Professional Documents
Culture Documents
To cite this article: Punya Sloka Ray (1964) The Vowels of Sanskrit, <i>WORD</i>, 20:3,
353-359, DOI: 10.1080/00437956.1964.11659827
Article views: 66
I. Preamble.
If the spoken language is accessible, then of course the syllables, that is
the chest pulses, are as much audible as are the segmentals. Why is it
necessary then, in a phonological analysis, to state any rules of association
between syllable number or syllable demarcation or syllable configuration
on the one hand and the segmental phonemes on the other hand? Only two
assumptions can provide this necessity: that syllable structures are to be
recorded in writing, and that by phoneme classes rather than by special
unit phonemes. The first assumption is again ambiguous, unless writing for
the sake of teaching pronunciation is distinguished from that for the sake
of a subsequent grammatical analysis of the language. The second assump-
tion too has to be clarified: is a preference to be justified from conveniences
of penmanship or typography or from-exigencies of a subsequent gram-
matical analysis?
It is a priori quite plausible that some aspect or other of syllables should
not be contrastive for a particular language. There is at least one language
for which syllable aspects cannot be contrasted, since these are entirely
predictable. This is one variety of standard Bengali. 1 Prima facie the
following class of segmentals might be suspected of being vowels:
I The data, not the interpretation, are from Dr. Suhas Chatterjee. The author's own
speech differs in these crucial cases: [dai]'midwife' [dai] 'responsible' [ca~] 'he wants it'
[cae] 'in the tea' [mee] 'girl', for which Chatterjee has respectively [daj], [da~i], [ca~].
[ca~] and [meee]. For further discussion: Suhas Chatterjee, A Study of the Relationship
between Written and Colloquial Bengali. Hartford Seminary Dissertation (1962), pp. 25-
29.
353
354 PUNY A SLOKA RAY
2. Statement.
According to the model proposed here, Sanskrit did not have any
vowels, but only a set of semiconsonants that could occur at syllable peaks
as well as at syllable margins. These were fa i u r 1/ and very infrequently
also frpf.
THE VOWELS OF SANSKRIT 355
2 AraQya Sik~a. 52. Quoted by Siddheswar Varma, Critical Studies in the Phonetic
Observations of Indian Grammarians (Delhi, 1961), p. 82.
356 PUNY A SLOKA RAY
Rule 10. A nonsyllabic fif after /Q/ and before a syllabic /i/ was a
lengthening cum fronting cum heightening of the latter. Such /ii/ was [i].
Rule 11. A nonsyllabic fuf after /Q/ and before a syllabic /u/ was a
lengthening cum backing cum heightening of the latter. Such fuuf was [0].
Rule 12. A nonsyllabic fa/ was a lengthening cum lowering of a succeed-
ing syllabic fa i uf. Thus fai au aa/ were [e 6 a], and faai aau/ could be
either [;!e !!O] or [~~i ~~u] or [re o].3
Rule 13. /i/ was front, fuf back, frf retroflex, /1/ lateral, and fa/ residu-
ally unqualified, all voiced vocoids.
To sum up, three allophones each are postulated here for /i uf, of classes
A, Band C; two allophones are postulated for /a/, of classes A and C; and
two allophones each are postulat~d for /r 1/, of classes A and B; class A
consisting of syllabic phones, class B consisting of nonsyllabic nonport-
manteau phones, and class C consisting of nonsyllabic portmanteau-
included phones.
3. Examination.
How far does the stated model go in covering the actually available data?
A complete answer to this question must require a great deal of hunting
through the reference texts. One of the prime functions of such a model as
stated here is just to provide the motivation for selective data gathering.
However, a preliminary and tentative examination should be performed
here if only to show the superiority of the stated model over the traditional
analysis.
A search in Monier Williams' Sanskrit-English Dictionary revealed the
following contradictory evidence.
[toya] 'water'. The model would permit [taviya] in the form ftauiia/ to
contrast with ftauia/ [tavya] 'able' and ftauiiias/ [taviyas] 'abler'. But as the
existing record stands, [taviya] was a free variant of [tavya]. That is, [toya]
is not generable, though attested.
Seven words with initial [yi] such as [yiyak~at] etc. The model permits
only initial [i]. These do not, however, constitute a big problem, since for
some of the same roots alternative derivations with [i] are also recorded,
such as [iyak~amaQa]. One is a proper name, and that too very late.
Eight words with initial [vu] such as [vu<;lita] etc. The model permits
only initial [v]. Five of these have alternants with initial [bu], and one an
alternant with [vru]. The remaining two are recorded only in very late
texts.
3 The current Eastern pronunciation of these as [a ~aQ]cannot be accommodated in
the proposed model. But there are other features in the current Eastern pronunciatio~
that are admittedly different from the earliest attestation, for instance the merger be
tween /5/ and/~/.
THE VOWELS OF SANSKRIT 357
Eleven words with initial [ur], such as [uras] [urvara] etc. The model
permits only [vr] or [vr]. A minimal pair actually occurs in [uraQa] 'ram,
sheep' and [vraQa] 'wound, hole'.
At this point what is surprising is not that a few contradictory pieces of
evidence could be discovered, but that so very few could be discovered
from quite a large dictionary. The number and the nature of these counter
instances are far from constituting a crucial mass of evidence.
A stronger set of objections would arise out of an internal examination
of the proposed model. The rules are too many, one may argue, and the
phonetic basis for them is not clear. They appear merely to summarize the
recorded data in a clever manner. 4 Such objections prove the exact opposite
of the intended, granted the absence of a zero accent. If there were no zero
accent, the location of an accent should indicate simply and regularly the
position of a syllable peak. The system of rules as outlined above could
then be regarded as a redundant system, a secondary instrument reinforc-
ing the identification of syllable peaks. And a redundant system should be
expected to be both a little complicated and just short of total success, as
almost but not quite competitive to the distinctive system.
But do we need to write the pitch accents for Sanskrit? That we do has
never been doubted, and can be shown by a series of minimal pairs. Let us
write the high pitch or udtitta with the acute accent, the low pitch or
anudtitta with the grave accent, and the high falling pitch or smrita with the
circumflex accent. Then-
That such minimal pairs were rare at the time of the codification may be
taken as proof that Sanskrit was at the time on the verge of becoming a
non-tonal language.
Little is gained in this discussion by bringing in the metrics, which
utilized a contrast between heavy and light syllables, the latter being
identified in the traditional analysis by a so-called short vowel not followed
by a syllable final consonant. Although classical Chinese verse was built on
a contrast between just two kinds of syllable pitches, even and changing,
that fact alone cannot be taken as sufficient evidence that classical Chinese
was a language with two tones only; it is by no means necessary that a
metric category should be homogeneous from the phonological point of
view. Again, it is not difficult to re3:_d off the transcription recommended
here the metric category of each syllable; for the accent sign marks the
syllable peak, which, if not followed by a syllable final and preceded im-
mediately by pause, a consonant, or a semiconsonant other than fa/, itself
preceded by pause of a consonant, would mark a light syllable. Further,
syllable number may be counted in the recommended transcription quite
easily through counting the accent marks. In effect, the analysis proposed
here amounts to elevating each accent into the status of a vowel and con-
sidering the associated sequence of segmental phonemes as its diacritic.
The argument so far stands on the facts that the tones were distinctive
and that a redundant system was also available for syllable peak identifica-
tion. It is only an indirect confirmation of the theory proposed here that its
adoption would immensely simplify the segmental morphophonemics,
without either increasing or decreasing the complexity of the tonal mor-
phophonemics. The greater simplicity is shown most clearly in regard to
apophony and sandhi. An example of apophony would look like this:
fuid uaida uaaidika/ [vid veda vaidika].
And examples of sandhi would be rewritten as: fnai+ana=naiana/
[ne+ana=nayana] fgaai+aka=gaaiaka/ [gai+aka=gayaka] jprati+
aasannai = pratiaasannaif [prati + asanne = pratyasanne] fatha + u=a thaW
[atha+ u =a thO].
But of course a complete automatization of sandhi would not be
achieved yet. For instance, a rule would still be needed that at word final,
and some times also at word initial, a geminated semiconsonant must
become single. e.g. fmaha:i + iis:i = mahaisa/ [maha +!Sa= mahesa].
It is not a valid alternative to avoid the dismissal of all vowels by ad-
mitting a semivowel additional to the ones listed in the tradition, a non-
syllabic counterpart to syllabic fa/ that one may perhaps write as fxf. Such
a hypothesis will merely complicate the segmental morphophonemics
without any compensating advantage towards a simplification of the tonal
morphophonemics.
THE VOWELS OF SANSKRIT 359
A potentially fatal contradiction to the theory presented here arises out
of a different interpretation of the same data. According to the thesis 6 de-
fended by H. S. Ananthanarayana, proto-Sanskrit must have had a word-
tone system, that is, contrasted pitches only on the stressed syllables and
not on the unstressed syllables. This implies a zero accent, namely for the
unstressed syllables. This thesis is independently strengthened by two pre-
sumptions, the fact that the actually attested tonal morphophonemics of
Sanskrit are complex enough for us to suspect error in the record, and the
fact that those modern Indo-Aryan languages which have been found to
exhibit lexically significant pitch distinctions, such as Panjabi or Garhwali
or Chittagong Bengali, have word-tone systems.
The analysis presented here differs from such speculations in one funda-
mental respect. It is a strictly synchronic structuralization of the actually
attested data on earliest Sanskrit, not a diachronic internal reconstruction
of an unattested proto-Sanskrit. Consequently, there is no motivation here
to make assumptions such as that since the svarita occurred more fre-
quently on morpheme boundaries it must have been absent in prehistory,
or that tonal morphophonemics and segmental morphophonemics had
been both not only uncomplicated at some period of prehistory but so at
exactly the same period of prehistory.
4. Conclusion.
Sanskrit was at the time of its codification a syllable-tone language
without vowels. That this fact could be missed by such profound analysts
as Panini and others may perhaps be explained by these considerations.
Given the rather low functional load of tonal distinctions and the relative
complexity of morphophonemic and intonational perturbations of tone
structure, it was inviting to disregard tones. Given the practical urgency for
speech correction and the absence of tqe concept of a portmanteau phone
shared between successive phonemes, it was safer to hug the phonetic
ground closely. The price paid was a tremendous complication of the
segmental morphophonemics.
Department of Linguistics
University of Chicago
Chicago, Illinois