbhis (from neut pres part of as <
*Hes) would be homophones! [BTW, note that Avestan has s^ for six, paralleling
.sa.t. So the word for six already identical to ruki-s in PI-Ir.] There is also
i.da:' vs ida:', but the etymology of i.da: is is obscure.


Just when I finished writing the above, I got the message in which
Pat Ryan wrote:

> [...] Modifying some apicals to a retroflex articulation helped disambiguate
> somewhat since they indicated a Nostratic root with [o] not [a] or [e];

Huh? You mean contrast of `Nostratic [o] vs Nostratic [a]/[e]' (whatever they
are) was preserved till the split of IA so the speakers could decide when to
change `apicals' to retroflex? Or do you mean PIE *o vs PIE *a/*e
(that is demonstrably false).



More information about the Indo-european mailing list