Вы находитесь на странице: 1из 35

Lingua 82 (1990) 277-31 ~.

North-t-lolland 277

HINDI gCHWA DELETION

Pramod Kumar P A N D E Y '~


Department of Linguistics. South Gujmat Umver~ity, Surat 395 007. lndio

Received February 1990; revised version September 1990

The purpose of this paper is to p~esent a non-linear account of Schwa Deletion ~a


Hindi. Ohala (1987) has argued that her linear analysis (Ohala (1983)) is superioE to
D'Souza's i1985) non-linear analysis, which yields obserqationally incorrect outputs. I try
to show that while D'Souza's ru~e of foot formation, on which Schwa ,Deletion is based,
and her theoretical assumptions need t,~ he_ ,'evised, all the facts relating to the pLc,,ume-
non have an elegant explanation within the prosodic phonological framework. The
various irregularities in the al~plicafion of the Schwa Deletion rule, which must be
stipulated in Ohala's account, ar~: predict~-~ by the independently motivated principles an~
assumptions of prosodic and iexical phonology. In ti~e end, data pertaining to an'~-
Antigemination (Odden (1988)) and related phenomena ace taken up to provide evidence
in favour of (possibly a weaker version of) the OCP.

1. |ntroducfion

Schwa Deletion is one of the most widely discussed processes in Hindi


phonology (Pray (1970), Narang and Becket (1971), Kalra (1976), Ohala
(1974, 1977, 1983, 1987), Shrivastava (1979), D'Souza (1985)). While there is
a general agreement on the factors influencing it, the precise statement of the
process is a controversial matter. The controversy has recently centered
around the linear versus non-linear cheract~r of the phenomenon. Ohala
(1987) contends that the nonolinear description of the phenomenon in
D'Souza (1985) yields observationaUy incorrect outputs, and is in no way
superior to her linear account (e.g., 1983). Her rule of Scbwa Deletion
(t983: 139-140) is stated as follows:

* In the presentation of this paper, l have benefitted from comments and sugestions from many
individuals, in particular, from D.M. Joshi and K.G Vijaykrishnan, and, espe~lally, from two
anonymous reviewers. I wot:Id ilke ~o e,xpress my gratitude to them.

0024-3d41/91/$03.50 (~) 1991 - - Ehevier Science Publishers g.V. ~North-Holiand)


278 P.K. Pander / Hindi schwa deletion

(!) Schwa Deletioz (linear version)

+ causual speechj
[ + normal tempo]
Conditio~l I. There may be no morpheme boundary in the environment
to the l::ft.
Condtion2. The output of the rule will not violate the sequential
,,ons,traints of Hindi.
C, mvention. The rule applies from right to left.

Oh:3la claims that the generalizations relating to the phenomenon are best
captured in the form of (I). The crux of her claim is that at least some of the
sequential constraints which function as output conditions must be stated in
liaear terms in order to yield correct outputs.
My purpose in th~,~ paper is to show that while D'Souza's prosodic rules
an_d th;:oretical ~.~sumptions about we!!-formedness constraints on segment
sequences as well as about rule application need to be revised, the linear
statement in (1) can be better formulated in non-linear terms"

(2) Schwa Deletion (non-linear version)

Z
/'\
o----~¢/Rs Rw
_1_
The non-linear rule 12) is devoid of all the conditions in (I), which are
predicted by the operational principles of" prosodic phonology. Besides,
among other gains of simplicity and descriptive adequacy discussed in
section 4, (2) is strictly local. I assure, that alternative non-linear expressions
of (2) are possible, as, for examp!e, ~it/qn the framework of moraic phono-
logy (e.g., Hyman (1985), Clemems (1988~). Nothing crucial however hinges
on a choice between these alternative expre,'sions.
The paper is organized as follows: section 2 pres,mts all the facts concern-
ing the phenomenon. Section 3 is devoted to, a non-linear account of the
phenomenon, with a brief discussion of the relevant theoretical underpinnings
P.K. Pandey / Hindi schwa deletion 279

~.,uc,,~ to tile interpret.::!_--~ -¢ (?)_ Section 4 presents a comparative evalua-


tion of the present analysis vis-~.-vis the analyses presented in Oh:da (1983),
and in D'Souza (1985). Section5 complements the investigation with
further support from prefixes and cases of OCP enforcement, wni,:h enhance
the plausibility of the explanation offered in the precedin?.: sections.
Section 6 presents the summary and implications of the analysis l,resented in
the paper.

2. Facts relating to Schwa Deletion

As is apparent from (i) and the alternating fonns in ~3), a is deleted when
flanked by VC{ on the !eft and CV on the right.

J g~'rl~C K.,I I U ~ ~ FII" UKf4,,~" I,.I I U~'~"

(a) pakaR catch pakRaa caught


macd prance maclaa pranced
laT:Jk hang (lntr.) laTkaa hanged (Intr.)
n~,kal copy naklii counterfeit
nazar sight nazro'b sights
naaTak drama naaxTkiiy dramatic
nagar ity naagrik citizen
saral easy, sim~le sarlaa (name of a female)
baahar outside baahrii external
(b) j~r,.gal forest jarag;ii pertaining to
forest, wild
bistar bedding bist(a)raa bedding
dorian view, pkiloaophy ~aar"~:'~k
I v " " ~
philosopher
kambal blanket kamb(~)[o~ blankets
marijarii (a name) mojajrii (a name)
oSTom eight(h) agTmii eighth day of
a fortnight

In (3b), the deletion of a results in certain three-consonant clusters which,


as compared to those in (Sb) bdow, are pm mitred in the language, e.g., -rjgl,
str, rgn, mbl, j~jr, etc.
(4) and (5) provide the negative contexts in which a does not delete. (4)
illustrates the blocking effects of the environmental conditions informing (!).
280 P.K. Pandey / Hindi schwa dcletion

These include more or less than one consonant m the following position (4b
and c), and more than two consonants in the preceding position (4d). (5a)
illustrates the effects of the negative Condition !, i.e., a preceding morpheme
boundary. (5b) contains examples with surrounding consonants which cannot
occur adjacently, e.g., -dt-, -tlb-, etc., as enjoined by Condition 2 of rule
(1).

(4a) pokoR pakoRtaa 'catch (i & 2sg.)'


noka: nakaicii 'one who copies'
naaTok naaTakkaar 'dramatist'
sarol soraltaa 'simplicity'
(4b) poiorlg 'cot' polong6"o 'cots'
ghom~ND 'pride' ghomaNDii 'proud'
sombondh 'relatioff sombondhii 'relative (N)'
aanond ~pleasure' aanondit 'pleased'
(4c) suor6"b 'pigs' * sur6'b
boRhoii 'carpenter' * boRhii
b~mboii 'Bombay' *bombii
l~khnouu 'Lucknow' *iakhnuu
(4d) !~ksmoN (a name) loksmoN6"o

(5a) a+somoy ~ul:timely' osomoy


compaa+kolii (a name) campaakalii
kaarii+g0r+ii 'craftsmanship' kaariigorii
kolaa+w0t+ii (a name) k~laaw~i
¢5b) audit 'habit" aadot~ *aadt~ 'habits'
po~oRaa ~side of a balance' * polRaa
m~tlob 'self interest' motlobii * m~tlbii ~selfish'
kosr~t 'exercise' k~sr~tii * k~srtii 'muscular'
mastak 'forehead' m~stok8"o *mzstk6-oo Yoreheads'
karw~T "::!~eside (in the k~rwoT~ * karwT~ 'the sides'
lying position)'
ojgor 'pytLon' ~jg~rii *ojg~,'ii 'pertainh~g
to py~" ms'
(1) also incorporates Lhe variables of normal tempo and casual style. In
allegro tempo, the rure may apply overlooking bomb the conditions noted
above, as for example,
P, K. Pandey / Hindi schwa deletion 281

(6a) kalaa+wot+ii kalaawtii


kaarii+gar+ii kaariigrii
o+samoy asmay
(6b) aadat+~ aatt~

On the other side, in a formal style, the rule may fail to apply:

(7) janataa 'public"


kamalaa (a name)

O~ala notes at least one non-trivial case where (1) fails to apply even thougt,
aH the conditions in it are met: it includes forms in which d is followe, fy
suffixes such as -iyaa and -i:m. Her discussion of this aspect of the p, oblen-, is
based on the experimental evidence for the psychological reality of ,he 9-
deletion rut.e:

(8) keesariyaa "saffron'


jhoopaRiyaa 'hut (dimin.)'
sanaatanizm ~

Ohala mentions only one example with the English suffix -izm, ard conjectures
the possibility of such suffixes in the language before which ~ fails to elide.
Some more examples where d is retained are the following:

(9) kamblinii 'lotus (fern.)'


khompbriyaa (a last name)
pipbriyaa (a place name)
pilbkhuaa (a place name)

Note that all the forms (excepting sanaatanizm) in (1~)and ('~) have a common
structure following a, namely, VCVV. If we leawz the nonce-word sanaata-
nizm out of discussion, the generalization about the forms in (8) and (9) needs
to be explained. Oha]a, adhering to the segmer, tal approa,:h, considers them
as cases of rule blocking: the affixes attached t,~ them block the application of
the a-deletion rule.
Note, however, that the retained anterJenult in (8) and (9) is regularly
stressed (cf. Kelkar (~q68), Pandey (1989)). There is a definite relation

l A nonce-form made up of [san,~:ton], a syneny~n for Hinduism, and the English sumn "ism'.
282 P.K. Pandey / Hindi schwa deletion

between stress and d-deletion that (1) ignores. The relation can be expressed
as follows"

(10) Only, but no t. all, unstressed o's are deleted.

It is the complicated nature of the conditions impinging on the non-


deletion of unstressed o that perhaps has led Ohala to ignore the relation
between stress and ~-deletion. Assuming for now that (10) is correct, and thus
its entailment thai r e s s e d a is never dropped, (8) and (9) show the regular
occurrence of ,~. ~o, then the statement in (1) must be amiss somewhat,
since it lacks a ge~ alization.
Having built ~ y case for yet arother analysis of a-deletion, I now wish
to show that the ~ e s s is based on the hierarchical representation of words.
I ,r .....
~ ~,,~t.~. ~,ish to ~ _. ~,~*'~"':--
that the ...... ,,u~ of,~ in apparently deletable positions
is on account of ~e factors of weii-formedness constraints on melodic
sequences and sy~! !e s~ructures, which prevent the adjacent occarrence of
certain segments, aa~d of the stratal organization of the foot construction rule
preceding a-deletion, which assigns different feet to prefixes.
It is useful to have in focus, at this point, the principal facts relating to a-
deletion, as exhaustively presented in Ohala (1983):

(ll) In order for ,~ to .~1~,~


(a) (i) o must be in the penalt position in a sequence of three vowels;
(ii) a must be followed by one and not more than one consonant;
(iii) the rule of d-deletion must apply from right to !eft;
(b) (~) ~here must not be a cluster of more than two consonants intervening
the first two vowels;
(ii) the deletion of a must not violate the sequential censtraints in the
language;
(c) (i) there mu~i ~ot intervene a morpheme boundary between a and the
preceding ~wel;
(ii) '.he const~-~ints in (b) and (c) do not hold for u~terances in allegro
tempo in casual style.

In the following sections, it will be shown that the facts in (a) in (1 l) relate to
the n~,v-!;~ear representation of forms, in (b) to prosodic wdl-formedness
constraints in the language, and in (c) (and (b) to the stratal and modular
P.K. Pandey / Hindi schwa deletion 283

applications of the rule, subject to the principles aad assu~,lptions of lexlcal


phonology.

3. A non-linear analysis

The non-linear rule of Schwa Deletion that was stated in (2) is reproduced
below:

(12) Schwa Deletion Rule

v,

, O/Rs Rw
2~
As mentioned earlier, the rule is expected to predict the correct output in (3).
and in (8)-(9). With the conditions in (i), prohibiting a-deletion in a large
number of cases, absent from it, (12) (= (2)) begs many questions regarding
the basis of its form, and its principled application in the language. It is in
order to answer such relevant questions that I shall now briefly present the
empirical and theoretical assumptions underlying (2), and the theoretical
principles governing its application so that all and only correct outputs
result.

3.1. Theoretical assumptions

3.1.1. Prosodic phonology


Metrical phonology (Halle and Vergnaud (1978), Kiparsky (1979),
McCarthy (1979), Selkirk (1980, 1984), Hayes (1981)), non-concatenative
morphology (McCarthy (1979, 1981)), and C-V phonology (Clements and
Keyser (1983), Clements (1985)), an offshoot of autosegmental phonology
(Goldsmith (1976)), recognize a hierarchy of prosodic units such as the
intonational phrase, the phonological phrase, the word, the syllable, the
onset, the rime, C-V elements, and melodic segments, and the hierarchically
ordered features composing melodies (e.g., Clements (1985)). The hier-
archical structure of words can be represented thus (see also McCarthy
(1984)):
284 P.K. Pandey / Hindi schwa deletion

(13) Word Wd

Foot Z

Syllable ~ 6 a a
I /k /N
Onset-Rime R O R O R O R
I I I I I I
CV-tier V C V C V V C V
I I I ! I ! I I
Melody, and i m i t e i ~ n
its compositional features [!l [i1 [i] [i] [i ] [il [i]

C-V phonology considers the C-V elements, which represent the syllabicity
of melodic units, to be directly associated with the o node, that is, it excludes
the internal hierarchical make up of syllables from representation. However,
whereas the C-V phonological representation may be adequate in most cases,
the hierarchical structure of syllables may be needed in the explanation of
phenomena which are sensitive to the quantity of syllables represented in
terms of rimes as branching (i.e., having two morae), or non-branching (i.e.,
having one mora). I therefore assume that phonological rules and constraints
may refer" to either forms of representation.
The vast literature on prosodic phonology has amassed sufficient evidence
(see e.g., Ito (1986)) to show that phonological rules which depend on
prosodic structure and prosodic structure formation rules are strictly
LOCAL, and DIRECTIONAL, i.e., they proceed from right to left or left to
right. In addition, all prosodic units, from melody upwards, are PROSODI-
CALLY LICENSED, i.e., constitute higher prosodic structure, subject to
extraprosodieity, which ignores peripheral elements of well-defined domains
(see Hayes (! 981)) at the surface level. Melodies which are not syllabified do
not appear phonetically.
The association lines, which connect the prosodic categories, are subject to
two main constraints given below.

(14) Prohibition Against Crossing Lines (Goldsmith (1976))


Association lines may not cross.
(~ 5) Linking°Constraint (Hayes (1986a))
Association lines in structural descriptions are interpreted as exhaustive.
.~.K. Pandey / Hindi schwa deletion 285

The Linking Constraint ensures that a rule may not apply more or less
generally than is required by its structural description. Thus if a rule applies
to ,nelodies that are singly linked w~ia the C eiemeii;.s, as in the following
rule, which spirantizes a non-geminate in the coda position, then it may not
apply to those melodies which are doubly linked.

C
(16) k ,

Thus the geminate k will not undergo the rule as it is doubly linked with the
C elements, e.g., C C
\/
k
This property of long segments to resist the application of a rule that a-priori
would be expected to apply to them is known as 'inalterability', and is
explained by Hayes by invoking the Linking Constraint. The constraint in its
original conception applied to structural descrptions containing aatoseg-
ments with multiple linking, but, as we shall see in 3.3.3, it has wider
applications.

3.1.2. Lexical phonology


The theory further assumes that the well-formedness conditions of prosodic
categories apply subject to the modular constraints of lexical phonology
(Kiparsky (19,82, 1985); Mohanan (1982, 1986)). One of the principles of
lexical phonology is Structure Preservation, which, in its strong form
(Kiparsky (1982, !985)), requires that well-formedness constraints in the
lexicon should not be violated in the representations of forms within the
lexicon. Postlkexical rules which are not structure preserving may, however.
violate these ,constraints.
Having broached these theoretical preliminaries, we now proceed to briefly
discuss the prosodic phonological facts of Hindi words, which, along with the
theoretical principles, have a bearing on the application of the Schwa
Deletion rule.

3.2. Lingz~dsticfacts of Hindi

In this section we are concerned with the structural facts relating to


melodies, syllables, and feet in Hindi words. The theoretical assumptions with
regard to these structures are not unanimousl3, agreed upon. I shall clarify the
286 P.K. Pandey / Hindi schwa deletion

main assum0tions underlying the present analysis as we proceed. I should


mention here that I assume that well-formedness constraints regarding the
sequential occurrence of melodic segments are expressed not only in terms of
syllabic structures, but also in terms of melodic sequences, subject to the
principle of locality. That is, the occurrence of a me!ody may be determined
by what precedes or follows it.

3.2.1. Alphabet
The lexical melodic alphabet of Hindi (which, in the case of consonants, is
larger than the underlying alphabet) is as follows (see also Pandey (1989)).
Vowels: i\ii, ee, ~e a~, ~\aa, a~ a~, oo, u~uu. Consonants: 2 p ph b bh t th d dh
T Th D D~ k kh g gh; c ch j jh; f s z g h; m n Nj~ rj; R Rh; I r; w y.

Constraints
Some of the melodic constraints in Hindi are as follows:

0 7 ) 3 (i) DhC (C = consonant)


(ii) *IR
(iii) *C C

placeJ L ~ placeJ
i.e., homorganic conson'mts with different voice features
may not occur adjacently.
(iv) *C C
[+ spr. gl. ] [+ spr. gl. ]
i.e., aspirated cons~.~ants may not occur adjacently.

3.2.2. Syllabication
Syllables form the crux of the presodic s'~ructures ol words. They critically
determine the concatenative arrangement of C-V and melodic elements, and
constitute the basis of foot structure assignment in words. Prosodic phonolo-
gists, however, do not hold a unified view about syllabification in grammars.

2 The dental and alveolar consonants, and /s/ and retroflex consonants often behave as
homorganics. /r/ has a special status: it may be homorga~ic with dcgtal, alveolar, and
postalveolar/retroflex consonants.
s Constraints (iii) and (iv) may be treated as cost-free language-specific implementations of the
universal OCP. For more on this, see section 5.2.
P.K. Pandey / Hindi schwa deletion 287

T_,,, this section I shall be concerned with only ,k~,,,.most important of those
issues which have a bearing on Hindi facts.
I shall subscribe to the view (Kiparsky (1979)) that syllabification takes
place in the root cycle of words. I shall further assume that, in Hindi, forms
are resyllabified both when skeletal elements are desyUabified as a result of
prosodic structure rules, and when morphological concatenation takes place.
In other words, syllabification is non-cyclic in Hindi.

3.2.2.1. Templates. Two approaches to syllabification have b ~ n proposed


in the literature, known as the rule-based approach (e.g., Kahn (1976), Hayes
(! 98 !), Steriade (1982)), and the template-based approach (e.g., Halle g~---I lillU

Vergnaud (1978), McCarthy (1979), Selkirk (1980), Ito (1986)). I follow the
latter approach and its assumption that syllabification is governed by both
universal as well as lar~uage-specific well-formedness conditions, subject to
the pfnciple of ie:~!cal phonology.
.As. mentioned
. . . . . . in. 3. !,. I. ,~,,m,,
. . . th,,t
•. , ,,,.,,~,,,a~,. o,.... , .... koo°a
v . , , o , , ' . . . ~ o,.~.,.,,..,, u,,~,.u ,..,!
u , , .. o. . ,. ,., ., , j refer
eithe: to the hierarchical structure of syllables or to the C-V phonological
structure~ or possibly to some other structure, such as the morale interpreta-
tion of syllables (Hyman (1985), McCarthy and Prince (1986)). Hindi requires
the specification of both the C-V phonological as well as the hierarchical
templatic structures.

(18) Hindi syllabic template: o[C V (V) C]

(18). is the
. maximal
. . core . syllable o,,r~;.~-,
~.... to f,Jrther extension. Minimally,
Hindi permits tr to dominate a singie vowel non-finaUy.
The hierarchical templates in Hindi are three, standing for three degrees of
syllable weight to which foot structure formation rules are sensitive:

(~.9) (a) Light (b) Heavy (c) Extraheavy

O R O R

0 R Rw
J'x
(

3.2.2.2. Well-formedness constraints. These are both universal as well as


language-specific.
288 P.K. Pandey / Hindi schwa deletion

Universal constraints
The fo!towing are generally accepted as universal constraints on syllabifica-
tion.

Sonority Sequencing Principle


The Sonority Sequencing Principle (Kiparsky (1979), Steriade (1982))
reguires a syllable to have one syllable peak, flanked by progressively less
sonorous melodie~ stretching up to the syllable margins.

Universai Core Syllable Condition (lto (1986))

(20) If C V

then o

The Onset First Principle (Ciements and Keyser (1983), also Selkirk (1982: 359))

(21a) Syllable-initial consonants are maximized to the extent consistent


with the syllable structure conditions of the language in question.
(21b) Subsequently, syllable-final consonants are maximized to the extent
consistent with the syllable structure conditions of the language in
question.

Language-specific constraints

(22) (i) [C C (morpheme initial)

nasj
i.e., morpheme-initially, the onset permits an obstruent + non-
nasal sonorant: Cr, Cl, Cw, Cy.
di) IC C (morpheme internaD
°[ -l sen] I
+ son
i.e., morpheme-internally, the onset permits an obstruent + so-
norant cluster: Cr~ Ci, Cn, etc.
P.K. Pandey / Hindi schwa deletion 289

(iii) (a) * (b) *

C C C C C C
i.e., in a three consonantal sequence neither the first two conso-
nants nor the last two uniquely belong to a syllable.
(iv) 6 a

C C
[a place] [a place]
i.e., a consonant may be ambisyllabic if it is homorganic with
the preceding consonant.

The effect of (iii) and (iv), in addition to that of (ii), is that only such three-
consonant ,.~u,,~,,
.!. ~ . . a.r e. p.m t n l:.t t e u-~ in which the middle consonant is ambisyiia-
bic, i.e., associated with two syllables, and in which the onset of the following
syllable has a sonorant as the second consonant. The condition for the
ambisyllabicity of a consonant is that it must be homorga~ic with the
consonant of the preceding syllable.
Recall that a-deletion takes place only when consonant sequences such as
ndr, 0gi, STm, etc. listed in (3), result; not otherwise, e.g., t!b, stk, jgr, etc.,
listed in (5). 4 A moment's reflection reveals the systematic difference between
the clusters of the two grc~upS: those of (5), but not of (3), violate the
constraints in Off) and (iv).
Constraint (iii) has many counterexamples in borrowed words from Sans-
krit, and especially, English, e.g., [inspekTor] 'inspector'. I am not prepared to
go into all the facts of weil-formedness constraints in Hindi-Urdu, but only
those that are general, and relevant to the problem at hand.

3.2.3. Feet
Foot forms in Hindi are of some complexity. I assume that they can be
explicated in the tree structure with the following revisions in the standard
tree theoretical assumptions (Hayes (1981)) proposed in Pandey (1999: 59-
60):
(23) Recessive node may not branch only in Quantity-sensitive systems.
(24) Recessive node may branch if the dominant node branches in Quan-
tity and Rhythm-scr, sifive systems.
4 For a detailed discussion see Ohala (1983: 128-130).
290 P.K. Pandey / Hindi schwa deletion

(25) Ternary feet are a result of either extrametricality, which ignores


endmost syllables, or QR-sensitivity, which prevents the endmost rime
from conjugating with its adjacent rime in a foot.

These revisions are proposed on account of the disadvantages of the


standard tree theory in describing a system such as Hindi, which is sensitive
to both quantity and rhythm, a QR-sensitive system, and not merely to
quantity. As a system sensitive to rhythm it has bounded s w feet, which
predict recurrent accents; its .~ensitivity to quantity permits it to have both
weak heavy syllables, if the strong syllable is heavy, as well as strong light
syllables, if the weak syllable is light (but not weak heavy syllables, if the
strong syllable is light). This character of Hindi feet is apparent from the rule
of core accentual patterns dis,cussed below.

3.2.3.1. Core patterns. The following rule predicts the core accentual pat-
terns in Hindi:

(26) From the trigger (i.e., word edge or accent o~l the right),
(a) accent the final if it it~ extraheavy, if not
(b) accent the penult if it is heavy or extraheavy, or if it is light and
flanked by light ,.yllables, or if it is the leftmost syllable; if not
(c) accent the antepenult.

There is no fixed relative prominence relation between feet. Thus whardhat-


~ whardhat 'event'. Some examples for (26) ar~: given below. ( ~ ) indicates
free variation in the relatiw: prominence between feet.

(27) Words G oss Words Gloss


(a) (i) sifftar siiar sombSndh relation
infiamd~ar j'-..) (a, last name)
(b) (i) daar6ogaa sub-inspector ~riiw~ast~w (a last name)
o f police
(ii) ~tithi ~uest 5num~,ti (~) permission
(iii) k~m~l (~ k~mSl) lotus s~maac~.ar (~) news
(c) 6) mShilaa Iady pip+riyaa (a place name)
(ii) sb0g~ti ~:ompany w~nSsp~ti vegetation
P.K. Pandey / Hindi schwa deletion 291

The foo:. formation rule in Hindi requires references to a minimally new set of
terms, namely, cognate rimes, i.e., rimes which have the same source in a
syllable, for example, the rimes in an extraheavy syllable, and absolute rimes,
i.e., rimes which are independent. The Foot Formation rule is stated as follows:

(28) Foot Formation (FF) (Pandey (1989: 62))


(a) Form a binary-branching s w foot on cognate rimes;
(b) Form a QR-sensitive, binary-branching s w foot an absolute rimes
from the right edge (if possible, i.e. if there are no cognate rimes in
the way.);
(c) Adjoin stray rimes to the left foot.

Notice that (28a) will erect s w feet on all extr.aheavy syllables in (27a).
Similarly, (28b) correctly predicts the penult pattern in (27bi). In (27ci), the
penult syllable cannot be strong, as the final is heavy. A foot on the
antepenult, however, is easily permitted. Notice that in all such eases where
the final is heavy, but the antepenult is light, accent falls on the antepenult, if
an antepenult is available. In the absence of the availability of an antepenult
syllable, as in (27bill), a Default Rule (DR) applies, which erects s w or w s
feet avoiding accent clash: k$mal ~ kamM, skmac :fiat (,,,) (* sam~ackar (~)).

Well-formedness constraints
(27cii) patterns are accounted for by .a well-fermedness constraint which
prevents the formation of feet on Eght syllables if preceded by a heavy
syllable:

(29 * Z

Rs Rw / R

For example, *sangkti, *wanasp&i. Another weil-formedness condition on


foot structures is the following:

(30) *Z
I
[... R...]
i.e., ~o degenerate ieet in polysyik~b!es.
292 P. If. Pandey / Hindi schwa deletion

For example, *sCtaar. *aalook. Note,, however, that monosyllables are


stressed, irrespective of their syllabic structure:

(31) z
I
[o]
Word tree
As there is free variatiol., between primary and secondary stresses within
~ords, this fact can be accounted [br by the following rule:

(32) Stress Interpretation Rule


Each foot has some degree of stress.

We can now give the derivations of some of the core patterns discussed above:

(33) (a) ~numbti (b) sSmaa¢ ""


a t', a i Rime Projection (RP) o aa aar s RP
]ial I
s w s w Foot Formation (FF) sw FF

s w Stress Interpretation Rule s ws w


x,/ DR
Wd s w
S~R, and
other rules
Wd
(c) ripbriyaa
i ~ i aa RP
J_HA/
WSWW

s Universal Stray
~/ Syllable Adju~ction Convention
Wd
s The horizontal bar connecling the twe rimes indicates cognate rimes or rimes belonging to the
same syllable, i.e., rimes of extraheavy syliables. See (19c).
P.K. Pandey / Hindi schwa deletion 293

3.2.3.2. Colloquial patterns. Besides the core patterns, there is a regular


optional pattern in colloquial Hindi, which retracts accent on the first syllable
of the word, subject to the QR-ser~sitive constraint (24). ~'or example,

(34) kSmalinii (~ kamSlinii)


dfiaroogaa ( ~ daar6ogaa)

(35) *sitaaraa
*w6naspati ~

3.2.3.3. The grammatical organization of $bot Foormation. In the ab~nce of


a detailed investigation of the play of morphological factors in the formation
of foot structures in words, we are on somewhat unsure grounds here. I
assume that Foot Formation takes place at the first lexical level, but not at
the level where inflectional affixation take~ place:

(36) Stem Stem + Future Gloss


gaa gfiaeegaa (3 sg.) *gaa6egaa sing
khaa khhaooge (2sg.) *khaa6ogee eat
col c$1u~gaa (1 sg.) *col~'~gaa move, go

Other inflectional affixes do not alter the shapes of forms in a metrically


significant way, for example~ g~ayaa 'sing+ Past', g~ataa 'sing+ Pres.
(3& I sg.)'; kiMab 'book ~, kiMab~ 'books'. Note, however, that deriv~tional
affixes do affect the stress pattern, e.g.,

(37) Stem Causative Gloss


dm~DR d~amRfianaa run
b~eTh b~a~Thfianaa sit
col colfianaa move

Certain am×es, however, yield different patterns, for example,

(38) Complexforms G/o~


(a) t/mkotwar (tfiakat + war) *taak5twar strong
bfia!akpan (b/~alak + pan *baal.$kpan childhood
jfiaduugar (jfiaduu + gar) *jaadfiugar magician

o Colloquial Hindi does not pe~lit short, final vowels (Pandey (!989: 50))
294 P.K. Pandey / Hindi schwe deletion

(b) hamlfiawar (h:~mlaa + war) attacker


laRbkpan (lbRok + pan) boyhood
sa~mdfiagor (s~bmdaa + gor) trader

Bharati (1988) contends that Hindi kas three strata of word formation in the
lexicon. Stratum I deals with roots and certain derivatiunal affixes, stratum 2
with subcompounds and derivafional affixes, and stratum 3 with co-compounds,
inflections, and some derivational affixes such as -wdr, -gsr, -pan, etc. 5he
does not consider all the evidence in support of her claims. 7 Her stuSy,
however, does provide strong evidence for at least two levels of morphologi-
cal and lexical phonological rules. Inflectional affixation and prefixation
(excepting those dealing with Sanskrit sandhi rules) seem to be taking place at
level 2. If that is so, then stress placement may not apply at this level. We
have seen this to be the case with 'he Future affix. Consider now ::he
following forms with prefixes (39) and some derivationai suffixes (40):

(39) Complex forms Gloss


+ maanow + ii [bmfianwii] inhuman
ku + samoy [kfisamby] bad time
bee + dhoRak [b~edhaRbk] unhesitating(ly~
owo + muulyzn [Swomfullyan] evaluation
up~ + niwee~ [fip(o)niw~eg] colony
op~ + bhron~ [$pabhr~ng] a corrupt form
of a word
znu + pmfiik [bnupmtiik] *z,nfipmtiik symbolic
odhi + niyom [5dhiniySm] *~dhiniyom rule
puraa + lipi [l~uraalipi] *pur~alipi ancient
manuscript

~ Take, for instance, the stratum 3 morphology consisting of the gerundive suffix -it (e.g.,
[paRhaaii] 'study'), -pan, co-compour~ds, inflections, etc. The reason fox' placing -ii at this stratum
is that it feeds co-compounds (e.g., [[paRhaaii] [likhaaii]] 'reading-writing, studies') which differ
from subcompounds at stratum 2. The facts of stress assignment however show that -ii differs
from -pan and other affixes in affecting a change in ~itress patterns: [mShgaa] 'costly', [mShg',iaii]
'high cost'. As -ii behaves differently from -pan and other inflectional affixes for stress
assignment, it should belong to a different stratum.
Co,~ider ~gain the nominal and adjectival (not the gerundive) suffix -ii (e.g.. safeed "white'/
s~feedii 'whiteness'; gulaab "rose'/gulaabii 'rosy', posited at stratum 1 since it feeds s~ratum 2
compounds [[[wyaapaar]ii]sa~gh] 'business group'). There are any number of examples showing
that the suffix i:. fed by subcompouads: [[jan] [sar.~ghlj "(a political organization)', [[[jani [sarjgh]ii]
'of Jan Sangh', [[suuryaj [wang]] '(a 'gotra')', [[[svurya] [wang]ill 'of Surya Wang'., etc.
P.K. Pandey / Hindi schwa deletion 295

(40) eekaakii + pan [eek~akiipan] *[eekaakiipan] loneliness


~rii + mat + ii [,~riimatii] Mrs or wife
ucakkaa + pan + aa [ucSkkaapSnaa] the quality of
swindling
jaaduu + gar + ii [jMduugSrii] *[iaad6ugarii] trickery or magic

Notice that all the cases in (39) show that prefixes are stressed. In some cases,
e.g.: [kdsamay], the structure of the word is such that stress would fall on
them regularly by the core foot formation rule. But in others, such as
[Sm~anwii], [pdraalipi] (,-,), the stress patterns are very different. The patterns
in such words can best be accounted for if treated as compounds: 8

(4 I) [[5] [[mfianaw]iil] [[5dhi] [niybmll


[[Spa] [bhrSng]] [[pfiraa] [lipi]]

Postlexically, the prefixes may concatenate as a single form, e.g., [su + gandh]
[sugandh] 'good smell', [sam + gooSThii] ~ [sarjgooSThii] 'meeting', etc.
The explanation for the prefixed forms seems to hold good for the suffixed
forms in (40) as well, as shown below:

(42) [[jb.aduu] [gbrii]] [[grii] [matii]] [[eek~akii] [pSn]]

With these oboervations with regard to foot formation, we are no,or prepared
to turn to the non-linear explanation of Schwa Deletion, as formulated in (2).

3.3. Schwa deletion

My main purpose in this section is to explain precisely how the process of


Schwa Deletion in Hindi takes place. It was claimed in section I that a non-
linear analysis of the phenomenon using (2) is preferable to a linear analysis
using (1), on the grounds (a) that the burden of conditions preventing the
process from applying is borne not by stipulated conditions in the rule, but,
in part by the structural description of the rule, and ;n part by the indepen-
dently metivated principles of non-linear phonology, and (b) that the non-
linear rule itself is much simpler and non-a~bitrary. In what follows. I shall
first justify the non-linear formulation (2) (= 12), and then suggest how the
rule fails to apply in those situations where it must.
a A la Hayes (1981) for a subset consisting of Greek prefix-stem words such as hdliccgr:~ph.
hdteron.Pm, kaldidoscdpe.
296 P.K. Pandey / Hindi schwa deletion

3.3.1
(2) captures the inductively arrived at generalization in (13) regarding d-
deletion, namely, that only an unstressed a is deleted.
There are two kinds of metrical environments for unstressed a: (a) follow-
ing a stressed syllable, and (b) preceding a stressed syllable. Only in the
former case does it meet the structural description of (2). Both are given
below for illustration:

(43) (= 3aii)
pbkaRaa
nbkalii
n~aTakhy ( ~ )
bb.ahorii

(44) (= 4aii, bii)


pakbRtaa
nak$1cii
polbrjg6-o
ghambNDii

The metrical structures of two of the forms in (43) are given below:

(45) (al Wd

Rs Rw Rw

~ k ~ R aa

(b) Wd

Y.,s J~w

Rs Rw Rs Rw
A l A I
aa T ~ k ii y
P.K. Pandey / Hindi schwa deletion 297

The final stray syllable in (45a) is adjoined to the left form by a universal
Syllable Adjunction Convention. The unstressed a's in (44) are weak syllables
adjoined to the word tree:

(46) Wd

Rw Rs Rw
I A A
P a k o R t a a

Note that given the QR-sensitive binary s w foot, on17 a word-initial a can be
outside it and thus not subject to deletion. This pa"tly explains why the
deleted schwa must be flanked by a vowel on both sides, as required by the
linear rule (1). The other part of the explanation is that a never occurs word-
finally in Hindi, as argued by Ohala (1983).
Rule (2) further predicts that stressed a's are never deleted. This is
corroborated by the forms in (44) and in (47):

(47) (= (9), (10)) pipSriyaa keesSfiyaa kamSlinii pilSkhuaa

The metrical structure of pipSriyaa in (33c) shows how d is stressed. Readers


will recall from the discussion of these forms around (10) and (11) that they
must be marked as exceptions in Ohala's linear account. The present analysis
predicts this fact by rule.

3.3.2
The structural description of a-deletion within the non-linear representation
in (2) includes all the environmental conditions in which a deletes. The
environmental conditions in (1), invoking the linear representation of forms,
are necesary but not sufficient; therefore the need to stipulate further
conditions. These, as claimed before, are predicted by different and indepen-
dent aspects of the present analysis.
First of all, the convention tha~ the rule must apply from fight to left is
wholly unnecessary in the present analysis. The directionality of the rule,
which forms one of the basic principles of prosodic phonology (see sec-
tion 3.1), is determined for the foot structure rule (see (26) and (28)), and
need not be specified for (2).
298 P.K. Pandey / Hindi schwa deletion

Condition 1 in (1), which prohibits the application of the rule in case a is


preceded by a morpheme boundary, as illustrated in (5a), can now be seen to
follow from the lexieal application of the foot formation rule at the level
where prefixed (39) and some suffixed forms (40) are treated as compounds,
as revealed by their accentual patterns. The metrical structure of these words
is such that they do not match the structural description of (2). a in these
forms is either stressed, or is not a part of the metrical foot, but is a~oined to
the word tree.
Further, as illustrated in (5b), Schwa Deletion fails to take place in those
instances in which the resultant forms violate the well-formedness constraints
of the language. Whereas the linear rule (1) must stipulate this condition, the
prosodic rule (2), a le×ical rule, is free from this burden. The condition
follows from the principle of Structure Preservation (see section 3.1), accord-
ing to which the lexical application ~f a rule must not violate well-formedness
constraints on both melodic sequences as well as syllabification.
An additional fact which underscores the descriptive adequacy of the
present analysis is that it predicts, rather than lists, the permissible sequences
in a principled way, as discussed above. In order to elaborate on the earlier
discussion, let us consider both the permissible as well as the aon-permissible
medial three-consonant sequences exemplified in (3b) and (5b), and listed in
Ohala (1983: 128-130):

(48) Permissible medial 3-consonant sequences


mbr, mbhl, ntr, ndr, ndhr, yicm, jajhr, .0jhl, yijr, NDI, rjgr, rjgl, rjgR,
rjghr, str, STm, tsn, rthn.
(49) _,Von-permissib!e medial 3-consonant sequences
~wT, rkT, rk~, rbt, lgm, drt, drk, dmt, rib, NTk, smn, smt, srt, jgr.

Recall that (22iii) prohibits three-consonant clusters in Hindi of which


either the first two or the last two uniquely belong to a syllable. (22iii),
however, is not violated if the association lines, following the Linking
Constraint, do nvt match it. (22iv) thus allows three-consonant clusters if C2
is doubly linked with nodes. Only a C2 homorganic with C1 meets this
condition, provided that it is followed by a sonorant, as required by (22ii).
Two Hindi-specific conditions thus license three-consonant clusters: first,
C3 must be a sonorant, and second, C2 must be homorga#c with Cl. Notice
*h~.t the cluster configurations in (48), but not in (49~. fulfil both the
conditions, for example,
P.K. Pandey / Hindi schwa deletion 299

(50)

C V C C C V V
l l l l l l l
j o 0 g 1 i i

(51)

C V C C C V V
l i l l i l l
m o t 1 b i i

Further, the syllabification of the clusters in (48), as exemplified in (50), fulfils


the universal constraints of Maximal Onset Principle (21), and Sonority
Sequencing Principle (20). The clusters in (49), on the other hand, violate
either the SSP (19) (and (22ii)), or the Hindi-specific Constraint (22iv).
Quite obviously, in the absence of the notion of the syllable, no significant
generalization about the well-formedness of (48), as against that of (49), can
be made. Thus Ohala's attempt (1983: 134-135) to explain the difference
between (48) and (49) in terms of major class features fails. She notes that
(48) has mainly nasal + stop + sonorant clusters. But there are also frica-
tive + stop + sonorant (str, stm), stop + fricative + sonorant (tsn), liquid +
stop + sonorant (rthn) clusters. All these cannot be systematically grouped
together in Ohala's terms. Besides, there is the problem of disallowing a
specific sequence which is labelled in terms of the same major class features as
other allowed ones. Thus, in order to rule out -lgm- (*balgm6-o), Ohala must
prohibit liquid + stop + nasal clusters. But in the face of the evidence of a
sequence such as orthn- (praarthnaa), she must treat one of the two as
exceptional. She admits that more specific constraints are necessary to rule
out the regularly non-permissible clusters. But she does not explain what
these specific constraints are like. As I have shown above these specific
constraints are syllable structure based.

3.3.3. There is one more contextual condition in (1) whose exclusion in (2)
requires clarification. The condition is that of the flanking C 12- - C environ-
ment.
There are clear reasons why not more than two consonants on the left, and
not more than one consonan{ on the right, are permitted. The reasons are,
firstly, that if there are more than two consonants preceding a, then a four-
300 P.K. Pandey / Hindi schwa deletion

consonant sequence would result, which is not regular in Hindi. Secondly, if


there is more than one consonant following a, then a heavy syllab,,~ we,_,!d
result, which would get stressed. The former possibility is ruled out in the
present analysis by ~h~ principle of Structure Preservation, and the latter by
the structural description of the rule.
What is not clear, however, is why there must be at least one consonant on
both sides. The consequence of not having this minimal condition satisfied
would be that, as illustrated in (4c), unacceptable forms such as *surO5
(< suar~), *bambii (< bambaii) would result. Recall that the deletion of a is
entirely dependent on the QR-sensitive foot structure in the language. It
teletes in rhythmi.ally weak positions, but without affecting the. total weight
of the word. M:v native speaker intuition about the latter condition is that it
is so general, like that of the failure of the rule to apply when certain
consonant seque~.ces result, that it must be independently predicted by the
theory, and not stipulated in the form of a rule. On a closer consideration of
the different aspects of the phenomenon, this is indeed possible.
Consider first the form of the rule. The focus of the rule has a, but not its
association with the Rime constituent. The context of the rule specifies that a
deletes when occurring as a segment singly associated with the weak rime of a
s w binary foot. The structural change of the rule only involves the deletion of
a, but not of the mora, indicated, in the form of representation adopted here,
by the non-branching rime:

(52) d~

Rs Rw Rs Rw
I I /N J I
k o m ~ 1 a a '~ ~ m ~ 1 a a

(53)

Rs Rw Rs Rw
[ I • deletio.___~n I [
s u ~ r 6"0 s u r oo

The difference between the ruled-in (52) and the ruled-out (53) forms shows
that the mora associated with a must eventually have an anchor. In the
absence of an anchor, the unassociated mora would get deleted. This,
eventually, does not happen.
P.K. Pande~ / Hindi schwa deletion 301

As mora is distinctive in the language, and in fact, forms the basis of foot
assignment and other lexical rules, such a: causativization (see (37)), it is
natural that its deletion when not warranted by the rule must be blocked,
followiag the principle of Structure Preservation.
The Linking Constraint (15) plays a cruciai role here. It provides for the
structural description of a prosodic structure rule to be so explicit as to
include all the association lines to which a rule applies. If the focus of a rule
includes only a part of the association lines of the structure on which it
applies then only that part is material to its application, as discussed above,
and clearly shown by Hayes (1986a) in an elaborate discussion of the
phenomenon of 'inalterability'.
The impossible form in (53) is thus predicted by the form of rule (2), as well
as by the principles of Structure Preservation and Linking Constraint. The
stipulation of the context C 2__C in rule (2) is therefore entirely unnecessary.

4. Comparison with other analyses

4.1

The main burden of the preceding sectior~ has b,en to show that the
present non-linear analysis is preferable to the linear analysis of Ohala on
many counts.
Firstly, the non-linear rule is devoid of arbitrary contexts. The linear rule
has no explanation for why a in order to be deleted must be flanked by the
segments specified in it. The context of the non-linear rule makes it clear that
a deletes only as the weak member of a disyllabic foot.
Secondly, the non-linear rule is strictly local, as compared to the linear
rule, which is not.
Thirdly, the non-linear rule is more general in that, in addition to the forms
covered in Ohala, it predicts the regular retention of a in forms such as
pip~nyaa, kamMinii, etc., which must be marked as exceptions in Ohala's
analysis.
Lastly, the non-linear rule is free from the stipulated convention and the
conditions governing the application of the linear rule.

4.2

While Oha!a's analysis of a-deletion is the most comprehensive, D'Souza's


302 P . K . Pandey / Hindi schwa deletion

analysis (1985) is the first non-linear account of the phenomenon. It is


therefore appropriate for me to compare it with, and thus justify, the present
one. D ' S o ~ a ' s rule of Schwa Deletion is similar to (2), with some differences,
which are discussed below.

4.2.1
D'Souza assumes the non-hierarchical structure of syllables in the revised
form of C-V phonology adopted in Archangeli (1984), in which the C-V
elements are substituted by unlabelled X slots. Syllabification is rule-gover-
ned, following Steriade (1982), and Clements and Keyser (t983). Her foot
construction rule is stated as follows:

(54) Foot Rule


Build left dominant, quantity sensitive binary feet beginning at the left
edge of the word. (D'Sotma (1985: 16))

Her rules ."f syllabification and of foot construction (54) give rise to represen-
tations such as the following:

(55) ,l, ,1 ,l ,i', ,~ ,1 ,l ,~


XXXX,~XXX XXXXXXX XXXXXXX
'0|1~||! ||l|ll! |l|gl|l
OillO|~| g|||lOI |||111!
bi s tora" 'bedding' ka: nta: 'wife' s~bok62": 'lessons"
O/ 0 0 0 ~ 0
I I I I
The syllables in (55) are represented above the melodies, and the feet below
~hem. q'he solid lines within syllables indicate pre-specified vowel-heads; the
dotted lines indicate all other elements incorporated by rule. At the foot level,
circles indicate strong syllables.
The rule of Schwa Deletion is stated thus:

(56) Schwa Desyllabifieation


1
X ..... >X / 0 ---

8 o /
(56) states that if a constitutes the weak branch of a fool the assocmtion
P.K. Pandey / Hindi schwa deletion 303

line marking it as the sylllable head is severed. The dissociation of ~ from the
syllabic node is to facilitate its reinstatement in case the con,;equent conso-
nant sequence cannot be syllabified, as in the forms such as pustak65 in some
variety and matlabii in all the varieties. The reinstatement rule is as follows:

(57) Schwa Reinstatement


i
x ~x / :(
I I
(x" denotes an unsyUabified slot)

A desyllabified schwa that is not reinstated is obviously phonetically not


realized. D'Souza ascribes the failure of schwa to delete across morpheme
boundaries, predicted by Condition 2 in Ohala's rule (1), to the erection of
.metrical
....... feet on morpheme ;.... ; ° o,,,~hi,=~ by
" ; t,,,l T,,.,:,;,-,I A . . , . . - , Rule ~=o,.
~Jo}.

(58) Ininal Accent Rule


Every morpheme initial syllable is the head of a foot.

Thus the retention of a in complex forms such as beepaRhaa, kalaawati, is


because of its morpheme initial position where it constitutes a foot by itself:

(59) be + poRh + a: ka:ri: +gor +i:


0 0 3 0 0 0 0

I I ! It I !
4.2.2
D'Souza's analysis clearly differs from the present one in many respects. I
shall briefly try to show below that in all those respects the present analysis is
preferable.
Most crucially, D'Souza's foot construction rules are somewhat 'ad hoe',
since they seem to be motivated mainly by the facts of the problem of Schwa
Deletion. I have tried to show in this analysis that the foot forms to which
Schwa Deletion is sensitive are independently motivated by the facts of accent
in Hind; (Pandey (1989)). A proof of the 'ad hoc' character of D'Souza's foot
rule (54) is the fact that it does not apply to the full r~r~ge of regular accent
patterns, for example, the patterns in ~47). It is interesting to note that like
(I), (54), too, must treat these forms as exceptions. Secondly, D'Souza has no
304 P.K. Pandey / Hindi schwa deletion

way of differentiating between heavy syllables, which are accented on!y


penultimately and alternately, from extraheavy syllables, which are ubiqui-
tously accented.
As in the present analysis, D'Souza predicts those cases as regular which
are accounted for by the stipulated Condition 2 in rule (1). These cases
violate the prosodic well-formedness constraints, as discussed in 3.3.2. They
are accounted for in D'Souza's analysis as cases of reinstatement, being
non-syllabifiable. Among the disallowed sequences violating prosodic well-
formednes conditie ~s, there is a subset consisting of sequences such as *dt
(*aadt~ < aadatb~ which do not involve syllables. As Ohala (1987) points
?ut, D'Souza's sylhb!e structure based constraints on consonant sequences
fail to account for the failure of a to delete in forms such as aadate~.
On the basis of this evidence Ohala argues that the constraint preventing
the adjacent occurrence of *dt being linear can be accounted for only
by a linear rule. Within the present analysis, which allows for well-
formedness constraints to hold at all possible levels (see • a ,,, the
.,.~),
exceptional form in aadat~ is in fact shown to be quite regular,
since the *dt seqLtence is prohibited from occurring at the melodic level
(see 3.2.1).
As in the case of sequences vi ~lating melodic constraints, there is nothing
in D'Souza's analysis to prevent the deletion of a in forms such as suar6"b
(> *st4r6"b), b~ nb,~:t(> *bambii), given her theoretical assumptions, criteria of
sy!labificatio~, and rules of Foot Formation (54) and Schwa Deletion
(56). This b~ clearly an undesirable result, and a problem for her
analysis.
The lexic~! p~,nological application of the Foot Formation Rule in the
present analysis is found to show th~,t me,~ prefixed and some suffixed forms
are treated as compounds, so that their metrical structures do not match the
structural descripfti.on of rule (2). D'Souza explains these form~ with the help
of the Initial Accent Rule (58), which is blind to Quantity-sensitivity, and is
clearly wrong in ~theface of the data in (27). The strategy undoubtedly helps
account for the prefixed fo._~rns, and thus dve:~ away with the stipulated
Condition 2 in ruLle (l). However it must consequently give rise to more
complex derivat ons. Morphemes which do not have initial accent, e.g.,
wicaar 'thought', tamaa~aa °~how', etc. must be shown to undergo initial
accent deletion. The present analysis does not need to complicate fie deriva-
tions in this way.
P.K. Pandey / Hindi schwa delelion 305

5. OCP and other imues

5.1

The main aim of this paper has been to provide a descriptively adequate
non-linear account of schwa syncope in Hindi in order to establish its
relative merit over a linear account. This should explain why I have not
gone into many other controversial issues, both analytical as well as
theoretical, pertaining to the problem. Thus, I have totally ignored the
question of sociolinguistic factors impinging on the process, taken up in
Ohala (1977). As Ohala has repeatedly shown, the process is optional in
at least a part of its manifestation. My goal, more precisely, has been
to explain the conditions on the deletion of sohwa when it does take
place.
I have also not taken up another controversial a s s e t of the problem,
namely, the deletion of schwa in morpheme-final position. Shrivastava (1969,
1979) has argued that in order to explain certain alternations such as [jal]
'water' and [jalaj] 'born of water, lotus', it would be necessary to posit an
underlying morpheme-final schwa in consonant-final morphemes - [[jola]ja-
]] > [jolaja] > [jalaj]. Ohala (i983: 148-149) has convincingly argued against
the final schv,a hypothesis. I assume the correctness of her explanation
of some of the possibly true cases as belonging to the loan phonology
of Sanskrit. I have not found any case of alternation in the native
i-iindi phono!ogy that crucially requires positing a morpheme-final
schwa.
There is however a case for final schwas in prefixes such a opo-, upo-, owa-,
etc. which have schwa-less alternant pronunciations: op-, up-, aw-, etc. Notice
that in all these cases (see (39)) the first syllable is stressed. The optional
deletion of the final schwa has a clear explanation in the prsent analysis: it
occurs as the weak member of a disyllabic foot, e.g.,

(60)

Rs Rw
I I
o p ~

and thus matches the description of (2).


306 P.K. Pandey / Hindi schwa deletion

5.2. OCP

One of the prosodic principles on which H;ndi Schwa Deletion has been
found to have some bearing (see McCarthy (1986: 259), Odden (1988: 465)) is
the Obligatory Contour Principle, originally proposed in Leben (1973) for
tonal phenomena (see also Goldsmi'h (1976)), and extended to apply to
melodic segments in McCarthy (1986). I have refrained from including it in
the theoretical discussion in section 2.1 because of its controversial nature as
a universal principle (see Odden (1988)2. I turn to it here to show how an
invocation to it leads to further simplification of the present a,.zeount; more
specifically, how it predicts the failure of Schwa Deletion to take place in
order to avoid geminates and homorganic s 9s in the output, e.g., aatataayii,
aadat~.
McCarthy (1986: 208) p~eposes the follow.~ng version of OCP as a univer-
sal principle: 'At the melodic !evei, adjacent identical elements are prohibit-
ed'. The principle was conceived of as operating as a Morpheme Structure
Constraint as well as as an output condition in the course of the derivation.
Yip (1988) further argues for its role in triggering rules. Ito (1986) and Yip
(J'988) also demonstrate its applicability t~, features or groups of features
which may appear on separate tiers subject f:. !anguage-spedfic constraints.
Melodic segments which are represented on multiple tiers are at some point
of derivation (presumably in the lexical phonology) brought together on a
single level of representation by Tier Confiation, which has the effect of
separating or merging any violations which remain.
The originality in McCarthy's proposals consisted in his explanation of
'Antigemination' exceptions to syncope rules. For example, Afar, a Lowland
East Cushitic language, has a rule of syncope 'that deletes an unstressed
vowel in a peninitial two-sided open syllable'. The ru!e fails to apply when
consonants on both sides of the potentially dele:able vowel are identical.

(61a) xamila xamlq 's~vamp grass (acc/nom.-gen.)'


dar~gu darg-i 'watered milk'
(61b) miDaD-i 'fruit'
danan-6 q/he was hurt'

(62) Afar Syncope


V --~ / # CVC-CV
[ - stress]
P.K. Pandey / Hindi schwa deletion 307

e e
I i
(63) [[c v c v c] *[[c v c c] v]
I I i I I I I I I
d a n a n d a n n

Odden (1988) argues against the OCP as a universal principle and tries to
show that it is an independent effect requiring independent explanation.
Among the many arguments he offers against the different aspects of Mc-
Carthy's theory of OCP, one consists of syncope rules which simply disobey
Antigernination. He cites the Hindi Schwa Syncope Rule as one such case.
His data are based on Bhatia and Kenstowicz (1972): /daanow+ii/ ,
[daanwii] 'demon',/kaanon+ii] , [kaannii] 'garden'. The rule of schwa
deletion, which accounts for the stem-final deletion of schwa in these in-
stances, but not stem-internally as in [waaraaNosii] 'Benares', was originally
proposed in Narang and Becker (1971):

(64) Schwa Syncope (Narang and Becker (1971))


a---~/VC C ÷V

Because of the suffixal condition in the rule, Odden rightly treats it as a


lexical rule. The geminate form [kannii] is thus a clear counterexample to
OCP.
My data and those in Ohala (1983) are, however, somewhat at variance
with those Odden considers. Note that Bhatia and Kenstowicz (1972: 212)
and Ohala (1983: 151) point out that forms like [waaraaNosii] with word-
internal schwa are rare. Indeed this form is an exception rather than a rule.
Even so it is found to exist alongside its common variant [waaraaNsii]. 9
Ohala (1983) has shown that (64) is inadequate on at least three counts: its
inability to explain syncope in a simple form such as [rojnii] ~ [rojanii] as well
as in the environment following consonant clusters as in [daar~nik] < [dor-
San], and the unlawful deletion in *[aadtt~] < [aadat~].
It is interesting to note here that Narang and Becker (1971) do concede the
presence of optional variants in certain forms such as [sundarii] ~ [sundrii] 'a
beautiful woman'. But they account for them as due to another optional rule
which is not the same as (64). Ohala (1983, also 1977) argues against positing
two rules in the grammar, one with obligatory and the other with optional
My wife and I both belong to Benares, and we have both the forms in our speech. The
syncopated form normally occurs more frequently.
308 F.K. Pandey / Hil~di schwa deletion

application. In the present analysis, Schwa Deletion (2) applies both in the
lexical and in the postlexical modules, giving rise to alternate forms such as
[aadot~] ~ [aatt~], [jananii] ,-, [2onnii] 'mother'. The lexical application of
the rule is subject to the melodic constraints and the syllable structure
constraints in (17) and (22) respectively. Note that (17) (iii) and (iv) are,
essentially OCP constraints operating at the place and laryngeal tiers respec..
tively. In the postlexical application of the rule these constraints as well as
Antigemination are not respected. Moreover, the postlexical application of
the rule, following a general constraint (see Kaisse and Shaw (1985: 6),
Pandey (1987)), is fully optional. In support of this observation, I have many
alternating forms to instantiate, e.g.,

(65) Uznonii] ~ [jonnii] 'that who gives birth, mother'


[aatotaayii] ~ [aattaayii] 'tyrannical'
[badotor] --, [bott~u] 'worse'

The forms without syncope are characteristic of normal tempo, as are the
following forms with syncope: [noklii] < [nokol]. [naaTkiiy] < [naaTak]. The
syncopated forms in (65) characterize fast speech. There are however some
frequently occurring words whose geminate forms are more common, for
example, [maannaa] ( ~ [maanonaa]) 'accept, believe, etc.' It is therefore likely
that occasional lexical violation of Antigemination does take place. The
enforcement of OCP at the underlying level, however, as McCarthy points
out, is strictly adhered to.
The evidence for :lie universalist claim for OCP that the Hindi facts
provide is somewhat like that for Structure Preservation, which is found (see
Mohanan (1989: 609)) to be occasionally violated at the higher strata in the
lexical modu|e. Considering the widespread evidence in its support, and
occasional vioXations of it, a weak version of the OCP seems better justified.
The nature of the exact formulation of the weak version is not clear to me at
this stage. "[he Hindi facts however seem to suggest the following partial
view: when Syncope has both lexical and post-lexical applications, :epresenta-
tions violating OCP may come to persist.
At any r~,Lte,the OCP-sensitive facts are in keeping with Ohala's account in
general. Her analysis does not contain any instance of'forms with identical
consonant~, flanking the site for Schwa Deletion, :uch as [jonanii~, but it dc~cs
include forms with flanking homorganic obstruents, such a,~ [aadot~] in
which schwa does not delete. She must account for it by the stipulated
Condition (2) in her rule. In the present account, it should be obvious by now,
P.K. Pundey / Hindi schwa deletion 309

the failure of (2) in these instances is on account of a language-specific


implementation of OCP. Even otherwise, I have treated these as predicted by
the well-formedness constraints, which is at some cost to the grammar, but
not to the account of Schwa Deletion. The OCP explanation has the happy
advantage of provid~.ng an entirely cost-free account of these exceptions.

6. Summary and conclusions

In section 2 we considered the facts impinging on the deletion of schwa in


Hindi. These included (i) the positive: features of surface phonological struc-
ture on which the rule appears to depend, and (ii) the negative features which
prevent its application, such as a morpheme boundary on the left, and
sequential constraints. With regard to (i), we had noted a fact, stated in (10),
which was ignored in Ohaia's linear account, but which has been crucial in
our investigation, namely: only, but not all, unstressed a's are deleted,
In section 3 we showed that all these facts have a straightforward explana-
tion within a prosodic phonological account. The foot-based rule (2) essen-
tially captures the insight regarding only unstressed schwas undergoing
deletion, as well as the right-to-left directionality of the rule. It was further
shown that at least some of the well-formedness constraints, for example,
those regarding three-consonant clusters, must be syllable structure based.
The !exical application of the rule was found to explain why it does not ~.pply
across prefixes which are added at a later stratum, and in those instances
which would violate the well-fon~edness constraints. The lz,tter are enjoined
from being violated by the principle of Structure Preservation. Other princi-
ples of prosodic phonology, such as ~he Linking Constraint, take ~:are of
specific exceptions to the rule. In 5.2 we considered tbe putative cases of OCP
violation, and found that while the underlying representation follows OCP,
the surface violations are on account of the optional postlexical appJicatior
of the rule, leading to alternate forms such as [ianonii] ~ [jonnii], [bodotar] ~
[bottar].
A comparison with Ohala's linear and D'Souza's non-linear ana~,ysis in
section 4 highlighted the preferable features of the present analysi.~,, They
consist, in the main, in the postulation of a simple, local rule of Schwa
Deletion. All the systematic irregularities in the application of the rule are
e×plained as owing to the independently motivated principles of prosodic and
lexical phonology. The analysis thus serves the two-fold purpose of providing
the most simple and elegant account of the phenomenon (and thus being an
310 P.K. Pandey / Hindi schwa delet/m+

obvious candidate for selection by the evaluation metric), and of offering


support for the theories of lexical and prosodic phonology, which form the
basis of its description. The paper further produces evidence that supports a
weaker version of the OCP, according to which the principle may be
occasionally violated by a productive rule with global application, such as
Hindi Schwa Deletion.

References

Archangeli, D., 1984. Underspecification in Yawelmani phonology and moxphology. Doctoral


dissertation, MIT, Cambridge, MA.
Bharati, S., 1988. Some aspects of the phonology of Hmdi and English. Doctoral dissertation,
C.I.E. & F.L., Hyderabad, India.
Chomsky, N., 1986. Knowledge of language. New York: Praeger.
Chomsky, N. and M. Halle, 1968. The sound pattern of English. New York: Harper and Row.
Clements, G.N., 1985. The geometry of phonological features. Phonology Yearbook 2, 225-252.
Clements, G.N., 1988. The role of the sonority cycle in core syllabification. Working Papers of
the Cornell Phonetics Laboratory, ,'~'o.2. Cornell University.
D'Squza, J., 1985. Schwa syncope and vowel nasalization in Hindi-Urdu: A non-linear approach.
Studies in Linguistic Sciences 5(1), 11-30.
Goldsmith, J., 1976. Autosegmental phonology. Doctoral dissertation, MIT, Cambridge, MA.
Distributed by IULC, Bloomington.
Halle, M. and J.-R. Vergnaud, 1978. Metrical s~ruetu.~esin phonology. Ms, MIT.
Hayes, B., 1981. A metrical theory of stress. Distributed by IULC, Bloomington.
Hayes, B., 1986. Inalterability in C-V phonology. Language 62, 321-351.
Hayes, B., 1989. Compensatory lengthening in moraic phonology. Linguistic Inquiry 20, 253-306.
Hyman, L.M., 1985. A theo,y of phonological weight. Dordrecht: Foris.
ito, J., 1986. Syllable theory in prosodic phonology. DocForal dissertation, University of
Massachusetts, Amherst, MA.
Kahn, k;+., 1976. Syllable-based generalizations in English phonology. Doctoral dissertation,
MIT, Cambridge, MA. Distributed by IULC, Bloomington.
Kalra, S., 1976. Schwa deletion in Hindi. Language Forum 2(!), 39-57.
Kiparsky, P., 1979. Metrical structure assignment is cyclic. Linguistic Inquiry 10+ 421-441.
Kiparsky, P., 1982. From cyclic phonology to lexical phonology. |n: H. van der HulsL N. Smith
(eds.), The structure of phonological representations, Part 1, 13I-175. Dordrecht: Foris.
Kiparsky, P., 1985. Some consequences of lexical phonology. Phonology Yearbook 2, 85-138.
Levin, J., 1983. Reduplication and prosodic structures. Ms, MIT.
McC+arthy, J.J., 1979. Formal problems in semitic phonology and morphology. Doctoral
dissertation, MIT, Cambridge, MA. Distributed by IULC, Bloomington.
McCarthy, J.J., 1984. Prosodic organization in morphology, In: M Aronoff, R.T. Oehrle (eds.),
Language sound structures, 299-317. Cambridge: MtT Press.
McCarthy, J.J., 1986. OCP effects: Gemination and antigemination. Linguistic Inquiry 17, 207-
263.
P.K. Pandey / Hindi schwa deletion 31 i

McCarthy, J.J. ann A.S. Prince, 1986. Prosodic morphology. Ms, University of Massachusetts,
Amherst, MA.
Mohanan, K.P., 1982. Lexical phonology. Doctoral dissertation, MIT. Distributed by IULC,
Bloomington.
Mohanan, K.P., 1986. The theory of lexical phonology. Amsterdam: Reidel.
Mohanan, T., 1989. Syllable structure in Malayalam. Linguistic Inquiry 20, 589-625.
Narang, G.C. -"nd D.A. Becker, 1971. Aspiration and nasalization in the generative phonology of
Hindi-Urdu. Language 47, 646--667.
Odden, D., 1988. Antigemination and the OCP. Linguistic Inquiry 19, 451-475.
Ohala, M., 1974. The abstractness controversy: Experimental input from Hindi. Language 50,
225-235.
Ohala, M., 1977. The treatment of phonological variation: An example from Hindi. Lingua 42,
1 i-76.
Ohala, M., 1983. Aspects of Hindi phonology. Delhi: Motilal Banarasidass.
Ohala, M., 1987. Schwa deletion in Hindi by linear and non-linear routes, Paper presented at the
XIV International Congress of Linguists, Berlin.
Pandey, P.K., 1987. Optional rules in lexical phonology. Paper presented at the XIV Interna-
,:A__,
u,. u~,l Cor~gressOl. .L. l.l .l g.U l S I S , Berlin.
Pandey, P.K., 1989. Word accentuation in Hindi. Lingua 77, 37-77.
Fray, B.R., 1970. Topics in Hindi-Urdu grammar. Research Monograph I. Berkley, CA: Centre
for South and Southeast Asia Studies, University of California.
Selkirk, E.O., 1980. The role of prosodic categories in English word stress. Linguistic Inquiry 10,
483-502.
Selkirk, E.O., 1982. The syllable. In: H. van der Hulst, N. Smith (eds.), The structure of
phonological representations, Part II. Dordrecht: Foris.
Selkirk, E.O., 1984. Phonology and syntax. Cambridge: MIT Press.
Shriv,~stava, R.N., 1969. A review of Studies in Hindi-Urdu l, by A.R. Kelkar. Language 45,
913-927.
Shfvastava, P_.N.. 1979. On capturing inaccessible mind: Further evidence for word-final schwa.
In: honour of D. Ko'stich. Belgrade.
Steriade, D., 1982. Greek prosodies and the nature of sylIabificatioa. Doctoral dissertation, MIT,
Cambridge, MA.
Yip, M., 1988. The obligatory contour principle and phonological rules: A loss of identity.
Linguistic Inquiry 19, 65-99.

Вам также может понравиться