Lingua 77 (1989) 37-73.

North-Holland 37

WORD ACCENTUATION IN HIND1

Pramod Kumar PANDEY *

South Gujarat University, Surat. India

Received April 1988; revised version September 1988

This paper presents an analysis of stress in Hindi using a new mode of foot construc-
tion, called the conjugational mode, which takes into account the relative weights of
rimes. This mode of foot construction is motivated by the system of stress in Hindi, which
is sensitive to both quantity and rhythm. A theoretical consequence of the present analysis
is that the tree analysis is found superior to a grid-only analysis. The tree theoretic
framework, however, must be revised in order to accommodate the present analysis. In
particular, the branchingness constraint must be restricted to be applicable to Q-sensitive
systems only, not to QR-sensitive systems such as Hindi. The essential claim of the paper
is orthodox: trees for the assignment of prominence relation to syllables, grids for
interpreting and evaluating their rhythmic form.

1. Introduction

The present paper proposes an analysis of stress in Hindi which introduces

a new mode of foot construction. The new mode, called the Conjugational
mode, erects a foot not on the basis of the absolute weight of the rime, but
the relative weights of rimes that constitute it. The justification for the new
mode lies in the disadvantage of the standard tree theory (Hayes (1981)) in
describing a system such as Hindi, which is sensitive to both quantity and
rhythm, and not merely to quantity. The essential feature of such a Quantity
and Rhythm (QR)-sensitive system is that it takes into consideration the
relative weights of rimes and not the absolute weight of a rime. The
explication of such a system, and the presentation of theoretical arguments
based on it, involves a clarification of the role of rhythm in generating

* I would like to express my gratitude to Bruce Hayes, and to two anonymous reviewers for
very useful suggestions and comments, and to Ashok R. Kelkar and K.P. Mohanan for advice
and help. I would also like to thank Manjari Ohala and R.N. Shrivastava for their comments on
a preliminary version of this paper, which was originally presented at the 1984 Fifth International
Phonology Meeting, Eisenstadt, and included in Pandey (1985).

0024-3841/89/$3.50 0 1989, Elsevier Science Publishers B.V. (North-Holland)

38 P. K. Pandey / Word accentuation in Hindi

metrical patterns. On the negative side, the analysis violates a number oif
claims in the theory, two of which are that heavy syllables may not be weak
in systems that are sensitive to syllabic weight, and that ternary feet are not
natural, but the outcome of an extrametricality rule. We try to suggest how
these claims could be revised so as to accommodate the analysis and still
make interesting predictions in general. The analysis of the QR-sensitive
system is finally used as the basis of an argument for the tree structure and
against the grid-only approach.

2. The variety examined, and the source of data

The data for the present investigation are drawn from the eastern variety of
standard Hindi (or KhaRi Boli). The division of standard Hindi into two
broad varieties - western and eastern (Grierson (1966), Harris (1966) Zo-
graph (1982)) - is based on the influence of regional dialects. Western
standard Hindi interacts with Hariani, Braj, Kanauji, and Bundeli; eastern
standard Hindi interacts with Awadhi, western Bhojpuri, Bagheli, and Chat-
tisgaRhi. There is a historical justification for this division, too. The western
group of dialects derives from the Sauraseni dialect of Middle Indo-Aryan,
and the eastern group from Ardhamagadhi Prakrit (excepting Bhojpuri,
which goes back to Magadhi).
In terms of accentual’ patterns, eastern Hindi also shows a systematic
difference between formal and colloquial varieties. In the absence of literature
on the systematic phonological differences between the varieties of standard
Hindi, I cannot go into the issue here. My purpose in broaching it is to
suggest that the data reported in this study form a homogeneous group, and
that the differences between the present (also Kelkar (1968) with minor
differences) and other descriptions of the phenomenon, for example, Moha-
nan (1979) (and see Ohala (1977)) could be on account of dialectal and
stylistic variation. Besides, these data are new within the metrical literature on
accent. Hayes (1981: 79-81) on the basis of which Hindi is classified with
systems such as Classical Arabic, Eastern Cheremis, and Huasteco (see Prince

1 From now on I shall use the term ‘accent’ for prominence in the abstract, and the term ‘stress’
for the actual degree of prominence on the surface. See Chomsky et al. (1956) for an implicit, and
Ladefoged (1971), Schane (1979), and Rischel (1983) for an explicit distinction in the usage of the
two terms along this line. I shall also use the term ‘accent’ or ‘accentuation’ for the phenomenon
in general.
P. K. Pandey / Word accentuation in Hindi 39

(1983: 77), Halle and Vergnaud (1987: 49)) does not include some of the core
patterns described here.
The main source of the data is the speech of twenty undergraduate and
graduate students and ten college and university teachers from four different
points of the eastern Hindi speech area: Lucknow, Allahabad, Varanasi, and
Jabalpur. The data consist of the pronunciation of 250 words selected on the
basis of syntactic, morphological, and phonological considerations. Thus the
words belong to major syntactic classes, are both simple and complex, and
cover, as far as possible, the full range of the number, weight and position
configurations of syllables. Thus trisyllabic words belong to one of the sixteen
possible types: L(ight) + H(eavy) + E(xtraheavy), L + L + L, H + H + H,
E+E+E... It may be noted that examples for all the structures were not
available for analysis.
The method of collecting the data was that of assigning the subjects the
‘linguistic manipulation task’ of pronouncing words typed in normal ortho-
graphic forms. Besides, both my wife and I are native speakers of this variety.
I have also drawn from earlier studies and my own experience of six years
which I spent largely in eastern Hindi speaking areas.
The data evinced a great deal of variation. The minimum percentage fixed
for accepting a pattern as regular was 70%. Patterns with at least 40%
regularity were considered as having a claim for accountability. Two types of
systematic variation were discovered. These are described here as core and
variant, or formal and colloquial patterns.

3. Facts

3.1. Vowels

Vowels in Hindi are short and long.

Short vowels: i[r], a[a], and u[u].
Long vowels: ii[i:], ee[e:], aze[ze:], aa[a:], DD[D:]~, and oo[o:].
Examples: (i) citaa ‘funeral pyre’, da1 ‘group’, ~unaa ‘heard’.
(ii) ciituu ‘cheetah’, ukeeluu ‘alone’
baa1 ‘bull’, dual ‘pulses’, kmr ‘morsel’
boom1 ‘bottle’, suunuu ‘deserted, quiet’
The distinctive features of the vowels are as in table 1.

2 Better [ax], the back counterpart of [a~:].

40 P.K. Pundey / Word accentualion in Hindi

Table I
Vowels of Hindi.

- Back + Back
- Round + Round

+ High i/ii ujuu

- High
- Low ee 00
+ Low aeie ajaa DD

3.2. Consonants

The articulatory chart of the consonants is given in table 2.

Table 2
Consonants of Hindi

Bilabial Labio- Dental Alveolar Retroflex Palato- Palatal Velar Glottal

dental Alveolar

Stop: voiceless
unaspirated P t T k
vl. aspirated P” t” T” k”
voiced b d D g
breathy vd. b” d” D” g”

Affricate:
vl. unasp. C

vl. asp. Ch

voiced
breathy vd. ‘h
J

Fricative:
vl. S

vd. z

Nasal : nl N P

Glide: W Y

Tap/Flap:
vd. r R
breathy vd. Rh

Lateral 1
P. K. Pandey / Word accentuation in Hindi 41

3.3. Syllable structure

We rely for the mode of representation of syllables on the relational theory

as presented in Kiparsky (1979) McCarthy (1979), and Hayes ( 1981).3
According to this theory in its broadest aspect, a syllable (o) consists of the
relatively weak constituent Onset (0) on the left, and the strong constituent
Rime (R) on the right,4 of which the Rime is an obligatory element5 The
rime may be branching or non-branching as below:

Ow Rs Ow Rs
I /A
a I 1 0 w

The geometrical representation of the syllable and its constituents is motiva-

ted mainly for the purpose of interpreting the moraic strength of syllables for
the languages which are sensitive to them. According to this model, a light
syllable has a non-branching rime, and a heavy syllable a branching rime.
A consideration of the factors involved in word accentuation in Hindi has
led analysts (e.g., Kelkar, Mohanan) to posit three degrees of syllable weight
- Light, Heavy, and Superheavy or Extraheavy. A light syllable has one
mora, a heavy two morae, and an extraheavy more than two morae. There
has been no effort, however, to show how the extraheavy syllable is to be
represented. Mohanan’s analysis takes care of it by final segment extra-
metricality, which reduces peripheral extraheavy syllables to heavy syllables.
A thorough investigation of the phenomenon, however, makes necessary the
representation of extraheavy syllables differently from the heavy and light
ones. The latter are represented below:

3 We are not considering the now widely acceptable and more elaborate structure with Nucleus
and Margin, and CV as lower level constituents, as they are not crucial to the description of
accent systems. For a recent theory of syllable structure see Ito (1986). See also D’Souza (1985)
for different representations of Hindi syllables.
4 Halle and Vergnaud (1980) also have an optional constituent, Appendix, on the right of the
rime.
5 The theory is in fact more detailed in its assumptions about the relative strength of segments
which form the terminal of the root. I exclude a discussion of it here (as of the Appendix above)
for a lack of direct relevance to the present discussion.
42 P. K. Pandey 1 Word accentuation in Hindi

OR OR OROR
I A I I IA
k i 1 aa ‘castle’

The extraheavy syllable is represented as in (3):

ORR
IN
J0 0r ‘noise’

The representation in (3) is after the representation of extraheavy syllables in

McCarthy (1979: 11) for Arabic, with a minor modification. This representa-
tion, as we shall show, is crucial to our description.

3.4. Syllable division

Syllables in Hindi divide (and segments disambiguate their syllabic consti-

tuency) on the following principles:

(44 When a vowel follows another there is syllabic division between them:
e.g., aa.i.ee ‘come (hon.)‘; taa.uus ‘peacock’.
(4b) An intervocalic consonant goes with the following vowel: V.CV e.g.,
ka.vi.taa ‘poetry’.
(4c) In a sequence of more than one intervocalic consonant the division is
made after the first: CVC.CV; e.g., candan ‘sandalwood’.

(4c) has several exceptions:

(9 There is no division in the sequence C : e.g., sa.preem ‘with

love’; su.jyooti ‘good (propitious) light’. If however the liquid or glide

is preceded by CC, then the cut is after the first C: C.C. : e.g.,

ab.bhyaas ‘exercise’; sum-mrid.dhi ‘prosperity’.

P. K. Pandey / Word accentuation in Hindi 43

(ii) In NC*C sequences (where N is a nasal hornorganic with the following

stop or sibilant C*, and C is a consonant other than liquid or glide) the
division may be after the C* (see Kelkar (1968: 25)); e.g., sans.thaa
‘organization’, caans.Zar ‘chancellor’.
(iii) Syllable division conforms to close juncture (i.e., pauses between
segments within the word)16 e.g., candr.maa ‘the moon’: candr=maa;
vand.naa ‘prayer’: vand=naa (where = stands for close juncture).

3.5. The underlying representation

Related to the problem of the representation of syllables is the question of

the level at which the metrical structure is created. The lexical and surface
representations of words in Hindi differ in their syllabic structures on account
of the processes of ‘a’ deletion (Ohala (1974) D’Souza (1985)) and consonant
gemination (Kelkar (1968: 6465)), among others. A few examples of the
surface forms of words showing a - 0 alternation are produced below:

(5) Verbs Past Stem Gloss

badlaa (masc.) badal ‘change’
samjhii (fem.) samajh ‘understand’
mahkaa (masc.) mahak ‘smell’

Nouns Plural Nom. Sing.

bahnE bahan ‘sister’
lapTEe lapaT ‘flame’
bootle?: bootal ‘bottle’

For the argument that a-less forms are on account of a-deletion and not a-

- 1
insertion see Ohala (1974).
Like a-deletion, consonant gemination causes the underlying sequence CC’
syllabic
(where C is any consonant and C’ is the latter being especially
[ -cons ’
y, v, r) to have the first consonant geminated:

(6) v@abs ‘belief, sattya ‘truth’, plittrii ‘daughter’

There are other contexts for consonant gemination, besides, but we shall

6 The evidence for the pauses is impressionistic. They can be heard in slow pronunciations of
the words, and may be replaced by the reduced vowel [a].
44 P. K. Pandey / Word accentuation in Hindi

not go into them here. The instances in (6) should suffice for the purpose at
hand.
Since syllable weight is an important factor in accent assignment in Hindi it
is necessary to specify the level at which the syllabic structures are erected in
derivation. In the absence of contrary evidence, we assume that stress rules
apply after rules that affect syllable structures. aavuffyak ‘necessary’, for
example, has regular accent on the penult syllable, which would be impossible
if the pregeminate form were accepted as underlying - aa.vafyuk (cf. 4cii) ~
since then accent would obligatorily be placed on the antepenult, as in the
following words with such a syllabic configuration: bntarim ‘inner’, kariminii
‘young woman’.

4. Analysis

4.1. The data and their description

We find two discrete groups of accentual patterns in our repertory ~ the

core patterns, and their optional variants. The latter characterize the collo-
quial variety and are dealt with at a later stage (section 4.2.2.2). The
following data represent the core patterns. Two general facts have determined
the arrangement. One is the similarity in the conditions surrounding accent
placement working leftwards from both the word edge and the rightward
accent. And the second is the positions that accented syllables occupy in the
word. (-) indicates free variation in the prominence relation between feet.
Thus, vacirdciat (-) ‘event’ should be read as vahrdriat - v&n-d&t.

(74 Words GIOSS Words Gloss

(1) sitaar (a musical Sam&t all
instrument)
aakaar shape sambLndh relation
vaardiat (-) event
(2) p8rivAar (-) family Gmaac$ar ( - ) news
zAniiib$ar (-) (a group of viahiyaat (-) useless, bad
islands)
inaamddar (-) (a personal
last name)

(7b) (li) tit’? date vistu thing

kadlaa black maHltii (a name)
Jaainti peace
P. K. Pandey / Word accentuation in Hindi 45

(lii) kalaa art kamal ’ lotus

(2) aCthi guest pravritti tendency
kiiTBaNu insects daaroogaa sub-inspector
adaalat court of law jakaartaa
aantulee (a personal kaartikeey (-) (a personal name)
last name)
cimpaanjii chimpanzee aandoolan (- ) movement
(3) piatinidhi (-) representative hindustainii (-)
parivartan change anubhuhti feeling, sensibility
paaritooJik award kaarpooreelan (-) corporation

(7~) (li) mahilaa lady antarim internal

(lii) sangati company Jaantanu (a personal name)
(2i) pipariyaa (a place name) deevariyaa (a place name)
mandaikinii (a personal name) aparaajitaa (a personal name)
(2ii) vanaspati vegetation diipaavali (a certain festival)

Note that most of the words in (7) are simple. Only such complex words
have been included whose syllabic configurations would easily predict the
patterns they have; eg, kiiT&Nu (< Skt /kiiTa+ aNu/) (cf. duarodgaa).
Complex words with transparent morphological structures have not been
included, as they are found to behave erratically with regard to accentuation,
and are thus treated separately in section 6.
Note also that the data include most types of disyllabic and trisyllabic
words (excepting EEE and EHL (E = extraheavy, H = heavy, L = light) types
which were not available at the time of data elicitation). Tetrasyllabic words
are limited, as all the possible syllabic configurations in simple words (or even
in complex words with opaque morphological structures) are hard to come
by.
If we now examine the data represented in (7) we find that there are two
factors involved in determining the accent contours:

(1) Syllable peight. Accentuation seems to be sensitive to three degrees

of syllable weight: light (CV), heavy (CVV or CVC) and extraheavy (any
other).
(2) SyZlabfe position. An extraheavy syllable is accented in all positions;
a heavy syllable is always accented in the penult position; a light syllable is
accented in the penult if it is flanked by light syllables, otherwise in the

’ Disyllabic words of L+ H syllabic structure have vacillating accent on either the final or the
initial syllable.
46 P.K. Pandey 1 Word accentuation in Hindi

antepenult position. If neither the final nor the penult are accented, antepe-
nult (heavy or light) is accented.

A consequence of the latter factor is that accent is placed within three

syllables from the right edge or accent. The above facts translate informally
as follows:

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

The statement does not, apparently, account for prominence relation between
accents, since it is generally variable. The tendencies which occasionally show
up in the data (cf. (7b3)) are briefly discussed in 5.3.2.

4.2. A new analysis

4.2.1. Preliminaries
Two approaches to constructing syllable and foot trees have been proposed
in the literature, broadly known as the template approach (e.g., Halle and
Vergnaud (1978) McCarthy (1980), and Selkirk (1980)) and the rule
approach (e.g., Hayes (1981) Steriade (1982)). In the standard metrical
theory of Hayes (1981) the rule for foot construction (see Hayes p. 48)8
erects feet on the absolute weight of rimes for Q-sensitive languages i.e.,
languages in which heavy syllables attract accent. In these languages, if weak
nodes dominate light syllables then feet are disyllabic, otherwise feet are
monosyllabic. In Hindi, although monosyllables are accented, polysyllabic
words have feet formed on the relative weights of constituent syllables, as
should be apparent from (8). A rule-based approach to foot construction in
Hindi, therefore, calls for a mode that takes into consideration the relative
weights of rimes rather than the absolute weight of a rime.
We shall now turn to consider those aspects of the phenomenon which
motivate the new mode of foot construction. Our main concern at this stage
is to formulate the basic principles which appear to govern the phenomenon.
In so doing we shall ignore the conventions of tree geometry developed in
Hayes, and consider only the primitives of the theory stated in (9).

a Henceforth, the page refences to Hayes are to his 1981 publication

P. K. Pandey / Word accentuation in Hindi 47

Pi) Metrical trees are binary-branching.

(9ii) Accent is explicated in terms of binary-branching trees with s w labels
on the sister nodes.

Besides, considering the wide variety of accentual systems examined in Hayes

(1981), and Prince (1983) we assume that, in addition to Quantity, languages
may also be sensitive to Rhythm. 9 Rhythm-sensitive systems generate
bounded feet with fixed s w or w s labels. How this is so will become clear in
a moment.
Metrical phonology (Liberman (1975), Liberman and Prince (1977) Prince
(1983), Selkirk (1984)) equates the organization of linguistic ‘rhythm’ with
that of ‘metrical pattern’ in music. Central to both, in fact, to any organiza-
tion of rhythmic pattern, is the recurrence of pulses.‘O Defining a pulse as
‘one of a series of regularly recurring, precisely equivalent stimuli’, Cooper
and Meyer (1960) explain its arrangement in ‘metrical patterns’ thus :

‘Meter is the measurement of the number of pulses between more or less regularly recurring
accents. Therefore, in order for meter to exist, some of the pulses in a series must be accented -
marked for consciousness - relative to others. When pulses are thus counted within a metric
context, they are referred to as bears. Beats which are accented are called “strong”; those
which are unaccented are called “weak”.’
(Cooper and Meyer (1960: 3). Quoted in Selkirk (1984: 10))

The recurrence of strong and weak beats takes place at a hierarchy of

levels. To illustrate, consider the following diagram with x’s standing for
pulses, and the underlining indicating accent. Strong beats at lower levels
occur at the next higher level.

(10)
X

X X

x x x x

~XXXXXXX
- _ _

p Prince (1983) as discussed in section 7 to follow, has endmostness as the third factor crucial
to the generation of accent patterns in the grid. Within the tree framework endmostness does not
play a critical role in the formation of foot structures. And so we ignore a discussion of it here.
Endmost patterns are generated by a Quantity-insensitive, Rhythm-insensitive, unbounded, left-
or right-branching foot formation rule. The ‘first heavy or initial’ patterns are generated by a rule
forming Q-sensitive, R-insensitive, unbounded, left- or right-branching feet.
lo The present discussion of rhythm in music is based on Selkirk (1984: 10).
48 P. K. Pandey / Word accentuation in Hindi

(10) is a grid-based representation of a metrical pattern. It comes fairly close

to the metrical grid representation of linguistic rhythm in Liberman (1975),
and Liberman and Prince (1977) into which we shall not go further here. A
detailed discussion of the grid-based representation of accent is presented in
section 7.
Within the metrical tree framework an R-sensitive metrical pattern (the
abstract pattern which mediates syntactic representations and the metrical
grid)’ 1 is generated in terms of recurring bounded feet with fixed s w or w s
labels. Note that R-sensitive feet must be bounded in order to guarantee the
recurrence of accent.
We are unable to say at this stage if bounded R-sensitive feet must be
binary only or can also be ternary in size. We shall, however, see that its
sensitivity to rhythm requires the foot construction rule in Hindi to erect
binary s w feet. But its simultaneous sensitivity to quantity gives rise to
ternary feet as well.

4.2.2. Informal discussion

4.2.2.1. Core patterns. Consider the patterns in (7b) (excepting 7blii) repre-
sented below:

(1 la) vidhi ‘method, law’ jantu ‘creature’

.,
cnnu ‘sugar’ aditi (a fem. personal
name)
sucaaru ‘smooth’ Jraddhaalu ‘devotee’
aavaaraa ‘loafer’ hifaazat ‘care, security’
(llb) anumati(-) ‘permission’ hindustaanii( -)
6nubhhuti ‘sensibility’ paaritoofik( - )
baaluusaahii( -) (a kind of sweets)

Because of the commonality of circumstances surrounding both final and

non-final accents the pattern in (11) can be designated as trochaic. From the
trigger the first syllable is unaccented, and the second is accented. The general
condition of syllable structure observed is that the accented penult syllable
has either equal or greater weight than the unaccented final syllable. When
that condition is not met, other accentual patterns result (12a = 7a;
12b = 7cl&c2i):

I1 Note the difference in the metrical phonological use of the term ‘metrical pattern’ from
its musical use.
P. K. Pandey / Word accentuafion in Hindi 49

(12a) (i) vivaah ‘marriage’ aasair ‘sign,

indication’
gaandhaar( -) (a place name)
(ii) itihais ‘history’
aaJutooJ( - ) (a personal
name) balaatkaar( - ) ‘rape’
bandanbaar( -) ‘a string of
leaves or flowers’ aalmaarii( -) ‘almirah’
(12b) suvidhaa ‘convenience’ kookilaa ‘cuckoo’
khamLriyaa (a place name) khampariyaa (a last personal
name
histiiriyaa ‘hysteria’

The principal core patterns considered so far are the penult (1 l), final (12a),
and antepenult (12b). Words of all the syllabic structures representing them in
(7) have been considered except of the kdua type in (7bli), and of the sriqgati
and vancispati types in (7clii & c2ii). The patterns in them have been included
among the core patterns on the ground of productivity. However, they will be
discussed separately in the following section. We shall try to demonstrate
below how all the three patterns are explained on the basis of a simple
principle of grouping of rimes and of the representation of syllables in (2) and
(3).
Let us consider the final, in fact, ubiquitous, accent on extraheavy syllables
first, An explanation for the special behaviour of extraheavy syllables is
available on the basis of their representation suggested in (3), consisting of
two rimes, and the above observation about the propensity of Hindi towards
trochaic accent. It is easy to see that they have an inherently trochaic pattern,
the left node being stronger than the right:

(13) aakaar
VI
SW

V
In fact, all the core patterns in (7) considered so far can be explained with the
help of a principle that seems to determine accentuation in Hindi, namely, the
principle of Conjugation, stated in a tentative form below:

(14) The principle of conjugation

From the trigger (i.e., word edge or accent on the right), conjugate the
50 P.K. Pandey / Word accenruaiion in Hind1

first rime with the second assigning them s^w labels, if the second rime
has equal or greater weight.

A foot within this analysis is defined in terms of its rimes which are in
accordance with (14). The rimes which directly constitute a foot are called
Conjugate rimes. A full classification of rimes according to their function in
metrical structures is proposed in section 5.2. Reverting to principle (14), note
that it is dependent on two factors - namely, rhythm and quantity. Rhythm
consists in the recurrent prominence on the left syllable reflected in the s w
labelling, and quantity resides in the sensitivity of the principle to the weight
of the conjugate rimes.
With the help of (14), we can now see how the antepenult pattern in (12) is
brought about. The final in it, being stronger in weight than the penult, does
not attach to the penult, but instead to the superordinate node dominating
the penult and the antepenult which together are equal to or stronger than
the final:

(15) mahilaa pipariyaa

I IV I I v
sww SW w

!!J y
The grouping of the final with the superordinate node dominating the
penult and the antepenult rimes, and of the penult rime with the antepenult
rime are both in keeping with the principle. Note that when it is not possible
to erect feet according to the principle, as for example, in disyllabic words of
L+ H type like kamal, then both the factors of rhythm and quantity are
found to compete for dominance, leaving uncertainty in accent patterns.
Words of L + H type have a vacillating pattern, as noted in footnote 7: krimal
N kamhl. However, such uncertainty does not exist in an analogous syllabic
configuration word-internally, as, for example, in ctimatkahr. In it the
branching foot on the final syllable prevents accent on the immediately
preceding syllable (in order to avoid accent clash). So the second accent
invariably falls on the initial syllable. In such cases, rhythm wins over
quantity. Accent on L + H disyllabic structures can be accounted for by the
following Default Rule:
P. K. Pandey / Word accentuarion in Hindi 51

(16) Default Rule

Form a binary branching s w or w s foot (avoiding accent clash).

The informal account presented thus far takes care of all the patterns
considered so far with the exception of the antepenult pattern reproduced in
(17).

(17i) saggati samprati

(17ii) vanaspati diipaavali
(17iii) Jaantanu

The contours in (17) offend the principle of conjugation according to which

accent would fall on the penult syllable on account of being equal in weight
to the final. In fact, the words originally had that pattern in Sanskrit: saygciti,
samprdi, paaNini. The pattern has been lost in all varieties of Hindi (except
the highly sanskritized variety), which now have the antepenult pattern
instead.
There are two ways of accounting for the contour in (17) - by modifying
(14) or with the help of an accent retraction rule which is needed in any case
for alternative variant patterns introduced in (18). We shall consider these
alternative accounts below, but shall reject them in favour of filters.

4.2.2.2. Accent retraction: colloquial patterns. Some of the core patterns,

(19) have, as noted above, an alternative optional pattern, (18) in colloquial
Hindi :

(18) saalaanaa
kamalinii
keesariyaa
(19) saalaanaa
kamalinii
keesariyaa

The pattern in (18) can be derived by an initial accent shift rule, which is
subject to the same condition as (14) as is apparent from the following
impossible patterns:

(20) *kinaaraa cf. saalaanaa (N saalaanaa)

*mariicikaa ‘mirage’ keesariyaa (N ktesariyaa)
52 P. K. Pandey 1 Word accentuation in Hindi

*suvidhaa samiti ‘committee’

*kamalinii anumati (-)

The condition in (14) and on the initial shift in (18) can be stated generally as
(21):

(21) The weight of the rime in strong position in a foot may not be less
than the weight of its conjugate in weak position.

Note that besides the patterns in (19) there is at least one more core pattern,
(23) which has a regular optional pattern, (22). (22) however differs from (23)
in segmental representation:

(22) aditii
apituu ‘but’
(23) aditi
apitu

The final long vowels in (22) are derived by rule in colloquial Hindi:

(24) Final Vowel Lengthening (FVL)

7 A
v-+v V/y]

FVL, like other rules affecting syllable structures, for example, Schwa Dele-
tion and Consonant Gemination, is extrinsically ordered before the accent
rules in colloquial Hindi. The rule is not extrinsically ordered with respect to
the other two segmental rules.
FVL is missing in formal Hindi which has the pattern in (25). Some more
examples of the formal colloquial difference with regard to final vowel length
are given below:

(25) Colloquial Formal

patii ‘husband’ pati
parantuu ‘but’ parantu
anumatii anumati
Jankaaluu Jaagkaalu
-Jankaaluu *Jan kaalu
P. K. Pandey / Word accentuation in Hindi 53

Although the pattern in (22) may be treated as a result of the initial shift rule,
its regularity, in fact, is on account of FVL.

(26) Formal Colloquial

aditi aditii
LLL Rime Projection m Rime Projection
s w Principle (14) sw Principle (14)
V V
aditii
LLY_
s w w later rules
V
S

Turning now to the data in (17), it appears that they could be explained in
terms of the initial shift rule, which is independently needed in the grammar.
As noted above, the original pattern in those words was penult, the antepe-
nult pattern being an innovation, perhaps a consequence of the shift rule
which seems to be applying obligatorily here. There are, however, two ways
in which (17) and (18) systematically differ.
Firstly, the patterns in (17), as compared to those in (18), are, as stated
above, obligatory, and this difference is not neatly statable in terms of a shift
rule. The obligatoriness of (17) is not on account of the initial syllable being
stronger in weight than the accented syllable, as evident from the pattern in,
for example, cDorcisiyaa, khamphriyaa, but rather of the antepenult position of
the heavy syllable (with penult and final light syllables). In the absence of an
independent explanation available for this fact, it cannot be stated in the
form of a shift rule.
Secondly, the retraction in (17) is not restricted to initial syllables, as it is in
(18). In order to accommodate the patterns in (17), for example, vanhspati,
diiparivali, we may order the shift rule after the foot formation rule within a
cycle. The difficulty with this proposal is that it would introduce considerable
confusion in the grammar, since the retraction rule would have to be allowed
optional application after the formation of each foot even in instances such as
the following: hindustarinii (-), bnumciti (-). Since patterns like *hindtistaa-
nii, *antimati do not exist, such a provision should not be admissible in the
grammar. Note that at any rate there is a clear difference between the feet of
54 P. K. Pandey / Word accentuation in Hindi

(17),withH+L+LorE+L+Lstructure,andthoseof(lS),withH+H+H,
L + L + L + H or H + L + L + H structure. Only the latter are optional.

4.2.2.3. ModiJication of the foot construction rule. An alternative account of

the pattern in (17) is possible by allowing its feet to be generated by an
extended version of the foot construction rule.

(27) Principle of conjugation (extended tentatively)

From the trigger, conjugate the first rime with the second assigning
them s^w labels if (a) the second rime has equal or greater weight, and
(b) the third rime does not have greater weight than the second.

In case the first rime fails to conjugate with the second, the latter, not being
stronger, would necessarily conjugate with the third.
While the extended principle of conjugation (27) has the effect of separat-
ing the obligatory pattern of (17) from the optional pattern of (18) it has
been made rather cumbersome with the addition of clause (b). As clause (b) is
needed to account for only one kind of foot structure, namely,. . H + L + L
(or E + L + L), we should look for a way in which that foot structure could be
explained separately from those created by clause (a), so that (29) can
preserve its natural, lawlike form.
There is all the evidence to believe that feet of H + L+ L structure have
become fixed in the language; although historically derived from the penult
pattern, they have come to have the non-alternating antepenult pattern. It
would therefore be better to state this fact in the analysis in order to bring it
closer to actuality.
We propose the following filter to take care of the obligatory pattern in
(17):

(28) *C

(28) prevents the formation of binary feet on non-branching rimes when

preceded by a branching rime in the antepenult position.

(29) *sangati (sangati) *vanaspati (vanhpati)

P. K. Pandey / Word accentuation in Hindi 55

In addition, the grammar includes another filter (30):

(30) *c
I
[...Ii...]
I
I.e., no degenerate foot in polysyllabic words.

which functions to prevent the formation of non-branching feet on single

syllables in polysyllabic words, as underlined below:

(31) *khcimp6riyaa *bdhaanaa *Mook ‘light’

Keeping the constraint in (21) in mind, the initial shift rule may now be stated
as follows :

(32) Initial Accent Shift Rule (IASR) (optional)

Retract accent onto the word initital rime.
Some examples of the derivation of accentual patterns on account of the
conjugational principle (14) and the constraint in (21) (to be formally stated
in the following section), the filters (28) and (30), and the IASR (32) are given
below :

(33) kamalinii daaroogaa

. ..
a aiu aa 00 aa
1 11 v R V V V R Rime Projection (14)
sww

F
kamalinii daaroogaa IASR (32)
IR =R
s www sww
56 P. K. Pandey / Word accentuation in Hindi

(34) *vibhiiJikaa *sitaaraa

.

i R i v 7 R Rime Projection (14)

SW w s w

vibhiiJikaa sitaaraa

s ww w SW w IASR (32)

(35i) *saggati (35ii) *bahaanaa

G R Rime Projection Ivv R RP

s w
Principie ( 14) Principle (14)
F Filter (28) --L F Filter (30)
V w s
Other rules Other rules
Wd Wd

5. Theoretical implications, apparatus, and rules

5.1. Theoretical implications

While all the patterns discussed in the analysis can be reasoned to naturally
follow the rules,i2 the rules themselves and the feet they generate are
counterexamples to at least three claims made in the standard (post-Hayes)
metrical theory, namely, (a) natural foot size, (b) uniformity of branching,
and (c) extrametricality, on which the other two claims are crucially depen-
dent.

I2 The rule for foot formation is formally presented in (49).

P. K. Pandey 1 Word accentuation in Hindi 57

5.1.1. Ternary feet

The ternary foot in words like mbhilaa, kamdinii is shown to be a result of
the principle of conjugation. There are two ways in which the ternary foot
has been treated in the literature. Halle and Vergnaud (1978) McCarthy
(1979, 1980) and Selkirk (1980) accept it as a taxonomic fact along with
binary and n-ary feet. They do not attempt to give a rational explanation of
it. As McCarthy (1979) observes in connection with the positing of ternary
feet in Cairene and Damascene Arabic:

‘It is easy to see why there might be binary and n-ary feet exclusively, but it is difficult to
understand why ternary feet should be allowed but not I-ary or 5-ary. Obviously this presents
no difficulties to our taxonomic survey, but it does militate against the construction of a
relatively elegant theory of foot structure.’ (McCarthy (1979: 45))

Hayes (1981) in his theory of natural metrical trees, marks ternary feet as
unnatural, and allows for only two unmarked sizes: binary and unbounded.
Ternary feet in his theory are merely surface structures, accounted for with
the help of extrametricality rules (or other features of abstract representation)
which exclude peripheral elements (segments, syllables, morphemes) from
metrical scanning in the erection of feet. The ternary foot in our analysis is a
natural result of the mora-sensitive conjugational mode of foot construction.

5.1.2. Unijormity of branchingness

Validity is claimed for rule-governed extrametricality in Hayes’ theory on
other universal grounds as well. A scrutiny of an inventory of foot types (see,
for example, Halle’s inventory in Hayes (p. 70)) reveals that in languages in
which the projection of rimes as branching matters (i.e., in Q-sensitive
languages), accent is assigned to all the branching rimes except the one in the
final position. Only in the final position is a heavy syllable labelled ‘weak’. If
the branchingness on the final syllable could be ignored (with the help of
extrametricality), then it is easy to see that foot structures can be shown to be
either left-branching or right-branching, depending on the direction in which
feet are assigned in a language: left-branching if feet are assigned from right
to left, and vice versa. The upshot of this provision is that using it Hayes is
able to put forward a restrictive universal theory of word accent systems in
terms of branchingness conditions in the following way.
In order to incorporate the conditions about uniformity of branchingness
on metrical trees, Hayes makes use of the abstract labels d and r for s and w
respectively. Trees have the labels d r or r d in the abstract. With the help of
these labels Hayes achieves an important restrictive result. It follows the
58 P. K. Pandey 1 Word accentuation in Hindi

statement of the law about recessive nodes (Hayes (1981: 41)) reproduced
below :

(36) Recessive nodes may not branch.

The immediate result of the law is that metrical trees are now uniformly
left-or right-branching as in (37); structures in (38) are prohibited (see Prince
(1983: 56)).

(37a) d... (37b) d

A A
d r r r r d

(384 A (38b)
*d r
A
d r d

d and r are later converted into s and w, with the marked option (e.g.,
Tiberian Hebrew, Hayes (p. 105)) d = s iff it branches.
Now consider the following data from eastern Hindi presented earlier:

(39a) saalaanaa (39b) hindustaanii( -) (39~) atithi

dandbolan( -) kaarpoorteJan( - ) anumati( -)
pipariyaa kilimanjaaroo( -) deevariyaa

The above data pose problems for two claims in support of the extra-
metricality thesis- (i) ternary and binary feet with weak heavy syllables occur
in both final and non-final positions, and (ii) these feet are accounted for
without the help of an extrametricality rule. The former poses a challenge to
the restrictive universal claim that ternary feet are unnatural, a mere surface
structure, and, to the uniformity of branchingness constraint which is now
taken to be almost axiomatic. In an effort to make the present analysis
conform to the standard metrical theory we shall find that while the claim
regarding bounded feet being binary in size can be maintained without any
serious damage, the notion of uniformity of branchingness as a universal
condition in the tree theory cannot be maintained. Consider the following
patterns rearranged from (39):
P. K. Pandey / Word accentuation in Hindi 59

(40a) saalaanaa (40b) atithi

hindustaanii( -) anumati

What is common to both the (a) and (b) patterns is the equivalence of weight
between the final and penult syllables, and the trochaic rhythm. Note that if
the penult has greater weight than the final, then accent is not affected e.g.,
sucaciru; but if the penult has less weight then the pattern changes: it is no
more penult but antepenult:

(41) kavitaa kamalinii kaaminii khampariyaa

It seems to us that the principle of conjugation stated in (14) essentially

captures the law that determines accentuation in Hindi. An extrametricality
analysis following (36) on the other hand, would be unrevealing in many
ways. Firstly, it would fail to show that the same rule determines accentua-
tion in words in both the sets in (40). Secondly, it would fail to account for
the antepenult pattern in (41) especially keeping in view the pattern in (40b).
Thirdly, it would fail to give a simple account of word-internal feet having
weak branching rimes, as, for example, in hindustabnii(-). Thus (36) must
either be abandoned or modified to accommodate the conjugational analysis.
We adopt the latter course.
On reflection concerning the law in (36), it becomes apparent that it has no
‘a priori’ justification. The law assumes the distribution of accent to be
identical with heavy syllables in Q-sensitive languages. In a language in which
accentuation is sensitive to both quantity and rhythm, it may not be attested.
The analysis of accent motivated here shows that this indeed is the case.
Hindi as described here is not a purely Q-sensitive language. It involves both
quantity and rhythm in foot structure assignment. In other words, it is a QR-
sensitive language. The explication of such a system, we have found, must
overrule the law about branchingness in (36), which can be abrogated as
applicable to Q-sensitive systems only, systems on which it is inductively
based. The law does not apply to a QR-sensitive system in which both
quantity and rhythm cooperate to form feet, and which is subject to a
constraint stated in (21) and restated in (43).
The theoretical findings of the analysismay be summarized in the following
statements:

(42) Recessive nodes may not branch only in Q-sensitive systems.

(43) Recessive nodes may branch iff the dominant node branches in QR-
sensitive systems.
60 P. K. Pandey 1 Word accentuation in Hindi

(44) 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.

Having revised the claims of the standard metrical theory in order to

accommodate the present analysis we now turn to the statement of the foot
construction rules.

5.2. Apparatus

In the presentation of rules, we shall use a new set of terms. It should be

useful to clarify the references of the terms, all of which denote a new class of
rimes.
The significance of the role of rimes in foot assignment has been generally
recognized. But so far they have been found to be crucially distinguishable in
terms of branchingness alone. As we have found compelling reasons to use
the conjugational mode of foot construction, and to refer to the representa-
tion of extraheavy syllables in (3) in this analysis, we find the classication of
rimes in terms of branchingness alone inadequate for the presentation of our
rules. We therefore propose here other (minimally) new distinctions between
rimes on the basis of both their structure and function in the formation of
foot and word trees. We find the distinctions useful in the statement of our
rules, and we think they could perhaps have some use in the general metrical
theory as well.
Rimes which directly constitute a foot are termed conjugate rimes. A rime
which does not conjugate with another rime in the formation of feet is a stray
rime. Stray rimes are adjoined to the metrical tree by the universal Syllable
Adjunction convention (see Hayes (1982: 235)). Stray rimes may be adjoined
to the foot if they have been scanned and excluded by rule of foot construc-
tion, or they may be adjoined to the word if they have not been scanned by
the foot construction rule. In the following representations

(45a) mahilaa (45b) coorasiyaa

sww w SW w

Wd
P.K. Pandey / Word accenluarion in Hindi 61

the weak rimes -aa in both the words, which are adjoined to the foot, are,foor
adjunct rimes, and the weak rime -DD adjoined to the word is a word adjunct
rime.
A distinction between conjugate and adjunct rimes is found to be useful
within the general framework of metrical theory which adheres to the claim
that feet are either binary or unbounded. The ternary feet in (45) are thus
shown to be a derived type. The present analysis explains ternary feet as those
which have the third syllable as an adjunct and not a conjugate.
The difference between foot adjunct and word adjunct rimes is that the
latter follow while the former are a part of the foot construction rules.
Consider the following evidence for the claim that the foot adjunct rime is a
part of the foot construction rule.
The presence of stray rimes within the scanning limit of feet prevents the
obligatory retraction of accent in words of (45b) type. The absence of a stray
rime within the scanning scope of the foot construction rule obligatorily
retracts accent in (46):

(46) saggati *sarjgati

s ww w SW
V V
J
S
1
S

In order to account for the feet in (46) and (45b) in a consistent way it would
be necessary to adjoin the stray rime to the conjugate rimes within the foot.
Only then is it possible to account for the obligatory retraction in (46). The
behaviour of the adjunct rimes in (45b) provides evidence against treating
ternary feet as mere surface structures, by which is meant that they are of no
phonological significance. We have seen that the ternary foot in (45b), as
compared to the binary foot in (46), prevents the inititial accent shift from
being obligatory.
On the basis of their structure in the syllable, rimes are also distinguishable
as absolute and cognate. Cognate rimes have the same source in the syllable,
and are thus inherently conjugate. Absolute rimes belong to different sylla-
bles. A distinction between cognate and absolute rimes makes convenient the
statement of foot construction rules which give precedence to cognate rimes
for conjugation. The distinction is brought out through the projection of
cognate rimes being related with a horizontal bar:
62 P. K. Pandq~ i Word accentuution in Hindi

(47) kaar ‘car’ havaalaat (-) ‘jail’

vl IVVl
The distinction makes it possible for the foot formation rules to operate
exclusively on the projection of rimes. The rule of foot formation which
required a reference to the extraheavy syllable earlier will now refer to it in
terms of its cognate rimes. The way the rimes of extraheavy syllables are
projected makes explicit their special status.

5.3. Foot formation and stress interpretation rules

5.3.1
The crucial rules involved in the erection of feet, which precede the filters
(28,30), the IASR, and word tree construction rules, can now be stated as
follows :

(48) Rime Projection (RP)

Project rimes of syllables.
(49) Foot Formation (FF)
(a) Form a binary-branching s w foot on cognate rimes.
(b) Form a QR-sensitive, binary-branching s w foot on 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.

5.3.2. Word trees

Thus far, we have dealt with the assignment of accent ‘per se’, and not
discussed the relation of prominence between accents, i.e., the assignment of
primary and lower degrees of stress. The reason for this is, as noted
repeatedly above, that the relative prominence between accents is variable.
We have indicated this variability throughout the data using (-) against
diaccentual words. There is a tendency however to favour prominence on one
of the syllables, as indicated in (7b3) and below:

(50) riandoblan ‘movement’, hhriartPendu (a personal name), itihiras ‘history’

It appears that the factor which tends to favour prominence between feet is
their total weight. This tendency, however, is in conflict with another ten-
dency towards word initial accent noted above, which may favour the
P. K. Pandey 1 Word accentuation in Hindi 63

assignment of primary stress on the first syllable (i.e., in the first foot). The
word tree formation rule being indeterminate at this stage in the development
of the language, it may be stated as follows:

(51) Stress Interpretation Rule (SIR)

Each foot has some degree of stress.

We can now give the list of rules that generate the stress patterns discussed in
the analysis:

(52) Rime Projection (48)

Foot Formation (49)
Initial Accent Shift Rule (32)
Filters on foot formation (28,30)
Default Rule (16)
Stress Interpretation Rule (51)

Some more examples of derivations using the rules are presented below:

(53i) sitaar vaardaat samaacaar

i aar aar aat a aa aar

1111 vI Iv\/1 R Rime Projection

Y sl”yl” Y F Foot Formation (49a)

sitaar vaardaat samaacaar

1111 vlvl ULln R Default Rule

w SW SW SW

Stress Interpretation Rule

Word Tree Formation
Wd Wd Wd

(53ii)
\ anumati hindustaanii hindustaanii
vvvv
I

LLLL JJJUR.P
SW SW swsw swsw

= m =F FF(49b)
s w S W W S

V SIR
Wd Wd Wd Wtf
64 P. K. Pandq 1 Word accentuution in Hindi

(53iii) sarjgati vanaspati sarTifikeeT

an ai a as a i ar i i eeT
v II JLlL J.(=!Q-RRP
s ww s wwsw

sangati vanaspati sarTifikeeT

v II JLlL VIIVI,
s ww w s ww s wwsw

J
V
S

Wd Wd Wd

6. Morphology and accent

We have not yet considered the role of morphology in accentuation. The

role is uncertain in many cases. It is not our purpose here to go into the
details of the nature of interaction between word-formation and accentua-
tion, as that will take us a little away from the present field. We shall be
content to show that while the patterns in morphologically complex words
can be explained once the nature of the interaction between word formation
and accentuation is clear, the rules of accentuation formulated in the present
analysis may not be revised. Observe the patterns in both the groups in (54)
(59, and (56):

(54a) taakatvar (taakat + var) ‘one who is strong’ *taakatvar

baalakpan (baalak + pan) ‘childhood’ *baalakpan
jaiduugar Cjaaduu + gar) ‘magician’ *jaaduugar
(54b) hamlaavar (hamlaa+var) ‘one who attacks’
1aRakpan (1aRak + pan) ‘boyhood’
soodaagar (soodaa + gar) ‘trader’
P. K. Pandey 1 Word accenruation in Hindi 65

(55) Stem Future Gloss

(4
gas gaaeegaa (III pets. sing.) *gaaeegaa ‘sing’
khaa kha’aoogee (II pets. sing.) *khaao’ogee ‘eat’
gir giriiGgaa (I pers. sing.) *girtigaa ‘fall’
(b) Stem Past + Caus. Gloss
dooR dovRa’ayaa ‘run’
baTh bazThaayaa ‘sit’
gir giraayaa
(56i) pr6tidhvani( -) (prati + dhvani) (56ii) pratiddhvani ‘echo’
pratikriyaa( -) (prati + kriyaa) pratikkriyaa ‘reaction’
p”alsvaruup (phal + svaruup) phalLssvaruup( -) ‘as a result’

(54) shows the derivational suffixes -var, -pan, and -gar behaving erratically in
(a) and (b). In (55) the Future non-causative and Past causative forms of
verbs have contrary patterns in (a) and (b) respectively. (56) has forms
without or with consonant gemination across morphemes in (i) and (ii)
respectively, affecting accent placement. In none of these cases is there a need
to modify the rules considered so far or to posit another rule.
We have not gone into the details of the problem of relation between
morphology and accent sufficiently to make it formulable. We do hope,
however, that further exploration in that area will clarify the nature of that
relationship without invalidating the principles arrived at in this analysis.

7. A grid-based analysis

7.1

The claim of the universal metrical tree theory of accent that tree geometry
uniquely determines the constraints on the form of stress rules has been
seriously challenged in favour of a grid-only theory by Prince (1983) and
Selkirk (1984).
A close look into the accent systems examined in Hayes shows that if we
accept the hierarchic prosodic levels and the device of extrametricality then
the assignment of accent is found to be subject to three independent factors -
endmostness (at the syllabic or foot level), rhythm (left or right node strong
in binary-branching trees), and quantity-sensitivity, all three of which, the
grid theorists claim, can be explicated without the tree theoretic apparatus
such as binary-branching trees, s/w labels, branchingness conditions, etc.
Prince translates the three factors into the following rules ~ End Rule, Perfect
66 P. K. Pandey / Word accentuation in Hindi

Grid, and Quantity-Sensitive rule, which, with parametric variation,13 may

generate the grid-structures representing the accent patterns in words. The
basic mechanism is to assign the terminal grid positions for all syllables and
then to apply a rule to give the next grid structure. We shall not go on to
exemplify the working of the grid-only apparatus. The reader unfamiliar with
the theory had best turn directly to the expositions in Prince or in Selkirk.
Note that the conjugational tree analysis of Hindi proposed above makes
reference to all the possible tree forms of feet in Hindi, not considered within
the grid-only theory. In fact, Prince justifies his move against the tree theory
on the ground, as noted above, that it does not exploit the full possibilities of
tree geometry. Consider, for example, the fact that in Q-sensitive languages
heavy syllables never occupy weak positions. Prince tries to show that the
property of heavy syllables to attract accent to themselves is the result of
some other factor than its tree representation - namely, its bimoraic (= bipo-
sitional) nature and the intrinsic prominence of the first mora, represented as
L in the grid.
However, within the present analysis we have seen that for a QR-sensitive
language such as Hindi the theory of foot form reduced to ‘w nodes may not
branch’ does not apply. In such a system labelling is independent of branch-
ing. Even the representation of the extraheavy syllable makes full use of the
potentialities of tree form. It is in the light of this that we propose to discuss a
grid theoretic account of the phenomenon. The possibility that we shall try to
explore is whether the grid-only apparatus can handle the data in which
heavy syllables do not enjoy an absolute status and are subject to similar
constraints of parity as light syllables. (In Hindi, as we have seen, a light
syllable can be strong if its weak conjugate is light, and a heavy syllable can
be weak if its strong conjugate is heavy.)

7.2. Analysis

7.2.1
It is expected that the algorithm of rules for Hindi proposed in Prince
(p. 77) on the basis of Hayes (pp. 79-81) does not generate the patterns not
included in Hayes. Our effort to formulate a set of rules within the grid-only
framework to explain the patterns in (9)14 has met with failure. However, the
following rules seem to work fairly well for a large part of the data:

I3 For example, PG may apply from Right to Left or L to R.

l4 The optional retracted pattern (18) of colloquial Hindi, however, is neatly explained in
section 7.2.3.
P. K. Pandey / Word accentuation in Hindi 67

(57) Quantity Sensitivity (QS) - with bipositional representation of heavy

syllables.
Perfect Grid (Right to left; trough) (PG(RL, tr)) - trough first Grid
Construction from right to left.

Note that the rules in (59) do not generate stress, but accents in the abstract.
It is assumed that in case of more than one accent, degrees of stress are
assigned by a word level End Rule. Since the End Rule lacks a definite
character we shall not discuss it further.
The rules in (57) give rise to derivations such as the following:

(584 X
kamalinii QS x PG(RL, tr) x x PG(RL, tr) x x
- > +
x xxxx xxxxx xxxxx xxxxx

(58b) X
keesariyaa 2 x x PG(RL, tr) x x x PG(RL, tr) x x x
. t
xx x x xx xx x x xx xx x x xx xx x x xx

The derivation of keesciriyaa shows that PG is with a Forward Clash

Override (FCO) option. The FCO in a rule provides a license to the system to
permit clash in the forward direction of the linear mapping of rules, in the
present case leftward.

7.2.2 The algorithm of (57) has a number of problems which are discussed
below along with the derivations of forms.

7.2.2.1. Number of levels in the grid. The derivations in (58) show that PG
must apply more than once. A consideration of the overall data however
reveals that it is not clear as to how many times PG must apply, and (b) at
what levels.
In (58) PG applies at the first and the second levels. However, in words like
p&iNdti( -) it must not apply at the second level since we shall then be
deriving fixed primary stress. On the contrary, in hindustaCi(-) it must
apply at the second level to determine the two accents in the word. Note that
in hindustah-nii(-) PG can apply further on the third level, but it should be
prevented from applying for the same reason as in the former word on the
second level.
The two instances point to a major problem in the grid-only theory - the
nature of correlation between the grid levels and the prosodic levels. We find
68 P. K. Pandey / Word accentualion in Hindi

that the depth of representation for feet in words with all heavy and all light
syllables which differs in the grid, (a), is the same in trees, (b):

(59a) pLriNati ( N ) x x
xx xxPG(RL,tr) xxxx
>
X X

hindustaanii (-) x x x x x x x x
xx xx xx xx QS xx xx xx xx PG(RL, tr) xx xx xx xx
- ,

(59b) pariNati (-) hindustaanii( -)

II WR
SW SW swsw
VVP VVP

Although they differ in syllabic structure the two words have identical
patterns. This is brought out accurately in the tree by means of the branch-
ingness of rimes, and identical hierarchical structures. The grids however
require an extra hierarchic level for the heavy syllables which do not evince
an absolutely special behaviour in the language.

7.2.2.2. Separateness of factors. The rules, as they are stated, do not explain
the behaviour of extraheavy syllables, which have ubiquitous accent. It is not
clear to me how they are to be represented. If treated as consisting of two
rimes - one heavy and the other light - as in the tree, then the following
pattern is easily derived:

(60) sitaar QS x PG(RL, tr), same

-
xxxx x xxx

But the pattern in a word of the structure aabh&r cannot be derived:

(61) X
*aLbhaar QS PG(RL, tr) same PG(RL, tr), x x
- ,
xx xxx xx xxx xx xxx

Recall that the PG must apply more than once and at different levels for the
pattern in (58). The rules in (57) and the representations of extraheavy
syllables as consisting of two rimes do not predict the correct pattern on
P.K. Pandey / Word accentuation in Hindi 69

aab%hr as they do in sit&r. What is being overlooked here is the intrinsic

ability of extraheavy syllables to be accented, which we found to be regularly
explained in terms of the tree representation.
A way out is available within the theory if we allow the QS rule to posit
three positions for extraheavy syllables on the depth parameter of grid also.
In that case extraheavy syllables can be shown to be ubiquitously accented:

(62) X
aabhaar QS x
-
xx xxx xx xxx

But even then the contour in the following word which has the same
representation at the syllabic level as aubhucir cannot be described for the
same reasons as the derivation in (61):

(63) X

*Jar)kaalu z x x PG(RL, tr) , same PG(RL, tr), x x

xx xxx xx xx x xx xx x

The pattern in faykadu ‘suspicious’ l 5 is critical in the context of the repertory

in (7a). Penult and antepenult patterns on light syllables (e.g., utithi, cinu-
m&i(-)), mbhiluu, pipciriyuu) and penult pattern on heavy syllables (e.g.,
kuciluu, duurodguu) justify both the QS and the PG(RL, tr) rules in (57). The
same rules, however, fail to predict the correct pattern in words of the
structure in (63).
What is the factor in a word like Juykucifu that the rules in (57) fail to
capture, but must capture? It is, as we have explained in relation to the tree
analysis, the situation in which the penult heavy syllable meets the demand of
both quantity and rhythm. If a restriction could be imposed for preventing
the PG from retracting accent from a syllable that already meets the
descriptions of QS and PG, as does the penult heavy syllable, then accent on
penult heavy syllables can in all circumstances be predicted. But in the
context of the rest of the data no such restriction on the PG is possible. What
the grid-only framework is capable of doing is to separate out the factors
which influence accentuation, but it has no way of combining the various

I5 Note thatJunk&lu has the optional antepenult pattern in colloquial Hindi, but with the final
vowel in the word long-J&&&u, like da&oogaa (-duarodgaa). finkablu, with the final short
vowel in the formal variety, does not have the antepenult pattern.
70 P.K. Pandey 1 Word accenruation in Hindi

factors into one. Trees, as we have tried to demonstrate, are inherently

capable of simultaneously capturing both the factors ~ quantity (through the
branchingness of rimes), and rhythm (through the binariness of foot trees).
But in terms of grids, on which positions are filled separately by different
descriptors and on which PG applies freely, quantity and rhythm cannot be
made to work in combination. As a result, the essential fact about the penult
heavy syllable (and the extraheavy syllable) ubiquitously drawing accent to
itself must remain obscure.
It is also a consequence of the separateness of various factors that the rules
in (57) give us the final pattern in disyllabic words of L + H type

(64) kamal E x
x xx x xx

The fixed final accent in such disyllabic words is an undesirable result, as

there is vacillation between the penult and final patterns, and in fact, the
penult pattern is commoner than the final. Deriving initial accent on it with a
shift rule would not be possible since the shift rule, as we have seen,
maintains the parity between the weights of the syllables involved. We have
found that the conjugational tree analysis predicts this situation attested by
the data.

7.2.3. Stress retraction

The optional initial accent pattern (18) is easily derivable if an End Rule is
formulated which picks the initial syllable for accent, and avoids clash. The
rule applies if PG does not apply; which means that the PG may be optional.
Since ER applies at the second level, corresponding to the level of C, having
stressed elements, as compared to the first level of o, having unstressed
elements, it may be characterized as End Rule (Initial, Foot) (ER (IJ)),
notwithstanding the fact that for Hindi the foot does not always correspond
to the second level. As accent is not retracted to the initial light syllable when
the peninitial syllable is heavy, the rule is not marked for the FCO option. All
the optional patterns are now accounted for.

(654 X
daaroogaa z x x x PG(RL, tr), DNA (same) ER(1, Z), x x x
xx xx xx xx xx xx xx xx xx
X

keesariyaa z x x PG(RL, tr), DNA (same) ER(1, C), x x

xx x x xx xx x x xx xxxxxx
P. K. Pandey / Word accentuation in Hindi 71

(65b) niJaanaa QS x x
-
xxx xx x xx xx

To the incomplete derivation in (b) ER(1, C) cannot apply in order to avoid

clash. Consequently, *n$aanaa is impossible, and so are *sitaar, *vivaahitaa
etc. The fact that the retraction phenomenon is neatly explainable in the grid
is as it should be. After all, the role that grids were originally seen to play was
to determine the conditions under which eurhythmic adjustment could take
place. It is not surprising therefore that in Hindi the phenomenon has a
simple explanation on the grid based parameters.

7.3. Grids and trees; conclusion

The regular QR-sensitive patterns of Hindi cannot be explicated within

either the standard tree theoretic framework with the branchingness con-
straint (36) or the grid-only framework. The reason why the present analysis
with the revised constraints (42, 43) succeeds is that it can unify the factors
which determine the concatenation of internally structured syllables into feet
and assign accent to one of them. We found that the grid-only theory is
incapable of achieving this result. It is possible, however, that a theory which
is a combination of both grids and trees (e.g., Halle and Vergnaud (1987)) has
the essential tools for explaining the phenomenon.
We are not prepared to go into the grid-and-tree theoretic account of the
phenomenon, as it would take the paper on other tacks, such as the
autosegmental representation of accent. The main purpose of the present
paper has been to address itself to the metrical theoretical claims primarily in
Hayes (1981) and Prince (1983). Considering the findings of our investigation,
it is, however, pertinent to note the relevant essential features of the grid-and-
tree theoretic representation in drawing to a conclusion.
The theory takes the basic insights from Prince (1983) and Liberman (1975)
regarding the grid representation of accent, but conceives of it as a result not
of the independent factors of quantity, endmostness and rhythm, but rather
of the concatenation of stressable elements into constituents and the marking
of the head in constituents. The theory also introduces the branchingness of
stressable elements, thus providing some tree structure to the grid, and
eliminating the need for a separate level of heavy syllables, as shown in the
following terminal level representation of an Eastern Cheremis word (Halle
and Vergnaud (1987 : 50)) :
72 P. K. Pandey / Word accentuation in Hindi

(Ir *) (* * * *) (* * Ir)
1 2 3 4 5 6 7 8 9

The theory does not deal with a system such as that of Hindi which makes
a vigorous use of tree forms, e.g., internal structure of ubiquitous extraheavy
syllables, and feet formed on the relative weights of constituent syllables. It is
expected that the theory will have to further adapt itself to the tree structural
representation in handling such a phenomenon. It should be noted in this
context that the tree structural representation of prosodic units (such as the
syllable, the metrical foot, the phonological word, and the intonational
phrase) is required at any rate in order to deal with phenomena responsive to
them (see e.g., Selkirk (1980) McCarthy and Prince (1985) Ito (1986)).
In conclusion, we would like to claim that the dependence of grids even
minimally on the internal structures of accent bearing elements reveals that
although grids can translate to a large extent the information that trees
provide regarding the relative prominence of elements in tree structures, the
two - trees and grids ~ play independent roles in phonological systems.
Hayes’ conjecture appears to be essentially correct (Hayes (1983: 391)), ‘. . .
trees are the domain in which prominence relations are assigned, while grids
are the means whereby the rhythmic form of trees is computed and evalua-
ted’.

References

Chomsky, N., M. Halle and F. Lukoff, 1956. On accent and juncture in English. In: M. Halle et
al. (eds.), For Roman Jakobson, 65-80. The Hague: Mouton.
Cooper, G. and L. Meyer, 1960. The rhythmic structure of music. Chicago, IL: Chicago
University Press.
D’Souza, J., 1985. Schwa syncope and vowel nasalization in Hindi-Urdu: A non-linear approach.
Studies in the Linguistic Sciences 15( 1). 1 l-30.
Grierson, G.A., 1966. A linguistic survey of India. Volume 1, Part I. Delhi: Motilal Banarasidas.
Halle, M. and J.-R. Vergnaud, 1978. Metrical structures in phonology. Cambridge, MA:
Massachusetts Institute of Technology. (Unpublished ms.)
Halle, M. and J.-R. Vergnaud, 1980. Three dimensional phonology. Journal of Linguistic
Research I, 83-105.
Halle, M. and J.-R. Vergnaud, 1987. Stress and the cycle. Linguistic Inquiry 18, 45584.
Harris, R. M., 1966. Regional variation in Hindi. Indian Linguistics 27, 5869.
Hayes, B., 1981. A metrical theory of stress. Doctoral dissertation, MIT, Cambridge, MA.
Revised version: Bloomington, IN. Indiana University Linguistics Club.
Hayes, B., 1982. Extrametricality and English stress. Linguistic Inquiry 13, 227-276.
P. K. Pandey / Word accentuation in Hindi 73

Hayes, B., 1983. A grid-based theory of English meter. Linguistic Inquiry 14, 357-394.
Ito, J., 1986. Syllable theory in prosodic phonology. Doctoral dissertation, University of
Massachusetts, Amherst, MA.
Kelkar, A.R., 1968. Studies in Hindi-Urdu, Part I: Word phonology. Poona: Post Graduate and
Research Institute, Deccan College.
Kiparsky, P., 1979. Metrical structure assignment is cyclic. Linguistic Inquiry 10, 421441.
Ladefoged, P., 1971. Preliminaries to linguistic phonetics. Chicago, IL: Chicago University Press.
Liberman, M.Y., 1975. The intonational system of English. Doctoral dissertation, MIT, Cam-
bridge, MA. Distributed by the Indiana University Linguistics Club, Bloomington, IN.
Liberman, M. and A. Prince, 1977. On stress and linguistic rhythm. Linguistic Inquiry 8. 249-
336.
McCarthy, J.J., 1979. Formal problems in Semitic phonology and morphology. Doctoral
dissertation, MIT, Cambridge, MA. Distributed by the Indiana University Linguistics Club,
Bloomington, IN.
McCarthy, J. J., 1980. A note on the accentuation of Damascene Arabic. Studies in the Linguistic
Sciences 10(2), 77-98.
McCarthy, J. J. and A. Prince, 1985. A prosodic theory of morphology. University of Massachu-
setts, Amherst. (Unpublished ms.)
Mohanan, K.P., 1979. Stress in Malayalam, Hindi, and Gujarati. Duplicated iv-vi. (Unpublished
ms.)
Ohala, M., 1974. The abstractness controversy: Experimental input from Hindi. Language 30,
225-235.
Ohala, M., 1977. Stress in Hindi. In: L. Hyman (ed.), Studies in stress and accent. Los Angeles,
CA: Southern California Occasional Papers in Linguistics, Vol. 4.
Pandey, P.K., 1985. Word accentuation in Hindustani English in relation to word accents in
English and Hindustani. Doctoral dissertation, University of Poona.
Prince, A. S., 1983. Relating to the grid. Linguistic Inquiry 14, 19~100.
Rischel, J., 1983. On unit accentuation in Danish and the distinction between deep and surface
phonology. Folia Linguistica XVII, 51-97.
Schane, S.A., 1979. Rhythm, accent and stress in English words. Linguistic Inquiry 10, 4833502.
Selkirk, E.O., 1980. The role of prosodic categories in English word stress. Linguistic Inquiry I I,
563605.
Selkirk, E.O., 1984. Phonology and syntax. Cambridge, MA: MIT Press.
Steriade, D., 1982. Greek prosodies and the nature of syllabification, Doctoral dissertation, MIT,
Cambridge, MA.
Zograph, G.A., 1982. Languages of South Asia: A guide. London: Routledge & Kegan Paul.