Thanks in part to its complex morphology, Sanskrit has many
word-internal consonant clusters. If OM and the syllabic theory of
sandhi are correct, then these should consist only of legal onsets,
possibly preceded by legal codas which undergo sandhi changes. And
this seems to be the case, if one assumes that a legal onset is one
that conforms to the general patterns seen at the beginning of words
or indicated by sandhi. By this criterion, I have identified the
following word-internal onsets, using a machine-readable copy of the
R.gveda (Ananthanarayana and
Lehmann 1970). Obstruents or h followed by a sonorant, s"ny,
s"m, s"y, s"r, s"rv, s"l,
s"v, s"vy, s.n., s.n.y,
s.n.v, s.n, s.m, s.y,
s.r, s.l, s.v, sn, sny,
sm, smy, sy, sr, sv,
svy, hn., hn, hny, hm,
hy, hr, hv, hvy. Voiceless stops
preceded by a sibilant (homorganic in the case of coronal stops):
s"c, s"cy, s"ch, s"p,
s.k, s.t., s.t.y, s.t.r,
s.t.ry, s.t.v, s.t.h, s.t.hy,
s.t, s.th, s.p, s.pr,
s.ph, sk, skr, st, str,
stv, sty, sth, sp, sph.
Voiceless unaspirated stops followed by a sibilant: ks.,
ks.n., ks.m, ks.y, ks.v,
ks, t.s, ts, tsn, tsm,
tsy, tsv, ps", ps, psny,
psv. Sonorant plus sonorant: l_h,
n.y, n.v, n.vy, ny, nv,
my, mr, ml, mv, vn.,
vn, vy, vr, vl. All of the extant
intervocalic clusters consist of either the above onset clusters, or
the same preceded by one of k, n", t., t,
n, p, m, r, l,
s, or the clusters n"k, rk, rt,
with the appropriate sandhi between the coda and the onset (recall
that the normal sandhi of r at low prosodic levels is no
change at all). These are exactly the word-final forms listed earlier
as inputs to sandhi. The only exception here is n"k in such
words as pr.n"kté `unites' (RV 10.95.9), whereas
k is deleted after a nasal word-finally
(pratyán"k -> pratyán", RV 9.80.3).
Perhaps the explanation is that n"k is legal word-internally,
since it falls within the same general pattern as rk and
rt (sonorant plus stop). On the other hand, n"kt is
a legal sandhi of n" plus t; possibly the original
k is basically deleted, but the text simply prefers to spell
the optional sandhi k when it agrees with a stop that is
etymologically justified (cf. imp. 3rd sing.
pr.'n.áktu, RV 1.84.1).
In all, there are over 400 different types of intervocalic
consonant clusters (including clusters found at the juncture of
compound words). The fact that all of them are explicable as onset
clusters or the sandhi output of permitted codas before onset clusters
is a strong indication that OM syllabification is operative in
Sanskrit.
The same result is obtained when one looks at specific
morphological operations. Some examples of the inflection of
j stems were given above, where it was seen that j
could be retained before vowels and sonorants, but not before
obstruents. Verb inflections also give information about the
syllabification of sequences of sibilant followed by stop. For
example, in the root dis"- `point', the s" is
converted to a stop when found in a coda, in such forms as
didid.d.hi (RV 2.32.6), where it converted to t.,
caused the imperative suffix -dhi to retroflex, and
assimilated. In the s-aorist ádiks.i (RV
5.43.9), it converted to k before s, causing
retroflection in the latter. But the s" remains as a
sibilant before t in the forms dídes.t.u (RV
7.40.2) and didis.t.a (RV 10.93.15; the change of
s"t to s.t. is a regular phonological change within
words). This seems to be a counterexample to the claim of Clements (1990:298-299, 317) that
core syllabification rules do not follow language-specific onset
templates but instead obey the Sonority Sequencing Principle, whereby
onsets must increase in sonority. In Clements's theory, sibilants and
stops are of equal sonority, and so s" should stay in a coda
before t or th, becoming a stop. It also calls into
question Steriade's finding (1982:312-321, followed by Cho 1990:70-71 and Vaux 1992:297) that sibilants in
Sanskrit do not syllabify with stops.
It might be objected that there are cases where a palatal converts
to a velar, even when it should be syllabifiable in an onset. For
example, ric- `leave' has a perfect participle
ririkvâ'm.sah. (RV 4.24.3); arc- `shine,
praise' has a nominalization arkah. `ray, hymn'. That these
are not coda-inspired depalatalizations can be seen from the fact that
the velars do not subsequently assimilate voice. The confusion arises
because the velars are actually the primary form. The palatals
historically arose from velars by sound change before front vocoids,
then spread analogically throughout the verb inflections, especially
the finite forms. So arka- can be straightforwardly derived
from ark- (or treated as an historical survival). Similarly,
the few cases in the verbal paradigms where c or j
are avoided before v or r are no doubt to be
considered survivals; presumably palatalization was held off longer in
this phonetically unfriendly environment.
In a similar vein, it has also been claimed that dn is not
a possible onset because the passive participle ending
-ná- causes preceding dental stops to turn into
n, as would be expected in sandhi (Vaux 1992:296): bhid- `split'
-> bhinnáh. (RV 1.32.8). But the same does not obtain
in words like udnáh. `waves' (RV 8.32.25). Perhaps
the participial paradigm is a survival from an earlier period when
tn and dn were not acceptable onsets. One can
easily imagine a diachronic state of flux between a system that
rejects these clusters for being homorganic and one that accepts them
by analogy with the other stop plus nasal sequences.