TARA MOHANAN

WORDHOOD AND LEXICALITY:

NOUN INCORPORATION IN H I N D I *

This paper examines the syntactic, semantic, morphological, and phonological proper-
ties of a certain type of noun-verb sequence in Hindi, and argues that it is an instance
of noun incorporation. The sequence must be analysed as a lexical category; yet
paradoxicaIly, verb agreement and negation show that this noun is on par with a
syntacticaUy independent argument. The paper proposes a solution to this dual behav-
iour of the noun by recognizing grammatical categories and grammatical functions
as belonging to distinct but co-present dimensions of syntactic representation. This
factorization of dimensions yields an account of the facts of Hindi Incorporation which
are not amenablc to analysis in terms of head movement (Baker 1988) or coanalysis
(Sadock 1991). When combined with the idea of different notions of wordhood, the
dual representation leads to a distinction between categoriaI word m-ld functional
word, central to the untangling of numerous issues surrounding lexieality.

This paper addresses a problem posed by a construction in Hindi that has

not been recognized in the literature on Hindi syntax, and which comes
under the rubric of Noun Incorporation. Noun Incorporation is the phe-
nomenon of a noun combining with a verb to form a single morphological
unit, while still retaining its syntactic status (Mithun 1984; Alien, Gardiner
and Frantz 1984; Hopper and Thompson 1984; Sadock 1980, I985, 1986,
1991; Knecht 1986; Baker 1985, 1988; Di Sciulto and Williams 1987; Rosen
1989; among others). If we assume that the verb and one of its arguments
form a lexical unit (N + V) in the Hindi Noun Incorporation Construction
(henceforth NIC), various semantic, syntactic, morphological, and phono-
logical characteristics of the construction follow as a natural consequence.
However, the verb can agree with that argument when agreement with
the subject is not possible. In order to account for verb agreement in NIC,

* The analysis in this paper began as a response to Paul Kiparsky's persistent and incisive
questioning, Joan Bresnan and K, P, Mohanan made enormous contributions to the paper
during various stages of its growth. Comments from N. S. Prabhu, Carol Georgopoulos, and
three anonymous reviewers have resulted in substantial improvements, I thank them all,
The paper has also benefited from presentations at the Stanford Linguistics Colloquium in
May 1990, and at the panel on Agreement at the 20th Annual Conference on South Asia
held at the Univers/ty of Wisconsin, Madison in November 1991. This study was done in
part with support from the Center for the Study of Language and Information, Stanford.
The judgements reported in this paper are largely my own. I am grateful to Akhil Gupta,
Purnima Mankekar, Ravi Oswal, Rajeshwari Pandharipande, and Alka Warrier, who have
generously offered their own judgements, increasing my confidence in my own judgements.
I thank Rajendra Singh for making me aware of the extent to which speaker judgements
can differ.

Natural Language and Linguistic Theory I3: 75-134, 1995.

(~) 1995 Kluwer Academic Publishers. Printed in the Netherlands.
76 TARA MOHANAN

the noun in the N + V compound must be analysed syntactically as an

argument on par with the other arguments of the verb.
This situation raises two problems. First, evidence from one set of facts
shows that the incorporation must be lexical, while another set of facts
shows that it is not lexieal. Second, a verb in Hindi agrees with one of its
sisters, but in NIC, the verb agrees with the incorporated noun, a
daughter. The intriguing question is, if the controller of verb agreement
must be a sister of the verb, how can the verb agree with its daughter?
These puzzles resolve themselves if, in the architecture of grammar, we
factor apart the internal organization of the verb's arguments and their
grammatical functions on the one hand, and their category structure on
the other. The factorization of different types of information into separate
dimensions of organization throws open the possibility of different con-
stituency relations at different dimensions of representation. The facts of
negation in Hindi support such factorization.
The paper is organized as follows. Section 1 introduces NIC in Hindi
in terms of a set of correlations associated with object case marking,
and outlines some theoretical assumptions used in the account of the
phenomenon. Section 2 argues for the treatment of NIC as a lexical unit
IN + V]v. Section 3 examines the facts of verb agreement, and shows that
agreement in NIC cannot be lexical. Section 4 discusses some of the
alternative analyses in detail. Section 5 deals with negative placement,
which provides support for this account of NI. Section 6 shows how the
account distinguishes different notions of wordhood in terms of well-
motivated dimensions of structure, and yields a promising way of resolving
the numerous debates on issues of lexicality.

1. INTRODUCTION

This section presents the puzzling asymmetries in Hindi which are analysed
as Noun Incorporation, and spells out the theoretical assumptions relevant
for the analysis.

1.1. The P u z z l e
In order to formulate the puzzle, let me first give some well-known facts
of object case marking in Hindi. Animate primary objects in Hindi bear
the accusative case clitic -ko: 1

1 T h e following abbreviations are used in the glosses:

N: Nominative NF: Nonfinite
 W O R D t l O O D AND L E X I C A L I T Y 77

(1)a. ilaa anil-ko uthaaegii.

Ila-N Anil-A lift-FU
Ila will lift up Anil.

b. *ilaa anil ut.haaegfi.

Anil-N
Inanimate objects have a choice between accusative (ACC) case, and
nominative (NOM) case which is morphologically unmarked, z If ACC, an
inanimate object must be definite. 3
(2) a. ilaa-ne haar-ko utflaayaa.
Ila-E necklace-A l~ft-PA
Ila lifted up the necklace.

b. ilaa-ne haar u.thaayaa.

necklace-N
Ila lifted up the/a necklace.
Thus, an ACC object must be animate or definite. Now consider (3), with
an ACC animate object, and (4), where the same object is NOM:
(3) ilaa bacct-ko khojtii rahtii hai.
Ila-N children-A search-HAB PROG be-PR
Ila keeps searching for the/some children.

E: Ergative PA: Past

A: Accusative PR: Present
D: Dative FU: Future
L: Locative HAB: Habitual
I: Instrumental PERF: Perfective
G: Genitive
M: Masculine SG: Singular
F: Feminine PL: Plural
The convention adopted in the word glosses is as follows. When a form uniquely expresses
one member in a contrasting pair, the member is included in the gloss. Thus, the specification
of number is indicated in the gloss if the singular and plural forms (or the NOM and
NONNOM case forms) of a noun are distinct on the surface. But the specification is omitted
from the gloss if the forms are identical.
2 I analyse subjects and objects without overt case marking as bearing nominative case. For
a defense of this position, see T. Mohanan (1993).
3 For details, see Srivastava (1969), McGregor (1972), T. Mohanan (1990, 1993) and the
references therein.
78 TARA MOHANAN

(4) ilaa bacce khojtii rahtii hai.

Ila-N children-N search-HAB PROG be-PR
Ila keeps children-searching (i.e. performing the act of search-
ing for children).
The only visible difference between (3) and (4) is the case marking on the
object. Yet, their interpretations as indicated in the sentence glosses are
different. Suppose we refer to the interpretation of (3) as "reading A",
and that of (4) as "reading B". How do we account for the correlation
between ACC/NOM case of animate objects and readings A/B?
To extend the puzzle further, consider the following contrasts. The
animate objects in (5)-(9) are ACC in (a), and NOM in (b):
(5)a. ilaa baccf-ko hamegaa/har jagahkhojtii
Ila-N children-A always every place search-HAB
rahtii hai.
PROG be-PR
Ila is searching for the/some children all the time/everywhere.

b. *ilaa bacce hamegaa/har jagah khojtii

Ila-N children-N always every place search-HAB
rahtii hai.
PROG be-PR

(6)a. ilaa chot.ebacc6-ko khojtii rahtii hai.

lla-N small children-A search-HAB PROG be-PR
Ila keeps searching for the/some small children.

b. *ilaa chotebacce khojtii rahtii hai.

Ila-N small children-N search-HAB PROG be-PR

(7)a. ilaa kis-ko k~ojtii rahtii hai?

lla-N who-A search-HAB PROG be-PR
Who does Ila keep searching for?

b. *ilaa kaun khojtii rahtii hai?

lla-N who-N search-HAB PROG be-PR

(8)a. ilaa lad.k6 aur la.dkiy6-ko khojtii rahtii hai.

Ila-N boys- and girts-A search-HAB PROG be-PR
Ila keeps searching for the/some boys and girls.
 WORDHOOD AND LEXICALITY 79

b. *ilaa lad ke aur lad kiygfikhojtii rahtii hai,

Ila-N boys-N and girls-N search-HAB PROG be-PR

(9)a. ilaa bacc6-ko khotii aur khojtii rahtii hai.

Ila-N children-A tose-HAB and search-HAB PROG be-PR
Ila keeps losing and searching for the/some children.

b. *ilaa bacce khotii aur khojtii rahtii hai.

Ila-N children-N lose-HAB and search-HAB PROG be-PR.
The only visible difference between (a) and (b) in (5)-(9), again, is that
the object is ACC in (a) and NOM in (b). Given that (4) is acceptable,
the second problem is: when an animate object is nominative, it cannot
be (i) separated from the verb ((Sb)), (ii) modified ((6b)), (iii) questioned
((7b)), or (iv) conjoined ((8b)). Nor can the verb be conjoined ((9b)). 4
The goal of this paper is to provide an account for the phenomenon
partially illustrated in (4). I will argue that the NOM object in (4) forms
a morphological unit with the verb, and has the structure in (10b), while
the ACC object and the verb in (3) have the structure in (10a):
(10)a. N b. V

OBJ
vl\ N
OBJ
V

1.2. Theoretical Background

The representation in (10b) contains the information that the noun forms
a lexical category with the verb, and is at the same time the object of
the verb. (10b) represents the substance of what has been called Noun
Incorporation (henceforth NI) in syntactic theory. I use the term NI,
following Sadock (1980, 1991), to refer to the phenomenon of a noun
stem exhibiting dual behaviour: it is a syntactic argument of a verb, but
morphologically part of that verb. 5

4 There are dialects of Hindi in which N O M animate objects may be more freely available
in such sentences. See, for instance, V e r m a (1971, p. 104).
5 That is, NI does not refer to a specific analysis of the phenomenon, e.g., the movement
of a lexical head noun into a verb, as in Baker (1985, 1988). For a historical perspective on
NI, see Sadock (1991: pp. 78-100).
80 TARA MOHANAN

Most analyses of NI, including the one proposed in this paper, agree
that there must be some representation in which the verb and the incorpor-
ated noun together form a lexical (X °) category ([N + V]v0). Sadock (1980
and subsequent work) and Baker (1985, 1988) additionally recognise the
synactic argument status of the incorporated nominal, which they repre-
sent as an NP sister of the verb. The N and the NP are related in Sadock
as parallel independent representations:

(11) VP
SYNTACTIC

NP V
]
N!
I
I

N V
MORPHOLOGICAL

In Baker, they are related in a single representation through the movement

of the head of the NP into the verb, leaving a trace:

(12) VP VP

:=>
NP V NP V

I
N
1
N
I
V

ti Ni V

The parallel representations in (11) and the output of head movement in

(12) express the same information as in (10b): the noun forms a lexical
category with the verb, but is also a syntactic argument of the verb. The
dotted line in (11) which connects the two types of information corre-
sponds to the coindexed trace in (12).
In contrast to Sadock and Baker, Mithun (1984), Di Sciullo and Willi-
ams (1987), and most recently, Rosen (1989) propose that NI is a purely
lexical phenomenon, without any syntactic consequences.
 W O R D H O O D AND L E X I C A L I T Y 81

My account of incorporation is essentially in the lexicalist tradition of

Lexical Functional Grammar and Lexical Phonology and Morphology, in
that it claims contra Baker that incorporation in Hindi is part of the lexical
module. Within this conception of grammar, the lexical module, where
words are formed, is distinct from the phrasal module, where words are
concatenated to form larger units. 6 The term phrasal in this paper has the
same meaning as postlexical in Lexical Phonology. It has no X-bar theor-
etic connotations.
Unlike Rosen and others, and like Sadock and Baker, this analysis also
claims that there must be some representation in which the incorporated
noun is syntactically independent, at least for languages like Hindi. Like
Sadock, it recognizes NI as simultaneously morphological and syntactic.
It differs from Sadock in that while the two representations in (11) are
those of morphology and syntax, the two relevant representations of my
proposal are those of grammatical categories on the one hand and gram-
matical functions on the other:
(13) CLAUSE
/x, G(RAMMATICAL) F(UNCTION)
T STR(UCTURE)

OBJ
!
PRED
|
,, I
I, ',
N
"x/ V
G(RAIVIMATICAL) C(ATEGORY) STR(UCTURE)

2. L E X I C A L I T Y OF N O U N I N C O R P O R A T I O N IN H I N D I

In this section, I propose a representation of the categorial structure for

the Hindi NIC, with syntactic, semantic, phonological and morphological
evidence for the analysis of the N + V sequence as a lexical category. I
argue, further, that the incorporation is part of the lexical module.

6 I hesitate to use the term syntactic instead of phrasal, because of the implication that no
word internal information is syntactic.
82 TARA MOHANAN

2.1. A Partial Analysis: N + V in N I as a Lexical Category

2.1.1. Overt Case Marking
As we saw in (2), an inanimate object in Hindi may be either NOM or
ACC. Let us assume that an animate object is assigned ACC case if it is
phrasal (lOa); it is N O M if and only if it is part of a lexical category (lOb).
The inability of the object in structure (lOb) to be ACC follows if we
assume that case markers in Hindi are clitics that yield phrasal categories,
not inflections that yield lexical categories. Evidence for this assumption
comes from conjoining. As I will show in section 2.1.5, lexical categories
cannot be conjoined in Hindi. Yet, case clitics can be attached to conjoined
nominals:
(14)a. madraas aur kalkatte me
Madras and Calcutta L
in Madras and Calcutta.

b. aurat5 aur bacc5 se

women and children I
from/by women and children.
Therefore, we must conclude that case clitics are attached to phrasal
categories. 7 The N in (10b) occurs within a lexical category. Since ACC
involves a case clitic, it cannot occur in (10b), unlike NOM.
We saw that (3) with an ACC object has reading A, while (4) with a
NOM object has reading B. Given our assumptions above, we can account
for this correlation between case marking and meanings by assuming that
reading B is licenced by the lexical structure in (10b), and reading A by
phrasal structrure in (10a). I will now go on to show how the above
analysis correctly predicts the behaviour of inanimate objects in Hindi.
Given the result that structure (10b) does not allow ACC case, and the
assumption that reading B is sanctioned only by (10b), it would follow
that inanimate objects can have either reading A or reading B if NOM,
but only reading A if ACC. This prediction is borne out by the following
data:

7 This is in contrast with a language like Malayalam, where case is indicated by affixes rather
than clitics. In a conjoined structure, for example, the affix is attached to every conjunct,
rather than to the entire conjoined structure.
 WORDHOOD AND LEXICALITY 83

(15)a. anil kitaabe becegaa.

Anil-N (M) book-N.PL(F) sell-FU.M.SG
A. Anil will sell (the) books.
B. Anil will do book-selling.

b. anil kitaab6-ko becegaa,

book-A.PL(F)
A. Anil will sell the books.
B. *

(16) a. anil-ne kitaab~ becii.

Anil-E(M) book-N.PL(F) selI-PERF.F.PL
A. Anil sold (the) books.
B. Anil did book-selling.

b. anil-ne kitaab6-ko becaa.

book-A.PL(F)
A. Anil sold the books.
B. *

(17) a. raam-ne lakdii kaatii,

Ram-E(M) wood-N (F) cut-PERF.F.SG
A. Ram cut (the) wood.
B. Ram did wood-cutting.

b. raam-ne lak .dii-ko kaataa.

wood-A(F)
A. Ram cut the wood.
B. *

The objects in (a) are NOM, and the sentences have two interpretations,
84 TARA MOHANAN

reading A and reading B. 8 The objects in (b) are ACC, and the sentences
allow only reading A . 9

2.1.2. Adjacency
As shown by (5b), the animate object must be adjacent to the verb if it
is NOM. This follows from the assumption that animate objects can be

s Given below are two more examples of inanimate nouns as N O M objects of sentences:
(i) anil-ne khaanaa pakaayaa.
AniI-E(M) food-N(M) cook-PERF.M.SG
A. Anil cooked food.
B. Anil did food-cooking.

(ii) raam kapde siitaa hai.

Ram-N (M) ctothes-N (M) sew-HAB.M.SG be-PRES
A. Ram sews/tailors clothes.
B. Ram does clothes-sewing. ( = Ram is a tailor.)
These sentences also have two interpretations. The different interpretations have correspond-
ing differences in word stress and word melody patterns. We will return shortly to these
correspondences.
9 The case restriction on the N in (10) is not on the case feature (NOM/ACC) but on the
presence of an overt case clitic. Other than N O M nouns, locative destinations are the only
cliticless arguments in Hindi. The locatives in (i) and (ii) have no clitic; those in (iii) and
(iv) bear locative case clitics:
(i) raam-ne apne bet e-ko ~ahar bhejaa.
Ram-E self-G son-A city send-PERF
A. Ram sent his son to the city.
B. Ram city-sent his son.

(ii) baecaa skuul gayaa.

chitd-N school go-PERF
A. The child went to school.
B. The child school-went,

(iii) raam-ne apne be te-ko gahar-m~ rakhaa.

Rarn-E self-G son-A city-L(IN) keep-PERF
A, Ram kept his son in the city.
B, *Ram 'city-kept' his son,

(iv) baccaa kursii-par bait.htaa hal.

child-N chair-L(on) sit-HAB be-PR
A. The child sits on a chair.
B. *The child 'chair-sits'.
Like (t5a) and so on, (i) and (ii) allow reading B: I assume, then, that cliticless locatives
can be part of (10b). In contrast, (iii) and (iv), in which the locatives bear case elitics, do
not allow reading B. I will not elaborate further on locatives that can be part of (10b).
 WORDHOOD AND LEXICALITY 85

NOM only in structure (10b). Since the noun and the verb form a single
lexical category in (10b), the noun cannot be separated from the verb.
This analysis predicts that reading B will not be available to inanimate
objects if they are not adjacent to the verb. This prediction is borne out
by the following data:
(18) a. anil kitaab~ hamegaa becegaa.
Anil-N(M) book-N.PL(F) always sell-FU.M.SG
A. Anil will always sell books.
B. *Anti will always do book-selling.
Like other South Asian languages, Hindi is a 'free word order' language
in that the predicate and its dependents in a clause are free with respect
to one another in their relative order. Thus, (18b, c) are acceptable ver-
sions of (15a): 1°
(18)b. kitaab~ anil becegaa.
book-N.PL(F) Anil-N (M) sell-FU.M
A. Anit will sell the books.
B. *Anit will do book-selling.

c. anil kitaab6 baazaar-m~ becegaa.

Anil-N(M) book-N.PL(F) rnarket-L sell-FU.M
A. Anil will sell the books in the market.
B. *Anit will do book-selling in the market.
As our analysis predicts, (18b) and (18c) disallow reading B. As predicted,
the result of scrambling the animate NOM object away from the verb in
(10b) is ungrammatical.

2.1.3. Modifier Stranding

Since the left member in structure (10b) is an N, not an NP, it follows
that the phrasal unit [modifier + N] cannot participate in structure (10b).
This explains why animate objects cannot take modifiers if they are NOM
(6b). This analysis predicts that inanimate objects cannot have reading B

to Entities or sequences that are not in their canonical order exhibit certain definiteness
effects in Hindi, The definiteness of the object in (18) follows from this condition, which I
will not discuss further in this paper.
86 TARA MOHANAN

if they contain modifiers. This prediction is confirmed by the following

data:

(19)a. anil-ne puraanii kitaab~ beciL

Anil-E(M) old-F book-N.PL(F) sell-PERF.F
A. Anil sold a/the old books.
B. *Anil did the selling of old books.

b. anil-ne acchaa khaanaa pakaayaa.

Anit-E good food-N(M) cook-PERF.M
A. Anil cooked good food.
B. *Anil did the cooking of good food. 11

c. raam bacc6-ke kapd. e siitaa hai.

Ram-N (M) children's clothes-N (M) sew-HAB.M be-PRES
A. Ram sews/tailors children's clothes.
B. *Ram does children's clothes-sewing.

The non-cooccurrence of object modification and reading B further leads

to the assumption that there is no representation in which the N in (10b)
is associated with an NP. 12

11 For some speakers, a sentence like (i) below, similar to (i) in note 9, allows reading B:
(i) anil acchaa khaanaa pakaataa hai.
Anil-N good food-N(M) cook-HAB.M be-PRES
A. Anil cooks good food.
B. Anil cooks well.
The reason for this ambiguity despite modification is that the modifier acchaa itself is
ambiguous between the adjective "good" and the adverb "well". Co-occurring with reading
B is the latter meaning, modifying the entire verb complex and not just the noun.
i2 An anonymous reviewer has expressed scepticism about this claim, and points out that
even though [[green car] driver] in English is unacceptable in ordinary situations, it becomes
acceptable in a situation where ([green car] is nameworthy, for example, under a gas-
rationing scheme where one's permission to fill up depended on the colour of one's vehicle.
However, rather than assuming that the NP green car can form a compound with another
noun if the phrase is nameworthy, I suggest that green car can become a compound noun
in English if nameworthy (like blackbird). If Hindi allowed compound nouns of the form
[A N], the analysis in this paper would predict noun incorporation constructions of the form
[[A N] V]v. However, I have not been able to find any [A N] compounds in Hindi.
The issue of nameworthiness is discussed in greater detail in section 2.2.
 WORDHOOD AND LEXICALITY 87

2.1.4. Wh-words and Pronouns

Given the widely accepted assumption that wh- words and pronouns are
phrasal categories (maximal projections), they cannot participate in struc-
ture (10b). 13 This explains why the animate NOM object cannot be ques-
tioned (7b). It also predicts that an inanimate object that is either a wh-
form or a pronoun cannot have reading B. This prediction is borne out
by (20a) and (20b):

(20) a. anil kyaa becegaa?

AniLN (M) what sell-FU.M.SG
A. What will Anil sell?
B. *What-selling will Anil do?

b. anil yah becegaa.

Anil-N (M) this sell-FU.M.SG
A. Anil will sell this.
B. *Anil will do this-selling.

2.1.5. Conjoining
Consider the examples in (21), which involve conjoining:

(21)a. laal kitaab~ aur laal kursiyg~

red book-PL.N and red chair-PL.N
red books and red chairs

b. laal kitaab6 aur kursiyfig

red book-PL.N and chair-PL.N
red books and chairs (i) chairs and red books
(ii) *red books and red chairs.

For the interpretation in (i), the structure of (21b) is s[s[red books] and
s[chairs]]. For the interpretation in (ii), it must be S [ A [ r e d ] N [ N [ b O o k s ]

13 U n d e r a DP analysis (Abney 1987), a wh- word or pronoun is not necessarily a maximal

projection. However, it is still not a lexical category, as required by (10b),
88 TARA MOHANAN

and N[chairs]]]. 14 The absence of this reading is accounted for if we assume

that only phrasal categories can be conjoined in Hindi. ~5
Given this independently motivated assumption, it follows that neither
the N nor the V in (10b) can be conjoined. This explains why (8b) and
(9b) are ungrammatical. Since the relevant expression in each example
involves conjoining, that expression cannot be a lexical category. How-
ever, only a lexical category can occur in (10b). Hence, (Sb) and (9b)
cannot be analysed in terms of (10b). However, an animate object can be
NOM only if it is part of the structure in (10b). Hence, a structure with
a conjoined animate NOM object is ungrammatical.

14 The data below illustrates this further:

(i) bahut badii kitaab6
a lot large book-PL.N
(a) very large books
(b) many large books

(ii) bahut bhaarii kitaab~

a lot heavy book-PL.N
(a) very heavy books
(t3) many heavy books

(iii) bahut bad.ii aur bhaarii kitaab~

a lot large and heavy book-PL.N
(a) [[very large] and [heavy]] books
(b) many [large and heavy] books
(c) *very large and very heavy books
In (i) and (ii), bahut may function either as an intensifier modifying the adjective as in (a),
or as a quantifier modifying the noun as in (b). In (iii), bahut may modify either the noun
as in (a), or the first adjective of the conjoined structure as in (b). However, the unac-
ceptability of interpretation (c) shows that it cannot modify the conjoined structure itself.
is That this is true for verbal units is illustrated by the following contrast:
(i)a. baccaa boltaa gayaa aur rotaa gayaa.
child-N speak-HAB go-PERF and cry-HAB goPERF
The child kept talking and kept weeping.

b. baccaa boltaa aur rotaa gayaa.

The child kept talking and weeping.
(ii)a. baccaa bol rahaa thaa aur ro rahaa t~aa.
child-N speak stay-PERF be-PA and cry stay-PERF be-PA
The child was "talking and was weeping.
b. *baccaa bol aur ro rahaa tUaa.
I claim that the ungrammaticality of (iib) is the result of conjoining two V's. Arguing for
this position calls for a detailed discussion of verbal morphology and the auxiliary system,
which is beyond the scope of this paper.
 WORDHOOD AND LEXICALITY 89

This analysis also predicts that if the verb or its inanimate object involve
conjoining, reading B will be ruled out. The prediction is confirmed by
(22a, b):

(22)a. anil haatr~ii aur ghod.e bectaa hai.

Anil-N elephants-N and horses-N seII-HAB be-PR
A. Anil sells elephants and horses.
B. *Anil does elephant- and horse-selling.

b. anil gho.de khariidtaa aur bectaa hal.

Anil-N horses-N buy-HAB and sell-HAB be-PR
A. Anil buys and sells horses.
B. *Anil does horse-buying and -selling.

2,1.6, Gapping
Consider the sentences below that involve coordination:

(23)a. anil g%.de khariidtaa hal aur raam gho.de

Anil-N horses-N buy-HAB be-PR and Ram-N horses-N
bectaa hai.
selt-HAB be-PR
A. Anil buys horses and Ram sells horses.
B. Anil does horse-buying and Ram does horse-selling.

b. anil ghodei khaliidtaa hai aur raam ~ bectaa

Anil-N horses-N buy-HAB be-PR and Ram-N sell-HAB
hai.
be-PR
A. Anil buys and Ram sells horses.
B. *Anil does horse-buying and Ram does - - -selling.

c. anil ghod.e --i aur raam haathii bectaai hal.

Anil-N horses-N and Ram-N elephants-N sell-HAB be-PR
A. Anil buys horses and Ram elephants.
B. *Anil does horse- -- and Ram elephant-selling.

Each of the conjuncts in (23a) may independently have reading B. How-

ever, reading B is unavailable in (23b) and (23c). In other words, if
90 TARA MOHANAN

either the object or the verb is gapped in a coordination construction, the

sentence disallows reading B.
This unavailability of reading B follows from (10b) if we assume the
Lexical Integrity Hypothesis which "prohibits syntactic reorderings into or
out of lexical categories" (Bresnan 1982b, 54). I6 This condition prevents
gapping into a lexical category, since gapping is a phrasal phenomenon.
In addition, the condition prevents part of a lexical category from also
being the direct daughter of a phrasal category (through movement or
coanalysis), a result that is supported by the absence of modifier stranding
described in section 2.1.3.

2.1.7. Summary: The Category Structure of Hindi NI

In sum, the asymmetries reviewed so far are accounted for by assuming
that Hindi has a construction in which a noun and a verb together form
a single lexical category. When a noun-verb sequence in a sentence such
as (16a) has reading A, it has the syntactic structure given in (24); when
it has reading B, it has the syntactic structure i n (25): 17
(24) S

I
anil-ne
I
kitaab6 becii (Anil sold books.)

16 Observe that the Lexical Integrity Principle is stated specifically on category information,
and will not extend, for instance, to grammatical function information. The significance of
this point will become clear later in the paper. Also see Simpson (1983, 1991, pp. 44-45)
for the idea within LFG that the principle holds only on constituent structure, not on
functional structure, and Bresnan and Mchombo (1993) for further support of the Lexical
Integrity Principle.
17 The V in these structures represents a,phrasal category, whether or not it is a maximal
projection. The ~r node could as well be labelled as VP without any empirical difference.
However, I avoid using the label VP, because a VP is characteristically viewed as a node
that contains the verb and at least all its non-subject complements. I assume that in Hindi
non-subject complements of a verb, including object NP's, are not dominated by V. I will
not defend this assumption here because it is not directly relevant to this paper.
 WORDHOOD AND LEXICALITY 91

(25) S

NP V

I
V

anil-ne
t
kitaab6 bec[i (Anil did book-selling.)

At the outset of the paper, I said that NI refers to the phenomenon of

a noun stem (a) being part of a morphological unit and yet (b) having
syntactic independence. In the preceding discussion, I gave syntactic evi-
dence for (a). Although I have not demonstrated the syntactic indepen-
dence of the noun yet, I claim that the structure in (25) is one of NI. In
what follows, I will call this construction the NI construction (or NIC) in
Hindi. NI in most languages discussed in the literature seem to invoh,e an
overtly visible difference between the NI construction and the construction
without an incorporated nominal. In Hindi, an NI construction has an
external appearance identical to that of its unincorporated counterpart.

2.2. Semantic and Pragmatic Correlates of N I

Let us first try to sharpen the distinction that I have appealed to so far
between readings A and B. Consider the truth conditions accompanying
the two readings of (26a):

(26)a. mohan chuttiyr-me va~kyum kliinar bectaa t~aa.

M o h a n - N holidays-in vacuum cleaner-N sell-HAB be-PA
A. Mohan was selling vacuum cleaners during the holidays.
B. Mohan was doing vacuum cleaner-selling during the hol-
idays.

b. usne do mahine-me ek bhii v~ekyum kliinar nahii

h e - E R G two month-in one even vacuum cteaner-N not
becii.
sell-PA
He didn't sell even one vacuum cleaner in two months,
92 TARA MOHANAN

(26b) and reading A of (26a) cannot both be true at the same time.
However, (26b) and reading B of (26a) are not logically contradictory:
they can be simultaneously true. 18 This contrast justifies distinguishing the
two readings.
Recall our assumption that reading B is sanctioned by structure (10b).
In addition to having different truth conditions, reading B requires that
the N in the N + V sequence be generic or 'non-referential': it can refer
only to the class of entities denoted by the noun, not to the individual
members of the class. ~9 This is illustrated'by the following examples:
(27) anil-ne tiin saal kapd.e dhoe.
Anil-E three year-N clothes-PL.N wash-PERF
A. Anil washed clothes for three years.
B. Anil did clothes-washing for three years.

(28) a. kapd.e bahut gande ho gaye t%.

clothes-N-PL a tot dirty-PL become go-PERF-PL be-PA-PL
The clothes had become very dirty.

b. anil-ne tiin ghan.t.e kapde dhoe.

Anil-E three hour-N-PL clothes-N-PL wash-PERF
A. Anil washed clothes for three hours.
B. *Anil did clothes-washing for three hours.
(27) allows both readings. In (28), however, because of prior mention in
(a), the object in (b) is definite, and reading B is impossible. We would
then predict that an animate NOM object with prior mention in the
discourse will be ungrammatical. This is borne out by (29):
(29)a. bacce sab k~o gaye the.
child-N-PL all lose go-PERF-PL be-PA-PL
The children had all got lost.

i8 I thank an NLLT reviewer for pointing my attention to this possible difference.

19 Genericity and non-referentiality have sometimes been seen as a characteristic of incorpor-
ated nominals. However, languages with more widespread, and sometimes obligatory, incor-
poration show that an incorporated noun is not necessarily accompanied by genericity or
nonspecificity (e.g., Mohawk (Mithun 1984), Nahuatl (Merlan 1976), Southern Tiwa (Allen,
Gardiner, and Frantz 1984) and so on). In Eskimo (Sadock 1980), an incorporated noun
may be specific, or even definite, and may introduce a new topic.
 WORDHOOD AND LEXICALITY 93

b. *ilaa-ne tiin ghan.te bacce khoje.

Ila-E three hour-N-PL chitd-N-PL search-PERF
In the context of (29a), the object in (29b) is definite. Therefore, reading
B is unavailable. Consequently, structure (10b) is unavailable. NOM case
on an animate phrasal object is ungrammatical.
Finally, the activity or process referred to by NI must be 'salient' or
'nameworthy' in what Hale and Keyser (1991, p. 13) call the 'cultural
encyclopedia' of the language users, z° Consider the contrast between A
and B in (30a-c): 2I
(30) A B
a. ghaas kaatnaa / becnaa dekhnaa
grass cutting selling seeing
A. cutting/selling grass seeing grass
B. grass-cutting/-seUing # grass-seeing

b. kapde becnaa / d%naa / siinaa phaad,naa/pahanna

clothes selling washing sewing tearing wearing
A. selling/washing/sewing clothes tearing/wearing clothes
B. clothes-selling/-washing/-sewing # clothes-tearing/-wearing

c. kitaab likt~naa / pad.hnaa/becnaa denaa / uthaanaa

book writing reading selling giving lifting
A. writing/reading/selling books giving/lifting books
B. book-writing/-reading/-selling # book-gi~Ang/-lifting
Take ghaas kaat.naa "cutting grass"/"grass-cutting" and ghaas dekhnaa
"seeing grass"/#"grass-seeing" in (30a). In most societies, cutting grass
is a salient activity in the conceptual system, while seeing grass is not.
Hence the former, but not the latter, can be an NIC in Hindi. ff we adopt
the suggestion that nameworthiness is a property of lexical items (Hale
and Keyser 1991, p. 13), the fact that the activity referred to by the Hindi
NIC must be salient or nameworthy in the conceptual system will follow
from the assumption that the N + V in NI is concatenated in the lexical
module.

20 B a u e r (1983, pp. 86-7) refers to this property as the "nameabitity" condition on lexical
items.
94 TARA MOHANAN

2.3. Phonological and Morphological Evidence: N I in the Lexical Module

I now turn to phonological and morphological evidence that further sup-
ports the representation of NI as a lexical category (lOb), and the claim
that NI is formed in the lexical module, rather than in the phrasal module.

2.3.1. Phonology
Consider the following scenario. A primary school teacher says to a class:
A frog e a t s . . . , pauses for an answer from the class, and completes the
sentence: ...flies. In other words, the teacher puts a planned meaningful
pause within the sentence. A similar pause, often used for dramatic effects
as well, can occur in the following sentence: A frog is a . . . f l y eater.
However, a pause between fly and eater in A frog is a f l y . . , eater is
unacceptable: it almost makes the sentence uninterpretable. Given that
fly eater is a compound noun, the above contrast in the effects of pauses
can be accounted for if we assume that planned meaningful pauses cannot
occur inside a lexical category.
In an object-verb sequence in a sentence in Hindi, a pause occurring
between the object and the verb blocks the reading B. This would follow
from (10b) together with the assumption that planned meaningful pauses
cannot occur inside a lexical category. 22
Another piece of evidence for the N + V in NI to be treated as a lexical
category comes from the facts of stress and word melody. Sub(ordinate)
compounds of the form [head + modifier] in Hindi have a single primary
stress and word melody, characteristic of single words, as opposed to
co(ordinate) compounds of the form [head + head + •..], which have as
many primary stresses and word melodies as they have stems, character-
istic of word sequences. Consider the asymmetry between the compounds
in (31a) and (31b), where (31a) is a subcompound and (31b) is a cocom-
pound:

(31)
a. j~ianakiinandan b. j~ianakiimindan
Janaki's son Janaki and Nandan
What (31) illustrates is a systematic correspondence between morpho-
logical construction types (subcompound vs. cocompound) and their

21 I am grateful to an anonymous reviewer for bringing these examples to my attention.

22 Planned meaningful pauses may occur between a nominal and its case clitic. This is
consistent with the prohibition of pauses within a lexical item. Recall that a case clitic in
I-Iindi is attached phrasally; this accounted for the impossibility of case marking on an
incorporated nominal, and the possibility of case marking a conjoined nominal.
 WORDHOOD AND LEXICALITY 95

phonological properties (stress and word melody). This correspondence

is expressed in the level ordered conception of lexical phonology and
morphology by associating the two constructions with sequentially ordered
strata or levels and defining the domain of the phonological patterning in
terms of these levels. In a theory of phonology and morphology that
does not admit sequential modularity, the correspondence is expressed by
associating the different construction types with different representational
units (e.g. prosodic stem, prosodic word) and defining the domain of the
phonological patterning in terms of these units. The choice between the
two strategies is not crucial for my purposes, z3
In the level ordered conception, the pattern in (31) is accounted for by
assuming that (i) subcompounding and cocompounding constitute different
lexical strata, (ii) the former is an input to the latter, and (iii) stress and
word melody assignments take place at the end of the subcompounding
stratum, z4 Under these assumptions, the subcompound in (31a) is assigned
a single stress and word melody, whereas each stem in the cocompound
in (31b) is assigned an independent stress and word melody.
The N + V in NI behaves exactly like a subcompound in that it has a
single stress and word melody, as in (32a):
(32) a. ---~M
gh6d.e becnaa
horse-PL sell-NF
to horse-sell

b.
gh6.de b6cnaa
horse-PL sell-NF
to sell horses

C.
safed gh6qe b6cnaa
white horse-PL sell-NF
to sell white horses

23 For a detailed discussion of the two strategies, and arguments against sequentiality, see
K. P. Mohanan (in press),
24 This is similar to the pattern observed in Malayalam (K, P. Mohanan 1982a),
96 TARA MOHANAN

d.
gh4d.6-ko b6cnaa
horse-PL.A sell-NF
to sell the horses

The contrast in stress and word melody in (32a) and (32b-d) follows if
the N and V in (32a) combine in the same lexical submodule as the
subcompound in (31a). In a non-sequential approach, the corresponding
hypothesis would be that subcompounds and the N + V in (32a) are of
the same morphological construction type.
Yet another piece of evidence that NI in Hindi takes place in the lexical
module comes from an optional phonological process of stem-final high
vowel shortening. A stem-final high vowel that is long when word-final
can optionally be short when not word-final, as shown by (33b):

(33) a.
jtianakii *j~ianaki
Janaki

b.
j~ianakiinandan j~ianakinandan
Janaki' s son

C.
j~ianakiimlndan *j~ianakinhndan
Janaki and Nandan

d.
j~ianakii-ne *j~ianaki-ne
Janaki-ERG

The stem-final ii in jaanakii may optionally be short in the subcompound

in (33b). However, it cannot be short word-finally ((33a)), or within a
cocompound ((33c)). The facts in (33) would follow if the domain of vowel
shortening is the subcompounding submodule, like stress and word melody
assignment. (33d) shows that the stem-final vowel cannot be shortened
when followed by a case clitic. Recall that the case clitic is attached in
the phrasal module. Hence, when shortening takes place, the long vowel
is final, and cannot shorten.
If the domain of vowel shortening is a subcompound, and the N + V in
NI is like a subcompound with respect to stress and word melody, we
 WORDHOOD AND LEXICALITY 97

predict that vowel shortening should be possible in the N + V in NI. This

prediction is borne out by the facts in (34):

(34)a. (i) mit.haaii banaaii. (ii) mit.haai banaaii

sweets-N make-PERF sweets-N make-PERF
made sweets.

b. (i) sabzii becii. (ii) sabzi becii

vegetables-N sell-PERF vegetables-N sell-PERF
sold vegetables.

This vowel shortening is not found in all speakers. However, for those
who can shorten the vowel, the process is systematic: when the vowel is
short, the noun and verb must have a single primary stress and word
melody, and must exhibit the syntactic properties discussed in section 2.1.
These facts follow if vowel shortening applies within the same lexical
submodule or morphological construction type as the subcompound.
Now, one might entertain the idea that the domain of stress, word
melody, and vowel shortening can be specified in terms of the notion
phonological word ~ la Sproat (1986), without appealing to the lexical
module or the corresponding morphological type. In order to evaluate
this proposal, it is first necessary to distinguish two meanings of the term
'phonological word'. In the sense of Liberman and Prince (1977), it refers
to a metrical structure created by putting together a number of feet. That
is, phonological word is the output of syllabification, foot tree formation,
and word tree formation. We refer to this concept as r~TRIC~ WORD
(Inkelas 1989, Cole 1990). For Sproat, the term phonological word refers
to a unit that acts as a domain in which phonological rules can apply,
including metrical rules like syllabification, foot tree formation, and word
tree formation. We refer to this concept as PROSODIC WORD. Thus, the
prosodic word defines the domain on which a metrical word is constructed;
in other words, it is the label for a morphological construction type. The
term 'phonological word' has been used in the literature indiscriminately
to refer to either metrical word or prosodic word.
Given this distinction between a phonological structure (metrical word)
and a domain in which the structure is built (prosodic word), the domain
of stress cannot be the metrical word, because metrical words are the
result of stress assignment. The metrical word cannot be the domain of
vowel shortening either. A noun with a case clitic forms a single metrical
word; together, they take a single stress and word melody: dahii-m~ "in
the yogurt"; sabzii-se "with the vegetable". Yet, the vowel in these nouns
98 TARA MOHANAN

cannot be short: *dahi-m~; *sabzi-se. Therefore, the domain of vowel

shortening cannot be captured in terms of the metrical w o r d y
The alternative is to state stress assignment, word melody, and vowel
shortening in terms of the prosodic word. If we assume that prosodic
words are the output of a particular sequentially ordered lexical submodule
(Inkelas 1989), our arguments that NI is in the lexical module remain
intact. If we do not assume sequential modularity, as in Sproat's (1986)
approach, prosodic wordhood is a way of representing a morphological
construction type. Thus, whether in terms of sequential modularity or
morphological construction types, noun incorporation in Hindi must be
analysed as belonging in the lexical module.

2.3.2. Morphology
Morphological evidence also points to the conclusion that NI in Hindi is
part of the lexical module: the derivational suffix -vaalaa, which may be
thought of as an agentive-occupational marker, can be attached to the
N + V sequence as illustrated in (35):
(35) gho.de becne vaalaa
horse-PL sell-NF
horse-seller
The sequence N + V-vaalaa in (35) takes a single primary stress and word
melody, and can undergo stem-final high vowel shortening. The object
nominal in (35) cannot be modified or conjoined, or take a case clitic.
Furthermore, it must be adjacent to the verb. These properties of the
N + V-vaalaa construction can be explained if we analyse the N + V as

25 An anonymous reviewer suggests that these facts can be captured in terms of the notion
metrical word if we represent, e.g. dahii-mg as a metrical word containing another metrical
word. ([w[wdahii]mg]). This proposal, however, violates the Strict Layer Hypothesis in
Prosodic and Metrical phonology which disallows the recursion of prosodic metrical units.
If we allowed metrical units to contain metrical units of the same level, there is nothing that
prevents representations such as the following:
SYLL SYLL

SYLL SYLL SYLL

SYLL SYLL SYLL SYLL

Given that such freedom is widely acknowledged to be undesirable, I take it that the
reviewer's proposal is unimplementable.
 WORDHOOD AND LEXICALITY 99

a n N I C . If so, N I C is a n i n p u t to d e f i v a t i o n a l m o r p h o l o g y . I s u g g e s t t h a t
t h e s t r u c t u r e o f (35) is as in (36):

(36) N

N SUFFIX

V SUFFIX

ghode bec- ne vaalaa

horse-seller
A c c e p t i n g t h e w i d e l y h e l d a s s u m p t i o n t h a t d e r i v a t i o n a l m o r p h o l o g y is
p a r t o f t h e l e x i c o n , w e m u s t also a s s u m e t h a t N I is p a r t o f t h e lexicat
m o d u l e . 26
T h e a t t a c h m e n t o f -vaalaa, t h o u g h fairly p r o d u c t i v e , is n o t u n c o n -

26 The derivational suffix -vaataa must be distinguished from the -vaalaa used as a relative
clause marker in (i), and as the marker of immediate future in (ii) (Verma 1971):
(i) mote pann6 aur gaandaar tasviir6 vaalii yah kitaab...
thick pages and grand pictures that book
That book, which has thick pages and grand pictures...
(ii) raam kaanfrens-ke liye k~lifomiya jaane vaalaa hai.
Ram conference-for California go about to is
Ram is going soon/about to go to California for a conference.
The form -vaalaa also appears in an N + vaalaa construction, where it is an occupation
marker, illustrated in (iii):
(iii)a. ghoqaa g~ode-vaalaa
horse a horse keeper/seller (51)
b. sabzii sabzii-vaalii
vegetable vegetable vender/seller (F)
c. nfithaaii mi~aaii-vaataa
sweets sweet-seller/marker (M)
d. t.%laa t%le-vaalaa
cart cartpusher/owner (M)
The -vaalaa in (iii) is deafly derivationak The -vaataa that appears in (35) shares with the
-vaalaa in (iii) all the properties we have discussed so far as being characteristic of lexical
elements, such as the genericity requirement, prohibition against case marking, vowel shor-
tening, and so on. In this respect, the -vaalaa in (35) and (iii) sharply contrasts with those
in (i) and (ii). Therefore, (35) cannot be treated as a relative clause.
100 TARA MOHANAN

strained. Thus, it is not difficult to coin novel forms like caawal khaane
vaalaa "rice-eater" and lakd.ii kaatne vaalaa "wood-cutter", from caawal
khaanaa "rice-eating" and takd.iikaatnaa "wood-cutting", respectively.
But coinages like gaad.ii dhone vaalaa "car washer" and ghod.e knariidne
vaalaa "horse buyer" are extremely awkward if not impossible, even
though the NI's gaadii dhonad "car-washing", and ghode khariidnaa
"horse-buying" are quite acceptable. Irregularities such as these support
the lexical status of NI in Hindi.

3 . V E R B A G R E E M E N T AND N I

Verb agreement in I-Iindi exhibits an alternation between subject agree-

ment and object agreement. A verb agrees with its object if and only if
the object is NOM and the subject is not N O M . 27 In this section, I will
outline the facts of verb agreement in Hindi, demonstrate that agreement
must take place in the phrasal module, and then explore verb agreement
in the NI construction, where the verb can agree with a lexically incorpor-
ated noun. Juxtaposing the analysis of NI as lexical and that of verb
agreement as phrasal (i.e., postlexical) forces us to re-examine and tease
apart different issues surrounding lexicality.

3.1. Verb Agreement: A Phasal Phenomenon

A verb in Hindi can agree only with a NOM argument. In (37) below,
the verb agrees in number, gender, and person with its NOM grammatical
subject (SUB J).
(37)a. ravii baalak-ko ut.haaegaa.
Ravi-N (M) boy-A(M) lifi-FU.M.SG
Ravi will lift up the boy.

b. niinaa baalak-ko ut.haaegii.

Nina-N (F) boy-A(M) tifi-FU.F.SG
Nina will lift up the boy.
In (38), the SUBJ is non-nominative (NONNOM), and the primary object
is NOM. The verb agrees with this OBJ:

27 In Hindi, subjects may bear non-nominative (ergative/dative/instrumental/

locative/genitive) case, depending on meaning. Objects, as we have seen, are either ACC
or NOM,
 WORDHOOD AND LEXICALITY 101

(38) a. ravii-ne ) rot.ii khaayii.

Ravi-E(m) bread-N(F) eat-PERF.F.SG
niinaa-ne j
Nina-E(F)
Ravi/Nina ate bread.

b. ravii-ne kelaa khaayaa.

Ravi-E(M) banana-N(M) eat-PERF.M.SG
niinaa-ne /
Nina-E(F) )
Ravi/Nina ate a banana,
In (39), the primary object is also NONNOM. The verb is in the default or
non-agreeing form: it bears the third person singular masculine inflection:

(39) -ne ] (baalak-ko ] ut.r~aayaa.

Iravii
Ravi-E(M) ~ ~boy-A(M)
niinaa-ne j ~baalikaa-ko
Nina-E(f) girl-A(f)
Ravi/Nina lifted up the boy/girl.
lift-PERF.M.SG

If the SUBJ and the OBJ are both NOM, then the verb agrees with the
SUB J, as in (40):

(40) a. ravii ( rotii ] khaaegaa.

Ravi-N(M) ~ br;ad-N(F) t eat-FU.M.SG
[kelaa
banana-M(M))
Ravi will eat bread/banana.

b. niinaa (ro.tii ] khaaegii.

Nina-N(F) ~bread-N(F) ~ eat-FU.F.SG
/ kelaa /
kbanana-N (M))
1~ TARA MOHANAN

Nina will eat bread/banana.

The generalization about verb agreement in Hindi recognized widely in
the literature (e.g. Kachru, Kachru, and Bhatia 1976, p. 86) is as follows.
If the SUBJ is NOM, the verb agrees with it. If the SUBJ is NONNOM,
and the OBJ is NOM, the verb agrees with the OBJ. If the OBJ is also
NONNOM, the verb is in the neutral form, namely, masculine third
person singular. In short, NOM case is a necessary condition for agree-
ment. The principle governing agreement may be stated as follows: 28
(41) Verb Agreement in Hindi
The verb agrees with its highest argument associated with NOM
case.
Thus, verb agreement in Hindi depends upon the NOM case on the
argument. N O M case is determined in the phrasal module. Therefore,
verb agreement is determined in the phrasal module, and not on the basis
of lexical information alone. 29
One could imagine a situation in some language where the case of the
arguments of a verb is entirely predictable from the le×ical entry of the
verb, and hence, the case requirement for agreement can be satisfied
lexically. But in Hindi this is untenable. First, for the same verb, the
subject may be either NOM or NONNOM, depending on the presence of
a modal:
(42)a. ravii bait h gayaa.
Ravi-N (M) sit-NF go-PERF.M.SG
Ravi sat down.

b. niinaa bait.h gaii.

Nina-N (F) sit-NF go-PERF.F.SG
Nina sat down.

28 The principle in (41) makes reference to the "highest argument": that is, it makes
reference to prominence relations of a particular kind, namely, those among the arguments
of a predicate. The principle could equally well be stated in terms of prominence relations
of another kind, namely, those among grammatical functions: The verb agrees with its
highest grammatical function associated with N O M case. The choice between these two
interpretations of "highest" does not in any way affect the point of this paper.
29 Despite significant differences, analyses available in the literature (for instance, Gait and
Wali 1989; Mahajan 1989) agree on the assumption that Hindi verb agreement is not a lexical
phenomenon.
 WORDHOOD AND LEXICALITY 103

c. (ravii-ko ) bait.h jaanaa pad.aa

j Ravi-N (M) [ sit-NF go-NF modal (obligation)-PERF.M
/niinaa-ko |I
k Nina-N (F))
Ravi/Nina had to sit down.

In (42a) and (42b), the SUBJ is NOM, and the verb agrees with it. But
when the modal of obligation pad.aa is present ((42c)), the SUBJ bears
dative case, and the verb fails to agree with it. 3° Given that agreement is
sensitive to the presence of NOM case, and that the modal can assign
dative case to the SUB J, it follows that agreement with the SUBJ is not
lexical. Since agreement with the SUBJ pre-empts agreement with the
OBJ, it follows that OBJ agreement is not lexical either.
Secondly, as stated earlier, an OBJ is NOM if inanimate, and ACC if
animate. Now, a verb may require that its object be animate or inanimate.
Or it may make no such demand: its object may be either animate or
inanimate. So the animacy of the OBJ nominal, and hence its case, can
be ascertained only after the OBJ and the verb have been put together.
In sum, we must conclude that verb agreement in Hindi takes place in
the phrasal module.

3.2. Agreement with the Incorporated Object

We have seen compelling evidence that in Hindi NIC, the verb and one
of its arguments form a single unit in the lexical module. We have also
seen that agreement between the verb and its argument cannot be lexical.
Let us now look at the facts of agreement in NI:
(43) a. anil kitaab~ beeegaa. (= (15a))
Anil-N (M) book-N.PL(F) seII-FU.M.SG
A. Anil will sell books.
B. Anil will do book-selling.

3o T h e modal does not form a word with the verb: it can be attached to a coordinate verb
structure:

(i) b a h u t baar u.thnaa aur bait.nnaa pa~aa,

a lot times rise-NF and sit-NF modal(obligation)-M.PERF
H a d to get up and sit down m a n y times.
Given our assumption that lexical categories cannot be conjoined, the coordinate structure
ut.hnaa aur bait.hnaa m u s t be a ~', not a V. Therefore, paS.aa is attached to a V.
104 TARA M O I--IAN A N

b. anil-ne kitaab~ bee//. (---(16a))

Anil-E(M) book-N.PL(F) sell-PERF.F.PL
A. Anil sold books.
B. Anil did book-selling.

In (43a), the verb agrees with the masculine singular SUBJ. In (43b), the
verb agrees with the feminine plural OBJ. This holds even when (43b)
has reading B, and is therefore an NIC.
Now, verb agreement is a relation between the entire verbal complex
and one of the arguments; it is not always manifested on the main verb.
In (44), the agreeing unit is not the main verb, but the whole verbal
complex: agreement is marked on the progressive and the be form:

(44) a. anil kitaab6 bec rahaa thaa.

Anil-N (M) book-N.PL(F) sell PROG-M.SG be-M.SG
Anil was selling books.

b. ila kitaab~ bec rahii thii.

Ila-N (F) book-N.PL(F) sell PROG-F.SG be-F.SG
Ila was selling books.

The agreement relation in (44a) is diagrammatically given in (45):

(45) NP V
a~nil --] kitaab~ bec rahaa thaa.
[Anil-N(M) I book-N.PL(F) sell PROG-M.SG be-M.SG

"Anil was selling books."

Under our analysis of NI, the incorporated noun is part of the agreeing
verb complex:

(46)
V

il-N(M~[ ] kitaab6 bec t -taa tl~aa.

book-N.PL(F) sell -tIAB-M.SG be-M.SG

t ......... T
"Anil used to do book-selling."
 WORDHOOD AND LEXICALITY 105

As shown by (43b), the agreeing verbal complex can agree with the OBJ
even when the OBJ is incorporated, that is, when it is internal to the
verbal complex. The agreement relation in (43b) is given in (47):

(47) --
il NP - ~ V
kitaab6 bec -I1.
book-N.PL(F) sell -PERF-F.PL
I ................

"Anil did book-selling."

This brings us directly to the two puzzles posed by the NI construction.

First, incorporation is lexical, and agreement is not; how can a verb agree
with an incorporated argument? Second, if verb agreement is between the
verbal complex and one of its sisters, how can the verbal complex agree
with an element internal to it? In what follows, we will see that solutions
to these problems demand the separation of two dimensions of syntactic
information, those of grammatical functions and grammatical categories.

3.3. A Formal Analysis: Factorization of Information

The restrictions on the NI construction involving nominal modification,
case clitics, adjacency, conjoining, gapping, pauses, stress and word mel-
ody, stem-final high vowel shortening, and vaalaa attachment motivate
the analysis of N + V in Hindi NI as a lexical category" formed in the
lexical module. At the heart of our explanation for these restrictions is
the category structure in (10b). The principle of agreement in (41), on
the other hand, makes no reference to category structure. Agreement in
Hindi is sensitive to the argument status of the nominal, and to NOM
case. The syntactic independence of the argument is not captured by the
category information in (10b), but by the grammatical-function infor-
mation under it.
Without stating it explicitly, what we have done in (10b) is to separate
category information from the predicate-argument and grammatical-func-
tion information required for agreement. Suppose we represent these
different types of information along different dimensions. This separation
is familiar in the literature, and has been argued for in Bresnan (1982a)
106 TARA MOHANAN

within Lexical-Functional G r a m m a r . 31 T h e conditions for agreement can

then be satisfied along the dimension of predicate-argument and grammati-
cal-function information, represented in (48) below as A R G S T R and G F
S T R above the sentence. 3z Two representations for (43b) are given below.
(48a) is the representation for (43b) as an N I C and (48b) is its unincorpor-
ated counterpart:

(48) a. CLAUSE b.

ARG ARG PRED

J"F'\CLAUSE

ARG ARG PRED ARG STR

SUBJ
I I i OBJ PRED I l l
ERG NOM PERF SUBJ OBJ PRED
ERG NOM PERF
---1 ....... t....... L--
anil-ne kitaab~ beNT - 1 ....... l....... l-- GF STR

anil-ne kitaab~ bec~i

........... I.......
N V
t- ---1..............~-- GC STR

Y NP NP V

V
S
!
NP V

S
E v e n though (48a) and (48b) have two distinct category structures, repre-
sented as G C STR, corresponding to the two readings of (43b), the
representation of A R G S T R and G F S T R for both readings is identical.
Observe that this representation satisfies the requirements for agreement.
Although incorporated on the dimension of G C S T R in (48a), the N is
an independent A R G in A R G STR, and an O B J associated with N O M

31 The separation of information is also found in works within GB (Riemsdijk and Williams
1981, Chomsky 1981, Hale 1983, K. P. Mohanan 1984, Zubizaretta 1987). However, gram-
matical function information in these works is also representcd in terms of category labels.
32 Following the framework in T. Mohanan (1990), I assume the separation of argument
structure information and grammatical-function information in the representations in (48).
I will not defend the separation here because it is not directly relevant for the purposes of
this paper.
 WORDHOOD AND LEXICALITY 107

case in GF STR. It is therefore a legitimate controller of agreement. In

short, the solution to the puzzles posed by NI in t-Iindi lies in an organiza-
tion of grammar which allows non-identity of the mother-daughter re-
lations with respect to different dimensions of syntactic information.
Given any parallel representations, it is necessary to specify the con-
straints that hold on the relation between them. First consider the interpre-
tation of anil-ne as the SUBJ and kitaab~ as the OBJ of the verb bec in
(48b). I assume that this association is governed by (49). The convention
of solid and dotted association lines notate the premises and consequences,
respectively, of an if-then conditional.
(49) Optional ARG PRED ARG STR
I
'
!F
]
,,
NP V GC STR

S
(49) is interpreted as: "If the GC STR is [NP, ~']s, and ~r is associated
in A R G STR with a PRED that has an ARG, then associate the NP with
the ARG. [Optional]" The A R G can bear any of the GFs of the PRED.
Thus, (49) allows any NP dominated by S to be associated with an argu-
ment of the predicate. 33
The interpretation of the GF of the N in NIC calls for an additional
principle, given in (50):

(50) ARG PRED ARG STR

J
NON-SUBJ GF STR
1I

V1 GC STR

V2

(50) is interpreted as: "If the GC STR is [N V1]v2, and V~ is associated

in A R G STR with a PRED that has a NON-SUBJ ARG, then associate

33 Within this conception, the grammar also requires principles that state the relationship
between grammatical functions on the one hand, and case, word order, nominal meanings
(e.g. animacy), and the like on the other.
108 TARA MOHANAN

N with the A R G . " (50) allows an N dominated by V to be interpreted as

a nonsubject argument of the predicate.
Within the formalism of LFG, these principles will be stated as the
annotations of phrase structure rules which state the mapping between c-
structure and f-structure. For example, (49) corresponds to the annotation
t GF = $ on the NP in a PS rule S ~ NP*, V. (50) corresponds to the
annotation ]' NON-SUBJ = ~ on the N in a PS rule V---~ N, V. (49)
states the effect of 'scrambling', that is, the independence of grammatical
functions from word order, which is stated in a movement theory in terms
of optional movement. Within the formalism of head movement (Baker
1988), (50) corresponds to the head movement of an N from an NP into
a V. As in the theories mentioned above, (49) and (50) are derived from
a universal theory that expresses the invariance and constrained valiability
in the relation between arguments and their grammatical functions on the
one hand, and their grammatical categories on the other. 34

3.4. Absence of Doubling in Hindi N1

The English compound stage-manage has the same GC STR as postulated
for Hindi NI: [N V1]v2. Nevertheless, as shown by the possibility of inde-
pendent phrasal objects, e.g., the show in John stage-managed the show,
the N in the [N V1] cannot be a syntactic argument of the verb. The
grammar of NI that we have sketched so far predicts that unlike what
happens in the English stage-manage, an incorporated object in Hindi
cannot co-occur with a phrasal object, under the assumption that a clause
cannot contain two instances of the same grammatical function (function-
argument biuniqueness within LFG). This prediction is borne out by (51c):
(51)a. anil jaanvar becegaa.
Anil-N (M) animal-N sell-FU.M.SG
A. Anil will sell animals.
B. Anil will do animal-selling.

b. gho .dr-ko anil becegaa.

horse-PL-A Anil-N (M) sell-FU.M.SG
Anil will sell the horses.

c. *ghod.6-ko anil jaanvar becegaa.

horse-PL-A Anil-N (M) animal-PL seU-FU.M.SG
Anil will do animal-selling of the horses.

34 See, for instance, Alsina (1993),

 W O R D H O O D AND L E X I C A L I T Y 109

(51c) is ungrammatical because its GF structure contains two PR.OBJ's,

animals and horses.
Rosen (1989) points out that natural languages fall into two types with
respect to modifier stranding as well as doubling in the NI construction.
Languages like Caddo and Mohawk allow an incorporated noun to be
modified by a 'stranded' word or phrase outside the verb complex (Sadock
1985, pp. 402, 407; Baker 1988, p. 93; among others). They also allow
doubling. In contrast, Polynesian and Micronesian languages disallow such
modification and doubling (Rosen 1989, pp. 311-2). Clearly, Hindi be-
longs to the latter class with respect to both these properties. The status
of this typological distinction and its theoretical consequences will be taken
up in the next section.

4. M O D U L A R I T Y AND R E P R E S E N T A T I O N

What we have seen so far is that Hindi has an N + V sequence which has
the crucial characteristics of NI: morphologically, it behaves like a single
word, but the N within this word is a syntactic argument. This section
examines how this duality is formally expressed in other accounts of NI
(Sadock 1980, 1985, 1986, 1991; Baker 1985, 1988), and articulates the
substance that the analysis in this paper shares with their accounts. It also
shows in what ways my proposals differ from those of Sadock and Baker.
In the analysis of NI in Hindi, we are faced with issues of both represen-
tation and modularity. The modularity issue can be stated as: is Hindi NI
part of the lexical module or the phrasal (=post lexical) module? As far
as the issue of representation is concerned, we have two questions. First,
does Hindi NI need to distinguish between the representation of infor-
mation relevant for the structure within words (lexicat) and that relevant
for the structure across words (phrasal)? Second, does the representation
of Hindi NI need to distinguish between grammatical categories and gram-
matical functions? We turn now to these issues.

4.1. The Lexicality of NI

To repeat, most accounts of NI acknowledge that the N + V sequence in
an NI construction exhibits lexical properties. This raises the question
110 TARA MOHANAN

what we mean by the term 'lexical'. 3s In order to clarify the issues of

lexicality, it is essential that we separate at least two different conceptions:
(i) 'lexical" as referring to a MODVLZin which representations are formed,
and (ii) 'lexical' as referring to a UNIT OF REPRZSENTATION.The claim that
NI in t-Iindi is lexical can be interpreted as either (52a) or (52b), or both:

(52)a. NI is part of the lexical module;

b. The N + V sequence in NI is a lexical category.

The claim in (52b) seems uncontroversial: all accounts of NI accept that

the lexical properties of NI require it to be represented as a lexical category
at some dimension of representation or stage in the derivation (see (11),
(12), and (13)). One of the claims of this paper is that, in addition, the
N + V sequence must be concatenated in the lexical module ((52a)). In
other words, the N + V sequence in the Hindi NIC is lexical in both
senses: it is a lexical category, and is formed in the lexical module. The
claim in (52a) is implicit in Sadock's analysis of NI: the level of representa-
tion in which NI is a lexical category, called the morphological representa-
tion, as distinct from the syntactic representation, encapsulates the auton-
omy of morphology required by (52a).
The statement in (52a) becomes a logical consequence of the statement
in (52b) if we assume that lexical categories cannot be created in the
postlexical module, and phrasal categories cannot be created in the lexical
module. Baker accepts (52b), but rejects (52a) by claiming that representa-
tions of phrasal categories may constitute the input to the formation of
lexical categories through head movement. Our evidence for (52a) comes
from the analysis of the phonological and morphological properties of NI:
it undergoes lexical phonological processes such as stress and word melody
assignment and final high vowel shortening. It is also an input to deri-
vational affixation, widely acknowledged to be a characteristic of the
lexical module. It is unclear how Baker's analysis of NI can account for
these facts.

35 During the early days of generative grammar, lexical was that which is idiosyncratic or
unpredictable in a language. The lexicon was therefore viewed as the list of morphemes,
specifying all their unpredictable properties. With Chomsky (1970), the lexicon came to
include not only idiosyncratic information, but also certain kinds of predictable information
such as the regularities of derivafional morphology and compounding. Over the years, the
lexicon has developed into a rich and organized module of the grammar whose output
captures the notion 'word' (e.g. Lieber 1980, Bresnan 1982a, K. P. Mohanan 1982, Kiparsky
1982).
 WORDIrlOOD AND LEXICALITY 11-1

4.2. The Syntacticity of NI

Sadock's lexical (=morphological) representation of ghod.e bec "horse-
sell" would be as in (53b), and the representation of the syntactic ob-
jecthood of the incorporated noun as in (53a):

(53)a. VP
/Q SYNTACTIC
NP V
g/'od.e bec
b.
",,/
NP V
LEXICAL (=MORPHOLOGICAL)

(53a) and (53b) are parallel representations: they do not bear an input-
output relation. Baker derives the representation in (53b) from (53a)
through head movement:

(54) VP

NP V

N V

tiii i

ghode bec
When (53) and (54) are juxtaposed, it becomes obvious that the coindexed
trace within a single dimension of representation in (54) performs the
same task as the linking of entities in two dimensions of representation in
(53): both express the idea that the nominal inside the verb is at the same
time the object of the verb.
Like Sadock and Baker, the analysis in this paper recognizes that the
incorporated noun has syntactic reflexes. In particular, it functions as the
syntactic object of a verb. The syntactic reflexes of NI are expressed in
112 TARA MOHANAN

the representation of its argument and GF structure, where the incorpor-

ated object ((48a)) is an argument on par with a phrasal object ((48b)).
In order to factor out the notational differences from the substantive
differences between the analyses, we must bear in mind that both Sadock
and Baker represent grammatical functions in terms of category labels
and dominande relations. The representations in (55a, b) are the counter-
parts of (53a, b) in my analysis: 36

(55) a. CLAUSE
GF STR

OBJ PRED
ghode bec
b. N V
GC STR

V
The above discussion reveals that the formal representations in (53), (54),
and (55) are roughly equivalent in recognizing the dual nature of NI: they
agree that NI is simultaneously 'lexical' (it constitutes a lexical category)
and 'syntactic' (it consists of a predicate and its object). Thus, the facts
of adjacency, prohibition against conjoining and against the occurrence of
case clitics, wh-words, and pronouns within NIC, are equally well ac-
counted for by the structure IN V]v in the three representations. Verb
agreement and the prohibition against doubling are accounted for
by the structure [NPV]vp in (53) and (54), corresponding to [ . . . OBJ
PRED]cLAUSEin (55).
A significant difference then is that while (55) factors out the representa-
tion of grammatical functions from that of grammatical categories along
two dimensions of syntactic representation, (53) and (54) conflate them
into a single representation. Empirical evidence for the former position
comes from the analysis of modifier stranding and gapping in Hindi.

The tree notation in (55a) is a notational variant of the bracket notation of functional
structure in LFG:
OBJ [gho .dell
PRED [bec] J
 WORDHOOD AND LEXICALITY 113

4.2.1. Modifier Stranding

As noted earlier, an incorporated nominal cannot be modified in Hindi.
One of Baker's motivations for analysing incorporation as head movement
in syntax is that in a number of languages, for instance, Mohawk, Onon-
daga, and Southern Tiwa, the incorporated noun can be modified by a
word or phrase that "remains morphologically outside the verb complex"
(Baker 1988, p. 93). An extension of the analysis to Hindi would allow
incorrect representations containing stranded modifiers such as in (56).
As far as I know, there is no constraint in Baker's analysis that accounts
for the inability of the N to leave behind a stranded modifier:
(56) S

NP VP

NP V

A N N V

anil-ne
4
pumanii
I
ti
I
kitaabfii
t
beNT.
Anit-E old-F bookl-N,PL selI-PERF.F
*Anil did old book-selling.

Thus, a head movement analysis must either assume that NI in Hindi is

not the result of head movement, or stipulate that NI in Hindi prohibits
modifier stranding.
The analysis of the incorporated N as being associated with an NP via
coanalysis (following Sadock 1985, 1991) faces a similar problem. Given
the basic premises of Sadock's theory, it is difficult to see how the repre-
sentation of NI as in (57) can be ruled out for Hindi:
 114 TARA MOHANAN

(57) V

A N V [MORPHOLOGICAL]

anil-ne
I
pur~nii
I
kitaab~
I
becii.

A N
I V [SYNTACTIC]

VP

Given what we have said about Hindi NI, neither (56) nor (57) are
legitimate structures: The noun kitaab~ in the N + V sequence is part of
a lexical category; its modifier puraani is outside the lexical category. By
the Lexical Integrity Hypothesis (see section 2.1.6), therefore, the internal
structure of kitaab~ becriis opaque to phrasal modification. 37 Within our
analysis, there is no representation in which the N in NIC is associated
with an NP.

4.2.2. Gapping
Recall that neither the N nor the V in NIC can be gapped in Hindi ((8b),
(9b)). Given our analysis of the N + V sequence as a lexical category, the
failure to gap follows from the Lexical Integrity Hypothesis. Since the
Lexical Integrity Hypothesis is stated on grammatical categories, and does
not hold on grammatical functions, syntactic phenomena that are stated
on grammatical functions, like verb agreement in Hindi, are irrelevant for
the principle. Given a syntactic movement analysis or the co-analysis of
morphology and syntax, where grammatical functions are represented in
terms of categories, the failure to gap remains unexplained.
One may think that the explanation lies in the N and the V being lexical
(X °) categories. However, as shown by the example in (58) below, there
is no constraint against the gapping of a lexical category in Hindi.

37 Sadock (1980) observes that modifier stranding and anaphoric reference in Greenlandic
Eskimo violate the Lexical Integrity Hypothesis. For a response, see Simpson (1991, pp.
226-237).
 WORDHOOD AND LEXICALITY 115

(58) anil-ne laal kaar becii aur raam-ne niilii khariidii.

Anil-E red car-N sell-PA and Ram-E blue buy-PA
Anil sold a red car and Ram bought a blue - - .

Hence the failure of the N and V in NI to gap cannot follow from these
constituents being lexical categories.
In sum, verb agreement and doubling show that the N in the Hindi NIC
corresponds to a syntactic argument. Yet, NIC does not allow modifier
stranding or gapping. This peculiar combination of syntactic and lexical
properties makes Hindi NI significantly different from the instances of NI
I have seen in the literature.

4.3. A Purely Lexical Analysis

Rosen (1989) claims that NI is purely lexical. She distinguishes two types
of NI, which she calls Compound NI and Classifier NI, which differ in the
following ways:

(59)
Compound NI Classifier NI
(a) the verb's valency is reduced x/ X
(b) NP OBJ's can co-occur with NI X ,/
(c) stranded modifiers are allowed X

Rosen claims that doubling (59b) and stranding (59c) are found only in
languages that allow null heads in the syntactic structure, independently
of NI. The properties of classifer NI follow from the combination of the
possibility of syntactic null heads on the one hand, and purely lexical NIs
on the other. The apparent objecthood of the incorporated noun is the
result of the incorporation of a non-argument co-occurring with a null
head semantically related to it.
Rosen's claim, then, is that the NI construction does not in itself have
any syntactic representation associated with it: the apparent syntactic
properties which have been attributed to the incorporated noun are the
effects of the independent existence of null heads in the language. Thus,
she rejects the substance of the representations in both (53a) and (55a).
Rosen's theory predicts a correlation between the possibility of doubling
(59b) and stranded modifiers (59c), both being consequences of the possi-
bility of null heads. This correlation has been questioned by Sadock
(1991), who shows that languages like Southern Tiwa and West Green-
116 TARA MOHANAN

landic allow stranded modifiers, but not doubling, and that Rosen's pro-
posed solutions to this problem are untenable.
Hindi NI is also incompatible with Rosen's typology. Hindi does not
allow null heads. Nor does NI in Hindi allow modifier stranding ((59c))
or doubling ((59b)). Therefore, Hindi NI should be compound NI. Yet,
there is no reduction of valency in Hindi NI in that NI does not make
a transitive clause intransitive: the incorporated nominal behaves as an
independent A R G with respect to verb agreement. Thus, the predicted
correlation between (59a) on the one hand, and (59b, c) on the other, is
incorrect.
In the light of these facts, Rosen's typology of NI in (59) must be
revised as (60):

(60)

a, N is an argument
b. Modifier stranding allowed
ABC
x
x
V
x
V
V
c. Doubling allowed X X X

Of the four classes in (60), type A is purely lexical (Rosen's compound

NI). But by the definition of NI we have adopted, this is not NI, but
simply compounding. Type D has the most characteristics normally associ-
ated with syntactic constructions (Rosen's classifier NI). Southern Tiwa
and West Greenlandic fall into type C, and Hindi into type B. In short,
the claim that NI is purely lexical, and that the incorporated noun has no
syntactic status, must be abandoned.
The central claims that I advance in this paper are that an analysis of
NIC in Hindi requires syntactic theory to recognize GF representation
and GC representation as separate though interacting syntactic representa-
tions, and that lexical categories are the output of the lexical module.
Now, the construction that has been called NI obviously allows certain
typological differences across languages. The analysis of NI in terms of
(55) captures the absence of modifier stranding, doubling, and gapping in
a language such as Hindi. Clearly, this analysis cannot be directly extended
to languages which do allow modifier stranding and/or doubling. A univer-
sal theory of NI should predict the range of typological variation. How-
ever, such a theory is not within the scope of this paper.
 W O R D H O O D AND L E X I C A L I T Y 117

5. NEGATION AND N I

Our analysis of NI hinges on the dual structure that separates category

information from grammatical function information ((55)), contra Sadock
((53)) and Baker ((54)). We found evidence for this separation in modifier
stranding and gapping. In this section, I will show that the requirements for
sentential negation in Hindi lend additional support to this dual structure.
Sentential negation in Hindi simultaneously requires in category structure
that the negative word be attached to the left of the verb, and in grammat-
ical function structure that it be adjacent to the predicate.

5.1. Phrasal and SententiaI Negation

Consider the facts of the placement and scope of the negative word nahTi
"no/not" in (61a, b):
(61)a. raam-ne ilaa-ko nahii dekhaa thaa.
Ram-E Ila-A not see-PERF be-PA
Ram had not seen Ila.

b. niinaa ravii-ko kitaab~ nahii bhej rahii thii.

Nina-N Ravi-D book-PL.N not send P R O G be-PA
Nina was not sending the books to Ravi.
The sentences in (61) are interpreted as instances of sentential negation
oniy if the tonic, or the nuclear stress of the sentence, is placed on the
word nah~E If the tonic falls on the phrase immediately before the word
naMh the result is the negation of that phrase, as illustrated below. The
tonic is indicated by underlining.
(62)a. niinaa ravii-kjo kitaab~ nahTg bhej rahii thii.
Nina-N Ravi-D book-PL.N not send P R O G be-PA
Nina was sending not the books (but something else) to Ravi.

b. niinaa kitaab~ ravii-ko nah~i b%j rahii thii.

Nina-N book-PL.N Ravi-D not send P R O G be-PA
Nina was sending the books not to Ravl (but someone else).

c. ravii-ko kitaab6 nfinaa nah~ b%j rahii thii.

Ravi-D book-PL-N Nina-N not send P R O G be-PA
Not Nina (but someone else) was sending (the) books to Ravi.
118 TARA MOHANAN

Phrasal negation as in (62) involves focussing the negated phrase: the

focussed phrase in (62a) is "books", and the sentence presupposes that
Nina sent something to Ravi, as is characteristic of the negation of a focus.
Similarly, a wh-phrase is allowed in a sentence containing phrasal negation
only if the negated phrase is not distinct from the wh-phrase, again a
characteristic widely acknowledged as a property of focus.
Contrary to what one might conjecture on the basis of (62), the negative
in phrasal negation is not necessarily adjacent to the verb, as shown by
(63a, b):
(63) a. ?niinaa-ne nahii ravii-ko kitaab~ bhejTi.
Nina-E not Ravi-D book-PL-N send-PERF
Not Nina sent the books to Ravi.

b. ?niinaa-ne ravii-ko nahri kitaab~ bhejii.

Nina-E Ravi-D not book-PL-N send-PERF
Not Nina sent the books to Ravi.
Although the examples in (63a, b) are a little odd, the oddness disappears
if the right context is provided, as in (64a, b), where the negated phrase
is a contrastive structure:
(64)a. niinaa-ne nah]7 ilaa-ne ravii-ko kitaab~ b~ejii.
Nina-E not Ila-E Ravi-D book-PL-N send-PERF
Not Nina but Ila sent the books to Ravi.

b. niinaa-ne ravii-ko nahii mohan-ko kitaab~ bhejii.

Nina-E Ravi-D not Mohan-D book-PL-N send-PERF
Nina sent the books not to Ravi but to Mohan.
None of the sentences in (63) and (64) can be interpreted as sentential
negation. In sentential negation, the negative is attached to the left of the
verb, while in phrasal negation, it is attached to the right of the negated
phrase. In order to account for this, we assume the following structures: 38

3s T h e structures in (65) assume for the sake of concreteness that the sister of nahi"i is an
X-bar category. Nothing in the analysis crucially hinges on this assumption: the sister of
nahfl m a y be an X-zero category.
 W O R D H O O D AND L E X I C A L I T Y 119

(65)a. Phrasal negation b. Sentential negation

2 V

X nah~ nah~g V
(65a) predicts that if the negated expression is the verb rather than the
entire clause, the negative should occur after the verb. Thus, in contrast
to the sentences in (61) which cannot be interpreted as instances of ne-
gation of the verb, those in (66) can be interpreted only as negation of
the verb:
(66)a. ?raam-ne ilaa-ko dekkaa nahi~.
Ram-E Ila-A see-PERF not
Ram didn't see Ila.

b. ?niinaa-ne ravii-ko kitaab~ bheji'i nah~L

Nina-E Ravi-D book-PL-N send-PERFnot
Nina didn't send the books to RaJd.
Like the sentences in (63), those in (66) are somewhat odd, because they
require special contexts. Once the contexts are provided, they are perfectly
acceptable :39
(67)a. raam-ne ilaa-ko dekhaa nahrf, bas yaad kiyaa.
Ram-E Ila-A see-PERF not enough memory-N do-PERF
Ram didn't see Ila, only remembered her.

39 In a contrastive structure, phrasal negation can also be achieved by placing the tonic on
the negated phrase and attaching the negative to the left of the verb, as in (i):
(i) niinaa-ne ravii-ko kitaab6 nahi~ b%j~'L
Nina-E Ravi-D book-PL-N not send-PERF
Nina didn't send the books to Ravi (someone else did).
in (i), the position of the negative indicates sentential negation, but the specific intonation
narrows the scope of the NEG to thc phrase that bears the nuclear sentence stress. It is
interesting to note that if the negated phrase is indicated by the tonic, the NEG cannot
occur to thc right of the verb:
(ii) *niinaa-ne ravii-ko kitaab6 b%ji'f nahrf.
Nina-E Ravi-D book-PL-N send-PERF not
In (ii), the scope of NEG must be the verb because of the position of NEG, but it must be
the underlined NP because of the tonic. The result is an uninter~retable sentence.
120 TARA MOHANAN

b. niinaa-ne ravii-ko kitaab~ bhejri nahi1, khud

Nina-E Ravi-D book-PL-N send-PERFnot self
jaa-kar dii.
go-NF give-PERF
Nina didn't send the books to Ravi, but took them to him
herself.
The structures in (65) also correctly predict that naM; can be sentence
initial if and only if it is immediately followed by the verb, and its interpre-
tation is that of sentential negation:
(68) a. nahii bhejii niinaa-ne ravii-ko kitaab~.
not send-PERF Nina-E Ravi-D book-PL.N
Nina did not send the books to Ravi.

b. *nahii niinaa-ne kitaab~ ravii-ko bhej/i.

not Nina-E book-PL.N Ravi-D send-PERF

c. *nahH ravii-ko kitaab6 niinaa-ne bhejii.

not Ravi-D book-PL-N Nina-E send-PERF

In (68a), the N E G is attached to the left of the verb, and satisfies the
structure for sentential negation ((65b)). In (68b, c), however, the NEG
has nothing to its left to satisfy (65a), and the element to its right is
not the verb, so it cannot satisfy (65b). The sentences are therefore
ungrammatical.

5.2. The Absence of Sentential Negation in NI

Given the requirement for sentential negation that N E G must immediately
precede the verb ((65b)), and the GC STR structure [N V]v for NIC
((55)), we predict that nahffcannot occur between the N and V in NIC.
We will see below that this prediction is correct.
Since the N and V in NIC form a verbal unit, one would expect senten-
tial negation in NIC with nahrf to the left of the N + V. As it happens,
however, this is not possible. The net result is that NIC does not permit
sentential negation. I will show that this gap is explained by the indepen-
dently required condition that the NEG must be adjacent to the PRED.
That is, Hindi requires two conditions for sentential negation: in terms of
categories, the N E G must immediately precede the verb; in terms of
grammatical functions, it must be adjacent to the predicate. The need to
 WORDHOOD AND LEXICALITY 121

distinguish VERB and PREDICATEin this manner justifies the separation of

the two dimensions of structure in (55).

5.2.1. Sentential Negation and NI

Consider (69), which is (15a) with nahiTbefore the verb:
(69) anil kitaab6 nab11 becegaa.
AniI-N (M) book-N-PL(M) not selt-FU.M
(i) Anil will not sell (the) books.
(ii) *Anil will not do book-selling.
Recall that (15a) had two interpretations, one of them being reading B,
associated with NI. Reading B is not available to (69). This is correctly
predicted by our assumptions about the structures of NIC ([N V]v in (55))
and of sentential negation ([nahi$9]v in (65b)): the latter cannot occur
inside the former: *[N[[naM~9]v]v]v.
Our analysis of Hindi NI also correctly predicts that if nah[~ occurs
between an animate NOM object and the verb, the result should be
ungrammatical ((70b)): if an animate object is NOM, it must be part of
NIC, as in (70c), but nahi~ cannot occur between the N and V in NIC:
(70)a. ilaa bacc6-ko nahii samhaaltii hal.
Ila-N children-A not rake care of-HAB be-PR
Ila doesn't take care of the children.

b. *ilaa bacce nahii samhaaltii hai.

Ila-N children-N not take care of-HAB be-PR

c. ilaa bacce samhaalti hai.

Ila-N children-N take care of-HAB be-PR
Ila babysits (as a job).

Now, given the NI structure [NV]v, it should be possible to attach

nahi~ to the left of the structure, as in (71) below, and get sentential
negation with reading B. However, the only interpretation available to
(71) is that of negation of the subject to the left of nahiT:
(71) ?anil nab11 kitaab6 becegaa.
Anil-N (M) not book-N-PL(M) sell-FU.M
(i) Not Anil (but someone else) will do book-selling.
(ii) *Anil will not do book-selling.
122 TARA MOHANAN

In order to explain why (71) cannot be an instance of sentential negation,

we must take a detour to look at the facts of nahff in complex predicates
in Hindi.

5.2.2. Negation in Complex Predicates

Hindi has a construction, illustrated in (72), in which a noun combines

with a verb to form a complex predicate: 4°

(72)a. anil-ne ilaa-ko yaad idyaa.

Anil-E(M) Ila-A(F) memory-N (F) do-PERF.M
Anil remembered Ila.

b. anil-ne ilaa-kii pra]amsaa kii.

Anil-E(M) Ila-G(F) praise-N (F) do-PERF.F
Anil praised Ila.

The noun yaad "memory" in (72a), and pra~amsaa "praise (N)" in (72b),
form complex predicates (henceforth CP) with the verb kar "do". They
are CPs because (i) they function as a single predicate semantically, and
(ii) the burden of determining the number of arguments in the clause, and
their case marking, is not borne by the verb alone, but is shared by the
noun.
Both NI and the CPs in (72) involve a noun + verb sequence that forms
a single unit. However, they differ in fundamental ways: (i) unlike the
noun in NI, the noun in the CP is semantically predicative, and can
contribute to the argument structure of the clause; and (ii) NI is formed
in the lexical module, and CP is formed in the phrasal module. I will
briefly discuss the relevant aspects of the A R G STR and GC STR of CPs
below, but will not defend these structures in this paper. 41
Evidence points to the CP in Hindi having the GC STR illustrated in
(73):

40 In the literature on complex predicates, the noun is often called the "host", and the verb
the "light verb" (see Cattell 1984).
41 A detailed analysis of the CPs in Hindi, and a defense of the structure of CPs assumed
in this paper, are given in T. Mohanan (1990, chapter 8).
 WORDHOOD AND LEXICALITY 123

(73)

N V

1
yaad kar
t (= remember)
memory do
N o w , t h e n o m i n a l h o s t s in t h e C P c o n s t r u c t i o n fall i n t o t w o classes with
r e s p e c t t o v e r b a g r e e m e n t : t h o s e t h e v e r b c a n a g r e e w i t h ( e . g . , (72b)),
a n d t h o s e t h e v e r b c a n n o t a g r e e with ( e . g . , (72a)). W e can f o r m a l l y
distinguish t h e t w o k i n d s o f N + V C P s in t e r m s o f t h e a r g u m e n t status
o f t h e N w i t h i n t h e C P (T. M o h a n a n 1990). T h e N is an a r g u m e n t in
(74a); t h e v e r b m a y a g r e e with it. It is n o t an a r g u m e n t in (74b); t h e v e r b
c a n n o t a g r e e w i t h it. 42

(74) a. Agreement possible 43 b. Agreement not possible

ARG PRED PRED ARG STR

pra~amsaa kar yaad kar

praise do memory do

N
l !
V
!
N
!V

V V GC STR
N o w to t u r n to t h e facts o f nahg p l a c e m e n t a n d s e n t e n t i a l n e g a t i o n in
t h e s e s t r u c t u r e s . G i v e n t h e s t r u c t u r e f o r s e n t e n t i a t n e g a t i o n in (65b), a n d
t h e G C S T R o f C P s in (73), it s h o u l d b e p o s s i b l e t o p l a c e nahff b e t w e e n
t h e n o u n a n d v e r b , as in (75a), o r to the left o f t h e e n t i r e C P , as in (75b):

42 In precisely those CPs where the verb can agree with its N host, the argument providcd
by the host bears genitive, instrumental or locative case, and a non-subject non-object
function. In contrast, in those CPs where the verb cannot agree with its host, the argument
contributed by the host bears either ACC or NOM case, depending on whether it is animate
or not, indicating that it is a primary object. The assumption that the nominal host in (74a)
is an argument, while that in (74b) is not, correctly predicts this set of correlations. The N
host that is an argument is the primary object of the clause ((74a)), so there cannot be
another primary object. A nominal that is v_otan argument cannot be an object ((74b)), and
the object function can be occupied by another argument.
43 At ARG STR the predicative noun pra~arnsaain (73a) is an ARG; at the level of semantic
structure it is an independent predicate with its own semantic dependents.
124 TARA MOHANAN

(75)a. V b. V

N V nahii V

nahii V N V

The structures in (75a) and (75b) are instantiated by (76a) and (76b)
respectively:

(76) a. anil-ne ilaa-ko yaad nahii kiyaa.

Anil-E(m) Ila-A(F) memory-N(F) not do-PERF.M
Anil didn't remember Ila.

b. anil-ne ilaa-ko nahi~ yaad kiyaa.

Anil-E(M) Ila-A(F) not memory-N(F) do-PERF.M
Anil didn't remember Ila.

The N in the CP in (76a, b) is one that the verb cannot agree with.
According to our analysis, it is not an argument of the verb ((74b)). What
is interesting is that with an N that can control agreement ((74a)), nahg
cannot be placed to left of the N for sentential negation ((77b)):

(77) a. anil-ne ilaa-kii pra~amsaa nahii kii.

Anil-E(M) lla-G(F) praise-N(F) not do-PERF.F
Anil didn't praise Ila.

b. ?anil-ne ilaa-kii nahii pra~amsaa kii.

Anit-E(M) Ila-G(F) not pra&e-N(F) do-PERF.F
Anil praise not Ila (but someone else).
*Anil didn't praise IIa.

We can account for the contrast between (76b) and (77b) by requiring the
NEG to be adjacent to the predicate in sentential negation. Consider the
relevant aspects of the structure of CPs in (76a, b) and (77a, b), given in
(78a-d):
 WORDHOOD AND LEXICALITY 125

(78) a. ARG PRED ARG STR

prasamsaa nahii
"praise'~ ~

GC STR

V
b. ARG PRED ARG STR
nahii prasamsaa
N ~ "praise" ] "dv" ]

GC STR

V
C, ..PRED ......... ARG STR
yaad nahii kii I
"memory" "do"
I

GC STR

V
d. PRED ARG STR
nahii yaad kii
v"mem°ry'-------~'
"d---~-
O''

GC STR

V
126 TARA MOHANAN

In each of the representations in (78), name" is attached to the left of the

V, as required by (65b). However, in the structure that corresponds to
the ungrammatical sentence ((78b)), nahff is not adjacent to the PRED.
In (78a) and (78d), it is clearly adjacent to the PRED. In (78c), it is not
non-adjacent, and therefore, the adjacency condition is not violated.
The two conditions for sentential negation can then be stated as in (79):
(79) In sentential negation:
(a) NEG is attached to the left of the verb in GC STR
(=(65b));
(b) NEG must be adjacent to the PRED in A R G STR/GF
STR.
(79a) is required to account for the ungrammaticality of (70b), and (79b)
to account for the ungrammaticality of (77b). If both (79a) and (79b) are
necessary in the grammar of Hindi, then the dual representations in (55)
and (78) are also necessary.

5.2.3. Absence of Sentential Negation in NI: An Explanation

Thus, NEG placement in Hindi is sensitive to two distinct representations.
The dual representation of NI along the dimensions of A R G / G F STR
and GC STR ((55)), coupled with the independently required conditions
on NEG placement in (79a) and (79b), correctly predicts an interesting
syntactic gap: sentential negation is impossible in NI in Hindi.
As explained earlier, the NEG cannot occur between the N and V in
the NIC, because it would create the ill-formed GC STR: this explains
why sentential negation is impossible in (69) and (70b).
(80) OBJ NEG PRED

kitaabg nahr~ becegaa

books NEG seI1-FUT

N V

V
 WORDHOOD AND LEXICALITY 127

The N E G cannot occur to the left of the NI either, because it would

create an ill-formed GF STR as in (77b), which violates (79b). This
explains why sentential negation is impossible in (71):
(81) * N~G OBJ PRED

F
nahrt
r
kitaabg
E
becegaa
books selI-FUT

N V

V
!
NEG V

5.2.4. The Need for Factorization: GC STR vs. GF STR

Our explanation for the impossibility of sentential negation in Hindi NIC
is not available to syntactic theories which do not separate categorial
structure and grammatical function structure as two dimensions of syntac-
tic representation. Note that both (79a) and (79b) are statements on
syntactic representations. Baker and Sadock distinguish between a syntac-
tic representation in which the incorporated nominal is an NP of VP, and
a morphological representation in which it is an N of V, but they do not
distinguish between the representation of grammatical categories and the
representation of grammatical functions. If objecthood is represented as
NP of VP, then it is difficult to see what prevents the placement of N E G
to the left of the NI in sentential negation, as in (82a) (head movement)
and (82b) (coanalysis):
128 TARA MOHANAN

(82) a. VP

NP V

NEG V

V
1 Head movement

N V

nahTi
I
kitaab~i
t
becegaa
b. VP

NP V

NEG V

kitaab~
1
nahii
1
becegaa
Coanalysis

nahii kitaab~t becigaa

N V

V
As a result, the explanation for the impossibility of sentential negation in
terms of (79a, b) is not available to them.

6. T H E C O N C E P T OF W O R D H O O D

Most debates on the Lexicalist Hypothesis (Chomsky 1970, Jackendoff

1972, Wasow 1977, Bresnan 1978, Lieber 1980, K. P. Mohanan 1982,
Kiparsky 1982, Sproat 1986, Baker 1985, Ackerman 1987, Ackerman and
 W O R D H O O D AND L E X I C A L I T Y 129

Webelhuth 1992, Bresnan and Mchombo 1993, and so on) center around
the relation between linguistic structure within words and across words.
These debates raise the question: "What is a word?"
Recent research has clearly shown that the term 'word' conflates a
number of notions that must be kept distinct, tf we use the term PHONO-
LOGICAL~VORD (P-WORD) to refer to units of a phonological domain, then
there are two types of phonological words, MZTRICALWORDand PROSODIC
WORD. Let us use the term SYNTACTICWORD (S-WORD) to refer to wordlike
units for the purposes of syntax. The expressions John's and isn't in
(83a, b) are single V-WORD'S, but not single S-WORD'S:

(83)a. John's coming here tomorrow.

b. Isn't he being promoted?

The representation of (83a) as (84) brings out the non-identity of P-WORD

and S-WORD:

(84) P-WD PHONOLOGY

John ~ coming here tomorrow

........ I............... f .................................................................

S-WD S-WD SYNTAX

The non-convergence of the two notions of wordhood in representations

like (84) are commonly acknowledged in the field, In sections 3 and 4,
we saw the need to separate syntactic organization at GC-structure from
syntactic organization at GF-structure. Given this factorization of dimen-
sions, and the potential non-identity of wordhood across dimensions, it is
reasonable to separate wordhood at GC-structure (GC-WORD) and GF-
structure (GF-WORD). The different notions of wordhood, then, are as
follows:

(85) metrical word

phonological word - -
-- prosodic word

r------ categoriaI word (GC-WORD)

syntactic word
-- functional word (GF-WORD)
130 TARA MOHANAN

The distinction between GC-WORD and GF-WORD allows us to

understand better the phenomenon of NI. In (48a), the relevant parts of
which are repeated below, the N + V is a single GC-WORD (V°). How-
ever, it is not a single GF-WORD: the N is an ARG, and the V is a
PRED.
(48)a. OBJ PRED GF STR
NOM PERF

......................... 1............... l ....

anil-ne kitaab~ bec~L

......................... t ............... t ....

N V GC STR

V
The structure of a type of complex predicates in Hindi in (73b) illustrates
the reverse case:
(73)b. PRED ARG STR/GF STR

yaid klar

N V

V GC STR
In this complex predicate, the nominal and the verbal are two different
GC-WORDS. However, together they form a single PRED. Hence they
constitute a single GF-WORD. 44
If the terms LEXICALand SYNTACTICrefer to relations within and across
words respectively, then we may say that Hindi NIC is lexical with respect
to GC STR, and syntactic with respect to GF STR. Conversely, Hindi
complex predicates are syntactic with respect to GC STR, and lexical with
respect to GF STR. Given this picture, whether a construction is lexical or
syntactic is not determinable unless the dimension of lexicality is specified.

44 See Ackerman (1987), Ackerman and Webelhuth (1992), Matsumoto (1992) for the need
to separate different notions of syntactic wordhood.
 WORDHOOD AND LEXICALITY 131

In order to avoid confusion resulting from ambiguity, I suggest that the

term lexical be reserved for relations within a GC W O R D . If this sugges-
tion is accepted, then Hindi NIC is lexical, while Hindi complex predicates
are not.
As discussed in section 4, the term lexical also involves an ambiguity
between a modular concept (lexical module) and a representational con-
cept (lexical category). It was our claim that Hindi NI is lexical in both
these senses. These two interpretations of lexical come together if we
accept the strong version of the Lexicalist Hypothesis, which I would like
to state as follows:

(86) Lexical categories cannot be created in the phrasal (=postlex-

ical) module, and phrasal categories cannot be created in the
lexical module.

7, CONCLUDING REMARKS

My goal in this paper has been to provide a detailed grammar of the

syntactic behaviour of NI in Hindi, and to outline a partial syntactic theory
that yields the grammar. The puzzle of Hindi NI is its dual behavior with
respect to lexicality. Fundamental to my account is the factorization of
two different types of syntactic information into two dimensions of struc-
turing, GC STR (category information) and GF STR (grammatical func-
tion information). The N + V compound is composed of two elements
which form a single morphological unit at GC STR, but retain their
syntactic independence at GF STR. The analyses of NI in both Sadock
and Baker recognise the dual behaviour of NI. Their representations,
however, do not attempt to separate two kinds of syntactic representa-
tions. As a result, the absence of modifier stranding, gapping, and senten-
tial negation in Hindi NIC have no analysis in these theories.
A natural consequence of the factorization is that a wordlike unit at
one level is not necessarily a wordlike unit at anothe) level. This non-
convergence of wordhood at different levels has been the source of a
number of controversies in linguistic theory, including debates surrounding
the lexicality of NI. The separation of GC STR and GF STR, together
with the recognition of the different notions of wordhood, leads to the
issue of the relation between wordhood and lexicality. I have claimed in
this paper that the notion of wordhood relevant for lexicality is the GC
W O R D . This proposal retains the basic insight underlying the lexicalist
hypothesis by formulating it as a relation between a lexical category and
132 TARA MOHANAN

the lexical module. In many of the debates on lexicality, arguments for the
lexicality of a construction are based on information that in our framework
belongs in GC STR; arguments for the non-lexicality of the construction
are based on information that belongs in GF STR. Given the perspective
outlined above, many of the points of debate become non-issues.

REFERENCES

Abney, S. P.: t987, The English Noun Phrase in its Sentential Aspect, unpublished Ph.D.
dissertation, MIT.
Ackerman, F.: 1987, Miscreant Morphemes: Phrasal Predicates in Ugric, unpublished Ph.D.
dissertation, University of California, Berkeley.
Ackerman, F., and G. Webelhuth: 1992, Topicalization and German Complex Predicates,
unpublished manuscript, Department of Linguistics, University of California in San Diego.
Alien, B., D. Gardiner and D. Frantz: 1984, 'Noun Incorporation in Southern Tiwa', HAL
50, 292-311.
Alsina, A.: I993, Predicate Composition: A Theory of Syntactic Function Alternations, unpub-
lished Ph.D. dissertation, Stanford University.
Baker, M.: 1985, Incorporation: A Theory of Grammatical Function Changing, Doctoral
dissertation, MIT.
Baker, M.: 1988, Incorporation: A Theory of Grammatical Function Changing, The Univer-
sity of Chicago Press, Chicago.
Bauer, L.: 1983, English Word-formation, Cambridge University Press, Cambridge.
Bresnan, J.: 1978, 'A Realistic Transformational Grammar', in M. Halle, J. Bresnan and
G. A. Miller (eds.), Linguistic Theory and Psychological Reality, MIT Press, Cambridge,
Massachusetts.
Bresnan, J. (ed.): 1982a, The Mental Representation of Grammatical Relations, MIT Press,
Cambridge, Massachusetts.
Bresnan, J.: 1982b, 'The Passive', in J. Bresnan (ed.) (1982a), pp. 3-86.
Bresnan, J., and S. Mchombo: 1993, 'The Lexical Integrity Principle: Evidence from Bantu',
ms. Stanford University and University of California at Berkeley, California.
Chomsky, N.: 1970, 'Remarks on Nominalization', in R. Jacobs and P. Rosenbaum (eds.),
Readings in English Transformational Grammar, The Hague, Mouton.
Chomsky, N.: 1981, Lectures on Government and Binding, Foris, Dordrecht.
Cole, J.: 1990, 'The Minimal Word in Bengali', in A. Hatpern (ed.), The Proceedings of the
Ninth West Coast Conference on Formal Linguistics', CSLI, Stanford, California, pp. 157-
170.
Di Sciullo, A. M., and E. Williams: 1987, On the Definition of Word, M1T Press, Cambridge,
Massachusetts.
Gair, J. W., and K. Wali: 1989, 'Hindi Agreement as Anaphor', Linguistics 27, 45-70.
Hale, K.: 1983, 'Wartpiri and the Grammar of Nonconfigurational Languages', Natural
Language and Linguistic Theory 1, 5-48.
Hale, K. and S. J. Keyser: 1991, On the Syntax of Argument Structure, Center for Cognitive
Science, MIT, Cambridge, Massachusetts.
Hopper, P. J. and S. A. Thompson: 1984, 'The Discourse Basis for Lexical Categories in
Universal Grammar', Language f)O, 703-52.
Inkelas, S.: 1989, Prosodic Constituency in the Lexicon, Doctoral dissertation, Stanford
University, Stanford, California.
 W O R D H O O D AND L E X I C A L I T Y 133

Jackendoff, R.: 1972, Semantic Interpretation in Generative Grammar, MIT Press.

Kachru, Y., B. Kachru and T. Bhatia: 1976, 'The Notion 'Subject': A Note on Hindi-Urdu,
Kashmiri and Panjabi', in M. K. Verma (ed.), The Notion of Subject in South Asian
Languages, South Asian Studies, University of Wisconsin, Madison Wisconsin, pp. 79-
108.
Kiparsky, P.: 1982, "Lexical Phonology and Morphology', in I.-S. Yang (ed.), Linguistics in
the Morning Calm, Hanshin, Seoul.
Knecht, L.: I986, 'Lexical Causatives in Turkish', in D. Slobin and K. Zimmer (eds.),
Studies in Turkish Linguistics, Benjamins, Amsterdam, pp. 93-I22.
Liberman, M. and A. Prince: 1977, 'On Stress and Linguistic Rhythm', Linguistic Inquiry
8, 249-336.
Lieber, R.: 1980, On the Organization of the Lexicon, Doctoral dissertation, MIT, Cam-
bridge, Massachusetts.
Mahajan, A.: 1989, 'On the A/A-bar Distinction: Scrambling and Weak Crossover in Hindi',
ms., MIT, Cambridge, Massachusetts.
Matsumoto, Y.: 1992, On the Wordhood of Complex Predicates in Japanese, Doetorat
dissertation, Stanford University, Stanford, California.
McGregor, R. S.: 1972, Outline of Hindi Grammar, Clarendon Press, Oxford.
Merlan, F.: 1976, 'Noun Incorporation and Discourse Reference in Modern Nahuatt', HAL
42, 177-191.
Mithun, M.: 1984, 'The Evolution of Noun Incorporation', Language 60, 847-895.
Mohanan, K. P.: 1982, Lexical Phonology, Doctoral dissertation, MIT, Cambridge, Massa-
chusetts.
Mohanan, K. P.: 1984, 'Lexical and Configurational Structures', The Linguistic Review 3,
113-139.
Mohanan, K. P.: in press, 'Organization of the Grammar', in J. Goldsmith (ed.), A Hand-
book on Phonological Theory, Chicago University Press,
Mohanan, T.: 1990, Arguments in Hindi, Doctoral dissertation, Stanford University, Stan-
ford, California.
Mohanan, T.: 1993, 'Case Alternation on Objects in Hindi', South Asian Language Review"
3,. 1-30.
Reimsdijk, H. van, and E. Williams: 1981, 'NP-structure', The Linguistic Review 1, 171-
217.
Rosen, S. T.: 1989, 'Two Types of Noun Incorporation: A Lexical Analysis', Language 65,
294-31.7.
Sadock, J. M.: 1980, 'Noun Incorporation in Greenlandic: A Case of Syntactic Word Forma-
tion', Language 57, 300-319.
Sadock, J. M.: 1985, 'Autolexical Syntax: A Proposal for the Treatment of Noun Incorpor-
ation and Similar Phenomena', NLLT 3, 379-440.
Sadock, J. M.: 1986, 'Some Notes on Noun Incorporation', Language 62, 19-31.
Sadock, J. M.: 1991, Autolexical Syntax: A Theory of Parallel Grammatical Representations,
The University of Chicago Press, Chicago.
Simpson, J.: 1983, Aspects of Warlpiri Morphology and Syntax, Doctoral dissertation, MIT,
Cambridge, Massachusetts.
Simpson, J.: 1991, Warlpiri Morpho-syntax: A Lexicalist Approach, Kluwer Academic Pub-
fishers, Dordrecht.
Sproat, R. : 1986, On Deriving the Lexicon, Doctoral dissertation, MIT, Cambridge, Massa-
chusetts,
Srivastava, M.: 1969, The Elements of Hindi Grammar, Motilal Banarsidass, India.
Verma, M. K.: 1971, The Stucture of the Noun Phrase in English and Hindi, MotitaI
Banarsidass, India.
Wasow, T.: 1977, 'Transformations and the Lexicon', in P. Cuficover, T. Wasow, and A.
Akin@an (eds), Formal Syntax, Academic Press, New York, pp. 327-360.
134 TARA MOHANAN

Zubizaretta, M-L.: 1987, Levels of Representation in the Lexicon and the Syntax, Foris,
Dordreeht,

Received 15 August 1991

Revised 31 March 1993

Linguistics Programme/Dept, of English Language & Literature

National University of Singapore
Kent Ridge, Singapore 0511
elltaram@nusvm,bimet