Hazout 1991 Thèse

VERBAL NOUNS: THETA THEORETIC STUDIES IN
HEBREW AND ARABIC
A Dissertation Presented
by
ILAN HAZOUT
Submitted to the Graduate School of the

University of Massachusetts in partial fulfillment
of the requirements for the degree of i
DOCTOR OF PHILOSOPHY
May 1991
Department of Linguistics
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Copyright by Ilan Hazout 1991
All Rights Reserved
HEBREW AND ARABIC
A Dissertation Presented
by
ILAN HAZOUT
Approved as to style and content by:
Elizabeth Selkirk, Chair
F. Higgins Member
Shmuel Bolozky, Member
Barbara H. Partee, Department Head

Department of Linguistics
This Dissertation is dedicated
to the memory of
NURIT HAZOUT
ACKNOWLEDGMENTS
I would like to thank the members of my dissertation

committee, Shmuel Bolozky, Roger Higgins and Lisa
Selkirk. Roger and Lisa read earlier drafts of this

dissertation and made many helpful comments. In
particular, Roger's comments on two earlier drafts were
extremely detailed and helpful and led to many
substantial improvements.
Thanks to Edwin Williams under whose direction most

of this work was written. The influence of his work can
be recocnized throughout this dissertation. The elegance

of Edwin’s work and his teaching style won my admiration
from the very beginning. His sympathy and encouragement
were important for the development of this work.

I have had some interaction with my fellow students
at the Umass linguistics department over the last four
years. This has always been an enjoyable experience and
I should therefore be thankful to Rose-Marie Dechaine,
Gingi Fu, Greg Lamontaigne, Reine Pinsonneault ,

Bernadette Plunkette, Roger Schwarzshild, Tim Sherer,
Karina Wilkinson and Sandro Zucchi.
I would have hoped that the completion of this
dissertation would bring some joy into my sister,

Nurit's, life. Sadly, she didn't live to see this
moment. This dissertation is dedicated to her memory.
ABSTRACT
HEBREW AND ARABIC
MAY 1991
ILAN HAZOUT, LICENCE, UNIVERSITE DE PARIS VIII
MAITRISE, UNIVERSITE DE PARIS VIII
Ph.d. UNIVERSITY OF MASSACHUSETTS
Directed by: Professor Elizabeth Selkirk
This dissertation is a study of a variety of
constructions in Modern Hebrew and Standard Arabic which
involve nominalization processes. Such constructions
manifest a certain mixture of verbal and nominal
properties and are analyzed as involving a verbal
subconstituent, a VP, governed by an underlying nominal
head, a nominalizer . The surface form of the deverbal
head of such constructions is . the output of a head
movement operation adjoining a verb to the nominalizer
which governs it. The properties and the differences
between the different types of nominalization
constructions are explained on the basis of ceratin
assumptions about the thematic properties, the argument
structure, of the different nominalizers that are
postulated. The heads of nominalization constructions
are morphological as well as thematic nominalizers in
that they provide, in addition to a particular •-
vi
morphological shape, an argument structure particular to
nouns. In this approach to verbal nouns, the mixed
properties of these costructions are derived from the

properties of underlying verbs and nouns occurring within
a particular configuration.
This approach to nominalizations is embedded within

a particular approach to thematic relations and argument
structure combined with theoretical techniques developed
in recent work within the Government and Binding theory,

in particular, the operation known as head movement.
Chapter 1 presents the main theoretical assumptions

and includes some proposals concerning the structure of
infinitival clauses and the phenomenon of obligatory
control. Chapter 2 is a comprehensive study of genitive
constructions in Hebrew and Arabic. Chapter 3 is a study

of Action Nominalization constructions and includes a
detailed argumentation in a favour of a non-lexicalist

approach. Chapter 4 investigates and compares the
properties of two types of infinitival constructions,
standard infinitives and the verbal gerund, a
construction which is particular to Modern Hebrew.

Chapter 5 studies the Agent Nominalization construction
and the Benoni relative, a construction which is analyzed
as involving a definite article functioning as a thematic
nominalizer and an abstract adjectival morpheme which
functions as a morphological nominalizer.
vii
TABLE OF CONTENTS
Page
ACKNOWLEDGEMENTS ........................................... V
ABSTRACT .................................................. vi
CHAPTER
I INTRODUCTION ......................................... 1
1.1 Head Movement, Morphology

and Syntax .................................. 11
1.2 Principles of Theta Theory ................. 23
1.2.1 Argument Structure, Theta

Roles and Theta Structure ............. 23
1.2.2 Types of Thematic Relatedness ........ 28
1.2.3 Infinitival Clauses
and Control............................ 32
1.3 Levels of Representation ................... 43
Footnotes to Chapter I................. 47
II THE SYNTAX OF GENITIVE CONSTRUCTIONS ......... ..49
2.1 Bound Genitives (The Construct State) ..... 51
2.1.1 The Data............................... 51

2.1.2 The Complement Approach
to Bound Genitives .................... 60
2.1.3 A Specifier Approach to Bound
Genitives: The Head Movement
Analysis ............................... 78
2.2 The Interpretation of Genitive

Constructions: Identification and
Classification .............................. 92
2.3 Sei Phrases and Double Genitives ......... 103
Footnotes to Chapter II ........................ 123
viii
Ill THE SYNTAX Of ACTION NOMINALIZATIONS 129
3.1 Introduction .......................... .129
3.1.1 Nominal Properties of Action

Nominalizations .................. .130
3.1.2 Verbal Properties of Action

Nominalizations .................. .134
3.1.3 Binding and Control.............. ..138
3.2 Active Action Nominalizations ........ ..141
3.2.1 The Head Movement Analysis ...... ..141
3 . 2 . 1 .1 More About the Facts ........ ,.141

3. 2. 1.2 The Syntactic Derivation
of Active Action
Nominalizations .............. ..148
3.2.2 Syntactic vs. Lexicalist

Approaches to Action
Nominalizations .................. ..163
3. 2. 2.1 A lexicalist Analysis:

First Version................ ,.164
3. 2. 2. 2 A Lexicalist Analysis:
Second Version............... ,.172
3. 2. 2. 2.1 The Distribution of

Adverbs and Adjective... ..174
3. 2. 2. 2. 2 Free Genitive Variants.. ..175
3. 2. 2. 2. 3 Subjectless Action
Nominalizations and
Control.................. ..181
3. 2. 2. 3 The Interpretation of Action
Nominalizations .............. . .186
3. 2. 2. 4 Causative Constructions
in Arabic.................... . .296
3.3 Subjectless Action Nominalizations... . .222
3.3.1 Introduction ..................... . .222

3.3.2 Arguments Against a
Passivization Analysis .......... . .232
Verbs Which Don't Passivize... . .234

Verbs Which Do Passivize ...... . .237
Anaphors ....................... . .241
Control Structures in Arabic.. . .241
ix
3.3.3 Burzio's Generalization? ............. 243
3.3.4 The Structure of Subjectless
Action Nominalizations ....... ....... 248
3. 3. 4.1 Bound Forms and Genitive

Case Assignment .................. 248
3. 3. 4. 2 Control .......................... 259
Footnotes to Chapter 3......................... 266
IV INFINITIVES AND GERUNDS ........................ 277
4.1 Introduction ............................... 277

4.2 Verbal Gerunds............................. 284
4.2.1 The Structure of

Infinitival Clauses .................. 284
4.2.2 The Structure of Verbal Gerunds ..... 289
4. 2. 2.1 A Sentential Analysis of

Verbal Gerunds (I).............. 289
4. 2. 2. 2 A Non-Sentential Analysis of
Verbal Gerunds .................. 295
4. 2. 2. 3 A Sentential Analysis of
Verbal Gerunds (II)............. 299
4.2.3 The General Analysis of Infinitives:

Some Implications .................... 319
4.3 Nominal Gerunds 329
4.3.1 Introduction ......................... 329

4.3.2 Nominal Properties of
Nominal Gerunds ...................... 332
4.3.3 Verbal Properties of
Nominal Gerunds ...................... 334
4.3.4 The Structure of Nominal Gerunds .... 338
Footnotes to Chapter IV .......................... 348
V THE SYNTAX OF PARTICIPIAL CONSTRUCTIONS ......... 352
5.1 Introduction ................................. 352
5.1.1 An Overview............................. 352

5.1.2 Participles As Verbs................... 358
5.2 Agent Nominalizations ....................... 363

5.3 Benoni (participial) Relatives .............. 393
Footnotes to Chapter V........................... 419
APPENDICES
APPENDIX TO CHAPTER II: ADJECTIVAL BOUND

GENITIVES
............................................ 421
APPENDIX TO CHAPTER III: THE PARTICLE "ET" AND
ACCUSATIVE CASE MARKING............................. 429
REFERENCES.......................................... 435
xi
CHAPTER I
INTRODUCTION
One of the major linguistic distinctions, a
distinction which has been made by both linguists and
philosophers since the time of antiquity, is the one
between verbs and nouns. Thus, Aristotle (De
Interpretations, Chapters 2-3) distinguishes between
nouns, which he characterizes as significant sounds which
have no reference to time and verbs which he
characterizes as "that which, in addition to its proper
meaning, carries with it the notion of time".
The almost unquestionable status of the distinction
between verbs and nouns is certainly justified in view of
its attestability throughout the languages of the world.
Thus, in a recent typological study, Schachter (1985)
observes that "the distinction between nouns and verbs is
one of the few apparently universal part-of-speech
distinctions". As Schachter notes, although the
universality of this distinction has sometimes been
questioned, "it now seems that the counterexamples have
been based on incomplete data, and that there are no
languages that cannot be said to show a verb-noun
distinction when all relevant facts are taken into
account" (underlining is mine [I.H.]).
Notwithstanding the fairly well established status

of the noun-verb distinction, the criteria to be employed
for part-of-speech classification and the verb-noun
distinction are somewhat less clear. The intuitively
most immediate criteria for defining verbs and nouns are
the familiar semantic, or notional, part-of-speech
definitions. Thus, the notional definition of noun would
be as a name of a person, place or thing whereas verbs
would be defined as words that express actions, processes
and the like. Clearly, the idea of time has always been
central to the notional noun-verb dichotomy, as can be
observed in Aristotle's definition of this distinction
cited above. However, as claimed by Schachter (1985), it
has been amply demonstrated in the linguistic literature
that the familiar notional definitions fail to provide an

adequate basis for part- of-speech classification since
there are many cases in which their applicability or
inapplicability is unclear.
Other criteria for part-of-speech classification
which seem to be fairly well established are grammatical

criteria. Grammatical criteria have to do with the
distribution of a word and the range of its syntactic

functions as well as the morphological and syntactic
categories for which it may be specified. Thus, we may
observe that in a sentence such as the following,
1- detectives chase criminals
the words detectives and chase can be shown to differ in

their distribution.
*2- chase detectives criminals
detectives and chase also differ in their functional

range. Thus, detectives but not chase can function as
the subject of a sentence. Morphologically, detectives
is specified for number but not for tense, whereas chase
is specified for both number and tense. These two words

are therefore assigned to two distinct part-of-speech
classes .
On the other hand, the words detectives and
criminals have highly similar distribution and functional
range ,
3- criminals chase detectives
and can be morphologically specified for the same type of

feature distinctions. They are therefore assigned to the
same part-of-speech class.
Clearly, the criteria employed for the assignment of
a certain word to a particular part-of-speech class are
often language particular rather than universal.
However, as noted by Schachter (1985) , the
characterization of a set of words as belonging to a

certain part-of-speech class mostly reflects universal
and crosslinguistically valid semantic considerations.
The distinction between nominal and verbal is also
standardly applied to the immediate syntactic environment
in which verbs and nouns occur. Thus, a Noun Phrase, a
collocation of words which is typically characterized as
nominal, may be said to inherit certain typically nominal
properties, or feature specifications (number, gender
etc.) from one of its members (its head noun) and to obey
certain collocation restrictions characteristic of noun
phrases. Equally, a Verb Phrase, a constituent "headed”
by a verb, may be said to inherit some of the properties
of its head (e.g. its external theta-role) and to obey
certain internal collocation restrictions which

distinguish it from a Noun Phrase.
To enumerate just a few of the more typical and

crosslinguistically observable collocation restrictions,
nouns may be typically accompanied by adjectives which
may be said to function as the modifiers of such nouns.
The occurrence of quantifier expressions and definite
articles is also typical NP internally. A noun, the head
of an NP, may also be accompanied by another NP which
would be marked with Genitive Case, or else governed by a
preposition, rather than being "directly" related to the
head noun. Verbs, on the other hand, may be accompanied
by a modifying adverbial expression. A verb may also be
accompanied by an NP which need not necessarily be

preceded (or followed, in some languages) by a
preposition. In such a case the NP may be said to be
directly related to (governed, or licensed by) the verb
and may, in languages such as Hebrew and Arabic, be
morphologically marked in a different way from a genitive
NP, namely, it may be marked with Accusative Case. Thus,
the characterization as nominal or verbal applies to

single lexical elements as well as to complex syntactic
constituents and may be fairly safely established on the

basis of certain crosslinguistically observed
regularities.
Verbal Nouns, a class of constructions with which

this dissertation is concerned, are of mixed nature in
that they manifest properties typical of verbs and VP's
as well as properties typical of nouns and NP’s. They,
thus, disturb the otherwise fairly clear cut distinction
between nominal and verbal constituents and for this
reason represent a major challenge to linguistic theory.
The main concern of this dissertation will be a class
of constructions, to be referred to as Nominalization
Constructions, of which the following are typical
examples (the designations H and A are short for Hebrew
and Arabic respectively and will be used throughout this
dissertation)
4- (H) Action Nominalization:
axilat Dan et ha-tapuax bi-mhirut

eating Dan OM the apple quickly
"Dan's eating of the apple quickly"
5- (H) Nominal Gerund:
cet-o ha-maftia Sei ha-nasi im alot
depart-his the surprising Sei the president with break

ha-Saxar
the dawn
"The president's surprising departure with the break

of dawn"
6- (A) Agent Nominalization:
muSarrib -u l-?awlaad-i 1-Haliib-a
make-drink-er-Nom the kids -Gen the milk -Acc

"The one who makes the kids drink the milk"
Let us briefly point out what the mixed nature of these
constructions consists of. First, all three
constructions above are NP's, as far as their sentence

internal distribution is concerned, that is, they relate
to their syntactic environment just like any other NP.
Their mixed nature relates therefore mainly to their
internal structure. Typical to all these constructions
is that the head noun, the noun which determines the
gender and number specification of the construction as a

whole (underlined in 4-6) , is morphologically related to
a certain verb. Moreover, just like a verb, the head of

such constructions may be accompanied by a direct object
marked with Accusative Case (examples 4,6) as well as by

other constituents which would normally be taken to be
the arguments of the corresponding verb. We may say that
the noun heading such a construction inherits the
argument structure of a verb. As examples 4,5 above

show, Nominalization constructions clearly allow the
occurrence of adverbs, a typical verbal property.

As to their internal nominal properties, let us just
note that the nouns heading these constructions are
always accompanied by an NP, an argument of the related
verb, to which they may be said to be related in a
genitive type of relation. This clearly shows on the
Case marking of the NP, or its head. Finally, as shown

in 5, the head of a nominalization construction may be
modified by an adjective, again, a typical nominal
property.
This mixture of properties is problematic to the

extent that the verb/noun distinction, as noted earlier,
is based mainly on the observation that a certain cluster
of properties is, in the general case, associated with a
certain class of elements to the exclusion of others.
However, in view of the mixed properties of verbal
nouns, it becomes, a t least, unclear to what extent our
view of the basic inventory of lexical elements in the
languages under consideration, as well as generally, can
be maintained or is in need of revision.
Thus, the main question facing us is how to explain,
or derive, this mixed nature of verbal houns. Here two
main approaches come to mind, the first would be to admit
that, besides verbs and nouns, the languages in question
also have a class of lexical elements that are of a mixed
nature, that is, elements which share properties with
both verbs and nouns and the constituents of which they
may be said to be the heads sh re some of the syntactic
properties of both NP's and VP’s. A second approach
would deny the possibility that there exist such things
as lexical elements of mixed catégorial nature. Rather,
the basic underlying inventory of lexical elements would
include, on thi” approach, only "strict" nouns and verbs.
Elements and constituents of mti.xed, verbo-nominal , nature
would, on such an approach, be derived rather than basic.
That is, they would only be the product of certain
processes to which verbs and nouns may serve as input.
Thus, we might say that in this, second, approach the
properties of verbal nouns are reduced to the properties
of verbs and nouns. We may therefore refer to this
approach as the reductionist approach and to the other,

opposing, approach as the non-reductionist approach.
The considerations in favour or against ope of these
two approaches may be of two main kinds . They can 1 be
general, theoretical or metatheoretical in character,

that is, they may involve considerations that would have
to do with a certain overall picture of the theory of
grammar and would involve notions such as simplicity,
elegance and the like. Thus, in making such i

considerations one might wish to claim that, given
certain notions of simplicity or elegance, a reductionist
approach is to be preferred as simpler or more elegant.
It is clear, however, that any discussion of tl is issue

must take into consideration mainly the empiriqal |
advantages of each of the approaches described above.

That is, given a certain set of basic assumptiqns
concerning the form of underlying representations and the

kinds of elements that may figure in such
representations , and given also the grammatical

mechanisms that may be assumed to be available for the
derivation of surface representations, the empirical
question would be how many of the known empiriqal facts
may be accounted for on the basis of each of ttyese two
main (the reductionist and the non-reductionist)

approaches .
This dissertation takes a rather clear position on
these issues. Chapters 3, 4 and 5 will present a
detailed argumentation in favour of a reductionist
approach to the analysis of each of the verbo-nominal
constructions to be considered. Moreover, the

argumentation in favour of this position will be strictly
empirical in character. Global considerations of

metatheoretical character will play no role.
As just stated, in a reductionist approach such as

the one defended in this dissertation only verbs and
nouns are assumed to exist and therefore only verbs and
nouns may be assumed to figure in syntactic deep
structures. It follows that the main task of the

analysis to be developed will be to show how the surface
manifestations of such constructions, in particular,
their mixed verbo-nominal properties are to be derived,
or explained. In developing such an analysis we will

make use of some recent theoretical developments. In
particular, we will make crucial use of recent proposals
concerning the interaction between syntax and word
formation. The centerpiece of these recent developments
is the transformational process known as head to head
movement. The elaboration and development of this
mechanism is due mainly to the work of Travis (1984) and,

most prominently, Baker (1985,1988). A second set of
theoretical assumptions we will adopt will include much
10
of the recent theoretical developments dealing with the
nature and distribution of theta roles and the
implications of a certain particular conception of theta
roles known as Theta-Theory, with respect to other
modules of the grammar. These recent developments are
due mainly to the work of E. Williams (e.g. Williams
1987a, 1987b, 1989) . The following sections present the
theoretical assumptions which will be the foundation on
which the analyses developed in the next four chapters
will be based.
1.1 Head Movement, Morphology and Syntax
This dissertation adopts and defends the view that
words may be the product of syntactic processes. Much
recent work in syntactic theory has been devoted to the
development of this idea. The work of Baker (1985,1988)
has been particularly influential in this respect.
The syntactic process of incorporation, applying to
two, or more, morphemes to create a morphologically
complex word, becomes possible if it is assumed that the
general transformation move- Alpha may also apply when
the bar level of the element to be affected by i t is
zero. This particular case of move-alpha is now
generally known as head movement.
11
Head movement is subject to generally assumed
constraints. Travis (1984) suggests a locality condition
on head movement in the form of the following constraint.
7- The Head Movement Constraint (HMC)

An X° may only move into the Y°
which properly governs it.
This constraint imposes a restriction on head movement in

terms of the configuration existing prior to its
application. Baker (1988) shows that the HMC can be

derived from the ECP, a condition which applies to the
output of a movement operation and imposes certain
constraints on the structure created by a movement. The

ECP is formulated as follows.
8- Traces must be properly governed
8 makes use of the notion of proper government which is
defined as follows.
9- A properly governs B iff A governs B and A and B are

coindexed
9, in turn, presupposes the notion of government which is

defined as follows,
12
10- A governs B iff A c-commands B and there is no
category C such that C is a barrier between A and B.
Note that government is only one part of the requirement

of proper government. The other part, namely, the
requirement of coindexation may be satisfied in two ways.

First, a proper governor may theta-mark a trace, in which
case the trace and its theta-marker are coindexed.
Otherwise a coindexation between a trace and its governor
may arise if the two elements are related to each other
by movement, in which case they are coindexed. It is
this second type of coindexation relation which is most
relevant for head movement and the relation between a

moved head and its trace. This relation is constrained
by the structural conditions which are specified by the
definition of government in 10, namely c-command and

barrierhood. I will follow Baker (1988) on these two
points. First, as Baker shows, the c-command definition
relevant for head movement is the one of Aoun and
Sportiche (1983) which is formulated in terms of maximal
projections as follows.
11- A c-commands B iff A does not dominate B and for any

maximal projection C, if C dominates A then C
dominates B.
13
To illustrate, in a configuration like the following, it
must be YP and not Y* which counts as the relevant node
for c-command.
12- YP
/ \
Y* XP
/ \ / \
Xi Y X ZP
I
ti
Clearly, it must be assumed that Y* does not count in
order for some sort of head movement to be possible at

all.
The second requirement in the definition of

government in 10 is that there be no barrier between a
governor and a governee. Baker's (1988) definition of
barrier goes as follows.
14
13 - Let D be the s m a l l e s t maximal projection containing
A . Then C is a barrier between A and B if and only
if C is a maximal projection that contains B and
excludes A and either
i) C is not selected
ii) The head of C is distinct from the head of D
and selects some WP equal t o or containing B.
To illustrate, consider the two following configurations.
14- a) XP b) XP
/ \ / \
X' X'
/ \ / \
X° YP X° ** * YP
/ \ \ / \ \
Yi X° Y' Zi X° Y*
/ \ / \
Y Y° '~~ * ZP
ti Z*
ti
15
In 14a Yi c-commands its trace inside YP, but YP is not
selected by X, therefore it constitutes a barrier and 14a

is ruled out by the ECP given clause (i) of 13. YP in
14a may be taken to exemplify the case of adjunct
clauses .
13ii, in Bakers's words, is a "minimality condition
expressing the fact that an intervening theta assigner

breaks a government path". Thus, in 14b ZP is selected
by Y° but since YP does not contain X, YP is a barrier to
government of the trace of Z by Z. This definition of

Barrierhood makes use of the notion of selection. Let us
consider first Baker's definition of this notion and then

motivate a certain deviation from it. Baker (1988)
defines selection as follows.
15 - A selects B if and only if:
i) A assigns a theta role to B

ii) A is of category C and B is its IP
iii) A is of category I and B is its VP
15 seems to be unsatisfactory in some respects.
Considering first 15i, to the extent that 15 as the
definition of selection makes part of the definition of

barrierhood and therefore, indirectly, of the general
definition of the relation of government, 15i seems to
cover a larger territory than necessary since the subject
16
of a sentence is assigned a theta role by a verb (or by
VP, or by I') but is not governed by it (or by VP, or by
I'). As to 15ii and 15iii, these clauses do not really
provide a definition of selection but simply stipulate
that the relation between an element of category C and IP
and the relation between an element of category I and VP
is one of selection.
In general , there are two ways in which lexical
requirements holding between a head and a constituent

governed by it are encoded in lexical representations,
namely, theta-role assignment and subcategorization
specifications. Now, as just observed, theta-role
assignment holds not only between a governor and its
complement but also between a verb (or VP) and its

subject. On the other hand, it is largely assumed that
subcategorization requirements hold only between a head

and its complement. Thus, within the domain dominated by
an X' node, subcategorization features are specified with
respect to any lexically required complement. It follows
that every complement which is assigned a theta-role is

also subcategorized for. Note, however, that the
opposite does not hold. That is, not every

subcategorized element is assigned a theta-role. This
is, presumably, the case with the relation between a

complementizer and IP as well as the relation between
INFL and VP. This dissertation is concerned with a
17
family of constructions which may be analyzed as
involving an underlying nominal bound morpheme which

stands to a VP in exactly the same kind of relation,
namely, it subcategorizes for a VP but does not assign a
theta-role to it. Thus, if this way to view things is

correct then selection can be reduced to
subcategorization. We can therefore simplify 15 to the

following statement.
16- A selects B if B satisfies subcategorization
requirements for which A is lexically specified.
Note now that an output configuration such as 14b would
still be possible even if all the constraints introduced

so far are assumed. This would be possible if Z in 14b
underwent a successive cyclic movement, adjoining first

to Y and then to X. This would yield the following
configuration.
18
17 - XP
/ \
X'
/ \
X YP
/ \ / \
Zi X Y*
/ \
Y ZP
/ \ / \
ti Y Z'
/ \
Z
I
ti
However, as Baker (1988) argues, this may be excluded if

one assumes that the rule of move-alpha may not move a
part of a word. The following filter may capture this
principle.
18 - *[x° ti ]
Given 18 the only possible successive cyclic application

of head movement to 14b is the one which results in 19.
19
19- XP
/ \
X’
/ \
X YP
/ \ / \
Y X Y'
/ \ / \
Z Y Y ZP
I / \
ti Z*
/ \
Z
I
ti
The output of an application of head movement , such as
[y Z Y ] in 19 above, may be an actual morphologically

complex word. But it can also be an abstract object
which is still to undergo a process of morphological

"interpretation" or "spelling out".
Most of the constructions to be studied in this
dissertation are cases for which some sort of further

morphological interpretation needs to be assumed. To be
more concrete, chapters 3, 4 and 5 will study
constructions of a general structure as the following.
20
20- a) NP b) NP
/ \ / \
N' =======> N*
/ \ / \
N VP N VP
1 / \ / \ / \
X V YP Vi XV YP
ti
In 2 0 a a nominal bound morpheme (X) s u b c a t e g o r i z e s for a

VP . The verb heading VP i s adjoined to X by the
application o f h ead movement a s shown in 2 0 b . For

reasons to b e d i s c u s s e d in chapter 3 (section 3. 2. 1.2) it
is necessary to assume that X in 2 0 a , b i s an a b s t r a c t
element rathe r than an actual morpheme w i t h a c o n c r e t e
morphological and phonological shape. Thus, X is an

element w hich combines with a verb to produce a unique
nominal form which i s lexically listed for the p a r t i c u l a r
verb in q u e s t i o n . This is not an i m p l a u s i b l e assumption

if one as su mes that the derived nominal form o f a
pa r t i c ular verb is pa r t of its lexical entry. It is
therefore available for "spelling out" at the n e c e s s a r y

point in the d e r i v a t i o n .
A configuration such a s 2 0 a in which a bound morpheme

(X) , a " n o m i n a l i z e r " , heads an NP and governs a VP w i l l
21
be referred to as a Nominalization Construction. The
application of head movement adjoining V to X as in 20b,

of which the output is an abstract deverbal head, will be
referred to as a syntactic process of Nominalization.

The following chapters will study a number of such
constructions. The differences between the different

Nominalization constructions will be shown to follow from
the properties of the particular nominal heads (the X of
20) to be postulated, in particular, their thematic, or
argument, structure. This will presuppose an overall
conception of the nature and distribution of theta roles,

in particular, a certain set of principles to be
referred to as theta theory. The following section is
devoted to these matters.
22
1.2 Principles of Theta Theory
The set of assumptions and principles presented in this
section are mostly adopted from the work of E. Williams, in
particular, Williams (1987a) , (1987b) and (1989) as well as
Di Sciullo and Williams (1987). However, Some of the ideas
suggested here, in particular, the ideas concerning the
structure of infinitival clauses and obligatory control
(section 1.2.3) are novel and were developed in an attempt to
deal with certain issues that are particular to
Nominalization constructions in Hebrew and Arabic. I should
therefore assume full responsibility for these ideas as well
as for the rest.
1.2.1 - Argument Structure, Theta Roles and Theta Structure
The basis for thematic relations is the argument
structure of predicates. The argument structure of a
predicate is a list of its theta roles (Agent, Theme, Goal
etc. ) with one of the theta roles distinguished as the
external theta role. The other theta roles are thus
internal. Nouns, just like verbs and adjectives, also have
an argument structure. In particular, nouns have an external
theta role designated as R (for
23
Reference) . When an NP is used predicatively this role is
assigned to the subject as in,
21- George is a president
Internal theta roles are assigned to NP's internal to the

first projection of the predicate of which they are lexically
specified as arguments. The external theta role may be
thought of as the head of the argument structure. It is
therefore passed up the X-bar projection and assigned to, or
becomes associated with, the maximal projection. We may say

that the external theta role of the head becomes the external
theta role of its maximal projection. The external theta

role is, thus, assigned "vertically". This is illustrated
below for VP's (22a) and NP’s (22b). We will adopt the
standard practice of marking an external argument by
underlining.
22- a) VPi b) NPi
/ \ I
V N’i
(Ai , B) I
Ni
(Ri ,. . . )
When an NP is used in a referential position, as an
24
argument, its external theta role, that is, the external R-
theta role of N which was vertically assigned to NP, is
coindexed with the particular theta role of the predicate

which is said to be assigned to it. Thus, theta role
assignment is a coindexation of two theta roles. This is

illustrated as follows.
23- VP
/ \
V NPi
A, Bi I
N’t
I
Ni
see a president
(Rt )
The coindexation is here expressed by the identical index (i)

associated with the internal theta role of the verb (B) and
the R role of its NP complement. This is to be interpreted
as expressing the idea that the reference of the verbal
theta-role is determined by the NP it is coindexed with, that

is, by the information provided by this NP. In this respect
the relation between a theta role of a predicate and an NP
(or its external, referential, theta role) is not symmetric.
Rather, it may be thought of as a relation between an anaphor
25
and its antecedent. We may say that a theta role "looks for"
reference which is provided by an NP which has reference
independently. Thus, instead of thinking of theta-role
assignment as a symmetric relation of coindexation, we will
follow Williams (1989) in viewing it as an asymmetric
relation of linking which may be
given expression by the use of arrows as in 24 below. We will

say that in such a state of affairs a theta role is
satisfied.
24- VP
/ \
V NP
(Aj,Bi) (Ri )
This conception of theta-role assignment may be taken

advantage of for an account of binding relationships (cf. in
particular Williams (1989) for a development of this idea) .
This dissertation, however, will be very little concerned
with issues of binding and therefore no detailed presentation

of a theta-theoretic binding theory will be undertaken here.
In general, except for vertical assignment (or "vertical
binding") , theta role assignment is subject to a severe

configurational restriction, namely, the "assigner" and the
"assignee" must stand in a relationship of sisterhood. This
may be claimed to hold for both internal and external theta
roles except that in the case of an external theta role the
relationship of sisterhood holds between a higher projection
and an argument NP, rather than between an NP and a governing

head. We can make this idea clear by illustrating what may
be referred to as the theta structure of a simple clause.
25- IP
/ \
NPi I 'i
1 / \
N 'i I VPi
1 / \
Ni V NPj
John (Ai , Thj ) 1

(Ri ) saw N'j
Nj
Mary
(Rj )
In 25 the theme role of see is assigned to Mary in observance

of the strict sisterhood restriction. The external (Agent)
theta role of the verb is vertically assigned to VP and then

to I’. It thus becomes the external theta role of I' and may
be assigned to John in accordance with the strict sisterhood
1
restriction .
27
As is clear from the structure illustrated above, we do
not assume, as has become common in much recent work in
syntax, a VP structure in which a subject is base generated
as specifier of VP. Rather, we adhere throughout this
dissertation to the assumption that a subject is external to
VP a t all levels of representation. This assumption will be
crucial for our analysis of nominalization constructions in
chapters 3 , 4 and 5.
1.2.2 Types of Thematic Relatedness
So far we have seen three types of thematic relations.
The core case of thematic relatedness is the relation between
a verb and an NP governed by it, that is, its internal
argument. This is, as was claimed, a relation between two
theta roles. Another type of thematic relation is the one
between a subject NP and its sister node (as in 25) . This
relation is similar to the first in that this also is a
relation between two theta roles except that in this case the
theta role assigner is not a head but, rather, a higher
projection. This relation is mediated, partly, by another
type of thematic relation, namely, vertical assignment (or
binding) of an external theta role from a head to its maximal
projection. This is the relation between the verb and VP in
25. This relation is natural given standard assumptions about
28
the relation between heads and their projections. Note,
however, that this is not exactly the case with respect to
the vertical assignment of the external theta role from VP to
I* in 25. I' is not a higher projection of VP in the X’-

theoretic sense of "being a projection of". Rather, the
relation between I' and VP in 25 follows from the relation
between I, the head of I', and VP. In this case I selects

VP, in the sense of the notion of selection as formulated in
16 above. However, given that I has no external theta role
of its own I ' has no external theta role either and the
external theta role of VP is free to be vertically assigned

to I* and become its external theta role. The external
theta-role is, thus, being "passed on" up to a point at which

it can be assigned to a sister.
Consider next the relation between the complementizer

that and the IP governed by it as in,
26- That John hit the ball
that in this case is not related to IP in the same way as I

to VP in 25 above. For one thing, that could never inherit
the external theta role of a lower constituent in such a way
that that external theta role would become its own external
theta role. This is because that has its own external theta
role. It is by virtue of this property that the sentential
constituent headed by that can function as an argument as in,
‘ 29
27- That John hit the ball is a surprise
However, that and I are similar in that none of them assigns

a theta-role to its complement. It follows that that is
incapable of "inheriting'* an external theta-role from the IP
governed by it and therefore the subject position of IP in

such a case must be occupied by an element which is capable
of being assigned a theta-role.
The nominalization constructions to be studied in
chapters 3, 4 and 5 will provide examples of the same type of

relation. In those constructions an abstract nominal
morpheme (the X of 20 above) selects a VP. Given, however,
that such nominal morphemes, like all nominal heads, have an

external theta role, the external theta-role of VP cannot be
vertically assigned to N’ to become its external theta role
but must find its satisfaction by a different strategy. We
will see soon how this may be done. Let us conclude this
section with a demonstration of the analysis of raising
constructions made possible within this framework. The main

assumption on which this analysis rests has to do with the
thematic properties of raising predicates. These are
identical to those assumed for I, namely, a raising predicate

has neither an external nor an internal theta-role. The
theta-structure of a standard raising construction would be
the following.
30
28- IP 2
/ \
NPl 1*2 1
I / \
John I VP2i
/ \
V IP11
I /. \
seems NP I *11
I / \
ei I VPi1
I / \
to Vi Adv.
work hard
In this configuration the external theta role of the verb
work is vertically assigned to VPi and then to I’i. At this
stage it is assigned to the NP trace in the subject position

of IPi . However, an NP trace, by definition, cannot satisfy
a theta role and, thus, the theta role is assigned to IPi in
its quest for satisfaction. It is subsequently assigned to

VP2 which is headed by seem, a raising predicate which has no
external theta role of its own, and ends up being assigned to

the subject John after being vertically assigned first to
1*2. A crucial feature of this analysis of raising is the
31
assumption concerning the nature of NP trace. The defining
properties of NP trace in the present framework are theta
theoretic rather than "binding theoretic". In the present
approach an NP trace is characterized as an element which
must be assigned a theta role but cannot satisfy it. The
distributional properties of NP trace, in particular,
certain locality restrictions applying to the relation
between an NP trace and its antecedent follow from its
assumed definitional properties combined with general
assumptions concerning the nature and distribution of theta
roles. This, in most cases, gives the same result as in a
standard, binding theoretic, approach to NP trace. But this
is not always the case. In particular, this view of NP trace
makes the prediction that i t should be possible for cases to
arise in which an NP trace has no antecedent occupying a
syntactic position. This is the case in infinitival clauses
headed by COMP, a matter to which we turn in the following
section.
1.2.3 Infinitival Clauses and Control
This section will present a brief discussion of
infinitival clauses and the phenomenon of obligatory control.
The analysis of obligatory control suggested here takes its
inspiration mainly from the work of Finer (1985) and Borer
32
(1989) and may be viewed as a theta theoretic implementation
of the insights of Borer (1989) . This section, however, is
only intended as a preliminary sketch. A detailed discussion
of control would go much beyond the intended scope of this
introduction.
We will follow Borer (1989) in assuming the non-existence
of the element PRO. However, unlike Borer (1989) who assumes
that it is small pro which figures in the subject position of
infinitives, we will assume that it is an NP trace which

occupies this position. The crucial feature of our approach
to infinitives will have to do with the complementizer
position and the particular element occupying this position

in (some) infinitives. We are concerned now with infinitival
clauses such as the sentential subject of 29a and the
complement of try in 29b.
29- a) to read this book is boring

b) John tried to read this book
The structure of the underlined part of 29a, b is the

following,
33
30- CP
/ \
c
/ \
COMP IP
/ \
NP Ii'
I / \
et I VPi
I / \
to Vi NP
I I
read this book
1» j
In 30 the external theta role of VP is vertically assigned to

I’ and from I’ to an NP trace in the subject position of IP.
Given the non-existence of PRO, an NP trace is the only
element that may occupy this position. In particular, empty
pronominal elements such as pro and expletive pro are
excluded from this position, pro is excluded from this
position since it requires to be "identified" by inflectional
morphology which is absent in this case (or else, pro is
simply not available in English) and expletive pro is
excluded due to the lack of Case to this position, an issue
which will be discussed in more detail in chapter 4.
34
Clearly, the lack of Case to this position is also what
excludes the occurrence of a lexical NP in it. Note that we
are here following standard assumptions with respect to the
clausal (as opposed to VP) structure of infinitives. This
view is standard within the Government and Binding framework
but otherwise far from being universally accepted. The

discussion in chapter 4 will provide what seems to be fairly
strong evidence in favour of this view.
Note now that our analysis of 30 is not yet complete
since, given that an NP trace can not satisfy a theta role,
it is not yet clear how the external theta role of VP is
satisfied in this structure. The crucial assumption here has
to do with the properties of COMP, in particular, its
argument structure. This may be represented as follows.
31- Argument Structure of COMP: (Ri , Rj )
Note that the element COMP in 31 is a particular, abstract,

lexical element and the assumptions about its thematic
properties are limited to this element alone. In particular,
COMP is different from that which selects a finite INFL (and,
therefore, IP) and, presumably, lacks an internal R role.
The element COMP which, by assumption, selects an infinitival

INFL (or IP) has an external theta-role R which makes it
possible for the CP of which it is the head to occupy
argument positions. It is in this sense that COMP may be
35
said to be a "nominalizer" . In addition, COMP also has an
internal R role which makes it possible for a "lower" theta-

role to be linked to it and in this way to be "satisfied".
Thus, a fuller picture of the theta-structure of infinitival
clauses such as in 29a, b would look as follows.
32- CPi
/ \
C’i
/ \
COMP IPj
Ri , Rj / \
NP I'j
I / \
ej I VPj
I / \
to Vj NP
I I
read this book
i, . .
In this structure the external theta role of read is assigned

to the NP trace but, being unable to be satisfied by it, it
is assigned to IP and ends up being linked to the internal
argument of COMP. At this stage, the formal operation of

theta role assignment has been accomplished so that the
36
external theta-role of the verb may be said to be satisfied.
If no other grammatical mechanisms or lexical requirements
are at work the external theta-role is construed as
arbitrary, or generic, in reference. This is the case in

29a. 29b, on the other hand, exemplifies the phenomenon of
obligatory control, another mechanism by which the reference
of the external theta-role of VP in a configuration such as

32 may be determined.
Obligatory control of an infinitival clause is a
lexically established relationship between an argument of a
governing control verb and the argument structure of COMP.

The lexical entry of the verb try would look as follows,
33- try; Agi , Xj
< COMF1 >
This lexical representation encodes a requirement that a
thematic index associated with COMP be coindexed, that is,

interpreted as coref erential with an argument of the
governing verb. I assume that no particular difficulties

arise from a lexical specification which makes reference to a
zero level element (COMP) rather than to its maximal

projection. A governing verb has access to the head of a
governed constituent and dependency relationships can

therefore exist between the two heads. Obviously, in the
37
actual structure COMP is the head of the maximal projection
CP, but that follows from X'- theory. The notation in 33 is
adopted, with some modifications, from Zubizarreta (1985).
In particular, unlike the notation suggested in Zubizarreta
(1985) , 3 3 mentions the types of theta-roles assigned.
Crucially, I will adopt one particular feature of this
notational device which relates to the external/internal
distinction. In a notation such as in 33 only internal
arguments, unlike external arguments, are associated with a
syntactic frame. Thus, the internal/external distinction is

2
encoded in lexical structure in terms of subcategorization .
This also makes it possible to establish unambiguously a
control relationship between COMP and any of the arguments of
the verb governing i t since each of the arguments may be
uniquely identified. Note that if this format of lexical
representations is strictly observed, in particular, the
asymmetry between external and internal arguments, then the
possibility of control of an infinitival clause occupying a
subject position is excluded since a lexical representation
of the form in 3 3 can not "look inside" an external argument.
Thus, a controlled constituent can only be a lexically
designated complement of a verb (or another lexical head) .

3
This seems largely to be true .
Now, as just mentioned, the notation in 33 is to be
interpreted as expressing the requirement that one (not
necessarily the external) of the arguments of COMP be
38
coindexed with the controlling argument of the higher verb.
However, COMP is associated with two thematic indices,
namely, its external theta-role and its internal theta-role
which happens to be coindexed with the external theta-role of

VP. Thus, the main task of the present account is to avoid
ambiguity in the control relationship. It is only the
internal, and not the external, theta-role of COMP which is
understood as being controlled. This, however, can be
established in a rather simple fashion. Looking at the

following structure in which the subject of the matrix verb
is the lexically designated controller,
I
34- IP
/ \
NPj VP
/ \
V CPi
i, i / \
COMP IPk
Ri ,Rk / \
ek I'k
I \
I VPk
suppose that the subject is construed with the external

argument of COMP, so that j=i. However, the external
39
argument of COMP is also the external argument of CP. Now,
CP being a lexically required complement of the verb, i t is
coindexed with the verb's internal argument. The result is
therefore that CP is coindexed with the subject NP. But this
is a simple violation of principle C of the binding theory.
Thus, CP may be thought of as referential in some sense and,
indeed, this is the way CP is viewed within the present
framework. But CP is certainly not anaphoric. It follows
that only a coindexation with the internal argument of COMP
can satisfy the control requirement while avoiding such a
violation, that i s , only a situation in which j=k is
possible .
Given this analysis, some of the well known properties
of obligatory control ( t o be distinguished from non-
obligatory control) seem to follow in a rather natural
fashion. The literature on obligatory control (e.g. Williams
(1980), Chierchia (1984)) usually enumerates a certain number
of properties taken to be distinctive of this relation. In
the following I will go through some of these properties in
an attempt to show how the present analysis might provide an
account for them.
First, the relationship between a controlling NP and a
controlled infinitive is strictly local. In particular, it
can never cross an IP boundary.
*35- Mary believes that John tried to kill herself
40
Within the present approach a controller of COMP can only be
a lexically designated argument of the control verb governing
COMP. That is, the controller and the controlled CP must be
coarguments of the same verb. Therefore, cases like 35 in

which the intended controller is not such an argument are
strictly excluded.
Second, a lexical NP cannot appear in the position of the

controlled subject (PRO in the standard theory of control)
36- *a) John tried Mary to leave

*b) John promised Mary John to leave
This, again, follows from the present analysis since a
control relation is a relation between the internal argument

of COMP and an argument of the governing verb. However, the
internal R argument of COMP must be satisfied, that is, it

must be assigned a theta role. This holds as an
independently motivated requirement. Namely, an R role, just
like an argument NP, must be thematically licensed, that is,

it has to be assigned a theta role. Now, the theta role
assigned to the internal R-role of COMP can only be the
external the external theta role of the lower verb. It

follows that the only element that can occupy the lower
subject position is an NP trace since this is the only
element which does not satisfy a theta role and, therefore,
makes possible its further upward propagation and assignment
41
to R. Any other nominal element will require this theta role
for itself and in this way block its upward propagation and
assignment to COMP.
Third, the controlling NP is uniquely determined.
*37- John promised Mary to wash herself
Given the notational device suggested here, the controlling
argument is lexically designated. In the case of promise

above, the subject (or Agent) is lexically specified as the
controlling element. Therefore, the anaphor in the
infinitival clause in this example couldn't be construed with
Mary as antecedent. A lexical entry for a control verb
couldn't specify two, or more, controlling arguments at one
and the same time since that will result in a lexical entry
in which both arguments will be specified as coreferential
and this is, simply, never the case.
Finally, Williams (1980) takes the requirement of c-
command between a controller and a controlled element to be
characteristic of obligatory control (his 0C3) . Thus, in 38a

below the preferred reading for the infinitival subject seems
to be with John as controlling its subject. But, as 38b
shows, an arbitrary reading is certainly possible for an
infinitival subject.
42
38- a) To leave would be John's pleasure
b) To leave would be a pleasure
Again, the c-command restriction follows naturally from the
present account. Under this account, the only position in
which a controlled infinitive wouldn't be c-commanded by its
antecedent is the subject position. But, as noted earlier,
the proposed format for encoding control relationships (cf.
33) is such that it excludes, in principle, control of a
sentential subject. This follows from the convention that
only internal arguments are associated in lexical entries
with subcategorization frames.
1 .3 Levels of Representation
An essential part of the view of the structure of
language within the generative tradition has always been the
distinction between different levels of representation. The
assumption of different levels of representation has been
originally motivated by the claim that certain linguistic
phenomena may be explained if one assumes that certain
elements in a sentence "belong", in some sense, to a
different position than the one at which they appear in the
actual spoken expression. It was therefore suggested that
such explanations could be achieved if one assumed the
43
existence of an underlying level of representation, a level
of "deep structure", from which a level of "surface
structure" could be derived by the application of various
movement operations. Standard work within the Government and
Binding framework (e.g. Chomsky (1981)) makes a distinction
between, at least, two levels of representation, namely, D-
structure and S-structure. The level of D-structure is
thought of as more basic in that it is this level which is
directly generated by the base in accordance with the general
principles of phrase structure (X' -theory). The mapping from
D-structure to S-structure is achieved by a transformational
component which consists of the general rule "Move-Alpha".
In particular, under the standard view, the mapping from D-

structure to S-structure includes both what is known as NP-
movement as well as A’-movement. NP-movement is movement
into argument (or A-) positions, that is, positions which can
bear grammatical functions and to which, in principle, a
theta-role may be assigned. A typical instance of this type
of movement is the transformation known as "raising".
A ’-movement, of which the most typical case is WH-movement,

moves an element to a non-argument position.
Van-Riemsdijk and Williams (1981) argue for the

existence of an additional, intermediate, level of
representation between D-structure and S-structure, a level

to which they refer as NP-Structure . In their model, NP-
44
movement maps D-structure to NP-structure and A'-movement
derives S-structure from NP-structure.

Williams (1986) goes one step further by claiming that
D-structure can be dispensed with if NP-structure is taken
to be the level at which A/A (as opposed to A/A') relations

are regulated. In the previous section it was shown how this
can be done if one adopts a certain set of principles which

characterize the nature of thematic relations and regulate
the distribution of theta-roles and NP-traces. Within such
an approach the distribution of NP-traces is governed by the

principles of theta-theory which apply at NP-structure. I
will adopt Williams' approach to the distinction between NP-

4
structure and S-structure . The term "underlying structure"
which will be often used in this dissertation is to be

understood as referring to NP-structure in the sense of
Williams (1986) . S-structure is derived from NP-structure by
the application of various movement processes such as WH-

movement, topicalization etc. all of which are instances of
A'-movement in the usual sense. Thus, the model suggested by
Williams (1986) does not get rid of all types of movement but
only of NP-movement.
This dissertation makes crucial use of the
transformational mechanism known as head movement which was
discussed in section 1.1 above. Clearly, head movement does
not put in relation two, or more, argument positions and in

this respect it may be viewed as another instance of A'-
45
movement. One might, therefore, say that head movement is a
surface phenomenon since it does not affect thematic
relations but, rather, applies to a level at which such
relations have already been established. We may therefore
consider head movement to be part of the inventory of

movement operations including WH-movement, topicalization
etc. by which S-structure is derived from NP-structure. As
will be seen, S-structure configurations may also be derived

by movement of NP 's . In such cases , however , movement is
purely motivated by the need to satisfy Case requirements and

the landing site of such movement is arguably a non-argument
position.
46
Footnotes to Chapter I
1
The reader may have noticed that the term predication is
being avoided here. Originally (e.g. Williams (1980)), the
predication relation has been assumed to hold between a maximal
projection (VP, AP, etc. ) and a subject NP. Note that, since I’
is not a maximal projection, the relation of theta-role
assignment between an I' node and a subject NP is not entirely in
accordance with the strict formulation of the rule of predication
in Williams (1980) . In recent work Williams (1990) suggests that
the relation between a subject NP and I’ may be thought of as the
relation between two heads in a doubly headed structure. Thus,
for instance, the sequence from Alabama to Louisiana in (i)a is a
doubly headed structure which may be represented as in (i)b (both
taken from Williams (1990) ).
(i) a) John played the banjo from Alabama to Louisiana

b) PP
/ \
PP PP
/ \ / \
P NP P NP
The idea is that a sentence is a doubly headed structure
([Nominative, Tense] P) of which one head is a phrase marked as
Nominative and the other is a phrase marked for tense. In such a
case both heads count as maximal.
(ii) [Nominative, T]P

I \
[N+Nom] P [V+Tense]P
This, however, leaves open the question concerning the relation
between a predicate and its subject in an infinitival clause. It
might be (as suggested to me by E. Williams (personal
communication)) that the same holds for infinitives with the two
heads differently marked. I will simply leave this issue open
and assume the mechanism of theta-role assignment by upward
propagation without making reference to the notion of
predication.
2
I am using two types of notation for lexical entries, the
"fuller" type such as in 33 which includes subcategorization
frames, and the "reduced" type without subcategorization frames
in which an external theta-role is marked by underlining. As far
as I can see, there is no conflict between these two types of
notation. In this work, the argument structure of lexical items
will be mostly represented by the reduced type. The fuller type
will only
be used for the discussion of control of Action Nominalizations
in Chapter 3 .
47
3
One does find cases in which a non-finite constituent in
subject position seems to be in some sort of control relation
with an element which is embedded inside the predicate ((i)a was
suggested to me by R. Higgins) .
(i) a) To perjure himself would ruin Edward

b) Killing herself will be Mary's end
It seems reasonable to assume that these are not cases of

obligatory control, that is, a control relationship which is the
result of a requirement by a certain verb and which must be
satisfied. Thus, in analogy to (i)a above, consider the
following examples.
(ii) a) Perjuring himself would ruin Edward
b) Raising taxes would ruin this country
c) Raising taxes would ruin this president
It seems that, for some reason, Edward in (ii)a must be
construed as a controller of the gerundive NP in subject
position. In (ii)b, however, such a construal is highly
unlikely. As to (ii)c, it seems to be ambiguous between a
control and a ”non-control" construal. Thus, it is possible,
and more likely, to interpret (ii)c with the president as the
one who raises taxes, but an interpretation in which some other
authority, say, the senate, raises taxes is not impossible. This
kind of freedom in control relationships is not attested in the
core cases of obligatory control and therefore we may assume that
examples such as (i) and (ii) above involve some other type of
mechanism.
4
Williams (1986) suggests also that the level of LF be
dissolved and its functions be reassigned to the two remaining
levels of NP-structure and S-structure. In this work, the level
of LF will be assumed, at least at one point, but not crucially.
Otherwise, questions related to LF will play no role with respect
to the main issues with which this dissertation is concerned and
therefore it will have nothing to contribute on these issues.
48
CHAPTER II
THE SYNTAX OF GENITIVE CONSTRUCTIONS
The main theoretical interest and most challenging
aspect of verbo-nominal constructions derives from the
mixed array of properties manifested by these
constructions. Most important among the nominal
characteristics of verbal-nouns is their internal NP-like
syntax, in particular, the possibility for elements
within the construction to be related to each other as
terms in a genitive construction and be subject to the
regularities characteristic of the syntax of any noun
phrase. ' Therefore, a reasonably adequate account of the
syntax of genitive constructions is, in my view,
indispensable for the understanding of the syntax of
verbal nouns .
Genitive constructions in Hebrew and Arabic appear
in three variants, of which the most widely discussed has
been the construct state ( henceforth the Bound
Genitive ) exemplified in 1 below. The example in 2 is
of a Free Genitive, a construction in which a head noun
is followed by Sei (of) NP . The NP in the Sei phrase
is interpreted as the "possessor" . The example in 3
shows a Double Genitive, a construction in which a clitic
pronoun is attached to the head noun, the first term in
49
the construction. The clitic pronoun obligatorily
"doubles" the NP in the Sel-phrase, that is, it must
match in features and be interpreted as coreferential
with the NP.
1- a) (H) mazkir ha-miflaga
secretary the party

"the secretary of the party"
b) (A) ra?iisu 1- Hizb-i
head the party
"the head of the party"

2- (H) ha-xaverim Sei ha-miflaga
the members Sei the party
"the members of the party"

3- (H) xavere-ha Sei ha-miflaga
members-her Sei the party

"the members of the party"
The bulk of this chapter, section 2.1, will be
devoted to a discussion of the bound genitive. Section

2.2 will present a discussion of some issues concerning
the interpretation of genitive constructions. Free
genitives and double genitives will be discussed in
section 2.3.
50
2.1 Bound Genitives (the Construct State)
The bound genitive construction has enjoyed a
considerable amount of attention in recent years and is
still the subject of much dispute. In the following I
will first present the basic data and then go into a
discussion of the different approaches in an attempt to
defend my own analysis.
2.1.1 The Data
The bound genitive is a construction which relates
two terms, namely, a head noun and an NP following it, in
a genitive type of relationship. Semantically, the
content of this relationship may be construed in many
different ways such as possessed-possessor, action-theme
or any kind of pragmatically or semantically conceivable
kind of relation.
4- a) (H) beyt ha-more
house the teacher
"the house of the teacher"
b) ( A ) bayt-u l-mu9allim-i
house • the teacher-Gen
51
5- a) (H) harisat ha-ir
destruction the city

"the destruction of the city"
b) (A) qatlu 1- waziir-i
killing the minister-Gen
"the killing of the minister"

6- a) (H) xoxmat ha-el
wisdom the god

"the wisdom of god"
b) (A) S’afaa? l-ma?-i
purity the water
"the purity of the water"
The main facts about this construction are of two

types. The first has to do with the syntactic
restrictions to which it is subject and the second with

restrictions on its interpretation.
To begin with the syntactic restrictions, those are
first, the impossibility of any intervening material

between the two members of the construction. Thus, in 7
an intervening adjective renders the structure
ungrammatical (compare with 1) .
7- *a) (H) xaverim ne?emanim ha-miflaga
members loyal the party
52
*b) (A) ra?iis kabiir ddawla
head great the state
An adjective may only appear following the second term in

the construction.
8- a) (H) xavre ha-miflaga ha-ne?emanim

members the party the loyal
"the loyal members of the party"
b) (A) ra?iis ddawlat-i al-kabiir

head the state-Gen the great
"the great head of the state"
In 8 the adjective modifies the initial (head) noun

but it may also modify the following NP or be ambiguous
between the two. The same fact may be illustrated with
relative clauses as in the following examples.
9- *a) (H) mazkir Se nivxar emeS
secretary that was elected yesterday

ha-miflaga
the party
*b) (A) ra?iis allad'i ?untuxiba ?amsi
head who was elected yesterday
ddawla
the state
53
10- a) (H) mazkir ha-miflaga Se nivxar
secretary the party that was elected

emeS
yesterday
"The secretary of the party who was elected
yesterday"
b) (A) ra?iisu ddawlat-i allad'i ?untuxiba
head the state who was elected
?amsi
yesterday
"The head of the state who was elected yesterday"
Likewise, a clitic pronoun attached to the first noun
counts as intervening material and renders the structure
bad.
11- *a) (H) beyt-o ha-mazkir
house-his the secretary

*b) (A) baytu-hu rra?iis
house-his the president
A second syntactic restriction applies to the
occurrence of definite articles in bound genitives,

namely, a definite article may not be attached to the
initial (head) noun but may only appear in the following
NP.
54
12- *a) (H) ha-mazkir ha-miflaga
the secretary the party

*b) (A) rra?iis ddawla
the head the state
As to semantic interpretation, the most striking and
often discussed fact about the bound genitive has to do

with the definite/non-def inite distinction. If the NP
following the head noun is definite then the structure as
a whole is interpreted as definite. Thus, in la,b
(repeated here as 13a, b) the NP's are interpreted as
denoting to the (unique) head/secretary of the party and
not to a head/secretary of the party (that is, one of
the heads/secretaries of the party) .
secretary the party

b) (A) raîiisu 1- Hizb-i
head the party-Gen

That the NP's in 13 are indeed definite can be
demonstrated by indicating the way they interact with
syntactic processes that are sensitive to the

definite/non-definite distinction.
55
Thus, in Hebrew an NP figuring as direct object must
be preceded by the object marker (OM) et if and only if
it is definite. The contrast in 14 shows that the NP
mazkir ha-miflaqa (the secretary of the party) as a whole
is perceived as definite. The situation in 15 is exactly

the opposite.
14- a) (H) rait-i et mazkir ha-miflaga
saw I OM secretary the party
"I saw the secretary of the party"

*b) (H) rait-i mazkir ha-miflaga
saw I secretary the party
15- a) (H) rait-i mazkir miflaga

saw I a secretary party
"I saw a secretary of a party"

*b) (H) rait-i et mazkir miflaga
saw I OM secretary party
One grammatical process in Arabic which is sensitive

to definiteness is Relative Clause formation. A relative
pronoun must occur as the head of a relative clause if

the NP modified by it is definite (16) . Otherwise, a
relative pronoun is absent as in 17 below.
56
16- a) (A) ra?ayt-u ra?iis-a 1- Hizb-i llad'i
saw I head ACC the party-Gen who
?untuxiba ?amsi
was elected yesterday
"I saw the head of the party who was elected
yesterday”
*b) (A) ra?ayt-u ra?iis-a 1- Hizb-i ?untuxiba
saw I head ACC the party-Gen was elected

?amsi
yesterday
17- *a) (A) ra?ayt-u ra?iis-a Hizb-en llad'i
saw I head Acc party Gen who
Tuntuxiba ?amsi
was elected yesterday

b) (A) ra?ayt-u ra?iis-a Hizb-en ?untuxiba
saw I head Acc party-Gen was elected

?amsi
yesterday
"I saw a head of a party who was elected yesterday"
In 16 it is "the head of the party" who was elected
yesterday and not just "the head" or "the party". The NP

as a whole is perceived as definite and thus dictates the
particular form of the relative clause. In 17, on the
other hand, it is an indefinite bound genitive
57
construction which is being modified and therefore a
relative pronoun must not appear.
Another relevant fact concerns modification of the
head noun by an adjective. An adjective functioning as a
modifier (as opposed to a predicate) must agree with the
modified head on definiteness as well as on number and
gender. An adjective modifying the head noun of a bound
genitive must be marked as definite if the head noun is
followed by a definite NP.
18- a) (H) mazkir ha-miflaga ha-zaken

secretary the party the old
"the old secretary of the party”
*b) (H) mazkir ha-miflaga zaken
secretary the party old
19- a) (A) mudiir-u 1-madrasat-i ttawiil-u
director-Nom the school -Gen the tall-Nom

"the tall director of the school"
*b) (A) mudiir-u 1-madrasat-i tawiil-u
director-Nom the school -Gen tall -Norn
In both 18 and 19 the NP’s "the party" and "the school"

are feminine. Therefore, the following adjectives must
be interpreted as modifying the head noun, which is
masculine. As the contrast between the a and the b
58
examples shows, the adjective must be preceded by a
definite article.
These are the main facts which any adequate account
of the bound genitive must provide an account for. The
analyses of the bound genitive which have been proposed
in the literature in recent years may be classified into
two main approaches, the complement approach and the
specifier approach. The complement approach, represented
mainly by the work of H. Borer (1984,1988), views the
relationship between the members in a bound genitive
construction as one between a head and its complement.
Under this approach the head and its complement are base
generated in their surface order and no movement
operations are needed to map the one into the other.
Under the specifier approach, on the other hand, the NP
(the second member of the construction) is base generated
as a specifier to the head. However, in order to derive
the surface word order some sort of movement operation
must be invoked. This may be achieved in several
different ways, a point on which different versions of
the specifier approach diverge. To the best of my
knowledge, a version of the specifier approach has been
available in the literature since a t least as early as
Aoun (1978) . More recent proposals within the specifier
approach make use of the DP analysis of NP structure put
forth by Abney (1986,1987). This particular treatment of
59
bound genitives is due mainly to Ritter (1986) . Slightly
different versions of it are presented in Fassi-Fehri
(1988) and Mohammad (1989) .
In the following discussion I will first review the
complement approach of Borer (1984,1988) and present my
arguments against it (section 2.1.2). I will then
proceed to a presentation of my own analysis, a variant
of the DP analysis (section 2.1.3).
2.1.2 The Complement Approach to Bound Genitives
Borer (1984) proposes that 20a, b have the structure
in 20c.
20- a ) ( H ) mazkir ha-miflaga
secretary the party
b) (A) za9iimu 1- Hizb-i
leader the party-Gen
"the leader of the party"
c) N*
/ \
N NP
leader the party
secretary
60
It is suggested that 20c is the configuration for the
assignment of genitive Case. In this configuration the

first node dominating the head noun dominates also the
complement NP. Thus, the complement NP may be said to be
governed, government being the condition for Case

assignment. Ungrammatical cases in which some material
intervenes between the head and its complement are
straightforwardly accounted for in this approach.
*21- N'
/ \
N' NP
/ \
N (AP/RCI
Thus, 21, with an intervening adjective (cf. examples

7,8) or a relative clause (cf. examples 9,10), is ruled
out as a violation of an assumed strict c-command
restriction on the government of the complement by the

head. Cases in which a clitic pronoun is attached to the
head (cf. example 11) may be explained as having either
of the structures in 22.
61
22- *a) N' *b) N'
/ \ / \
N' NP N-cl NP
/ \
N-cl pro
22a incorporates an assumption that a clitic is always
associated with an empty pronominal category which it may

be said to "identify" (assign features ,content etc. ).
22a is thus ruled out on the same grounds as 21. In 22b

no pro is present and therefore no configurational
explanation may be invoked for its ungrammaticality. It
is however possible, within the framework assumed by
Borer, that a clitic must be assigned Case. If this is

so then the clitic in 22b may be said to absorb the Case
feature assigned by the head and thus deprive the
complement NP of Genitive Case. The structure is thus

ruled out as a violation of the Case filter.
Before proceeding into a discussion of the empirical
difficulties faced by the complement approach, it is
important to note a difficulty raised by it which is more
conceptual than empirical in character. The notion of

complement is generally used to refer to a syntactic
constituent standing in some privileged relationship to a
lexical head. Such a constituent, if present, is said to

satisfy requirements for which the head is lexically
62
specified. The syntactic realization of such a
relationship is assumed to be the configuration of
government. That is, a configuration which obeys severe
restrictions including, at least, sisterhood and possibly
linear adjacency. However, none of this holds with
respect to the core cases of the bound genitive
construction. In general, any noun can stand in a

genitive relationship to another noun and any noun can
simply stand on its own. No lexical restrictions are at
work in this respect. Moreover, the relationship of
being "the complement of" is not necessarily a one to one

relationship. Thus, a lexical head may be associated
with more then one constituent which may be said to be
its complement as, for example, in 23.
23- (A) sallama 1-waziir -u naa?ib-a -hu
handed the-minister-Nom aide -Acc-his

rrisaalat -a
the letter-Acc
"The minister handed his aide the letter"
However, only one NP may ever stand in a bound genitive

relationship to a head noun.
63
24- a) (A) tasliim 1-waziir -i naa?ib-a-hu
handing the minister—Gen aide -Acc-his

rrisaalat -a
the letter-Acc
"The minister's handing his aide the letter"

b) (A) tasliim naa?ib-i-hu rrisaalat-a
handing aide-Gen-his the letter-Acc

"The handing of the letter to his aide"
*c) (A) tasliim 1- waziir-i naa?ib-i-hu
handing the minister-Gen aide -Gen-his
rrisaalat-a
the letter-Acc
Note also that the kind of semantic relationship

that can be entertained by two terms in a bound genitive
may vary from the different kinds of argument-type
relationships to adjective-noun-like modification

relationship. Such a variation in the mode of semantic
construal is completely absent when it comes to the core
cases of constructions involving the head/complement
relationship such as, for example, verbal
complementation. Thus, the relations of "being a

complement of”, on the one hand, and the relation of
"being the second member in a bound genitive

construction" seem to belong to two fundamentally
different categories. It seems to me that this point
64
alone casts serious doubts on the complement approach.
However, in the rest of this section I will concentrate
on its empirical inadequacies.
Coming back now to the empirical issues, the

complement approach runs into serious problems when it
comes to explaining the ungrammaticality of cases in
which a bound genitive is preceded by a definite article

(cf.12). A configuration such as the one in 25 (proposed
in Borer (1984)) would, in fact, count as a violation of
the assumed conditions on the assignment of genitive

Case, but there is no motivation for the assumption of
such a structure.
25- N'
/ \
N*
/ \
Det N
25 implies the existence of a phrase structure rule such
as: N' --- > Det N , but such a rule seems to have no
motivation, neither on general theoretical grounds, since

it implies that Determiners may be base generated under
N' rather than as specifiers of NP immediately dominated
by the maximal projection, nor is it justified on
language specific grounds since Hebrew and Arabic are
65
both assumed to be, and in fact are, head initial
languages .
Another attempt to deal with this problem, still
within the complement approach, is a suggestion made by
Shlonsky (1987) to the effect that Hebrew (and Arabic)
NP's simply lack a specifier position. Following this
view the examples under consideration would have the
structures as follows.
26- a) NP b) NP
I I
N' N’
/ \ / \
N NP N NP
leader party leader the party
Note that, given this view, the definite article in

the NP complement of 26b couldn't possibly occupy a
specifier position in that NP since such a position, by

assumption, simply does not exist. It therefore must be
assumed that this definite article is the realization of
a definiteness feature specified, say, as [+Def ] , which
is present in the underlying feature composition of the
noun in question, together with other features such as
number, gender and so forth.

As a fact which seems to support such a view,
66
Shlonsky notes that elements such as quantifiers and
numerals, which typically figure as specifiers of NP in

languages such as English, seem to figure as nominal
heads in Hebrew.
27- a) (H) SloSet ha-yeladim
three the children

"the three children"
b) (H) SloSt-am
Three-them
"the three of them"
28- (H) ha-kol
the all
"everything"
In 27a the numeral "SloSa" (three) seems to function as a

head of a bound genitive since it undergoes the
phonological modification associated with this
construction. In 27b a clitic pronoun is attached to the

numeral "SloSa" and in 28 a definite marker "ha"
precedes the quantifier "koi" (all) . These facts are
brought in support of the claim that there is no need to
postulate a specifier position for Hebrew NP's. Rather,

these quantifier expressions would occupy the head
position of an NP.
67
An obvious objection that may immediately be raised
concerns the, by now familiar, examples repeated below.
29- *a) (H) ha-mazkir ha-miflaga

the secretary the party
*b) (H) ha-mazkir miflaga
the secretary party
Shlonsky's suggestion fails to account for the

ungrammaticality of the examples above. By assuming that
the definite marker occupies no syntactic position this
analysis makes the claim that 29 a,b have exactly the
same structure as the grammatical 26 a,b above. That is,
definiteness being a feature, and just a feature, there
is no reason why the head noun wouldn't be specified as
[+Def] as well. Next, observe the two following

examples.

secretary the party
b) (H) mazkir Sei ha-miflaga
secretary Sei the party
"a secretary of the party"
According to the view under consideration, 30b is assumed
68
to be the result of the application of an optional rule
of Sei insertion to the underlying structure 26b, which
is the structure of 30a (this is so given standard
assumptions about Sel» my assumptions are, however,

different, as will become clear later). However, 30a and
30b differ clearly in their interpretations, the one
I
(30a) being definite and the other (30b) indefinite. The
element Sei being a "dummy" Case marker, it is hard to

see how it could contribute to this difference in
I
meaning. Rather, one may consider the possibility that
30a and 30b are associated with two different underlying
structures, but these are not available under this
analysis. In fact, this objection holds regardless of

whether Sei is assumed to be base generated or, rather,
inserted at a later stage of the derivation for the
purpose of Case assignment. In either case it would be
I
implausible to assume that the difference in
interpretation between 30a and b derives from the

particular semantic contribution made by Sei. Thus ,
since, except for the presence or the absence of Sei , the
structures are identical, this difference in meaning
remains unexplained.
A third and last remark concerns the need to
postulate a specifier position in Hebrew NF's. There are
quantifier expressions that clearly fail to act like head
69
nouns. Thus, "harbe" (many) and ”kama" (some) cannot, for
example, have clitics attached to them.
31- a) (H) harbe
many
*b) (H) harbe-hem
many -them
32- a) (H) kama
some
*b) (H) kama -hem
some-them
In 33 a,b "harbe" (many) and "kama" (some) function as

regular quantifier expressions. Not being nominal heads
(as shown in 31, 32) , the most natural assumption would
be that they occupy a specifier position.
33- a) (H) harbe yeladim

many children
b) (H) kama yeladim

some children
A numeral like "SloSa" (three) , as already mentioned,

does have the possibility of having a clitic attached to
it. Note also that one finds constructions in which
70
numerals do not undergo the phonological modification
characteristic of heads of bound genitives (compare 27a) .
34- SloSa yeladim
three children
Now, "harbe" (many) , "kama" (some) and "SloSa” (three)
are in complementary distribution with the definite
marker "ha" .
35- *a) (H) harbe ha-yeladim
many the children
*b) (H) kama ha-yeladim
some the children
*c) ( H ) SloSa ha-yeladim
three the children
Moreover, exactly like the definite marker "ha", "harbe"
(many) , "kama" (some) and "SloSa" (three) cannot precede
the head of a bound genitive. Compare the following
examples with those in 29.
36- *a) ( H ) harbe yalde ha-kita
many children the class
*b) (H) kama yalde ha-kita
some children the class
71
*c) (H) SloSa yalde ha-kita
three children the class
It is therefore fairly plausible to assume, first, that

Hebrew (and for that matter Arabic) NP’s do have a
specifier position of the kind that it would be necessary
to postulate for 33,34, and, second, that the definite
marker "ha" occupies this very position, as would be
necessary in order to explain the facts in 35. Thus, the

analysis of bound genitives, in particular, a unified
account of the facts in 29 and 36, must have something to
1
do with this position .
Note that the proposal made by Shlonsky (1987) ,

along with its syntactic problems, still does not provide
(and, I assume, does not claim to provide) an account of
the interpretation facts. That is, the fact that, if the

"complement" NP is definite then the NP as a whole is
interpreted as definite. Borer (1988) presents an
attempt to provide an account for this fact making use of
the idea that definiteness may be viewed as a feature and

a special percolation mechanism by which the feature
[+Def] may percolate up and mark the NP, as a whole, as
definite. There is no need here to go into the details

of Borer's (1988) account which is embedded within a
general approach to morphology and the interaction
between morphology and Syntax and is designed to provide
72
support for such an approach. This general issue is
beyond the scope of the present discussion. The main
ideas of that account are the following. First, word
formation operations are possible both before D-structure

as well as after D-structure (that is, "in the syntax") .
A word formation operation applying to a certain domain
defines that domain as a word and thus makes it possible

for certain processes, such as feature percolation
processes which are characteristically word internal, to

take place. To make it possible for definiteness to mark
the NP as a whole the only additional assumption needed
is that definiteness in Hebrew is a feature. This
assumption is justified on the ground that Hebrew has
definiteness agreement. As has already been observed, a

head noun and a modifying adjective agree in
definiteness.
37- a) (H) ha-bayit ha-gadol/*gadol neheras

the house the big big was destroyed
"The big house was destroyed"
b) (H) bayit gadol /*ha-gadol neheras

house big the big was destroyed
"A big house was destroyed"
Thus, agreement processes are assumed to be processes
which involve features. Therefore, if definiteness is
73
subject to such processes then i t must be viewed as a
feature. Generally, the most typical type of information
that is assumed to b e encoded by the device of features
is "things" such as number and gender and, indeed, a head
noun and an adjective must agree in number and gender as
well .
38- *a) ( A ) al-bayt-u al-kabiirat-u
the house (ms) the big (fm.)
*b) (H) ha-bayit ha-gdol-im
the house (sg) the big -pl
Borer's percolation mechanism may be illustrated as
follows .
39- NP NP/ Def
NP/ Def
NP NP/ Def
Def Def
Thus, the configuration in 39 is defined as a domain in
which word formation operations are possible and this
allows the definiteness feature to percolate up.
The immediate question that comes to mind at this
point is what happens to other features . This question

is entirely justified since this whole approach is based
on the analogy between definiteness and other features.
Obviously, it is only definiteness which is ever allowed
to percolate up. Thus, it seems as though the
transparency to feature percolation which is achieved by
allowing for word formation rules to apply in the syntax
is rather selective, allowing certain types of
information in and filtering other types of information
out. The picture becomes even more suspicious once we
note that exactly the same phenomenon, that is, the
difference in behaviour between definiteness on the one
hand and other features on the other, may also be
observed in a language like English. Consider the
following examples (English makes no gender distinction

so only definiteness and number are relevant here) .
40- a) the man 1 s book [+Def, -Pl]

b) the man's books [+Def, +P1]
41- a) the men ’s book [+Def, -Pl]
b) the men's books [+Def, +P1]
42- a) a book [-Def, -Pl]
b) some books [-Def, +P1]
As the feature indications above show the specification
75
of the NP as a whole with respect to definiteness is
always determined by the material occupying its specifier

position. On the other hand only the head noun seems to
be relevant with respect to the overall number

specification of these NP’s. Thus, the NP's "the man"
and "the men" are [+Def, -Pl] and [+Def, +P1]
respectively and perceived as such by the syntactic

processes sensitive to these distinctions (number
agreement, definiteness effect, etc.). However, in the
cases above it is only the [+Def] feature of these NP’s
which seems to "percolate up". In this, English NP's

occupying the specifier position are rather strikingly
reminiscent of the Hebrew NP's assumed to be complements

in Borer's analysis of the bound genitive. It seems
rather obvious that something more general is here at
work than the, rather ad hoc, percolation mechanism
2
proposed by Borer .
Note that the difficulties pointed out so far for
the complement approach are all internal in character,
that is, they concern the syntax of bound genitives
considered in isolation from any other aspects of a given
construction in which it may occur. However, as already
stated, the main interest in the syntax of bound
genitives in the present context derives from the fact

that it constitutes one, important, aspect of the syntax
of verbo-nominal constructions. Therefore, any
76
satisfactory analysis of bound genitives must be
compatible with the general analysis of verbo-nominal

constructions such as Action Nominalizations (cf. chapter
3), nominal gerunds (cf. chapter 4) and Agent
Nominalizations (cf. chapter 5). A careful examination

of these constructions will reveal that the complement
approach to bound genitives is, quite clearly,

incompatible with any reasonable analysis of Action
Nominalization constructions, let alone the Nominal

Gerund and Agent Nominalizations. If this is indeed so,
then it constitutes another, fairly compelling, argument
against the complement approach. For a detailed
discussion of this particular issue with respect to

Action Nominalizations see footnote 4 to chapter 3.
77
2.1.3 A Specifier Approach to Bound Genitives:
The Head Movement Analysis
The analysis of bound genitives outlined in this
section makes use of an idea, due mainly to Abney
(1986,1987), as well as Fukui and Speas (1986) and
others, according to which constituents that have
traditionally been taken to be NP's should be viewed as
being headed not by a noun but rather by a functional
head, D( eterminer) , which takes an NP as its complement
in the same way that COMP is assumed to be the head of
CP and to have IP as its complement and INFL is assumed
to be the head of IP and have a VP as its complement.
The application of this approach to the analysis of
bound genitives is due mainly to Ritter (1986) . Variants
of Ritter's original analysis are developed in Fassi-
Fehri (1988) , Muhammad (1989) and Cohen (1990) . In the
following I will follow Ritter's proposal in most of its
details, deviating from it only as seems necessary in

3
order to overcome some of its main problems . In
developing this analysis I benefitted considerably from a
suggestion of Ur Shlonsky (personal communication) .
Under the present proposal, the underlying structure
of 43a, b is the one illustrated in 43c.
78
43- a) (H) mazkir miflaga
secretary party
"a secretary of a party"
b) (A) ra?iis Hizb-en
head party-Gen
"a head of a party"
c) DP
/ \
D'
/ \
D NP
! / \
POSS DP N'
/ \ I
D’ N
/ \ I
NP secretary
! head
N*
1
N
I
party
In the configuration above secretary/head is the head of
79
an NP embedded in a DP and party is a DP occupying the
specifier position of that NP. The head position of the
matrix DP is occupied by the abstract morpheme POSS.

POSS may be specified as [Def] . A DP headed by a [Def]
POSS is definite and a DP headed by a POSS unmarked for
definiteness is simply non-definite. In the
configuration above, the head (D) position of the DP

party is empty. In particular, there is no definite
article occupying this position. The position is

therefore indefinite by default and therefore also the
lower DP as a whole.
POSS is also a Case assigner. That is, the source of
genitive Case in a bound genitive is POSS rather than the

head noun and, crucially, this feature must be
discharged. It is also assumed that the relation of
genitive Case assignment is just like the relation of

nominative Case assignment, which holds between INFL and
a subject NP, in that this relation too holds under
agreement or, to use Chomsky’s (1986) expression, feature
sharing. The difference is that in this case the only
feature relevant for agreement is [Def]. However,
agreement being a process which applies between heads and
their specifiers, the assignment of genitive Case cannot
apply directly between POSS and party in their underlying

4
positions, as shown in 43c . There is, however, a way
out of this situation which consists of a head movement
80
adjoining the head noun to POSS and (following a
suggestion of Ur Shlonsky (personal communication) ) a
certain process of reanalysis as illustrated below.
44- DP
[Gen] Di NP
Ni [Gen] D DP N'
secretary [Gen] POSS NP N

head
N' ei
party
In 44 the head noun secretary/head moves and adjoins to

the element occupying the governing D position, leaving a
trace at the head position of NP. I will assume a
process of reanalysis by which the D node which dominates
the complex head N+POSS inherits the index of N. The
result of this process is a complex head which inherits

the features originating from POSS, that is, genitive
Case and, possibly, a definiteness feature (in case that
81
POSS is specified as [Def]) , and carries an index which
is identical to the one carried by the trace of N. It
follows that the complex head may now be construed as

antecedent of the trace of N and, since antecedence
implies the inheritance of features of the antecedent by
its trace, it also follows that the trace of N in 44
inherits the feature content of Di . In particular, the

trace of N inherits a genitive Case feature and a,
possible, specification for definiteness. Thus, it is
the trace of N in 44 which governs and assigns Case to
the DP in its specifier position. Since neither the
trace nor DP are definite in this case, the requirement
of agreement is satisfied and the DP party is licensed by
the Case filter.
As to the surface realization of such a
configuration, the sequence N+POSS is realized at the
surface as the bound form. The difference between the
bound and free forms is most visible with the plural and
feminine forms of nouns. 45 and 46 below demonstrate the
alternation /im/ — >/e/ for plural in Hebrew and /a/ —
>/at/ for feminine in Arabic (as well as in Hebrew)

respectively .
45- a) (H) mazkire miflagot

secretaries parties
"secretaries of parties"
82
*b) (H) mazkirim mifl ago t
secretaries parties
46- a) (A) ra?iis-at-u Hizb-en
head -Fm-Nom party-Gen
"a head of a party"

*b) (A) ra?iis-a Hizb-en
head -Fm party-Gen
The derivation of the definite variant of 43a, b is

identical except for the presence of a definite article
in the lower DP.
secretary the party

b) (A) ra?iis-u 1 - Hizb-i
head -Norn the party-Gen
83
c) DP DP
/ \ / \
D NP ====> Di [Def] NP
1 / \ / \ / \
POSS DP[Def] N* Nt D DP [Def] N*
[Def] / \ î 1 1 / \ 1
D NP N secretary POSS D NP N
1 1 1 head [Def] 1 I |
ha/al N' secretary ha/al N' et

1 head 1
N N
1
party party
The definite article ha/al and the morpheme POSS are, in
this analysis, two separate, independently existing,
elements. In particular, the definite article is not

assumed to be the surface realization of an underlying
feature [Def]. The definite article ha/al and POSS, both

being heads of category D, are in complementary
distribution with respect to the head position of DP.

Obviously, the definite article is definite, that is, it
is marked as [Def] and the DP of which it is the head is
marked as [Def] as well. Thus, in 47c, the trace of N

inherits Case and definiteness specifications from the
complex head Di and, given that the DP it governs is also
84
definite, Case assignment may apply. Thus, Case
assignment to a definite DP is possible only if POSS is

specified as [Def] , that is, if the matrix DP is
specified and interpreted as definite.
Some of the observations made earlier with respect
to ungrammatical cases of bound genitives which involve
material intervening between the head noun and the
following NP receive a natural explanation in this

analysis. Thus, an adjective modifying the moved head
would be simply left behind and could never intervene
between the two head nouns. An interesting implication

of this analysis is that it should be possible for two
adjectives to occur, one modifying the head noun and the

other modifying the noun in Spec. Such cases are
somewhat marginal in Hebrew but seem to be perfectly

natural and acceptable in Arabic. The structure and
derivation of 48a, b,c is illustrated in 49 below.
48- ?a) (H) menahelet beyt ha-sefer ha-xadaS
director (fm) the school the new

ha-mecuyenet
the excellent (fm)

"The excellent director of the new school"
85
b) (A) mudiir-u 1-madrasat-i 1-jadiidat-i
director— Nom the school -Gen the new -Gen
1-mumtaaz -u
the excellent-Nom
"The excellent director of the new school"
c) (A) za9iim-u 1-firqat-i ssaRiirat-i
leader-Nom the group -Gen the small-Gen

al Raniyy-u
the rich -Norn
"The rich leader of the small group"
49- DP DP
/ \ / \
D NP ====> Di NP
| / \ / \ / \
POSS DP N* Ni D DP N’
/ \ / \ 1 / \ / \
D NP N* AP POSS D NP N' AP
| / \ | 1 / \ \
ha/al N' N ha/al N* ei
/ \ / \
N* AP N* AP
N N
86
Switching the order of adjectives in these examples
renders them completely unacceptable. Note that the
nouns in these examples are different in gender and
therefore there is no ambiguity about which adjective is

supposed to modify which noun.
50-*a) (H) menahelet beyt ha-sefer ha-mecuyenet
director (fm) the school (ms) the excellent(fm)

ha-xadaS
the new (ms)
*b)(A) mudiir-u 1- madrasat-i al-mumtaz -u
director-Nom the school-Gen the excellent-Nom
al-kabiirat-i
the big -Gen
*c)(A) za9iim-u 1-firqat-i 1-Raniyy-u
leader-Nom the group-Gen the rich -Norn

ssaRiirat-i
the small-Gen
Note that even though 48a is somewhat marginal, 50a is

completely impossible, a contrast which is accounted for
by this analysis.
The present account also allows for a
straightforward derivation of multiply embedded cases
such as the following,
87
51- a ) (H) roS kvucat ha-kiconiyim
head group the extremists
"the head of the group of the extremists"
b) ( A ) ra?iisu firqat-i 1-mutaTarif iin
head group -Gen the extremists-Gen
"the head of the group of the extremists"
The structure and derivation of 51a, b is as follows.
52- a) DPi b) DPi
/ \ / \
D NPi Di NPi
1 / \ / \ / \
POSS DPa n* Nt POSS DPa N'
[Def] / \ 1 1 [Def] / \ 1
D NPz N head D NP 2 ei
/ \ 1 / \
DPa N * head DPa N’
/ \ 1 / \ 1
D NPa N D NPa N
ha/al N' group ha/al N* group

i
1
N N
extremists extremist
88
52- c ) DPi
/ \
Di NPi
/ \ / \
Ni D dp2 \
1 1 / \ N’
head POSS Dj \ 1
[Def] / \ np2 ei
Nj D / \
1 | DPa N'
group POSS / \ 1
[Def] D NPa N
I I I
ha/al N* ej
extremists
As the structures above show, 51a, b are derived by two
simple applications of head movement of the type
illustrated earlier for the non-embedded structure. It
is interesting however to observe how the interpretation
with respect to the def inite/non-def inite distinction is
computed in these structures. As mentioned above, a
definite D P is one of which the head position is occupied
89
by a definite element. That can be either POSS specified
as [Def] or the definite article ha/al . In 52a DPi must
be headed by a [Def] POSS since only POSS can provide the
genitive Case feature necessary for DP2 . Now, the POSS
heading DPi and the element heading DPa must (indirectly)

agree on definiteness. Of the two conceivable elements
that could occupy the head position of DPa , only a
definite POSS could satisfy this requirement since only
POSS carries the genitive Case features necessary for the
licensing of DPa . Next, looking at 52c, could a [Def]

POSS occupy the head position of DPa in order to satisfy
the agreement requirement between ej and DPa ? The answer
is "No" since POSS must discharge its genitive Case

feature and there is no DP further down to which this
feature could be assigned. It follows that the head of
DPa must be a definite article since this is the only

element that can satisfy the agreement requirement if
Case is to be assigned to DPa . Thus, in a multiply

embedded bound genitive it is only the last noun that may
be preceded by a definite article. It also follows from
the above suggestions that it is impossible for a non
embedded DP to be definite without an overt definite
article occupying its D position (ignoring right now
inherently definite elements such as proper names and
pronouns). Such a situation could result if a D
90
position was occupied by a definite POSS instead of a
definite article as follows.
53- DP
/ \
D NP
I I
POSS N'
[Def] |
In 53 POSS has nowhere to discharge its genitive Case

feature and the structure is therefore ruled out. For
the same reason, an isolated occurrence of a noun in a

bound form is impossible. Such a form may be the result
of the following configuration.
54- DP
/ \
Di NP
/ \ I
Ni D N*
I I
POSS N
I
ei
54 is ruled out for the same reason as 53.
91
2.2 The Interpretation of Genitive Constructions:
Identification and Classification
In dealing with the syntax of genitive constructions
it seems useful to make a certain distinction between two
main types of semantic relationship that may hold between
two members of a genitive construction. The first is a
possessive type of relationship, where "possessive” must
be understood in a very large sense. This is the kind of
relationship that is understood to hold between the two
members in the following examples.
55- a ) ( H ) beyt ha-more
house the teacher
b) ( H ) manhig ha-miflaga
leader the party
"the leader of the party"
c ) (H) xalon ha-bayit
window the house
"the window of the house"
56- a ) ( A ) baytu ttilmiid* -i
house the student -Gen
"the house of the student"
92
b) (A) ra?iisu ddawlat -i
head the state-Gen
"the head of the state"

c) (A) muHarriku ssiyyaarat-i
engine the car -Gen

"the engine of the car"
In the a examples above, the entities referred to by the
two nouns are most naturally understood to be related to
each other in the relation of possession. In the b and ç

examples, on the other hand, a different relationship
seems to hold. However, common to all these cases is an
interpretation involving two entities that are related to

each other in some way. The kind of relationship
involved is mostly determined by pragmatic factors, such
as general knowledge of the world and the like. I will
follow traditional terminology in referring to this type

of semantic relationship as Identif icatory . In an
identificatory interpretation the "possessor" half of the

pair answers the question "which", that is, the reference
of the NP as a whole is determined through the reference
of the "possessor".
A second type of semantic construal is exemplified by
the following NP's.
93
57- a) (H) metos krav
aircraft battle
"a battle aircraft"
b) (H) saxkan hagana
player defence
"a defence player"
c) (H) eSet mehandes
wife engineer
"the wife of an engineer"
58- a) (A) ka?su Hamr-en
glass wine-Gen
"a glass of wine"
b) (A) kursu xaSb-en

chair wood-Gen
"a wooden chair"

c) (A) qit9atu laHm-en
piece meat-Gen
"a piece of meat"
In these examples no interpretation is available in which

two separate entities stand in a relationship to each
other. Rather, the only available interpretation is one

in which one term, the second, restricts the reference of
the other. It is the same kind of relation that holds
between an adjective and a head noun, that is, a relation

of modification. In contrast to the identif icatory
94
relationship, in this case the genitive noun specifies,
or provides an answer to a question relating to, the
kind of something. The semantic relationship between the
two elements is such that the denotation of the
construction as a whole is a subset of the denotation of

the head alone. We will again follow traditional
terminology in referring to this mode of construal as

Classificatory ♦
In general, and this may be said with respect to

both Hebrew and Arabic (cf. Cowell 1964) there is a very
strong tendency to impose a classif icatory interpretation

on bound genitives of which the second term is
indefinite, whereas definite bound genitives are freely
ambiguous between the two types, depending, of course, on
the plausibility of the interpretation. Here it is
important to note that the process by which

classificatory bound genitives are formed is extremely
general and new combinations may be formed quite freely.

The interpretation that such forms could receive would be
strictly compositional in the same way that the meaning
of a noun and the meaning of an adjective are
compositionally combined to produce an overall meaning.

It is therefore not necessary to assume that forms such
as in 57,58 are compounds that are generated in the
morphological component, just as no such assumption is
95
made with respect to adjectives. Indeed, it seems
desirable to maintain the same general analysis of bound

genitives suggested here for classif icatory genitives as
well and this for the following reasons.
First, classificatory bound genitives observe exactly
the same restrictions with respect to the occurrence of a
definite article as do identificatory bound genitives.
Compare the following ungrammatical examples with 57a and

58a respectively.
*59 - (H) ha-matos krav

the aircraft battle
*60 - (A) al-ka?s-u Hamr-en
the glass wine-Gen
To the extent that these restrictions seem to require an
explanation in terms of syntactic configuration and
derivation when dealing with identif icatory
constructions, it seems reasonable to assume that the

same should apply when dealing with classif icatory
constructions. Note also that, just like adjectives, it

is possible to conjoin genitive nouns which are
interpreted in the classificatory manner.
96
61- a ) (H) metos krav ve hafcaca
aircraft battle and bombing
"an aircraft for both battle and bombing"
b) (H) saxkan hatkafa ve hagana
player offence and defence
"a player of both offence and defence"
c ) (H) mivne zxuxit u varzel
construction glass and iron
"a construction made of glass and iron"
Finally, we may note that, in Hebrew, the classif icatory
mode of construal is possible not only with bound
genitives but also with Sel-phrases.
62- a) (H) bayit Sei ec
house Sei wood
"a house of wood" ( a house made of wood)
b) (H) moadon Sei studentim
a club Sei students
"a club of students" ( a students club)
Moreover, i t is also possible for an adjective to
intervene between the head and the Sel-phrase.
97
63- a) (H) bayit gadol Sel ec
house big Sel wood
"a big wooden house"
b) ( H ) moadon yaSan Sel studentim
club old Sel students
"an old students club"
In the following slection it will be suggested that Sel-
phrases are adjuncts and that, being adjuncts, they may
be adjoined to a syntactic structure a t any syntactic
level. In any case, whatever analysis of Sel-phrases one
adopts, it seems highly unlikely, in view of examples
62,63, to assume that a classif icatory construal in the
case of Sel-phrases is due to certain semantic
combinatorial mechanisms that are exclusively associated
with morphological, as opposed to syntactic,
configurations. That is, i t is unlikely to assume that
the classifica oryl construal is exclusively associated
with semantic interpretation procedures that accompany
structure building processes at a level "earlier" than
the syntax, namely, the morphology. Rather, the
classif icatory mode of construal must be assumed to be
able to apply at a stage higher than the word level.
Given this fundamental distinction between these two
types of interpretation, i t might be useful to wonder how

i
they might be reflected in syntactic representation. As
98
just observed, the fundamental difference between these
two modes of construal is that in one case the genitive
NP is construed as a term in a relationship, whereas in
the other case it is construed as a modifier. Now,
within the set of assumptions we have adopted concerning

the nature of thematic relationships, for an NP to be a
term in a relationship means for the external referential

role of that NP to be coindexed with a thematic index
available in the argument structure of a predicate. In

the case of genitive constructions, the type of thematic
relationship holding between the two terms is
indeterminate. It is mostly pragmatically inferred. I
would like, therefore, to propose the idea that nouns, or

at least those nouns which are capable of standing as
9
heads in an identificatory relation , have as part of
their argument structure, in addition to the usual
external theta-role R, an internal thematic index

unspecified for thematic content. The argument structure
of a noun would, thus, look as follows.
64- (Ri , j)
Thus, at the level of syntactic representation, for a

genitive NP to be incorporated into the overall structure
as a term in a thematic relationship, means for that NP
to be coindexed with the internal argument in the
99
argument structure of the head noun. The identificatory
relationship in a bound genitive such as 65a may be

represented as in 65b.
65- a) (H) beyt ha-ikar
house the farmer
"the house of the farmer"

b) House [The farmer]
Ri ,j Rj
Note that the relationship between the two elements in 65

is not symmetric. It seems that the best way to
characterize this relationship is as one between a head
noun X and a genitive NP dominated by the maximal

projection of X (at least at some point in the
derivation) . This is restrictive enough to establish the
asymmetry between the two elements and would in fact be
necessary for the case of genitive adjuncts (Sel-
phrases) . Semantically, the mode of combination
associated with the identificatory construal of genitive

constructions may be represented by the following
formula.
66- Xy X p THEx [P (x) & Q(y,x)]
Applied to 65b this gives the following representation
100
where Q is most naturally construed as the possession
relation.
67- THEx [House (x) & Q(The Farmer, x)]
The operation associating two elements in the

identificatory relationship, as in 65b, is not obligatory
and if it does not take place then the classif icatory
mode of relationship is construed. In a classif icatory
structure only the head noun is ’’strictly” referential
while the modifying noun is not. This point may be

easily appreciated when one considers definite
classif icatory structures such as the following.
68- a) (H) saxkan ha-hagana

player the defence
"The defence player"
b) (H) metos ha-krav
aircraft the battle

"The fighter aircraft"
Nothing in the interpretation of 68a involves reference

to a contextually unique "defence" and, likewise, nothing
in the interpretation of 68b involves reference to a

unique "battle". Rather, the NP’s in 68 refer to a
unique player of a certain kind and a unique aircraft of
101
a certain type respectively. Thus, an NP which has not
been incorporated as a term in an identificatory relation
may still be construed, in the manner of an adjective, as
an expression of type <e,t>. In a case where two terms
in a bound genitive construction are not construed in the
identificatory relationship the thematic index in the

argument structure of the head noun remains without a
referential value. Therefore, it is free for
coindexation (or rather, linking) with another NP. Such

a situation arises when a head noun is related to two
genitive NP's, one which is related to it as a bound

genitive and another in a Sei phrase.
69- (H) beyt ha-ikar Sei Dina
house the farmer Sei Dina
In this example, unless Dina and "the farmer" are
interpreted "genitively" ("the farmer of Dina"), the only
other interpretation available would be the one in which

"the farmer" is a "classif icator" (that is, "a farmer's
type of house") and Dina is a possessor .
102
2.3 Sel-Phrases and Double Genitives
Here we will start first by considering cases of
what was referred to earlier as the Free Genitive. Since
constructions involving Sei. or some equivalent of i t do
not occur in Arabic ( a t least not in Standard Arabic) ,
the discussion in this section will be limited to Hebrew.
70- (H) ha-mazkir Sei ha-miflaga
the secretary Sei the party
My assumption with respect to Sel-phrases is that they
are adjuncts. Here I am using the notion of adjunct in
the sense of Lebeaux (1988) . In that approach adjuncts
may be absent from the tree structure at the level of D-
structure and be adjoined to it at a later stage in the
derivation. As will be seen, this particular assumption
will prove essential for the analysis of nominalization
constructions .
Sei phrases show the "anti-reconstruction" effects
first observed by Van-Riemsdi jk and Williams (1981) and
analyzed by Lebeaux as cases of adjunction. Anti-
reconstruction effects may be exemplified by the contrast
between the grammatical pair 71a, b and the ungrammatical
pair 72a, b.
103
71- a) (H) eyze sefer Se Dina kanta hi haxi
which book that Dina bought she the most
ohevet
likes
"Which book that Dina bought does she like the most"
b) (H) eyze sefer Se Dan katav ata XoSev Se
which book that Dan wrote you think that

hu haxi ohev
he the most likes
"which book that John wrote do you think he likes
the most"
72- *a) (H) eyze sipur Se Dan haya be-amerika hu
which story that Dan was in america he

himci be-acmo
invented in himself
* "Which story that Dan was in america did he

invent himself"
*b( (H) eyze teana Se Dan bicea et ha-peSa
Which claim that Dan committed OM the crime
ata XoSev Se hu asuy lehkxiS
you think that he may deny

*"Which claim that Dan committed the crime do you
think that he may deny"
The contrast between 71 and 72 derives, under Lebeaux's
approach, from the fact that 71a, b involve relative
104
clauses which are taken to be adjuncts whereas 72
involves argument clauses. The contrast between 71 and
72 may be explained if one assumes that argument clauses
must be present at D-structure (NP-structure, in our

terms) whereas adjunct clauses may be adjoined to the
tree at a later stage in the derivation, crucially in
this case, after Wh-movement has taken place. Given the
additional assumption that principle C of the binding
theory applies at all levels of representation, including
D-structure, the ungrammaticality of 72 may be explained
as a violation of principle C under the intended
interpretation in which Dan is coreferential with hu
(he) . This is because the clause containing Dan must be

present at D-structure in a position in which Dan is c-
commanded by hu (he) , that is, the original complement
position. This effect is circumvented in 71a, b since in
these examples there is no level of representation in
which the relative clause occupies a position which would

give rise to a principle C violation.
We may now note that the same effect observed for
relative clauses in 71 and 72 may also be observed for

Sei phrases in the examples below, thus, providing
support for their treatment as adjuncts.
105
73- a ) ( H ) eyze sefer Sei Dina a t a xoSev Se hi haxi
which book Sei Dina you think that she most
ohevet
likes
"Which of Dina ' s books do you think she likes i the
most"
b ) ( H ) eyze sefer Sel Dan ata XoSev Se hu
which book Sel Dan you think that he
katav haxi maher
wrote the most fast
"Which of Dan's books do you think he wrote the
fastest"
Thus 70 will h a v e the following structure in which a Sel-

6
phrase is adjoined a t the N * -level .
106
74- DP
/ \
D NP
I / \
the N*
/ \
N* Sel-Ph
I / \
N Sel NP
I / \
secretary the party
An NP/DP in a Sel-Phrase may be integrated into the
overall meaning of the NP containing it just like an

NP/DP in a bound genitive, that is, it may be related to
the head noun in either of the two main modes of noun-to-

noun construal, identificatory and classificatory ,
observed in the previous section. The pairs of free and

bound genitives below are identical in meaning.
75- a) (H) ha-bayit Sei Dina (identif icatory)

the house Sei Dina
"Dina's house"
b) (H) beyt Dina (identificatory )
house Dina
"Dina's house"
107
76- a) (H) bayit Sel ec (classificatory)
house Sel wood

"a house made of wood"
b) (H) beyt ec (classificatory)
house wood
"a house made of wood"
It may, therefore, be concluded that Sei does not make

«
any semantic contribution in such cases and should be
analyzed as a "dummy" Case marker. This, however, should

not be interpreted as a claim that Sei (or, rather, a
phonetically identical variant of it) may not sometimes
function as a "semantic" preposition. It

is plausible to assume that Sel-phrases in predicate
position are of this kind.
77- (H) ha-bayit ha-ze Sei Dina
the house the this Sei Dina
"This house is of Dina"
("this house belongs to Dina")
In this example, it is not the external referential
theta-role of Dina which is assigned to the subject "this

house", but rather the external theta-role of the PP
"Sel-Dina", establishing a relationship of possession
between Dina and "the house".
108
As was mentioned earlier, in cases in which both a
bound genitive NP and a "Sel-NP" are present, only one of
the two may stand in an identif icatory relation to the
head, and, therefore, the classif icatory interpretation
must be imposed on the other. This is the cape in the
following example where the relations are indicatted by
the arrows.
classif icatory
78- (H) beyt ha-mora Sei Dina
identificatory
house the teacher Sei Dina
"Dina's teacher-house"
Of-course, it is possible for Sei Dina to be related as a
possessor argument of ha-mora (the teacher). Thiis is ,in
fact, the more likely construal of this example but is
not the relevant one a t this point. Cases such as 78
seem to be subject to a very strong semantic restriction
which is the following. The NP in the Sel-phpase must be
construed as an "identif icator" rather than as a
classificator (modifier) . More generally, thp |
restriction seems to be that in a classif icatory Irelation
the modified element may not be complex, that is, in our
case, consisting of more than a single noun. This is why
cases such as the following are unacceptable.
109
*79- (H) beyt Dina Sel ha-mora
house Dina Sel the teacher
In trying to assign 79 an interpretation, one way to go

would be to construe ha-mora (the teacher) as the
possessor argument of Dina, but this is unlikely for the
same reason that 80 is unlikely.
*80- (H) Dina Sei ha-mora
Dina Sei (of) the teacher
Another way to go would be to construe the teacher as a

possessor argument of the head bayit (house) and Dina as
a classif icator . But this is an unlikely interpretation
for a proper name. The remaining option is to construe
Dina as a possessor and "Sel-the teacher" as a
classificator , but this, again, seems to be unfeasible by
virtue of the above mentioned restriction, namely, that
in a classificatory relation the modified element may not
be complex. Note that this restriction applies even if

the "would be" classif icator is indefinite, indefinites
being most likely construed as classificators .
*81- (H) beyt Dina Sei mora

house Dina Sei (a) teacher
110
Thus, to repeat, our explanation for the ungrammaticality
of 79 is the following: The only construal available for
the NP in the Sel-phrase is as an "identificator"

(possessor) with respect to the head noun. However, that
would force a classif icatory construal on Dina, which is
impossible, Dina being a proper name. In our terms, the

internal argument position in the argument structure of
the head noun is "taken" by Dina and there is no way to
assign an interpretation to Sel-the teacher.

Note now that exactly the same state of affairs as
in 79 holds in the following example where a pronoun
replaces Dina. Here, again, a pronoun could never be
construed as a classif icator and must, therefore, be
construed as standing in the identificatory relation to

the head.
82- (H) beyt -a Sei ha-mora

house-her Sei the teacher
"The teacher's house"
82 exemplifies the much discussed phenomenon of clitic-

doubling. Thus, the teacher and the pronoun which is
attached as a clitic to the head bayit (house) , must be
interpreted as coref erential. Obviously, that this is
the only interpretation that could ever be assigned to 82
is now clear, given our discussion of 79. The question
111
is, rather, what is the mechanism that makes it possible
for the Sel-phrase to be assigned an interpretation, that
is, to be "licensed**, given that the internal argument of
the head is "taken* 1 and, second, what is it that
distinguishes languages which have clitic-doubling from
those which don't.
We might, in fact, start from the second question by
suggesting that this variation among languages involves
the notion of "levels of representation". Here the idea
would be that, in a language that does not allow for
clitic doubling, the counterpart of a Sel-phrase in that
language must be "licensed" in the syntax.
Generally, an element is licensed if it can be
incorporated into a well formed representation given
general principles and mechanisms available in the
grammar. An overall conception of the grammar in which a
distinction between levels is assumed, makes i t possible
to provide explanations by attributing the applicability
of certain principles or the application of certain
mechanisms to particular levels of representation. Thus,
as was demonstrated in the discussion above, certain
observations about the syntax of adjuncts may be
accounted for by the assumption that adjunction
operations are free to apply at a certain "later" level
of representation rather than a t another "earlier" one.
The explanation proposed here for the phenomenon of
112
clitic doubling and the linguistic variation observed
with respect to this phenomenon relies on the distinction
between, on the one hand, the levels of representation in
which strictly syntactic processes are assumed to take

place (NP-Structure, S-Structure) and the level of
representation which provides the input to semantic

interpretation, namely LF, on the other. The claim made
here is that it can be shown that the mechanism which
makes clitic doubling possible is basically semantic in
nature. The idea would be that a language in which
clitic doubling is not attested is a language in which

the counterpart of a Sel-phrase must be licensed at the
level of S-structure or NP-structure, whereas a language

in which clitic doubling is attested is one in which the
counterpart of a Sel-phrase may be licensed at LF7 .
The question is then what does this mode of

licensing consist of. Here we may adopt an approach
similar to the one generally suggested for the treatment
of resumptive pronouns. In particular, I will follow

Chomsky's (1982) proposal with respect to resumptive
pronouns according to which resumptive pronouns are
construed as bound variables by a mechanism of
coindexation applying at the level of LF. Such a
mechanism, according to Chomsky, must be restricted to

the level of LF since NP's base generated in an A'
position do not have an index at S-structure.
113
Allowing for free indexing at S-structure would lead to
base generation of violations of island constraints.
Following the spirit of this approach, we will assume an
LF mechanism by which an NP in a Sel-phrase is coindexed
with a pronoun within the NP immediately dominating it.
Being coindexed with a pronoun, a Sel-NP is properly

licensed as part of a well formed representation which
may serve as input to a process of semantic
interpretation to be described soon.
Following Lebeaux's (1988) treatment of relative
clauses, we may now introduce the following constraint
applying in English.
83- An Of-Phrase must be licensed at S-structure.
Given these assumptions, we may now propose an

explanation for the ungrammaticality of 84 below along
the following lines.
*84- her house of Mary
In this example, the only way for Mary in of Mary to be

licensed would be for it to have its referential theta-
role coindexed with the thematic index in the argument
structure of the head noun (a classificatory construal of
Mary being impossible. Compare, however, "her house of
114
snow", "my friend of many years"). However, the
classif icatory construal being impossible for pronouns,
that thematic index is taken by the pronoun for which
too this is the only available mode of licensing. Thus,
since an of-phrase in English must be licensed in S-
structure, prior to the availability of any LF
mechanisms, 84 is ruled out.
Hebrew, on the other hand, allows for a Sei-Phrase
to be licensed at a later stage, that is, at the level of
LF. Thus, a language like Hebrew can make use of the
kinds of mechanisms that are available at this level of

representation, in particular, the mechanism of
coindexing applying between a genitive Sel NP and a
possessive pronoun. The resulting LF representation is
then subject to generally available mechanisms of
interpretation.
One such typical semantic mechanism is lambda

abstraction, an operation which makes it possible to
transform a semantic expression of any kind into a
predicate and in this way to make it possible for a
certain element to be incorporated into the overall
interpretation as an argument to that predicate. An

obvious condition for the application of this mechanism
is the availability of a variable in the semantic
translation of the expression over which the operation of
lambda abstraction applies, that is, the availability of
115
an element inside that expression which is "translatable"
into a variable of the required type (the required type
being <e>, in this case). Thus, at the level of semantic
representation a clitic doubling configuration such as 82

may be represented roughly as follows.
85 - 7)x [ x's house ] (Dina)
In this expression Dina is interpreted (and semantically

"licensed") as the argument of a lambda-expression formed
by "lambda abstracting" over "her house". Thus, the

pronoun "her" enters the semantic interpretation as a
bound variable. Obviously, such an interpretation is

only available for a pronoun having the status of an
argument (in the sense of the identification/
classification distinction suggested in this chapter) ,
from which the impossibility of 81 is derived.

A rather striking confirmation for this analysis,
that is, the view that it is the logical representation

of the structures in question and, in particular, the
status of a pronoun in such a representation which is
responsible for the acceptability of clitic doubling

constructions comes from examples such as the following.
116
86- a) (H) bayit Sel ec
House Sel wood
"a house of wood" (a house made of wood)
b) (H) beyt ec
house wood
"a house of wood" (a house made of wood)
c) (H) beyt -o
house-his
"his house"
*d) (H) beyt -o Sei ha-ec
house-his Sei the-wood
In both 86 a and b "ec" (wood) is interpreted as a
classificator . Semantically, it is a predicate, that is,

an expression of type <e,t>. The mode of combination of
a construction such as 86c may be represented as follows.
87- Ap y THEx [P(x) & Q(x,y)]
The resulting representation after the filling in of the
specific lexical contents for the variables in 87 is as

follows, where the relation Q is most naturally construed
as the possession relation.
88- THEx [ Q(x,he) & House (x) ]
117
The pronoun in 88 is a term in a relation involving two
entities. It is semantically of type <e>. It is an

expression such as 88 which would be the input to the
semantic mechanism involved in clitic doubling. The
point at which the incorporation of the adjunct starts
may be roughly illustrated as in the following
representation (where i is a variable binding the

pronoun) which would be the logical structure of 86d.
89- 2\i [ THEx [ Q (x,hei ) & House (x) ]] (wood)
Note now that there is no way in which "ec" (wood) in 86d
could be incorporated into the overall interpretation as

represented in the above formula since, being an
expression of type <e,t>, it is of the wrong type to be
construed in this way with the bound variable in 89.
Therefore, 89 is ruled out as a case of type mismatch.
Thus, the possibility or impossibility of a clitic
doubling construction seems to be sensitive to the
semantic relation holding between the NP in the Sel-

phrase and the rest of the expression, the relevant
distinction being that between identification and

classification, a distinction which is basically semantic
in nature.
Further confirmation for this approach to clitic
doubling is provided when one observes the interaction of
118
such constructions with anaphoric relations. Consider
the following pairs.
90- a) (H) xaver Sei acmo
friend Sei himself

"a friend of himself"
*b) (H> xaver -o Sei acmo
friend-his Sei himself
91- a) (H) tmuna Sei acmo
picture Sei himself

"a picture of himself"
*b) (H) tmunat-o Sei acmo
picture-his Sei acmo
As the contrast between the a and the b examples above
makes clear, the reason for the ungrammaticality of the b

examples couldn't be that the thematic, or argument,
structure of the nouns xaver (friend) or tmuna (picture)
is such that they cannot provide the necessary antecedent
for the anaphor. Some kind of antecedent must be assumed
to be available in the a examples and therefore also in

the b examples. Under the present account of clitic
doubling, the anaphors in the b examples may only be
incorporated into a well formed representation at the

level of LF as A* binders of a bound variable.
Therefore, principle A of the binding theory to which
119
anaphors are subject is violated at all syntactic levels
of representation since at none of these levels is the

anaphor integrated as an argument into the overall
structure.
Thus, clitics in this approach are syntactically and

semantically just like any other pronoun and no special
stipulations need to be made. In particular, the
"clitichood" of the clitics in the constructions under
discussion is irrelevant to any of the syntactic or

semantic issues that seem to be involved and seems to be
relevant only to issues of morphology. Linguistic

variation with respect to clitic doubling is in this
approach reduced to a minimum, namely, the distinction

between licensing at different levels of representation,
a distinction of which the necessity has been
demonstrated elsewhere in any case. To conclude,

the structure of a clitic-doubling construction such as
in 82 would be the following.
120
92- DP
/ \
D'
/ \
D< NP
/ \ / \
Ni D DP N'
1 1 1 / \
house POSS NP N* Sei the teacher
N' N
1 1
N ei
her
As the above configuration demonstrates and given the

discussion so far, a clitic doubling construction is one
in which a pronoun stands as a second term in a bound
genitive construction with the head noun and with a Sel-
phrase adjoined to that construction. Given the thematic
structure of nouns as discussed above, the NP in the Sel-
phrase must be "parasitic" on the pronoun for its
licensing and interpretation. We may therefore say that

the so called double genitive is a construction which
involves the cooccurrence of the bound and the free
121
genitive. Thus, there are, in fact, only two basic types
of genitive constructions, the bound and the free
genitive, and a derived, or combined, one, namely, the
double genitive.
122
Footnotes to Chapter II
1
Note that the discussion at this stage is neutral with
respect to the issue concerning a DP analysis of NP's as opposed
to the more traditional NP structure. The present arguments hold
regardless of this issue since what they show is that some
syntactic position to the left of the head of an NP must be
assumed.
2
Needless to say, Borer's analysis is not supplemented by any
explicit theory about percolation. Note that the percolation
mechanism as illustrated in 39 is a clear deviation from the
standard assumption that features, or a t least those features
which d o , percolate up a projection path rather than from a
complement of a head to a node which is a projection of that
head. To the best of my knowledge, there is no explicitly
elaborated theory of percolation within the GB framework. I
suppose that an eventual theory of percolation will have the form
of a set of principles which specify feature cooccurrence
restrictions in such a way as to regulate the distribution of
information in a tree structure, that is, a t least, the type of
information which is generally assumed to be encoded by means of
the device of features (such an explicit and detailed theory has
been developed within the GPSG framework, cf. Gazdar e t al.
(1985)). With respect to the present issue, one might only
wonder a t this point whether such a theory of percolation would
allow for the percolation of definiteness across verbal nodes as
well as nominal nodes. Thus, Borer (forthcoming) assumes the
following configuration to underlie an Action Nominalization
construction.
(i) NPi [+Def]

/ \
SPEC N'i [+Def]
I \
Ni VP [+Def]
/ \
NPa [+Def] VP
I / \
Na [+Def] V
/ \
V NPa
In the surface structure derived from the configuration above Ni

and NPa form a bound genitive. This is assumed by Borer to
involve movement of N 2 to Nt . Now, since all the observations
which apply to bound genitives elsewhere apply also to bound
genitives in Nominalization constructions, I suppose that in this
case too the percolation mechanism of Borer (1988) must be
assumed to apply, allowing for the feature [+Def] to percolate
123
from NP2 to NPi . This, as shown in <i), will create a sequence
of nodes along the path from NPa to NPi , including VP, that are
marked as [+Def ] . This seems somewhat problematic since, in
general, definiteness is a distinction which is relevant to the
nominal rather than to the verbal domain. But, obviously,
nothing can be concluded one way or the other in the absence of
any explicit theory of percolation.
3
In Ritter's (1986) analysis the underlying structure of a
bound genitive as in (i) would be the configuration in (ii).
(i) - a) (H) beyt ha-more

house the teacher
(ii) - DP
/ \
D NP
[+/-POSS] / \
[+/-Def] (DP) N'
I I
house N
I
teacher
In (ii) the head D is specified for the feature [+/-POSS] which,

if specified as [+] , can assign genitive Case, and for the
feature [+/-Def ] which, if specified as [+] , is realized as a
clitic. Moreover, if [Def] is specified as [+] then the DP is
interpreted as definite. (i) is derived by a cliticization of
[+def] to the DP in the Spec position of its NP complement,
followed by a head movement of the head noun teacher to D. The
movement of the head noun is assumed to be obligatory in order
for the feature [POSS] to be morphologically supported. This
analysis encounters some serious difficulties which relate mainly
to the issue of definiteness both at the level of surface
realization as well as on the level of interpretation.
As to surface realization, looking at the configuration in
(ii), we see that the specifier position of the complement NP is
occupied by DP. However, if this DP happens to be definite then
the cliticization of the higher [+Def] to it would result in a
sequence of two definite articles, clearly, an undesirable
result. Ritter (personal communication) suggests that this may
be avoided by assuming a filter which rules out such sequences,
but this would be an ad hoc move to make.
A more serious problem arises with respect to
interpretation. It is assumed by Ritter (1986) that a bound
genitive is interpreted as definite if its head D is specified as
[+Def] and as indefinite otherwise. However, the surface
realization of the definite article in an example like (i) could
124
originate either from the head D or from the DP in the specifier
of the complement NP. That is, (i) is predicted to be ambiguous
between the definite and the indefinite interpretations. This
prediction is obviously wrong.
« Given the theory of government adopted here (cf. Chapter 1)

and in particular the notion of Barrierhood employed in this
theory (see 13 in Chapter 1) , POSS in 43c certainly governs the
lower DP. I am, thus, deviating from the assumption made in
Hazout (1988) (and adopted by Shlonsky (1990)) that NP blocks
government in such a configuration.
0
Although the great majority of nouns manifest a syntactic
behavior which is in conformity with the proposal represented in
64, one does find some variation in this domain. Thus, pronouns
have no internal thematic index in their lexical representation
and the same is true (except for some marginal cases) of proper
names. The examples in (i) below are obviously possible and
those in (ii) are not.
(i) a) his house

b) Mary's house
(ii) *a) he of the house
*b) Mary of the house
6
The evidence above and the similarities with relative
clauses, as opposed to sentential complements, clearly support
this treatment of Sel-phrases. Note that there would arise some
difficulties for any treatment which takes Sel-phrases to be
complements. Under such a treatment Sel-phrases are assigned
genitive Case, of which Sei is the surface marking, by virtue of
being governed by a head noun. An analysis of this kind must
provide some special stipulations about the distribution of Sei.
I will explain what this means by looking at Shlonsky *s (1990)
analysis of bound and free genitives.
Shlonsky (1990) assumes that the free genitive involves a
complement DP whereas a bound genitive involves a DP in specifier
position. Thus, the free genitive in (i)a will have the
underlying structure in (i)b (ignoring irrelevant details).
(i) a) (H) ha-harisa Sei ha-maxanot

the destruction Sei the camps
"the destruction of the camps"
125
b) DPi
/ \
D NP
I / \
ha N*
/ \
N DP 2
I /
harisa D NP
I I
ha max a not
(i) b is taken by Shlonsky to be a configuration in which DP2 is

assigned genitive Case by the governing N (or its trace, if N
moves to D) . Genitive Case is realized by means of the
preposition Sei. On the other hand, the bound genitive in (ii)a
below is derived by an adjunction of N to NP as illustrated in
(ii) b.
(ii) a) (H) tmunat ha-yalda

picture the girl
"the picture of the girl"
b) DPi
/ \
NP
/ \
Ni NP
1 / \
tmuna DPz N'
/ \ 1
D NP N
1 1 1
ha N ei
1
yaldt1
The adjunction operation (which is adopted, with some

modifications, from Hazout (1988)) in (ii)b is motivated by the
assumption that the head N does not govern its specifier prior to
the application of movement and therefore must adjoin to NP, a
position from which it governs DPa and assigns Case to it. In
this case no "Sel-insertion" takes place.
One wonders at this point what it is about this analysis
that distinguishes between the two cases in such a way that makes
it possible to determine when the preposition Sei occurs and when
it doesn't. If the crucial factor is government of DP by N then
the prediction made by this analysis is that Sei should always
(or never) be "inserted". This is because both configurations
are configurations of government of DP by N with the same two
elements participating, namely, N and DP. After all, this is
the motivation for the adjunction operation, that is, to move N
126
to a governing position with respect to DP. Thus, to distinguish
between the two cases, the distinction must be made literally in
terms of complement and specifier.
Considering this possibility , note first that the
distinction between specifier and complement couldn't be
understood as a thematic distinction since one finds common
minimal pairs such as the following in which the two terms in the
construction seem to stand to each other in the same thematic
relation.
(iii) a) (H) beyt ha-mora

house the teacher
"the house of the teacher”
b) (H) ha-bayit Sei ha-mora
the house Sei the teacher
(iv) a) (H) harisat ha-ir
b) (H) ha-harisa Sei ha-ir
the destruction Sei the city
So the distinction between specifier and complement must be

configurational. Configurationally, a complement is a sister to
the head whereas a specifier, at least in a base generated
structure, is a sister to X* . In a structure resulting from
adjunction such as (ii)b the specifier is governed by the head
but is not a sister to it.
Now, the problem with this analysis is that i t makes one
clear prediction with respect to cases for which i t is possible
to show that the DP which is the second member in the genitive
relation couldn't possibly be a complement ( a sister to the head)
at any derivational stage. The prediction is that in such cases
only a bound genitive, rather than a free genitive, should be
possible. Cases of this sort are fairly common. The following
are typical examples.
( v ) a) (H) harisat ha-cava et ha-maxanot

destruction the army OM the camps
"the army's destruction of the camps"
b) (H) ha-harisa Sel ha-cava et ha-maxanot
the destruction Sei the army OM the camps
"the army's destruction of the camps”
The examples in ( v ) are instances of what we refer to here as the

Action Nominalization construction. Chapter 3 is dedicated to a
detailed discussion of this construction. One of the main
conclusions of that discussion is that the two NP's in such a
construction must be asymmetrically related to each other and
must also be differently situated with respect to the head.
This is necessary in order to establish an unambiguous mapping of
127
Case markings to the two HP's (Genitive, Accusative) and also in
order to establish an asymmetry in anaphoric relationships
between them (see in particular the discussion in sections
3. 2. 2.1 and 3. 2. 2. 2 as well as footnote 4 to Chapter 3). It
follows from these considerations that the first, the genitive,
NP (ha-cava in these examples) couldn't be a complement at the
underlying structure of such examples. It couldn't also be a
complement in any derived structure since there is no movement
operation which would have as output a configuration in which the
"Spec-NP" is a sister to the moved head. The prediction is then
that in such cases Sei couldn't occur, that is, that an example
like (v)b should be impossible. But this is clearly false.
On the other hand, when it comes to cases such as (i)a above
in which the genitive NP is arguably, on the basis of thematic
considerations, a complement, the prediction is that only the
free genitive should occur. This prediction, again, is false.
(vi) harisat ha-maxanot

destruction the camps
"the destruction of the camps”
It is therefore clear that the specifier/complement distinction

has no relevance to the respective distribution of free and bound
genitives and, therefore, some other way must be found to
distinguish between the two construction types.
7
This type of explanation is suggested by Lebeaux (1988) for
the variation between languages which have correlatives and
languages that don't. A language with correlatives is, to use
Lebeaux's terminology, a language in which "a relative clause
linker need not be saturated at S-structure” . Semantic and
syntactic treatments of correlatives along the lines suggested
here for clitic doubling may be found in Bach and Cooper (1978)
and Srivasta (1988).
It also seems reasonable to assume that the same type of
mechanism is involved in the licensing of left-dislocated
elements as in i below. The well formed semantic representation
that would make i possible would be the one in ii.
i) John, I like him

ii) x [ I like x] (John)
128
CHAPTER III
THE SYNTAX OF ACTION NOMINALIZATIONS
3.1 Introduction
In this chapter I will propose an analysis of the
Action Nominalisation construction in Hebrew and Arabic.
From the point of view of the theoretical questions
raised by this construction a classification may be made
into two main types, namely, an Action Nominalisation
construction in which the full argument structure of an
"underlying" verb is lexically present (la) and one in
which the subject, or rather, the external argument of
the verb in question i s missing (lb). This view of the
facts will also dictate the overall organisation of this
chapter. Section 3.2 will be concerned with active
Action Nominalisations and section 3.3 with their
subjectless counterparts.
1- a ) (H) axilat Dan et ha-tapuax
eating Dan OM the apple
"Dan's eating the apple"
129
b) (A) aklu ttufaaHat-i
eating the apple-Gen
"the eating of the apple"
This section, however, will be devoted to a presentation
of the main facts for which the analysis described in the
following sections attempts to provide an account. The
main theoretical interest of Action Nominalizations, in
my view, derives from their "mixed" verbo-nominal nature.
This is to say that they manifest certain syntactic
characteristics some of which are normally considered
typical of both the external and the internal syntax of
NP's and some of which are normally taken to involve a
constituent headed by a verb. Our presentation will
start with facts concerning the first type of properties,
that is, the nominal.
3.1.1 Nominal Properties of Action Nominalizations
First among the nominal characteristics of Action
Nominalizations is their distribution. Action
Nominalization constructions have the regular
distribution of NP's. Thus, one finds Action
Nominalizations as subjects (2) , objects (3) and objects
of prepositions (4) .
130
2- a ) (H) harisat ha-cava et ha-ir hayta
destruction the army OM the city was

axzarit
cruel
"The army’s destruction of the city was cruel"
b) (A) dahaSa-ni iRtiyaal-u zaid-en
surprised me assassination-Nom zaid-Gen
al waziir -a
the minister-Acc
"Muhammeds assassination of the minisrter
surprised me”
3- a ) (H) ha-bamay hisrit et harisat ha-cava et
the director filmed OM destruction the army OM
ha-ir
the city
"The director filmed the army’s destruction of
the city"
b)(A) ra?ayt-u iRtiyaal-a zaid-en
saw -I assassination-Acc zaid-Gen
al waziir -a
the minister-Acc
"I saw zaid's assassination of the minister"
4- a ) (H) dibar-nu al harisat ha-cava et ha-ir
spoke-we about destruction the army OM the city
"We spoke about the army's destruction of the city”
131
b) (A) fuji?t -u bi- Rtiyaal -i zaid-en
surprised -I with-assassination-Dat zaid-Gen
al-waziir -a
the minister-Acc
"I was surprised by Zaid's assassination of the
minister"
Second, as can be seen in the arabic examples in 2-4, the
head of an Action Nominalization construction is a
typical noun in that it shows Case marking depending on
the position it occupies. Thus, the noun iRtiyaal
(assassination) is marked for Nominative, Accusative and
Dative in 2, 3 and 4 respectively.
A third type of facts characteristic of the internal
syntax of Action Nominalizations is the assignment of
genitive Case to an NP ("logical" Subject or Object) by
the head noun. The examples in 2-3 all have a subject
(agent) standing in the bound genitive construction to
the head. In modern Hebrew the head and the subject may
also appear in the free and double genitive
constructions as follows.
5- a) (H) Free Genitive:
ha-harisa Sel ha-cava et ha-ir
the destruction Sei the army OH the city
"The army's destruction of the city"
132
b) (H) Double Genitive:
harisat -o Sel ha-cava et ha-ir
destruction-his Sel the army OM the city
"The army's destruction of the city"
This "opening" of the structure in the free and double
genitives makes it possible for an adjective modifying
the head noun to intervene between the head and its
subject, again a typical nominal property.
7- a ) (H) ha-harisa ha-maftia Sel ha-cava et
the destruction the surprising Sei the army OM
ha-ir
the city
"The surprising destruction of the city by the army"
b) (H) harisat-o ha-maftia Sel ha-cava et
destruction-his the surprising Sei the army OM
ha-ir
the city
"The surprising destruction of the city by the army"
133
3.1.2 Verbal Properties of Action Nominalizations
First among the verbal properties of Action
Nominalizations is the occurrence of Accusative Case
marking on a direct object. Accusative Case marking is
typical, and is normally taken to involve the government
of an NP by a verb (see also Appendix to this chapter).
All the examples in 2-4 show the marking of accusative
Case on a constituent which would correspond to what is
standardly designated as a direct object in a sentence.
Thus, if one wishes to argue that the constructions in
question do not involve the syntactic presence of a verb
one would still have to account for the striking
similarity which holds between the way a verb in a
sentence is related to its arguments on the one hand and
the way a head noun is related to the corresponding
elements in an Action Nominalization construction. This
correlation holds, in fact, with respect to other
complement types that a verb might have. In general, a
nominalized head manifests the same argument structure as
its corresponding verb. Thus, in the following examples
an arbitrary requirement by a verb for a certain type of,
semantically empty, preposition is shown to be observed
by the corresponding nominal form.
134
9- a ) ( H ) Dan viter al ha-misra
Dan "gave up" on the position
"Dan gave up the position"
b) ( H ) vitur -o Sel Dan al ha-misra
"giving up" -his Sel Dan on the position
"Dan’s giving up of the position"
10- a ) (H) Dan histalek min ha-misra
Dan left from the position
"Dan resigned the position"
b) (H) histalkut-o Sei Dan min ha-misra
leaving -his Sei Dan from the position
"Dan’s resignation from the position”
In. Arabic, Action Nominalisations with nominal heads
that are related to double object verbs take the same
number of arguments and show the same dative alternation
characteristic of the corresponding verbs.
11- a ) (A) ?a9ta zaidun al-walad-a ttufaaHat-a
gave zaid-Nom the boy -Acc the apple-Acc
"Zaid gave the boy the apple"
b) ( A ) ?i9ta zaid-en al-walad-a ttufaaHat-a
giving zaid-Gen the boy -Acc the apple-Acc
"Zaid's giving the boy the apple"
135
12- a) ( A ) ?a9ta Zaid ttufaaHat-a li-1- walad-i
gave zaid the apple-Acc to the boy -Dat
"zaid gave the apple to the boy"
b) ( A ) ?i9ta zaid-en ttufaaHat-a li-1- walad-i
giving zaid-Gen the apple-Acc to the boy-Dat
"Zaid's giving of the apple to the boy"
The occurrence of adverbs is also characteristic of
verbal constituents. Again, Action Nominalization
constructions in both Hebrew and Arabic allow freely for
the occurrence of adverbs.
13- (A) ?akl zaid-en ttufaaHat-a bi-sur9a
eating zaid-Gen the apple-Acc quickly
"Zaid's eating of the apple quickly"
14- (H) harisat ha-cava et ha-kfar be-axzariyut
destruction the army OM the village cruelly
"The army's destruction of the village cruelly"
It seems that the occurrence of adverbs in these
constructions is not licensed simply by the action or
process type of interpretation associated with these
constructions, but is, rather, syntactically
conditioned. This may be evidenced by the fact that,
although the nouns in question may appear all by
themselves (15) , the occurrence of an adverb is not
136
allowed unless it is accompanied by the rest of the
complement system required by the corresponding verb.
15- a ) (H) ha-harisa nimSexa Saatayim
the destruction lasted two hours
b) (A) kaana 1- iRtiyaal-u waHSiyy-an
was the assassination-Nom barbaric-Acc
"the assassination was barbaric"
*16- *a) (H) ha-harisa be-axzariyut
the destruction cruelly
*b) (A) al-iRtiyaal bi-sur9a
the assassination quickly
These are then the main nominal and verbal
characteristics of Action Nominalization constructions,
obviously, the main question raised by this interesting
mixture of properties is what assumptions must be made
about its underlying syntactic structure and what
properties need to be attributed to the lexical items
involved in order to derive its surface manifestation.
This way of presenting the question already indicates the
two main strategies that come to mind. The first would
be to postulate a VP or sentence-like structure from
which the surface forms would be derived by the
application of certain syntactic operations. This
strategy would necessitate no (or almost no) additional
137
assumptions about the properties of the individual items
(verbs, nouns etc. ) involved. An alternative strategy
would make minimal assumptions with respect to syntactic
structure and derivation and would attempt to explain
some of the facts by attributing certain, exceptional,
properties to some lexical items, such as, for example,
the ability to assign Accusative Case attributed to a
certain class of nouns. We may refer to these two
approaches as the syntactic and the lexicalist approach
respectively. In this work the syntactic approach will be
argued for.
3.1 .3 Binding and Control
Less directly classifiable as either
characteristically nominal or verbal are the facts
concerning binding and control relations in Action
Nominalization constructions. To start with binding, the
relationship between an anaphor and its antecedent in an
Action Nominalization construction shows the same
asymmetries that are observed in a sentence between the
subject and the object position. Thus, a subject NP can
serve as the antecedent of an anaphor in the object
position but not the other way around.
138
17- a) (H) Dan ohev et acmo
Dan loves OM himself
"Dan loves himself"
*b) (H) acmo ohev et Dan
himself loves OM Dan
The contrast between the two examples in the following
pair is the same as the one above.
18- a) (H) ahavat Dan et acmo
love Dan OM himself
"Dan’s love of himself”
*b) (H) ahavat acmo et Dan
love himself OM Dan
As to control phenomena in Arabic, an Action
Nominalization construction may figure as the complement
of a control verb just like an infinitival clause in
English or in Hebrew. Obviously, in this case the
subject of the deverbal form is not overtly realized but
is understood as "controlled" by the governing verb.
19- (A) Haawala Zaid-un ?akl-a ttufaaHat-i
tried Zaid eating-Acc the apple-Gen
"Zaid tried to eat the apple”
139
Example 19 also demonstrates an important £act concerning
Action Nominalisation constructions in which the subject
is missing, namely, in such cases an object must be
assigned genitive Case and not accusative Case as in the
fuller forms.
*20- (A) (al-) ?akl ttufaaHat-a
(the) eating the apple-Acc
By themselves, the facts concerning binding and
control are not readily interpretable. However, it would
certainly be desirable if one could show how they
correlate with or follow, if they do, from an analysis of
this construction that would be motivated by the rest of
the facts.
As mentioned at the beginning of this presentation,
there are two main variants of Action Nominalization
constructions, those in which a (logical) subject is
missing and those in which it is present. I will refer
throughout this work to the first type as the
" subjectless " and to the second as the "active" Action
Nominalization. The discussion in the next section will
be devoted to active Action Nominalizations .
140
3.2 Active Action Nominalizations
3.2.1 The Head Movement Analysis
3. 2.1 .1 More About the Facts
Before going into the actual discussion let us
elaborate a little more on the facts. The verbal
properties of active Action Nominalizations are basically
two, the assignment of accusative Case and the
admissibility of adverbs, both of which are exemplified
in 21.
21- (H) axilat Dan et ha-tapuax bi-mehirut
eating Dan OM the apple quickly
"Dan's eating of the apple quickly"
As to the first property, I will adopt the general
assumption (which will be amended later, cf. section
3.3.4) that accusative Case is assigned to an NP in the
context: [vp V ] , that is, accusative Case is assigned
by a verbal head within a particular configuration. As
later discussion will show, such an assumption must be
made and therefore the object NP in 21 must be assumed to
satisfy this structural requirement a t least at some
stage of the derivation.
141
As to the admissibility of adverbs, two points are
to be mentioned. The first concerns the kind of adverbs
which are admitted. In particular, while manner and time
adverbials may freely occur in these constructions,
sentential adverbs are strictly excluded. 22 below shows
sentences with sentential adverbs at different positions
in the sentence. The Action Nominalizations
corresponding to these sentences (23) are all bad.
22- a) (H) lelo safek Dan katav et ha-avoda
doubtlessly Dan wrote OM the work
b) (H) Dan karov le-vadai axai et ha-tapuax
Dan probably ate OM the apple
c) (H) ha-oyev behexlet haras et ha-ir
the enemy definitely destroyed OM the city
23-*a) (H) ktivat Dan et ha-avoda lelo safek
writing Dan OM the work doubtlessly
*b) (H) axilat Dan et ha-tapuax karov le-vadai
eating Dan OM the apple probably
*c) (H) harisat ba--oyev et ha-ir behexlet
destruction the enemy OM the city definitely
24 and 25 below are the corresponding paradigms for
manner and time adverbials. These types of adverbs are
perfectly admissible as demonstrated in 25.
142
24- a) (H) Dan katav et ha-avoda bi-mehirut
Dan wrote OM the work quickly
b) (H) Dan axai et ha-uga be-nimus
Dan ate OM the cake politely
c ) (H) ha-oyev haras et ha-ir emeS
the enemy destroyed OM the city last night
25- a) (H) ktivat Dan et ha-avoda bi-mehirut
writing Dan OM the work quickly
"Dan's writing of the work quickly"
b) (H) axilat Dan et ha-uga be-nimus
eating Dan OM the cake politely
"Dan's eating of the cake politely"
c ) (H) harisat ha-oyev et ha-ir emeS
destruction the enemy OM the city last night
"The enemy's destruction of the city last night"
Thus, if it is assumed (as it generally is) that
manner and time adverbials are attached to the tree
structure a t the VP level (or are at least internal to
VP) , whereas sentential adverbs are attached higher on
the tree structure, then the ungrammaticality of the
examples in 23 may be taken to indicate that an
underlying verbal constituent to be postulated goes up to
the VP level and no further. This obviously presupposes
that the required configuration is indeed available at an
underlying level of representation in order for such
143
distinctions as the one between the VP and the IP-level
to be possible.
The claim that the distribution of adverbs in Hebrew
Action Nominalizations is syntactic rather then semantic
in nature is further supported by the following facts.
As was already mentioned above, adjectives can follow the
head noun in free and double genitives. Thus, compare
the following paradigm with 25 above.
26- a )(H) ha-axila ha-menumeset Sel Dan et ha-uga
the eating the polite Sei Dan OM the cake
"Dan's polite eating of the cake"
b) (H) axilat-o ha-menumeset Sel Dan et ha-uga
eating-his the polite Sei Dan OM the cake
"Dan's polite eating of the cake"
Note, however, that in contrast to adjectives, it is
impossible for adverbs to follow the head noun in free
and double genitives. Thus, compare the following
paradigm with 25 above.
27-*a) (H) ha-ktiva bi-mehirut Sel Dan et ha-avoda
the writing Quickly Sei Dan OM the work
*b) (H) axilat-o be-nimus Sel Dan e t ha-uga
eating-his politely Sei Dan OM the cake
144
*c) (H) harisat-o emeS Sel ha-oyev et
destruction-his last night Sel the enemy OM
ha-ir
the city
To complete the picture, note that adjectives as opposed
to adverbs, are bad in final position. Compare the
following paradigm with 25 again.
28-*a) (H) ktivat Dan et ha-avoda ha-mehira
writing Dan OM the work the quick
*b) (H) axilat Dan et ha-uga ha-menumeset
eating Dan OM the cake the polite
*c) (H) harisat ha-oyev et ha-ir ha-axzarit
destruction the enemy OM the city the cruel
Thus, adverbs and adjectives are in complementary
distribution with respect to final position and the
position immediately following the head. This may also
be demonstrated by examples in which both an adjective
and an adverb occur as in the following. Switching the
order between adjective and adverb as in 29b below is
totally inconceivable.
145
29- a ) (H) harisate-nu ha-axzarit et ha-ir
destruction-our the cruel OM the city
in alot ha-Saxar
with break the dawn
"Our cruel destruction of the city with the break of
dawn"
*b (H) harisate-nu im alot ha-Saxar et
destruction-our with break the dawn OM
ha-ir ha-axzarit
the city the cruel
These distributional facts may be summarized as follows.
30- (Det) N (cl) (Adj) (Sei) NPi NP2 (Adv)
Note that, of the options available in 30, only a subset
occurs in Standard Arabic. That is, as far as Standard
Arabic is concerned, only the bound genitive form occurs
and therefore the possibility of an intervening adjective
as in the free and double forms in Hebrew is not
available. In 30, NPi and NP2 are the subject (agent)
and object (theme) respectively and the distribution of
adverbs and adjectives is as indicated. In this sequence
NP2 is assigned accusative Case while NPi is genitive.
Concerning the respective distribution of adverbs and
146
adjectives, the ungrammaticality of the examples in 27
(adverbs following the head) must be taken to show that
the admissibility of adverbs in Action Nominalizations is
not a fact that has simply to do with some kind of
semantic compatibility. Rather, it is a syntactic fact
that has to do with the principles that determine which
elements can be attached at which positions in the tree
structure. This is not to be interpreted as a claim that
the admissibility of adverbs has nothing to do with the
event type of interpretation associated with Action
Nominalizations. Indeed, some such correlation, or even
dependency, seems very likely. But it is clear that the
interpretation of Action Nominalizations as denoting an
event is simply not sufficient as an explanation of the
severe restrictions on the distribution of adverbs within
the construction.
Thus, all of the facts considered above point
clearly towards the necessity of assuming an underlying
configuration of which a substructure is a verbal
projection of a certain level (the VP level) . Such a
configuration will make possible a direct account of the
above observations concerning the distribution of adverbs
as well as the facts concerning Case marking.
147
3 . 2. 1 . 2 The Syntactic Derivation of Active Action
Nominalizations
Here we will be concerned first with the core cases
of Action Nominalizations, that is, cases in which the
head noun and the subject form a bound genitive as in 30
below. These cases are common to both Modern Hebrew and
standard Arabic. Free and double genitives which are
special to Hebrew will be examined later.
31- a ) (H) axilat ha-yeled et ha-tapuax bi-mehirut
eating the boy OM the apple quickly
"The boy's eating the apple quickly"
b) (A) aklu 1-walad-i ttufaaHat-a bi-sur9a
eating the boy-Gen the apple-Acc quickly
"The boy's eating the apple quickly"
Let us first present the actual derivation assumed here
and then go into further details. The underlying
configuration assumed for 31 a,b is the following.
148
32- DP
/ \
D NP
I / \
POSS NPi N*
I / \
the boy N VP
I / I \
NOM V NPa Adv
I I I
eat the quickly
apple
In 3 2 a nominal bound morpheme, NOM, figures as the head
and subcategorizes for a VP headed by a verb of which any
lexical requirements must be satisfied within the domain
of its government. The NP which is interpreted as the
logical subject of the underlying verb is base generated
in the specifier position of NP. The actual deverbal
form of the head noun is derived by a head movement of
the underlying verb to NOM. The result of this movement
is the following configuration.
149
33- DP
D NP
POSS NPi N
N® VP
N V NP 2 Adv
NOM
The circled subtree in the above configuration is the
derived head of the construction. The trace left by the
verb is assumed to inherit the properties of the verb, in
particular, its Case assigning features. The next step
in the derivation is the movement operation generally
assumed as part of the derivation of bound genitives (cf.
Ch. 2). The result of that process is the following.
150
34- DP
/ \
D NP
/ \ / \
N POSS NPi N'
/ \ 1 / \
V N the N VP
1 1 boy 1 / 1 \
eat NOM e V NPa Adv
the apple
The two main characteristics of this construction are
directly accounted for by this derivation. First,
accusative Case assignment is a manifestation of the
government of the direct object by the verb (or by its
trace). Second, adverbs are licensed in the
configurations 32-34 as subconstituents of VP. Given the
above configurations, it is only VP adverbs that may
occur, and this only in the position following the NP
object, if one assumes that linear adjacency is a
requirement on government and Case assignment.
As will be soon realized, this analysis relies on
properties attributed to the abstract nominal head NOM to
derive many of the properties of this construction.
However, before going into this issue, a word must be
151
said about the morphological realization of the derived
head in 33. As is well known, verbs in both Hebrew and
Arabic appear in a certain number (roughly, 7 or 8 in
Hebrew and 1 2 or more in Arabic) of different
morphological patterns known as Binyanim. Generally, a
certain pattern has a unique nominal i zed form.
Otherwise, if more than one nominal form is associated
with a verb form, each single verb in that particular
class will be associated with a unique nominal form. A
possible difficulty for the present analysis is that it
raises the possibility that one will have to postulate a
different NON morpheme depending on the verb form
appearing as head of VP in a configuration like 32.
Making such an assumption, however, would raise another
difficulty, namely, how to account for the dependency
between the form of the verb and the particular
morphological form of NOM, given that both are inserted
independently. However, no such difficulty arises if
one simply assumes NOM to be an abstract form rather than
an item having an actual morphological shape. Being an
abstract form, NOM may be thought of as some sort of
function which maps a verb onto a unique nominal form
which is lexically listed for that verb.
Some examples of Binyan patterns with the corresponding

1
verbal and nominal forms are the following .
152
35- Hebrew
Binyan pattern verb form nominal form
a) CaCaC SaVaR SViRa
break breaking
b) CiCeC DiBeR DiBuR
speak speaking
c) hi-CCiC hi-LBiS haLBaSa
dress (trans) dressing
d) ni-CCaC ni-SBaR hiSaVRut
be broken being broken
e) hit-CaCeC hit-KaTeV hitKaTVut
correspond correspondence
36- Arabic
Binyan pattern verb form nominal form
a) (Form I) CaCaCa NaQaLa NaQL
transport transportation
b) (Form II) CaCCaCa SaRRaBa taSRiiB
make drink making drink
c) (Form III) CaaCaCa QaaWaMa muQaaWaMa
resist resistance
d ) (Form IV) ?aCCaCa TaRSaLa 7iRSaaL
send sending
e) (Form V) ta-CaCCaCa ta-BaYYaNa taBaYYuN
become clear becoming clear
f) (Form X) ?ista-CCaCa 7ista9MaRa ?isti9MaaR
colonize colonization
153
Having clarified the morphological issue, we may now
come back to the syntactic analysis. As stated earlier,
the role played by NOM is central and many of the
properties of the construction may be explained by making
certain assumptions about the nature of NOM. Aside from
being a bound morpheme taking a VP complement, NOM has
the properties of a nominal head. Given the fact that an
Action Nominalisation construction may occur as an
argument of a verb and refer to events processes and the
like, we may assume (following ideas of E. Williams) that
NOM has a referential theta-role R which is assigned up
("vertically") to the dominating NP node. We will also
assume that NOM has an additional referential theta role
R, which is necessarily an internal one. Thus, a lexical
representation of NOM may look roughly as follows, where
37b encodes its subcategorization frame.
37- a) Argument structure of NOM: (Ri , Rj )
b) Subcategorization frame of NOM: [ VP]
Note that NOM is always situated in a configuration i n
which it may be assigned the external theta role of VP.
NOM is the sister of the VP subcategorized by it. In
this case it is the internal R-role of NOM which enters
the relation of theta role assignment with VP. Now, just
as it is the case with any instance of theta role
154
assignment, here too the relation is not symmetric.
Rather, it is an asymmetric relation of linking. It is
the external theta role of VP which is dependant on Rj
for its referential content rather than the other way
around. Therefore the linking of both theta roles
results in a thematic structure such as illustrated
below.
38- N’i
/ \
Ni VPj
I / \
NOM V NPk
Ri > Rj
The syntactic process represented above is one of theta-
role assignment in the usual sense in which theta-role
assignment is understood in this work. Thus, as far as
the formal operation of theta role assignment is
concerned, the external theta role of VP is satisfied.
Moreover , the operation in 38 must apply since the
external theta-role of VP must be assigned and, most
importantly, it couldn't be vertically assigned to N* and
become its external theta-role since N* already has an
external theta-role, namely, the external theta-role of
NOM. Thus, the assignment of theta-roles as indicated in
155
38 is the only available option. This simple point will
turn out to be of crucial importance as our presentation
proceeds .
The only unusual aspect about the situation
illustrated in 38 is that the internal theta role of NOM
has no referential content. In the normal cases in which
an R-role figures as the external argument of a noun this
referential, or descriptive, content is provided by the
noun itself. Thus, the noun man associated with an
external theta role R corresponds to the expression "man
(x) '* in logical notation. Now, clearly, the referential
content of Rj in 38 does not remain unspecified since it
is the NP in specifier position which ends up being
interpreted as the subject. Thus, it remains to
establish the relationship between the internal argument
of NOM and the NP situated in the specifier position of
the NP of which NOM is the head (cf. 32). This, however,
may be achieved in the way that this is done in any
genitive construction in which the two members of the
construction are related in the identif icatory mode of
construal, as described in Chapter 2, section 2.2. That
is, the external theta-role R of the genitive NP is
coindexed with the internal thematic index of the head
noun, that is, the internal argument of NOM. We can,
therefore, say that the subject NP is assigned a theta-
role through the mediation of NOM. That is, the subject
156
theta-role which was assigned first to NOM nay be said to
have been passed on (although not externally, or,
vertically) rather than being "absorbed" by NOM. This
state of affairs may be represented as follows.
39- NPt k
/ \
NPa N'k
R t / \
N VPi
1 / \
NOM V NPa j
Rk ,Ri i,j
Note that in a case where an NP is present in specifier
position the operation (coindexation) illustrated in 39
nust take place since this is the only way for this NP to
be thematically licensed. If no NP is present in this
position then no ungranunaticality results and the
internal argument of NOM is construed as arbitrary in
reference. An additional mechanism by which the internal
R-role of NOM may be assigned referential content is the
one of obligatory control to be described later in this
chapter .
This is then the analysis of the core cases of
Action Nominalization constructions involving the bound
157
genitive. Given this approach, the analysis of the free
and double genitive variants which are special to Hebrew
is a straightforward natter. Note that just as is the
general case in genitive constructions, here too, a Sel-
phrase may be adjoined to the tree structure at a later
stage in the derivation. This is in line with Lebeaux's
(1988) general approach to adjuncts adopted in this work
and applied to the treatment of Sel-phrases. The Sel-NP
is construed in the identif icatory relation with the head
noun NOM by having its external R theta-role coindexed
with the internal theta role available in the argument
structure of NOM, which is at this point assigned
(coindexed with) the external theta-role of VP. The
result is the free genitive variant of 39, of which the
two relevant derivational stages are illustrated in 40
b,c.
40- a ) (H) ha-axila Sel Dan et ha-tapuax
the eating Sei Dan OM the apple
"Dan's eating of the apple"
158
b) NP-structure
NPk
/ \
Det N* k
1 / \
the Nk VPi
1 / \
NOM V NPj
Rk , Ri 1
eat i , j
c) S-structure DP
/ \
D - NP
1 / \
the N'
/ \
N' VPi
/ \ / \
N Sei V NPj
/ 1 Dan? 1 1
V N Ri e the apple
1 1
eat NOM
Rk , Ri
159
The structure of the corresponding double genitive
construction is a combination of the free and the bound
form. Here it is the pronoun in the Spec-of-NP position
which is thematically linked with NOM and it is to this
underlying structure that a Sel-phrase is adjoined. 41b
is the configuration resulting from this adjunction prior
to the head movement of V to NOM and of the movement of
the resulting head noun to D.
41- a) ( H ) axilat-o Sei Dan et ha-tapuax
eating-his Sei Dan OM the apple
"Dan's eating <of the apple"
b) DP
/ \
D NP
1 / \
POSS NP N•
1 / \
hei N' \
/ \ VPi
NOM Sei- Dan / \
Rk ,R1 V NPj
i,j
The relationship between "Dan" and the doubling clitic
pronoun in this structure follows from the general
160
analysis of clitic doubling proposed in Chapter 2
(section 2.3). The pronoun is coindexed with the
internal argument of NOM in the same way as this is
achieved with a "full" NP in the bound genitive variant.
The Sel-phrase is incorporated into the overall
interpretation by the proposed LF mechanism.
Binding relationships between NP2 and NPa ,
coarguments of the underlying verb in the configuration
in 39, are directly accounted for. NP2 , being assigned
the external theta-role of VP, stands to NPa , an internal
argument, in the same asymmetric relationship that would
exist between a subject and an object in a sentence. The
contrast in 42 is thus explained.
42- a ) (H) reiyat Dan et acmo
seeing Dan OM himself
"Dan's seeing of himself"
*b) (H) reiyat acmo et Dan
seeing himself OM Dan
In this respect there is no difference to be expected
between the three different variants of genitive
constructions. The following examples confirm this
expectation.
161
43- Free Genitive
a) (H) ha-reiya Sei Dan et acmo
the seeing Sei Dan OM himself
*b) ( H ) ha-reiya Sei acmo et Dan
the seeing Sei himself OM Dan
44- Double Genitive
a ) (H) reiyat-o Sei Dan et acmo
seeing-his Sei Dan OM himself
*b) (H) reiyat-o Sel acmo et Dan
seeing-his Sei himself OM Dan
162
3.2.2 Syntactic vs. Lexicalist Approaches to
Action Nominalizations
The account outlined in the previous section may be
referred to as syntactic since it assumes a syntactic
process by which the noun head of an Action
Nominalization construction is derived, the head being a
single word as far as morphological and phonological
criteria are concerned. Taking this line, we were able
to derive many of the properties of the construction,
some of which are “configurationally sensitive" (for
example, binding and the distribution of adverbs) and
some of which have to do with the properties that may be
attributed to lexical items by virtue of being members of
certain word classes (for example, Acc. Case assignment
by a verb). The burden of providing an explanation is
put in the syntactic approach mainly on the properties of
the configuration and the syntactic operations that it
makes possible. However, as mentioned at the beginning
of this chapter, this is not the only conceivable
approach. Another way to go might be to attempt to
derive the facts by attributing certain properties to
lexical items. For example, the property of being able
to assign accusative Case may be attributed to a certain
class of nouns. Such an approach might be desirable from
the point of view of a certain general theoretical
163
conception which wishes to exclude the possibility of any
word formation processes in the syntax. I n this section
I will examine two possible versions of a lexicalist
approach to Action Nominalizations, in an attempt to
demonstrate their inadequacies (sections 3. 2. 2.1 and
3. 2. 2. 2). In sections 3.2.2. 3 and 3. 2. 2. 4 I will
present what seem to be two fairly compelling arguments
in favour of the syntactic analysis outlined in the
previous section. These arguments will be based on
certain facts concerning the interpretation of Hebrew
Action Nominalisation constructions and on an analysis of
causative constructions in Arabic respectively.
3. 2 . 2 . 1 A Lexicalist Analysis ~ First Version
A first possible version of a lexicalist analysis of
Action Nominalizations would be represented by the
structure in 45b which is assigned by this analysis to
45a below.
45- a ) (H) harisat ha-oyev et ha-ir
destruction the enemy ON the city
"the enemy's destruction of the city"
164
b) N*
/ \
N NP NP
1 1 1
destruction the the
enemy city
The particular analysis represented by the tree structure
above is proposed in Doron (1989) . The discussion in
this section will concentrate on general issues
concerning the validity of this approach. Some problems
that are internal to Doron* s particular proposal are
discussed in footnote 3. A slight variant of this
approach due to Borer (1984) is discussed and rejected in
footnote 4 to this chapter.
In 45b the two arguments of the head noun are .
strictly. c-commanded and thus governed and assigned Case
by it. It is assumed by Doron that nouns such as the one
figuring as head of 45b have the special, lexically
specified, property of being able to assign accusative
Case. Such nouns are referred to in Doron (1989) as
"strong" nouns.
One might say that the set of assumptions
represented by the configuration in 45b is the minimal
and the most simplistic one to make and, as I will show,
its inadequacies are rather striking. Nevertheless, I
165
assume that it still deserves to be examined even if only
for the purpose of illustrating the nature and the
complexity of the problems involved.
The central assumption in this analysis is the one
about the existence of what are referred to as "strong"
nouns. This assumption makes it possible to derive the
fact of the assignment of accusative Case to the
(logical) object without being forced to assume a more
elaborate sentence or VP-like structure. Obviously,
this is a legitimate assumption to make and the
comparison between this analysis and a syntactic analysis
should center mainly on the explanatory gains made by
each one. That is, how many of the facts associated with
the Action Nominalization construction may be explained
on the basis of this assumption.
Taking this line of reasoning, the most striking
property of the configuration in 45b is that both
arguments are situated in the same structural
relationship to the head (immediate dominance by the
first branching node) and are, thus, also symmetrically
related to each other.
As to the empirical facts, we may start with those
concerning the order of constituents in the construction.
Thus, one might raise the legitimate question as to why
the two arguments, the genitive NP and the accusative NP,
16 6
appear in the order as in 45 (Genitive-Accusative) rather
then in the reverse order as follows.
*46- (H) (ha-) harisa et ha-ir ha-oyev
(the) destruction OM the city the enemy
Here an obvious answer suggests itself, namely, it is
fairly plausible to assume that an NP standing in a bound
genitive relation to a head noun must be adjacent to it
whereas such a requirement need not, presumably, be
observed by a direct object which is assigned accusative
Case. Note that by proposing this answer we are ignoring
a range of facts concerning the syntax of bound genitives
for which the structure in 45b can provide no account.
These include the well known restriction on the
occurrence of an initial definite article in a bound
genitive. But this problem may be ignored at the moment
for the sake of the argument. Note also that by implying
that no adjacency requirements hold with respect to the
relationship between the head and the (logical) object
one raises the expectation that other types of material
may intervene between the two, say, adverbs. In this
case, again, the following contrast remains unaccounted
for.
167
47- a) (H) harisat ha-oyev et ha-ir be-axzariyut
destruction the enemy OM the city cruelly
"The enemy's cruel destruction of the city"
*b) (H) harisat ha-oyev be-axzariyut et ha-ir
destruction the enemy cruelly OM the city
We may ignore this problem as well for the sake of the
discussion and come back to the issue of word order.
Note now that the answer suggested above, namely, that a
genitive NP must be adjacent to the head noun does not
yet solve the problem but simply sets the stage for
another, more fundamental, question, namely, why it is
that the (logical) object, or theme, may only be marked
with accusative Case and the (logical) subject, or agent,
may only be marked with genitive Case (in case, of
course, that both arguments are present), that is, why
the following example is impossible with an
interpretation identical to that of 45a ("The enemy's
destruction of the city").
*48- (H) harisat ha-ir et ha-oyev
destruction the city OM the enemy
The problem is that there is nothing about the structure
in 45b that makes it possible to establish a unique
configurational relationship between the head and each
168
one of the arguments and in this way to explain the
difference in Case marking between the two arguments as a
manifestation of two different configurational
relationships. Now, if such an unambiguous mapping from
logical (Subject , Object) or thematic (agent , theme) roles
to Case marking is impossible on the basis of
configurational relationships, some other kind of
mechanism must be invoked. It seems that the only
alternative left is a direct, one-to-one, mapping from
logical, or thematic, roles (taken as primitive notions)
to case marking in the way that such mechanisms are
assumed to exist in so called "non-configurational"
languages. Thus, we will have to assume that, at least
as far as the syntax of NP's is concerned, Hebrew and
Arabic are "non-configurational". A mapping from logical
or thematic roles to case marking for the two languages
might look as follows.
49- Logical/Thematic role Case marking
Subject/Agent --------> Genitive
Object/Theme --------> Accusative
Is it, however, the case that such a, one to one, mapping
really exists ? Going from left to right in 49, a
(logical) object is indeed assigned accusative Case, but
169
only if the (logical) subject is present. Otherwise, it
is assigned genitive Case and is still, to be sure,
interpreted as the theme. Going from right to left,
genitive Case is assigned to the (logical) subject, but
if the subject is missing then it must be assigned to the
object. We see then that there is no unambiguous,
unique, mapping from logical, or thematic, roles to Case
marking, and there is no way in this approach to exclude
48. If 48 cannot be ruled out on the basis of some non-
conf igurational mechanism, it becomes necessary to assume
a configuration in which each of the arguments stands in

2
a unique structural relationship to the head .
Obviously, 45b couldn't be that configuration.
The same type of problem arises with respect to
anaphoric relationships in these constructions. In 45b
the two NP's are in a relationship of mutual c-command.
Therefore, given a standard version of the binding
theory, no account is available on this approach for the
following contrast.
50- a ) (H) harisat ha-enoSut et acma
destruction the humanity OM herself
"The destruction of humanity by itself"
*b) (H) harisat acma et ha-enoSut
destruction herself OM the humanity
170
Rather, given condition C of the binding theory, both 50a
3
and 50b are predicted to be ungrammatical . It is
therefore obvious that this version of a lexicalist
approach suffers from serious inadequacies at the most
elementary level. Indeed, it accounts for none of the
relevant facts, including the assignment of accusative
Case to the object. This fact is not explained on the
basis of any generally known facts or principles but is
simply derived from a certain stipulation about the

4
properties of "strong" nouns .
The important lesson to be drawn from this
discussion is the following. Accusative Case assignment
is not a property that may simply be attributed to a
lexical head of a certain kind. It is the property of a
lexical head within a particular syntactic configuration
and in this sense it may be said to be configurational.
Therefore, any syntactic analysis of a construction
involving Accusative Case marking must provide the
required configuration.
171
3. 2. 2. 2 A Lexicalist Analysis - Second Version
One might think of another, more elaborated analysis
within a lexicalist approach. Such an analysis may be
represented roughly as in the following configuration.
51- DP
/ \
D NP
/ \
NP N*
I / \
the enemy N NP
I I
In the configuration above each of the arguments stands
in a unique structural relation to the head and in this
way this analysis, at least, establishes a certain
asymmetry between the two NP's. The surface structure
corresponding to 51 would be derived by the movement
operation assumed to apply generally in the derivation of
bound genitive constructions. The output of this
operation would be the following.
172
52- DP
/ \
D NP
/ \ / \
Nt D NP N'
1 1 1 / \
destruction POSS the N NP
enemy 1 1
et the city
It may also be said in favour of this analysis that the
asymmetry it establishes between the two NP positions
makes it possible to account for binding relationships in
this construction.
Note that, in fact, at least within the theta
theoretic framework assumed here, this analysis does not
yet solve the problem. This is for the following reason.
Given the framework assumed here, the head noun in 51
would have three arguments, an external R role and two
internal roles, say. Agent and Theme. However, this
theory makes only a binary distinction between external
and internal roles. It follows that the two internal
roles are of an equal status so there is still no way to
ensure that the Agent role will be mapped onto the Spec-
of-NP position and that the Theme will be assigned to the
complement position. This is, however, a highly theory
173
internal argument and is therefore far from being
sufficient to settle the issue.
In the following I will present several other
arguments against this analysis, some of which may be
considered more conclusive and some of which less.
However, there do seem to be two fairly conclusive
arguments against it, or any other lexicalist analysis,
which will be presented in the next two sections.
3 . 2. 2 . 2.1 The Distribution of Adverbs and Adjectives
It is generally assumed that the distribution of
adverbs is conditioned by syntactic factors such as the .
availability in the tree structure of a VP node (at least
for time and manner adverbials) and possibly other
factors. If these assumptions are accepted, then it can
be claimed that a configuration as in 50,51 does not
provide the right environment for an adverb to occur.
Otherwise it would become necessary to argue that the
distribution of adverbs is conditioned by semantic rather
then syntactic factors. This, however, was argued
against in section 3. 2. 1.1 in which evidence was provided
showing that the distribution of adverbs, in particular,
the respective distribution of adverbs and adjectives in
174
Action Nominal izat ion constructions, is due, most
likely, to syntactic rather then semantic factors.
3. 2 .2. 2. 2 Free Genitive Variants
Cases in which the subject NP appears in a Sel-
phrase in the free genitive construction as in the
following example raise some serious difficulties for
this version of lexicalist analysis.
53-* (H) ha-hafcaca Sel ha-cava et ha-ir
the bombardment Sei the army OM the city
"The army's bombardment of the city"
The problem with 53 is that the approach under
consideration incorporates the assumption that the
assignment of theta-roles is achieved unambiguously by
mapping the "theme” onto the complement position and the
"agent" onto the specifier position. However,
since in this case the specifier position of NP is
occupied by the subject as in 52 above the only way to
derive the surface order in 53 is by an operation of head
movement adjoining the head noun to D. This raises at
least two serious problems as follows.
Starting with the, possibly, less serious one, the
175
head position of DP may be occupied by a definite
article. It is therefore not clear whether such a
movement would still be possible. Shlonsky (1990)
suggests that such a movement is not only possible but
also necessary. In particular, Shlonsky (1990) assumes
that the definite article "ha-" is an affix which needs
to be supported. For this requirement to be satisfied
the head noun must move to D and adjoin to "ha-". The
structure of 53 would thus be the following in which the
definite article is an affix attached to the head noun.
54- DP
/ \
D NP
/ \ / \
D N NP N’
ha 1 1 / \
the bombar- Sei N NP
dment the 1 1
army e the city
It seems that there is some data which makes this
analysis highly implausible. Thus, it is possible to
have numerals instead of the definite article in examples
corresponding to 53.
176
55- a ) (H) SaloS hafcacot Sel ha-cava et ha-ir
three bombardments Sel the army OM the city
"Three bombardments o£ the city by the army”
a ) (H) arbaa nifnufim Sel Dan et ha-yad
four wavings Sei Dan OM the hand
"Four hand wavings of Dan”
It is questionable whether the numerals in the examples
above may be viewed as affixes rather than free forms and
whether the sequences Numeral+N in these examples can be
taken to be derived words rather than units of a higher
level. Note that the numerals in these examples are in
their free rather than bound forms (SaloS as opposed to
SloSa and arbaa as opposed to arbaat). It seems
reasonable to assume that in general numerals and the
definite article occupy the same position since they are
in complementary distribution as demonstrated by the
following pairs.
56- a ) (H) SaloS mexoniyot
three cars
* b) (H> SaloS ha-mexoniyot
three the cars
57- a ) (H) arbaa sfarim
four books
177
b ) (H) arbaa ha-sfarim
four the books
A more serious problem to this analysis relates to such
examples as the following i n which the head is modified
by an adjective.
58- (H) ha-hafcaca ha-maftia Sel ha-cava et
the bombardment the surprising Sei the army OM
ha-ir
the city
"The army's surprising bombardment of the city**
The problem is that the present analysis simply can't
generate such cases. This is because the operation which
moves the noun up to D applies to the head alone and must
therefore leave a modifying adjective behind as
illustrated below.
178
59- DP
/ \
D NP
/ \ / \
(ha) Nt NP N*
/ \
N* NP
/ \
N AP
et
Thus, the best that the present analysis can do is
generate ungrammatical sequences such as the following
(compare with 58).
*60- (H) ha-hafcaca Sel ha-cava ha-maftia et
the bombardment Sei the army the surprising OM
ha-ir
the city
Within the analysis proposed in this work, in particular,
the approach to adjuncts adopted here (due t o Lebeaux
(1988)), cases such as 58 pose no problem. Thus, it may
be assumed that Sel-phrases are adjoined to the tree
179
structure at a later stage in the derivation, say, S-
structure, as follows.
61 - NP-structure S-structure
DP DP
/ \ / \
D NP -----> D NP
1 / \ 1 / \
the N' the N’
/ \ / \
N' NP N' NP
/ \ 1 / \ 1
N AP the city N' Sei the
1 / \ the city
bombardment N AE’ army
bombardment
Now, it couldn’t be the Sel-phrase which is (directly)
assigned the subject theta role in this construction
since its configurational positioning does not observe
the requirement of asymmetry between subject and object.
So it must be some other element. This issue is handled
within the approach advocated here by making certain
assumptions about the role played by NOM as a "receiver"
180
of a theta-role, but this makes a non-lexicalist approach
necessary.
3. 2. 2. 2. 3 - Subjectless Action Nominalizations
and Control
An analysis such as the one represented in 51 , 5 2
runs into serious problems when it comes to subjectless
Action Nominalizations. Here we are referring to cases
such as the following.
62- a ) (H) haxzarat ha-taSlumim bi-mehirut
return the payments quickly
"the return of the payments quickly"
b) (A) ?aklu ttufaaHat-i bi-sur9a
eating the apple-Gen quickly
"the eating of the apple quickly"
I n these examples the object stands in the bound genitive
construction to the head noun and therefore must be
assumed to occupy the specifier position of NP. In this
state of affairs no other structural position is
available for an additional argument. However, that the
existence of such an additional argument must be assumed
may be demonstrated by the fact that subjectless Action
181
Nominalizations allow for the occurrence of an anaphor
and therefore must be assumed, at least under a standard
approach to binding, to provide a position for an
antecedent .
63- (H) reiyat acmo ba -mar?a
seeing himself in the mirror
The facts concerning control of subjectless Action
Nominalizations in Arabic point in the same direction.
64-a) (A) yuriidu zaid-un naql-a 1-kitaab-i
wants zaid-Nom transportation the book -Gen
?ila bayruut
to beirut
"zaid wants to transport the book to Beirut"
b) (A) yuriidu zaid-un al-intiqaal -a ?ila
wants zaid-Nom the transportation-Acc to
bayruut
beirut
"Zaid wants to be transported to beirut"
As the facts in 64 show, control of subjectless Action
Nominalizations in Arabic is just like control of
Infinitives in Hebrew or English. That is, it is
insensitive to the type of theta-role assigned to the
182
controlled element. Rather, it is a control of a
position. Presumably, under a standard approach, the
controlled element would be an empty NP, PRO, situated
somewhere i n the configuration. However, it would have
to be situated in such a position as to be able to serve
as the antecedent of an anaphor as in the following
example.
65- (A) yuriidu zaid-un qatl-a nafs-i -hi
wants zaid-Nom killing-Acc self-Gen-him
"Zaid wants to kill himself"
These structural requirements cannot be met by the
configuration in 51,52 if the anaphor in 6 5 is assumed to
occupy the specifier position of NP.
There is still another possibility to be considered
with respect to the control phenomenon, namely, one which
assumes a theta theoretic approach to control, rather
than a standard, "structuralist" approach. Under such an
approach, the control relation would hold not between a
controlling NP and a controlled element occupying a
specific structural position, but rather a relation
between a controller and a uniquely identified thematic
role. It is my contention that such an unambiguous
relationship cannot be established within the approach
183
under consideration. This can be demonstrated with the
following example.
66- ( A ) Sajja9 -tu zaid-an 9ala l-?akl -i
encouraged-I zaid-Acc on the eating-Dat
"I encouraged zaid to eat"
In this example it is unambiguously the agent of "eating"
which is controlled, although the verb "eat" is lexically
specified for both an "agent" and a "theme". To
establish an unambiguous relationship between a
controller and a theta-role, one of the theta-roles would
have to be assigned a unique status. Here, the only kind
of distinction available is the external/internal
distinction since notions such as "agent" or "theme" are,
as demonstrated above, irrelevant. However, given that
the noun ?akl (eating) has, like any other noun, an
external referential role R, the two remaining "agent"
and "theme" roles are both internal and there is no other
way to single out one of them for the purpose of control.
In section 3. 3. 4. 3 below it will be demonstrated how the
non-lexicalist analysis developed so far can handle the
control phenomenon. This will be done by relying,
crucially, on the thematic properties, the argument
structure, of NOM and the way it interacts with the
overall configuration. But this necessitates the
184
assumption of such an element as NOM and such an
assumption implies a non-lexicalist analysis.
As may be concluded from this discussion, this
second version of a lexicalist analysis of Action
Nominalizations faces a number of serious problems, most
of which have to do with the fact that the structure
assumed by this version does not provide enough
"configurational space" so as to make possible an account
of a certain number of facts. However, the main
characteristic of a lexicalist analysis, the one to which
it would owe its name, is the assumption that the head
noun in an Action Nominalization construction is not the
product of a syntactic process but is rather lexically
inserted at the base generated structure in its final
surface form. In section 3. 2. 2. 4 I will outline what
seems to be a rather strong argument against this
assumption based on the analysis of causative
constructions in Arabic. Before going into this issue,
however, I will present in the next section an argument
in favour of the syntactic approach based on the
interpretation of Action Nominalization constructions in
Hebrew.
185
3. 2. 2. 3 The Interpretation of Action Nominalizations
Here we will be concerned with pairs such as the
following
67- a) (H) harisat ha-romaim et ha-ir
destruction the romans OM the city
"The romans* destruction of the city"
b) (H) harisat ha-ir Sei ha-romaim
destruction the city Sei the romans
"The romans' destruction of the city"
Under any version of a lexicalist approach to Action
Nominalizations both examples above would be assumed to
involve a lexically derived head noun, with the only
difference between the two variants being the different
positioning and Case marking of the NP's related to it.
This is the approach taken by both Dor on (1989) and Borer
(1984) . For the sake of the discussion I will
concentrate in what follows on the particular details of
Doron's analysis of the pair in 67. But, as will become
clear, the argument presented in this section undermines
the validity of any lexicalist analysis regardless of its
particular details and provides strong support in favour
of the syntactic analysis proposed in this work.
186
Doron (1989) assigns the a and b examples in 67 the
structures in 68 a and b respectively.
68- a) N*
/ I \
N NP NP
I I I
destruction the the city
enemy
b) NP
/ \
N' NP
/ \ I
N NP Sei the enemy
I I
In Doron's words, 68a is claimed to be "a process NP with
an internalized argument" whereas 6 8 b is "a process NP
with an external argument". Thus, both NP's are taken to
be process NP's with the only difference being that the
agent theta-role is assigned internally in one case and
externally in the other. 68 a,b are, on this view, what
one may refer to as "transformational variants". That
is, in both cases it is, allegedly, the same array of
187
theta-roles which is lexically available, the only
difference being the structural position onto which the
agent theta-role is "mapped" in each one of these
variants. This alleged difference between the two
variants would be similar to the difference which is
assumed to exist between, say, an active sentence and its
passive counterpart. That is, both sentences tell
basically the same story with some slight differences of
emphasis. The main argument of this section is that
this view of the difference between 67a and 67b is
totally erroneous. Rather, 67a and 67b are different in
both interpretation and structure and an explanation of
the difference between the two involves, necessarily, the
distinction between syntactic (67a) and lexical (67b)
nominalizations. Thus, if the argument to be presented
is valid then a syntactic process of nominalization of
the kind advocated here must be assumed as part of the
analysis of 67a.
There is a subtle, but still visible, difference of
interpretation between 67a and 67b. 67a denotes a
specific, temporally defined, event in which the Romans
destroyed the city. 67b, on the other hand, is most
felicitous in a context in which the destruction of the
city by the Romans is compared or contrasted with the
destruction of the city by somebody else, say, the
crusaders. Here the NP is interpreted as referring to
188
the manner in which the d e s t r u c t i o n was executed rather
than to the event i t s e l f . Corresponding differences may
be observed with pairs such as the f o l l o w i n g .
69- a) (H) reiyat Dan et Dina
seeing Dan OM Dina
"Dan 's seeing <

of D i n a "
b) (H) reiyat Dina Sei Dan
seeing Dina Sei Dan
"Dan's seeing of Dina"
The most likely interpretation o f 69a is one in which i t
r e f e r s to the event of Dan's seeing Dina. 69b , on the
other h a n d , i f i t i s assigned any interpretation a t a l l ,
i s more l i k e l y to be assigned a g e n e r i c type o f
i n t e r p r e t a t i o n in which reference is made t o D a n ' s way o f
viewing Dina in g e n e r a l .
A d m i t t e d l y , these contrasts are rather subtle and
may not be agreed upon by some s p e a k e r s . It would
therefore be advisable t o concentrate on p a i r s in which
the contrasts are c l e a r and undeniable. The f o l l o w i n g i s
such a pair.
70- a ) <H> harisat pikasso et yeruSalayim
destruction picasso OM Jerusalem
" P i c a s s o ' s destruction o f Jerusalem"
189
Reproduced with permission of the copyright owner. Further

Reproduced reproduction prohibited
Further reproduction permission..
without permission
prohibited without
b) (H) harisat yeruSalayim Sel pikasso
destruction Jerusalem Sel picasso
"Picasso's destruction of Jerusalem"
70a can only have the event interpretation by which it
refers to the event of the destruction of Jerusalem by
picasso (say, a roman emperor) . For 70b, on the other
hand, the most likely interpretation is radically
different. In a pragmatic context in which picasso is
widely known as a famous painter it is most likely
construed as referring to a picture depicting the
destruction of Jerusalem which was painted by picasso.
We may refer to this as the authorship interpretation.
Note that in an imaginary context in which "picasso" is
the name of a roman emperor an interpretation of 70b in
which picasso is the destroyer of Jerusalem, although
very unlikely, is not completely excluded. To make such
an interpretation more likely we may replace "picasso" in
70b by the name of the real destroyer of Jerusalem.
71- (H) harisat yeruSalayim Sei titus
destruction Jerusalem Sei titus
"Titus* destruction of Jerusalem"
Here an agent (destroyer) reading of "titus" is indeed
more likely. Thus, pragmatic factors such as background
190
Reproduced
Reproduce Further reproductio
d with permission of the copyright owner. Further reproduction prohibited without
n prohibited permission..
without permission
knowledge and the like seem to play an important role in
the interpretation of 70b, 71 and similar examples,
giving rise to widely varying interpretations. Note now
that, as has often been observed, such a variation is
typical of the interpretation of constructions which
involve "noun to noun" relationships. Thus, in the
following example
72- John's book
John may be understood as standing in any of a wide range
of possible relationships with respect to the book such
as ownership, authorship and so forth. The only
limitations in this respect are pragmatic in nature.
Similar observations were also made by Williams (1982)
with respect to english derived nominal constructions
such as the following which may have, among others, the
interpretation indicated below.
73- Your destruction of Rome =* Your account of the
destruction of Rome.
Returning now to the pair in 70, note that for 70a,
unlike 70b, the "authorship" interpretation is completely
excluded. Rather, "picasso" in this example may only be
understood as the agent of "destruction". Pragmatic
191
prohibited without permission.

reproduction prohibited
Further reproduction
Repro duced with permis
Reproduced permission the copyright
sion of the owner.. Further
copyright owner without permission.
factors such as background knowledge and the like are
strictly excluded from influencing the interpretation of
such examples. Thus, in observing the difference
between 70 a and b, a clear generalization seems to
emerge which may be formulated as follows:
Whenever the "object" is marked Accusative the
"subject" must be interpreted as the agent (or
whatever type of theta-role happens to be the
external theta-role of the corresponding verb)
It is clear that any approach to the analysis of these
constructions which allows only for the possibility that
the head noun in both variants in 70 is a lexically
(rather then syntactically) given element is completely
incapable of making sense of such a generalization.
Generally, if destruction and picasso in 70 a and b are
lexically given nouns then one would expect the freedom
of interpretation typical of noun-to-noun constructions
to be attested in both variants. Obviously, this
semantic difference couldn’t be attributed to a
contribution of the Case markers which happen to be
different in each one of these variants.
If 70 a and b had the structures in 6 8 a and b
respectively, with the distribution of theta roles as
claimed by Doron, then the attested difference in
192
Reproduced with permission of the copyright owner. Further reproduction prohibited without
reproduction prohibited permission.
without permission.
interpretation is, again, totally mysterious. Thus, on
Doron's account, "picasso” would have to be interpreted
as the agent of "destruction" in both variants and this
is clearly false.
These facts and the observation stated above receive
a straightforward explanation within the analysis of
Action Nominalisations and the general view of genitive
constructions advocated in this thesis. Thus, 70b is a
typical instance of a construction in which two nouns
("destruction" and "picasso") are construed in an
identification relation (with Jerusalem being construed
as a classificator) . The interpretation of such a
construction is typically open to pragmatic influences.
70a, on the other hand, is the output of a syntactic
process of nominalization and has the following
underlying structure.
193
prohibited without permission.
74- DP
/ \
D NP
I / \
POSS NP N’
I / \
Picasso N VP
I / \
NOM V NP
I I
destroy the city
The obligatory agent interpretation of "picasso" in this
case derives from the obligatoriness of the assignment of
the external theta role of VP first to NOM and then from
NOM to the NP in SPEC. Thus, the interpretation of the
subject in this case is strictly determined by syntactic
factors to the exclusion of any pragmatic factors.
Clearly, in this configuration the object NP
("the city") may only be assigned accusative Case. In
this way the correlation between the interpretation of
the "subject" and the Case marking of the object receives
a fairly simple account which is clearly impossible under
any version of a lexicalist approach.
The difference between the two structures in 67 is
clearly reflected in other syntactic properties. Here
194
the most striking difference between the two has to do
with the phenomenon of clitic doubling. Thus, in a
structure such as 67b, if the object (theme) NP is a
pronoun then it must be coreferential with the NP in the
Sel-phrase.
75-*a) (H) harisat-a Sei ha-romaim
destruction-her Sei the romans
b) (H) harisat-am Sei ha-romaim
destruction-their Sei the romans
"The destruction of the romans"
Obviously, clitic doubling is completely out of question
for 67a
*76- (H) harisat-ai et ha-iri
destruction-hen OM the cityi
An analysis such of the type suggested in Doron (1989)
has nothing to offer on this point. On the contrary,
given the assumption that the two NP's are independently
assigned a theta role in both variants, the observed
facts come rather as a surprise. Of-course, it is
possible for a pronoun to be coreferential with an NP
which is configurationally placed in a certain way with
respect to it but, on this account, one sees no reason
195
why it should be obligatory in one case while strictly
excluded in the other.
3. 2 . 2 . 3 Causative Constructions in Arabic
The discussion in this section will be concerned
with data from both Standard literary Arabic (A) and what
is largely designated as Syrian Arabic (SA). Marking for

0
each of the examples will be given correspondingly •
Note that dialectal variation is irrelevant for the main
point that this section is intended to make and therefore
no difficulties arise from our use of data from
Standard Arabic and Syrian Arabic alternately in the
following discussion.
Typical examples of causative constructions are
pairs such as the following.
77- a ) (A) 9allama zaid-un al- qasm-a haad'a
teach zaid-Nom the class-Acc this
1-kitaab-a
the book -Acc
"Zaid taught the class this book"
196
b) (A) 9allama zaid-un haad'a 1-kitaab-a
teach zaid-NOM this the book -Acc
li- 1- kasm-i
to the class-Dat
"Zaid taught this book to the class'*
78- a ) (A) Sarraba zaid-un al- awlaad-a al-
make-drink zaid-Nom the children-Acc the
Haliib-a
milk -Acc
"Zaid made the children drink the milk"
b) (A) Sarraba zaid-un al-Haliib-a li- 1-
make drink zaid-Nom the milk-Acc to the
?awlaad-i
children-Dat
"Zaid made the children drink the milk"
As the translations show, in the a examples above the
verb is followed by two accusative marked NP’s whereas in
the b examples the order of the two NP's is reversed, the
first of which is marked accusative and the second is
governed by the dative preposition li. I will refer
throughout the discussion to the first type (examples a )
as the "double accusative" and to the second type
(examples b) as the "dative" (or "dative causative").
Mushaweh (1986) and Hoyt (1989) enumerate a certain
number of facts which suggest a complex hierarchical
197
structure for the double accusative and illustrate some
striking differences between the two variants. We may
start with some facts concerning binding.
In a double accusative construction only the first
accusative NP may be anaphorically related to the subject
of the sentence.
79- a) (SA) samma9-na ba9da-na 1-Ranaani
make hear-we selves-our the songs
"we made ourselves hear the songs"
*b) (SA) samma9-na al-awlaad ba9da-na
make hear-we the children selves-our
*”we made the children hear ourselves"
Equally, the first accusative NP may be an antecedent to
the second accusative NP, but the reverse does not hold.
80- a) (SA) Hanna 9addad aHmadi Haalui
John made count ahmed himself
"John made Ahmed count himself"
*b) (SA) Hanna 9addad Haalui Ahmadi
John made count himself Ahmed
In the dative construction, on the other hand, the
accusative NP, which corresponds to the second accusative
198
NP in the double accusative variant, may be anaphorically
related to the subject.
81- (SA) Hannai 9 add ad Haalui la-aHmad
John made count himself to Ahmed
"Johm made Ahmed count himselfi "
We may summarize the binding facts as follows,
82- a ) Double Accusative: NPi V NP2 NPa
b) Dative : NPi V NPa P NP2
In 82a NPi can be the antecedent of NPa and NPa can be
the antecedent of NPa . NPi and NPa may not be
anaphorically related. In 82b NPi can be the antecedent
of NPa .
Another set of facts concerns the possibilities of
cliticization in double accusative and in the dative.
Using the patterns in 82, we may say that in a double
accusative, if NPa is a pronoun then it can (in fact,
must) be cliticized on the verb (82b) but this is
impossible for NPa (82c).
83- a ) (SA) fahham aHmad hind ddars
made understand ahmed hind the lesson
"Ahmed made hind understand the lesson"
199
b) (SA) fahham-a aHmad ddars
made understand-her aHmed the lesson
"Ahmed made her understand the lesson"
*c) (SA) £ahham-o aHmad hind
made understand-it ahmed hind
In 83a NP2 and NPa are two "full fledged" non-pronominal
(feminine and masculine respectively) NP's.
In the corresponding dative, however, it is
perfectly possible for NPa to be cliticized on the verb.
84- (SA) fahham-o aHmad la-hind
made understand-it ahmed to hind
"Ahmed explained it to hind"
One more set of facts may be provided before we proceed
to present an analysis. As observed by Hoyt (1989),
double accusative sentences with a modifying adverb are
ambiguous between two possible construals of the adverb,
as indicated in the following examples.
200
8 5- (SA) Hanna massak xaalid 1-Sunta bi-beit 1-
John made hold khalid the suitcase in house the
jiraan
neighbors
"John made khalid hold the suitcase in the
neighbors house"
i ) John caused the following: khalid held the
suitcase in the neighbors house.
ii) It was in the neighbors house that John made
khalid hold the suitcase.
86- (SA) Hanna ?akkal xaalid 1-sanduiS bi-tariqa
John made eat khalid the sandwich in manner
biSaa9
rude
"John made khalid eat the sandwich in a rude
manner"
i ) khalid *s actions were rude
ii) John's actions were rude
Hoyt (1989) also notes a rather interesting contrast
between double accusative sentences and semantically
equivalent sentences of which the main verb is not a
morphological causative. In the following example with
the causative verb "cause to die" the same ambiguity is
observed with respect to an accompanying adverb as in
85-86.
201
87- (SA) Hanna mawwat xaalid bi- bila9 1-
John caused to die khalid with swallowing the
mikrofilm
microfilm
"John made khalid die by swallowing the microfilm"
i) Khalid swallowed the microfilm
ii) John swallowed the microfilm
However, in the following example, with the non-causative
verb "kill", only one construal is possible for the
adverb, namely, the one in which it is related to the
subject.
88- (SA) Hanna qatal xaalid bi-bila9 1-
John killed khalid with swallowing the
mikrofilm
microfilm
"John killed khalid by swallowing the microfilm"
i ) John swallowed the microfilm
I will present first the analysis I propose for the
double accusative and show how it accounts for the facts
enumerated above. The underlying structure of the double
accusative in this analysis is the following.
202
89- IP
NPi VP
NP2 VP
0
NPa
In this structure the abstract verbal element CAUSE
subcategorizes for a VP and an NP object to which it
assigns a theta role. The surface structure
corresponding to 89 is derived by a single application of
head movement adjoining the lower verb to the morpheme
CAUSE.
90- IP
/ \
NPi VP
/ I \
V NPa VP
/ \ / \
Vi CAUSE V NPa
et
203
In the configuration above, both NPa and NPa are
governed, either by a lexical verb or by the trace of a
verb and are, in this way, assigned accusative Case.
Within the binding theory assumed here
(cf. Williams (1987a, 1989) ) the binding facts with
respect to 89 are directly accounted for. The basic
principle guiding this approach to binding may be roughly
stated as follows:
Two elements X and Y may be anaphorically related if
they are coarguments (or coarguments of
coarguments) .
In 89 NPi and NPa are coa rguments

c of the verb CAUSE and
may, therefore, be anaphorically related. The same holds
with respect to NPa and NPa as indicated by the arrows in
89. Since NPi and NPa are not coarguments they may not
be anaphorically related. Note that the two verbs in 89
constitute two separate thematic (argument-) complexes
and therefore no ungrammaticality results from the fact
that NPa is assigned a theta-role by both verbs a t the
same time. This is a standard case of what are known as
structures of secondary predication exemplified in
English by the following sentences.
91- a) John considers himself sick
b) John considers Mary crazy about herself
*c) John considers Mary crazy about himself
204
These examples have the following structure,
92- IP
/ \
NPi VP
/ I \
V NPa AP
I / \
consider A (NPa )
In 91c NPi cannot be the antecedent of NPa just as in the
Arabic double accusatives above. The other options for
anaphoric construal (NPi -NP2 , NPa -NPa ) are possible as in
91 a,b.
The cliticization facts are also directly accounted
for. Thus, as far as cliticization phenomena in Hebrew
and Arabic are concerned, a pronoun may only be
cliticized onto a lexical head (V,P,N) directly governing
it. In the configuration in question (89,90) only NPa is
governed by the verb CAUSE-V. Obviously, NPa has no
governing lexical verb to be attached to. The contrast
between 83 b and ç is, thus, explained. As pointed out
to me by Lisa Selkirk, it is interesting to note that
this explanation of cliticization possibilities in
causative constructions fits nicely with the general view
of cliticization advocated in this work. Cliticization,
205
at least in the constructions under investigation, is a
"late" process which belongs, most probably, to the
morphology. Thus, the head movement operation assumed
here as part of the derivation of causatives provides the
input, the necessary lexical head, which must be
available for cliticization to take place.
The facts about the ambiguity in the interpretation
of adverbs are also directly accounted for. An adverb
may be attached a t each one of the VP nodes as follows,
each construal corresponding to one of the possible
interpretations .
93- IP
/ \
NPt VP
/ I Adv
V NPa VP
I / \*'"'"''Xdv
CAUSE V NPa
Clearly, as the pair of examples below shows, a reversal
of order of the two accusative NP*s is completely out of
question (under the same interpretation) . This fact is
naturally accounted for given the hierarchical
configuration assigned to this construction under the
present analysis. In this configuration each of the two
206
NP's is governed by a different verbal head, the two
verbal heads being strictly linearly ordered with respect
to each other.
94- a ) (A) Sarraba zaid-un l-?awlaad-a 1-
made-drink Zaid-Nom the kids -Acc the-
Haliib-a
milk -Acc
"Zaid made the kids drink the milk”
*b) (A) Sarraba zaid-un 1-Haliib-a
made-drink Zaid-Nom the milk -Acc
1- ?awlaad-a
the kids -Acc
Thus, the analysis of the double accusative is rather
straightforward and simple. It relies crucially,
however, on the assumption of a "double-VP"
configuration, the two VP*s being headed by the morpheme
CAUSE and a verbal form respectively. The surface
causative verb must be, on this analysis, the output of a
syntactic derivation. If this assumption is made, then
all of the observed facts are explained.
We may turn now to the dative causative. Here a
rather striking observation has been made by a number of
authors with respect to a certain semantic difference
between the double
207
accusative and the dative. Thus, Cowell (1964) reports
as follows:
"In some cases.... the first object of the causative
may be replaced by a la-phrase and put after the
remaining object: Sawwaf jeruHak le-ddaktoor "show
your wounds to the doctor” samme9 darsak la-?abuuk
"recite your lesson to your father”. The use of a
prepositional complement with the causative in lieu
of a first object generally implies a certain
idiomatic specialization with respect to the
underlying simple verb: samma9 meaning "to recite”,
kattab meaning "to dictate" etc." (Pg. 241)
Similar facts are observed by Mushaweh (1986) for
standard Arabic as well as Syrian Arabic. Thus, while
the interpretation associated with the double accusative
is strictly compositional and can be paraphrased as
"make, or force, someone to do something”, a dative
construction may receive idiomatic and derived meanings.
On the basis of this observation Mushaweh concludes that
the two variants of causative constructions couldn't be
transformationally related, that is, they couldn't be
derived from the same underlying structure. This seems
to be the right conclusion. However, there is no need to
assume that the two variants are completely unrelated.
They are visibly related, in fact identical, as far as
208
the morphology of the main verb is concerned, and they
are certainly related also in interpretation. As Mushaweh
(1986) notes, a dative causative may also have the
strictly compositional interpretation in addition to the
idiomatic and derived interpretation.
One might say that the ability to undergo semantic
drift is the property "par excellence" of lexical items.
If this is so then one couldn't avoid the conclusion that
the causative verbs occurring in dative constructions are
single lexical items inserted, as they are, at the level
of NP-structure. But this seems to be a fairly natural
possibility. That is, if the morpheme CAUSE and the
whole array of verbal forms are available in the lexicon
anyway, then there is no reason to expect that the option
for them to be combined in the lexicon, rather then in

6
the syntax, shouldn't be open . If this is indeed the
case then new lexical items are formed which may freely
assume derived and idiomatic meanings. Thus, the
resulting picture seems to be that in Arabic both options

7
are available . This implies that the dative causative
construction involves a single verbal head and a single
VP.
As to the effect of the lexical operation by which a
causative verb is derived, this may be described as
consisting of the two following components: i ) addition
of an external agent theta-role, ii) internalization of
209
the external theta-role of the verbal form. The
resulting verb has only one Accusative Case feature to
assign and therefore, in order to allow an additional
object to occur, a special mechanism is invoked by which
a "dummy" Case marker li is inserted. Similar facts seem
to hold crosslinguistically. (for similar observations
see a discussion of Romance causatives in Zubizarreta
(1985)).
The structure of a dative causative construction may
therefore look as follows,
95- IP
/ \
NPi VP
/ I \
CAUSE-V NPa li-NPa
The facts observed at the beginning of this discussion
with respect to the dative construction may now be
directly accounted for. Thus, NPi and NPa can be
anaphorically related simply by virtue of being
coarguments. As to the cliticization facts, NPa in the
configuration above is governed by the lexical head
CAUSE-V and can be cliticized onto it (cf. 69).
Note that any analysis which derives both variants
of causative constructions from an underlying structure
210
involving a bi-clausal (or "bi-VP") configuration can
provide no explanation for the differences between the
two. Such an analysis is in fact proposed in Hoyt
(1989), based on ideas of Larson (1988) for the analysis
of the double object construction in English. The
following are the D and S-structures proposed by hoyt
(1989) for the two causative variants.
96- a) Double Accusative: D-structure
IP
/ \
I'
/ \
I VPi
/ \
NPi V
/ \
V VP2
I / \
CAUSE NPa V’
/ \
V NPa
C-C-C
(verbal root)
211
96- b) Double accusative: S-Structure
IP
/ \
NPi » I'
/ \
I VPi
1 / \
CAUSE-V ei V'
/ \
V VP2
1 / \
e NP2 v
/ \
V NPa
212
9 7 - a ) Dative Causative: D-Structure
IP
/ \
I’
/ \
I VP
/ \
NPi V
/ \
V VP
I / \
CAUSE V
/ \
V NPa
/ \
V NPa
I
C-C-C
213
97- b) Dative Causative: S-Structure
IP
/ \
NPi i I’
/ \
I VPi
I / \
CAUSE-V ei V’
/ \
V VP2
1 / \
e NPa j V'
/ \
V1 PP
/ \ / \
V NP P np2
1 1 1
e ej la
There is no need to go into the derivational
complexities of this analysis. In any case, it is hard
to see how the different underlying and surface
structures attributed to the two variants under this
analysis could correspond to the differences in
214
interpretation observed earlier, in particular, the
difference between a compositional and an idiomatic
interpretation observed by Cowell (1964). On Hoyt's
analysis, the difference between the two causative
variants would correspond to the difference between the
two corresponding variants involving a dative shift verb
in Arabic or in English ("give Mary a book" vs. "give a
book to Mary") . However, if there is any difference in
meaning between two such variants, it certainly does not
correspond to the idiomatic vs. compositional difference
observed above. In any case, on Hoyt's (1989) approach
one would expect to find at least a certain regularity in
the meaning difference between the two variants, a
regularity which would be due, on this approach, to a
possible semantic contribution of the preposition la
(to). Such a regularity, however, is non-existent.
Thus, at least in this respect, the analysis
proposed here is to be preferred. Hoyt's analysis makes
also the claim that the two causative variants have the
same hierarchical structure, namely, both variants have a
bi-clausal structure and therefore it should be expected
that some of the empirical facts that may be explained
on the basis of this structural property will be observed
for both variants. The main relevant fact here is the
one concerning adverbial modification. The ambiguities
observed by Hoyt (1989) (cf. examples 85-87) certainly
215
support a bi-clausal (or "bi-VP") analysis, but these
ambiguities, as Hoyt fails to note, are observed only for
the double accusative and not for the dative
construction. Thus, the following dative causative
example allows only for the interpretation in which Hanna
is using a knife to make khalid eat the sandwich (compare
with 85) . An interpretation in which khalid is using
the knife is not available.
98- (SA) Hanna ?akkal 1-sandwiS la-xaalid bi-
John made eat the sandwich to khalid with-
ssikiini
the knife
"John made khalid eat the sandwich with the knife"
The analysis proposed here makes the right prediction and
is, again, to be preferred.
The main conclusion of the discussion so far seems
to be the following. In the double accusative
construction the causative verb is the output of a
syntactic process whereas in the dative causative the
main verb is the output of a lexical process. This
distinction between processes at the lexical as opposed
to the syntactic level was shown to be crucial for an
explanation of the difference between the two variants as
well as the properties internal to each one.
216
The implications of this conclusion with respect to
the analysis of Action Nominalisation constructions may
now become clear. Here too, both causative variants are
attested as in the following double accusative/dative
pairs .
99- a) (A) ta91iim zaid-en al- qasm-a haad'a 1-
teaching zaid-Gen the class-ACC this the
kittab-a
book -Acc
"Zaid's teaching of this book to the class"
b) (A) ta91iim zaid-en haad'a 1-kittab-a li-1-
teaching zaid-Gen this the book-Acc to the
qasm-i
class-Dat
"Zaid's teaching of this book to the class"
100- a ) (A) taSriib hind-en al- ?awlaad-a al-
making-drink hind-Gen the children-Acc the
Haliib-a
milk -Acc
"Hind's making of the children drink the milk"
b) (A) taSriib hind-en al-Haliib-a li- 1-
making-drink hind-Gen the milk-Acc to the
?awlaad-i
children-Dat
"Hind's making of the children drink the milk"
217
It follows from the analysis argued for above that at
least the double accusative variants (the a examples) in
99,100 must involve a syntactic derivation of which the
head noun is an output. The head nouns in 99,100 may be
viewed as the combination of the three morphemes NOM,
CAUSE, and V. However, at least for the case of double
accusatives, the combination CAUSE-V must be viewed as
syntactically derived, that is, it must have at least the
sub-structure 101a and therefore may only be assumed to
have the underlying structure 101b.
101-a) VP b) NP
/ 1 \ / \
V NP 2 VP NPi N*
1 / \ / \
CAUSE V NP 3 N VP
1 / 1 \
NOM V NP2 VP
1 / \
CA USE V NPa
It 1nay therefore be concluded that, since causative
constructions of the double accusative variant occur in
Action Nominalizations , the option of a syntactic
derivation for Action Nominalization constructions must
be assumed to be open in principle and therefore there is
218
good reason to assume that this is, indeed, the actual
derivation of active Action Nominalizations in general,
given the empirical advantages of such an assumption.
Some of the more basic facts about Action
Nominalizations involving the double accusative receive a
straightforward account given an underlying structure
such as 101b, whereas such facts would remain unexplained
under any lexicalist approach to these constructions.
One such fact concerns the order of the two accusative
NP's in 101b. A reversal of order of NPa and NPa in this
configuration is strictly ungrammatical (under the same
interpretation) as in the following pair corresponding to
the grammatical 100a and 101a respectively.
102-*a) ( A ) taOliim zayd-en haad'a 1-kitaab-a
teaching zaid-Gen this the book -Acc
al- qasm-a
the class-Acc
*b) ( A ) taSriib hind-en al-Haliib-a al-
making-drink hind-Gen the milk -Acc the
?awlaad-a
children-Acc
A reversal of order of NPa and NPa is possible only if
the dative option is taken. These word order facts are
identical to the facts attested in the corresponding
219
sentential structures (compare with 94) and receive a
natural explanation if the same underlying configuration
is assumed. A lexicalist analysis of 99a and 100a might
assign them something like the following configuration
and would have nothing to say about the respective order
of NPa and NPa .
103- NP
/ \
NPi N*
/ I \
N NPa NPa
Another basic fact which follows naturally from the
present analysis concerns the subjectless variants of 99a
and 100a. Generally, in the absence of a subject no
accusative Case may be assigned to the object, which
therefore must surface in a bound genitive construction
with the head noun. Note now, that in the case of double
accusative Action Nominalizations only NPa may be
affected by this process, that is, only NPa may appear as
a genitive NP in the absence of a subject.
104- a ) (A) ta91iim al-qasm -i haad'a 1-kitaab-a
teaching the class-Gen this the book -Acc
"teaching the class this book"
220
*b) (A) ta91iim haad'a 1-kitaab-i al-qasm -a
teaching this the book -Gen the class-Acc
105- a ) (A) taSriib al-?awlaad -i al-Haliib-a
making-drink the children-Gen the milk -Acc
"making the kids drink the milk”
*b) (A) taSriib al-Haliib-i al-?awlaad -a
making-drink the milk -Gen the children-Acc
A detailed discussion of the phenomenon by which the
absence of a subject results in the impossibility of
assigning accusative Case will be undertaken in section
3.3.4. It is sufficient for the present argument to note
that a lexicalist type of configuration such as the one
in 103 can provide no explanation as to why only NPz
should be affected by the absence of a subject. In that
configuration no distinction can be made between NPz and
NPa (Note that this issue is clearly different from that
concerning the configurational distinction between
subject and object which was discussed earlier). On the
other hand, in a configuration such as 101b NPz and NPa
are governed by two different verbal heads and there is a
clear asymmetry in their configurational positioning
with respect to each other. This makes it possible for
whatever principles are involved in the process in
question to distinguish between the two NP's and to
affect NPa while "ignoring” NPa .
221
3.3 Subjectless Action Nominalizations
3.3.1 Introduction
Subjectless Action Nominalizations are those
constructions in which the (logical) subject, or the
external argument of the corresponding active verb, is
missing as in the following examples.
106- a ) (H) ktivat ha-avoda
writing the work
"the writing of the work”
b) (A) ?aklu ttufaaHat-i
In this introductory section I will first review some of
the data which seems to point towards a mixed verbo-
nominal nature and, therefore, motivate a syntactic
analysis corresponding to the one suggested for active
Action Nominalizations in the previous section.
The first relevant fact has to do with the
occurrence of adverbs.
222
107- a ) (H) ktivat ha-avoda leat
writing the work slowly
b ) (A) ?aklu ttufaaHat-i bi-sur9a
eating the apple-Gen quickly
Thus, both the Hebrew and the Arabic variants allow
for the occurrence of adverbs. In this respect we may
also observe the following data from Hebrew,
108- a ) (H) bniyat ha-binyan tox Svuayim
construction the building within two weeks
"The construction of the building in two weeks”
b ) (H) harigat ha-nasi be-ofen pitomi
killing the president in manner sudden
"Killing the president suddenly"
c ) (H) harisat ha-ir be-ofen axzari
destruction the city in manner cruel
"Destroying the city cruelly"
These examples should be contrasted with the "bare"
nominals in the pairs below.
109- a ) (H) ha-bniya
the construction
*b) (H) ha-bniya tox Svuayim
the construction within two weeks
223
110- a ) (H) ha-hariga
the killing
*b) (H) ha-hariga be-ofen pitomi
the killing in manner sudden
111- a ) (H) ha-harisa
the destruction
*b) (H) ha-harisa be-o£en axzari
the destruction in cruel manner
The carefully chosen verbs corresponding to the nouns in
108-111 are all strictly transitive. Obviously, like any
other verb they may have the corresponding de verbal nouns
which appear in 109a-llla. These nouns, however, may not
cooccur with an adverb alone (109b-lllb) , simply because
of a general restriction on the distribution of adverbs,
and they obviously do not require an object like their
corresponding verb. Thus, the grammaticality of 108 a-c
could only be explained by invoking a more complex
syntactic structure in which a VP node makes it possible
for an adverb to occur. Conversely, the ungrammaticality
of 109b-lllb could also be explained as involving
structures in which the lexical requirements of an
underlying verb for an object is not satisfied as it is
in 108 . Thus, nouns derived from non-transitive verbs
or optionally transitive verbs admit adverbs, without
"asking for" an object, as in the following examples.
224
112- a) (H) ha-halixa leat nimSexet z m a n rav
the walking slowly " t a k e s a long time"
"walking slowly t a k e s a long time"
b) (H) ha-axila bi-mehirut goremet keeve
the eating quickly causes aches
beten
stomach
" E a t i n g quickly causes stomach a c h e s "
Another verbal c h a r a c t e r i s t i c of Hebrew subjectless
Action Nominalizations r e l a t e s to the d i s t r i b u t i o n o f by-
phrases. Unlike E n g l i s h , by-phrases in Hebrew cannot
occur in what we may r e f e r to as " p u r e " nouns. T h u s , the
Hebrew counterparts of 1 1 3 are s t r i c t l y excluded.
1 1 3 - a ) a f i l m by Felini
b ) a play by Shakespeare
114-*a) ( H ) seret al-yede felini
film by Felini
* b ) ( H ) maxaze5 al-yede Shakespeare
Play by Shakespeare
By-phrases do occur in p a s s i v e sentences (115) and
s u b j e c t l e s s Action N o m i n a l i z a t i o n s (116).
225
115- (H) ha-uga ne ex la al-yede Dan
the cake was eaten by Dan
116- (H) axilat ha-uga al-yede Dan
eating the cake by Dan
"The eating of the cake by Dan"
Thus, in this respect subjectless Nominalizations pattern
with sentences and differ from pure nouns.
The probably most compelling argument in favour of a
syntactic analysis of subjectless Action Nominalizations,
along the same lines suggested earlier for their active
counterparts, comes from the interaction between these
constructions and the syntax of Arabic causative
constructions. As was observed in the last section
( 3 .2. 2. 4), causative constructions do occur in
subjectless Action Nominalizations in both the double
accusative and the dative variants. I will repeat here
the relevant examples for convenience.
117- a) (A) ta91iim 1-qasm -i haad'a 1-kitaab-a
teaching the class-Gen this the book -Acc
"The teaching of this book to the class"
b) (A) ta91iim haad'a 1-kitaab-i li- 1- qasm-i
teaching this the book -Gen to the class-Dat
"The teaching of this book to the class"
226
* c ) (A) ta91iim haad'a 1-kitaab-i 1- qasm-a
teaching this the book -Gen the class-Acc
118- a ) (A) taSriib l-?awlaad-i 1-Haliib-a
making-drink the kids -Gen the milk-Acc
"Making the kids drink the milk"
b) (A) taSriib 1-Haliib-i li- l-?awlaad-i
making-drink the milk-Gen to the kids -Dat
"Making the kids drink the milk"
* c ) (A) taSriib 1-Haliib-i l-?awlaad-a
making-drink the milk-Gen the kids -Acc
looking simply at the surface of things, if Accusative
Case marking is taken to be an indication of government
by a verbal element (the a examples), then we are led,
again, to the assumption that some such element is
under lyingly available. But this implies that a
syntactic derivation must be postulated by which the
actual nominal form of the head is derived. Such an
assumption receives further support from the ç examples
above which show that a simple reversal of order of the
two NP's is impossible. The same restriction was shown
to apply in the corresponding sentential structures. A
reversal of order is possible only in the dative variant
(the b examples). It would therefore seem necessary to
assume the same type of causative substructure for these
constructions as the one suggested earlier for causative
227
constructions occurring in sentences and in active Action
Nominalisations. This will make possible a unified
account of the same phenomenon across all construction
types, certainly a desirable result.
I assume that it is justified, on the basis of the
above considerations, to assunie an analysis of
subjectless Action Nominalisations which postulates the
same underlying structure as the one suggested for active
Action Nominalisations in section | 3.2. Obviously, the
analysis will have to prove capable of accounting for
further facts. Among these f ctsi an extremely
intriguing one relates to the phenomenon of control in
subjectless Action Nominalisations in Arabic already
alluded to in this chapter. Typical examples of control
structures in Arabic are the following.
119- a) (A) Haawala zaid-un | naql -a 1-
tried zaid-Nom trànsportation-Acc the
kitaab-a ?ila bayruut
book -Acc to beirut
"Zaid tried to transport the book to beirut”

1
b) (A) Haawala said-un al- ?intiqaal-a ?ila
tried said-Nom the transportation-Acc to
bayruut
beirut
"Zaid tried to be transported to beirut”
228
As the translations of these examples suggest, the
control relationship established between the higher
subject Zaid and the unexpressed subject of the Action
Nominalizations is not sensitive to thematic role
distinctions. Here we might make use of a distinction
between two types of control relations discussed i n
Williams (1985) , namely, the distinction between control
of a theta-role and control of a position. Control of a
theta-role is exemplified in English by pairs such as the
following (from Williams (1987)).
120- a) John underwent an operation
b) John performed an operation
In 120a the verb "undergo" dictates a control by "John"
of the theme of "operation" whereas in 120b the verb
"perform" dictates a control of the agent of "operation"
by "John".
Standard cases of obligatory control of infinitival
complements lack this kind of sensitivity as in the
following pair.
121- a) John wants to see the movie
b) John wants t o be seen
121 a,b are, thus, cases of control of a position. The
229
Arabie examples in 119 are cases of control of a position
too. Standard syntactic approaches to control assume it
to involve some sort of anaphoric relationship between
the controlling NP and an empty NP (mostly PRO, sometimes
pro) occupying the subject position of an infinitival
clause. It is hard to see how such an approach may be
applied to the Arabic construction given the assumed
distributional properties of PRO and what we know about
the syntax of Action Nominalizations . The main
difficulty here has to do with the assignment of Case in
these constructions, in particular, the fact the object
in such a construction is in the bound genitive
configuration with the head. It therefore must be
assumed to be situated in the specifier position and
would therefore leave no place for an empty NP, either
PRO or pro, to occur. Moreover, any treatment of the
control issue should provide an explanation for the
observed difference between Hebrew and Arabic. Thus,
given that control of Action Nominalizations is possible
in Arabic but not in Hebrew and the apparent identity in
syntactic structure of Action Nominalization
constructions in the two languages, it is plausible to
assume that this difference between the two languages
couldn't be explained on the basis of structural
distinctions, or distinctions having to do the
distribution of elements which occupy structural
230
positions but rather, as will be suggested here, on the
basis of distinctions made at the lexical level. The
relevant lexical distinctions are, to be sure, reflected
in the syntactic configuration but this while observing
the assumed structural similarity. Such an outcome would
be natural, given that the phenomenon of obligatory
control is typically associated with the properties of
individual verbs. This issue will be addressed in
section 3. 3. 4. 3.
Another fact concerning subjectless Action
Nominalizations has to do with the, already mentioned,
possibilities of Case marking in these constructions,
namely, the fact that in the absence of a subject only
genitive Case may be assigned to the remaining NP (or
NPa in the corresponding subjectless double accusative
Action Nominalization) . This raises the natural
assumption that something like a process of passivization
may be involved in the derivation of subjectless Action
Nominalizations. The following section provides some
evidence against such an assumption.
231
3.3.2 Argumenta Against a Passivization Analysis
This section is concerned with contrasts as in the
following pairs in which the a examples show accusative
Case marking on the (logical) object NP whereas in the b
examples it is in a bound genitive construction with the
head noun.
122- *a) (H) ha-harisa et ha-ir
the destruction OM the city
b) (H) harisat ha-ir
"The destruction of the city"
123- *a) (A) ?aklu ttufaaHat-a
eating the apple-Acc
Given the analysis proposed in section 3.2, in
particular, the assumption that the underlying structure
of Action Nominalization constructions contains a VP
headed by the related verb, an obvious possible
explanation for this contrast comes to mind, namely, that
the underlying verb in these constructions has undergone
a lexical process of passivization and is therefore
232
deprived of its ability to assign accusative Case to its
object. Under such a view, the relevant steps in the
derivation of 122b 123b would be roughly as follows.
124- a) DP b) DP
/ \ / \
D NP D NP
1 / \ 1 / \
POSS N’ ====> POSS NPi N*
/ \ ! \
N VP N VP
1 / \ 1 / \
NOM Vr NPi NOM V ei
In 124a no Case is available for NPi . Therefore, the
only way for it to be assigned case iss by moving to the
specifier position in which it can be assigned Case by
the morpheme POSS at the head-* of- DP position in the
manner usual for bound genitives. This is then the
passivization analysis. In the following subsections I
will present four different arguments against this
suggestion, three of which rely on data from Hebrew and
one on data from Arabic.
233
3. 3. 2.1 Verbs Which Don't Passivize
There are verbs in Hebrew which may figure as
transitive verbs in active sentences but cannot be
passivized. The two following pairs show the verbs calax
(cross) and hitxil (begin,start) respectively. 125a and
126a are the active sentences and their b counterparts
are the ungrammatical ]passives.
125- a) (H) ha-cava calax et ha-teala
the army crossed OM the canal
*b) (H) ha-teala niclexa al-yede ha-tsava
the canal was crossed by the army
126- a ) (H) ha-saxkan hitxil et ha-hacaga
the actor started OM the play
*b) (H) ha-hacaga hutxela al yede ha-saxkan
the play was started by the actor
Naturally, one would like to know why these verbs resist
passivization. It is hard to see any morphological or
phonological reason for it. Thus, for example, the verbs
Salax (shalakh) (send) and salax (pardon) each form a
phonologically minimal pair with calax (cross) and are
identical to it in their morphological pattern. Both
verbs, however, appear in the passive forms niSlax (was
sent) and nislax (was pardoned) respectively. Neither
234
can one find any semantic explanation for this fact.
Thus, a verb like hifsik (stop), which belongs to the
same semantic class of verbs as hitxil (begin, start)
(aspectuals) and manifests the same variety of uses
(transitive, raising constructions etc. ) and has also
the same morphology, can be passivized as in the
following pair of examples.
127- a) (H) ha-saxkan hifsik et ha-hacaga
the actor stopped OM the play
b) (H) ha-hacaga hufseka al-yede ha-saxkan
the play was stopped by the actor
It is hard to see in what sense two verbs like calax
(cross) and hitxil (begin, start) could be claimed to
belong to the same class, except in the sense of not
being able to undergo passivization. It seems,
therefore, that this is an arbitrary lexical property, as
lexical properties happen to be, a fact which might be
expressed by simply attributing to the lexical entries of
these verbs the feature [-PASSIVE] . Assuming the
analysis proposed above for "passive" Action
Nominalisations, it is these lexical items that would be
lexically inserted in tree structures such as 124a.
Having the feature [-PASSIVE] , the "passive" Action
Nominalizations of these verbs are predicted to be bad in
235
the same way that the passive sentences 125b and 126b
are. The following pair of examples shows this
prediction to be wrong. Crucially, the a and the b
examples allow the occurrence of an adverb and a by-
phrase, respectively, and therefore, by the logic of the
argument so far, must be syntactically derived.
128- a) (H) clixat ha-teala bi-zehirut
crossing the canal carefully
"The crossing of the canal carefully"
b) (H) hatxalat ha-hacaga al-yede ha-lahaka
starting the play by the group
"The starting of the play by the group"
At this point it might be argued that even if the feature
[-PASSIVE] is admitted to figure in the lexical
representation of these verbs, it only relates to their
morphology. If this is so, it should be possible for the
lexical items in question to undergo all the operations
involved in passivization except for the morphological
one and, in this way, for the forms in 128 to be
derived. This, however, leads to the prediction that, at
least for these verbs, it would be possible to derive
sentences with the active morphology on the verb but with
passive interpretation. This is clearly wrong.
236
129- *a) (H) ha-teala calxa
the canal crossed
b) (H) ha-hacaga hitxila
the play started
129a, of which the intended meaning would be: "The canal
was crossed", is completely impossible. 129b can only
mean: " The play started" and not: "The play was started
by someone".
3. 3 . 2. 2 Verbs Which Do Passivize
A converse argument can be made as follows. As
observed by Berman (1978), one finds i n Hebrew verbs
which take a prepositional complement, an example of
which is the verb hiSpia (influence) .
130- (H) Dan hiSpia al Dina
Dan influenced on Dina

It
"Dan influenced Dina
Interestingly, the verb in 130 can be passivized and the
complement of the preposition al (on) become its derived
subject.
237
131- (H) Dina huSpea al-yede Dan
Dina was influenced by Dan
If a passive verb like the one in 131 was the input to
the nominalization process proposed here, one would
wrongly predict the NP in 132 to be grammatical .
*132- haSpaat Dina al-yede Dan
influence Dina by Dan
Here it is important to note that the pair of examples
130-131 which shows the disappearance of a preposition in
the passive variant is far from being an isolated case.
Rather, it seems to be an instance of a general
phenomenon by which a preposition may be dropped as part
of the general process of passive formation. It is
likely that the possibility for a preposition to be
dropped has to do with its semantic contribution. That
is, a preposition can only be dropped if i t has no
semantic import and its presence is arbitrarily required
by a certain verb. The following is a list of some
verbs which show the same behavior as hiSpia (influence) .
238
133- a) verbs taking the preposition be (in) :
tipel be (take care o£,deal with) , bagad be
(betray) , baat be (kick) , tamax be (support) .
b) verbs taking the preposition al (on) :
diveax al (report) , hexlit al (decide)
hixriz al (declare) , himlic al (recommend) .
c) verbs taking the preposition le (to) :
hirSa le (allow, permit) , azar le (help, support)
kara le (call) .
The following sets of examples show the same contrasts
illustrated above for hiSpia (influence) . 134 shows the
verb tipel (deal with) with the preposition be (in) and
135 shows the verb kara (call) which takes the
preposition le (to) . '
134- a) (H) Dan tipel ba- inyan
Dan dealt with in the matter
"Dan dealt with the matter"
b) (H) ha-inyan tupal (al-yede Dan)
the matter was dealt with (by Dan)
c ) (H) ha-tipul ba -inyan
the dealing with in the matter
"The dealing with the matter"
239
*d) (H) tipul ha-inyan
dealing with the matter
133- a ) (H) ha-memSala kar?a la -cibur
the government called for to the public
lehoSit ezra
to hand help
"The government called the public for help"
b) (H) ha-cibur nikra lehoSit ezra
the public was called for to hand help
"The public was called for help"
c ) (H) ha-kria la- cibur lehoSit ezra
the call to the public to hand help
"The call to the public for help"
*d) (H) kri'at ha-cibur lehoSit ezra
call the public to hand help
The b examples above show passive sentences in which the
prepositions be and l e respectively are dropped. The ç
examples are the corresponding Action Nominalizations in
which the preposition is retained while in the
ungrammatical d examples it is missing.
240
3. 3. 2. 3 Anaphora
The passivization analysis faces a serious problem
with structures involving anaphors.
136 -*a) (H) acmo nira
himself was seen
b) (H) reiyat acmo be-ofen barur
seeing himself in manner clear
"Seeing himself clearly"
Assuming the analysis in 124, in particular, the fact
that it contains a VP, we would rather expect 136b to be
bad, in the same way that a passive sentence with an
anaphor in subject position (136a) is. This is, however,
wrong, as the above pair shows.
3. 3 . 2 . 4 Control Structures in Arabic
The already mentioned phenomenon of control of
Arabic Action Nominalizations raises another serious
problem for a passivization analysis. The argument here
is strikingly simple. Consider the following example.
241
137- (A) yuriidu zaid-un ?akl-a ttufaaHat-i
wants zaid-Nom eating-Acc the apple-Gen
"Zaid wants to eat the apple"
In this example the object "the apple" is assigned, and
can only be assigned, genitive Case. However, 137 can
only mean that "Zaid wants to eat the apple". It is hard
to imagine what would be the paraphrase of a passive
construal of this sentence. It certainly doesn't have
the interpretation: "Zaid wants the apple to be eaten"
(in which, in any case, zaid is not a controller) . In
this, 137 is identical to the corresponding English
infinitive in which an indisputably active form of the
verb eat occurs.
138- John wants to eat the apple
It is therefore hard to see what it would mean to say
that the embedded subjectless Action Nominalization in
137 is the output of a passivization process. In fact,
to the extent that passivization means the non-assignment
of the external theta role of a verb, it is possible to
develop a detailed argument showing that this is not
what happens in subjectless Action Nominalizations . To
this extent, the argument in the next section should be
viewed as completing the argument of this section.
242
3.3.3 Burzio* s Generalization ?
The impossibility of accusative Case marking in the
absence of a lexical subject is strongly reminiscent of a
general phenomenon observed among others by Burzio (1981)
and formulated by him as the following generalization.
139 - Burzio* s Generalization
If a verb does not assign an external theta role
then it does not assign accusative Case.
Although the statement in 139 was originally put forth as
a simple observation about the individual properties of
lexical items, it has been suggested by a number of
authors (e.g. Larson (1988)) to assign it the status of
an active principle of the grammar which may be taken to
participate in different grammatical processes (e.g.
passivization, dative shift). The issue of Hebrew (and,
therefore, Arabic as well) Action Nominalizations did not
remain unaffected by this trend. Thus, generalizing
Burzio *s generalization to both verbs and nouns, the
suggestion has been made by Ritter (1986) , and adopted by
Doron (1989) , that it is exactly this generalization
which is at work in the Action Nominalization
constructions under consideration, making it impossible
for the head noun to assign accusative Case in the
243
absence of a subject. It is the purpose of this section
to show that this view is wrong.
Note first that it is not just the absence of a
lexically realized subject which is here at stake.
Obviously, the absence of a lexical subject in a standard
infinitive is not sufficient to trigger the application
of Burzio’s generalization.
140- a ) (H) Dan roce leexol et ha-tapuax
Dan wants to eat OM the apple
*b) (H) Dan roce leexol ha-tapuax
Dan wants to eat the apple
Rather , it is the non assignment of an external theta-
role which would trigger the application of the
generalization regardless of whether the theta-role would
be assigned to a lexically overt or non overt NP. Thus,
an argument intended to show that burzio's
generalization is not at work in subjectless Action
Nominalizations would have to proceed in two steps as
follows. First, to show that the theta-role which would
normally be the external theta-role of the verb or noun
in question is assigned. Second, to show that it is
assigned as an external theta-role (having the status of
an external theta-role) and, therefore, externally.
Here, again, the control phenomenon associated with
244
Arabie subjectless Action Nominalisations proves to be
extremely helpful. We may, again, observe the following
standard example.
141- (A) yuriidu zaid-un ?akl-a ttufaaHat-i
wants zaid-Nom eating-Acc the apple-Gen
"Zaid wants to eat the apple"
It seems that the agent theta-role of ?akl (eating) in
this example is assigned. That is, it is syntactically
available for participation in the control relation and
therefore must be assumed to be present in some form or
another. Moreover, the controlled agent theta-role is
also syntactically present for the purposes of binding,
serving as an antecedent to an anaphor as in the
following example.
142- (A) yuriidu zaid-un qatl-a nafs-i -hi
wants zaid-Nom killing-Acc self-Gen-his
"Zaid wants to kill himself"
For a theory which views control as a relationship
between a controlling NP and a PRO subject, the agent
theta-role would have to be assigned to an ungoverned PRO
and, therefore, necessarily, structurally external.
For a , possible, theta-theoretic analysis of control
245
in which a control relationship is established between a
controlling NP and a thematic role, a minimal
requirement, in order to avoid ambiguity in the control
relation, would be for the controlled theta-role to be
assigned a special status, that is, to be distinguished
from other theta-roles for which the noun or verb in
question is lexically specified. Within such an
approach, the only available distinction between theta-i
roles is the external/ internal distinction (note that, lin
any case, it is in terms of this distinction that
Burzio's generalization is formulated) and, for a theta-
role to be unique in status, it would have to be external
since there can, in principle, be more than one internal
theta-role but only one external. Clearly, the control
relationship observed in arabic Action Nominalizations is
unambiguous and, as demonstrated by the following pair,I
insensitive to thematic distinctions but only to I
"position" .
143- a ) (A) Sajja9t-u zaid-an 9ala 1- ?akl -i
encouraged-I zaid-Acc on the eating-Dati
"I encouraged Zaid to eat"
246
b) ( A ) yuriidu zaid-un al- intiqaal-a ?ila
wants zaid-Nom the transportation-Acc to
bayruut
Beirut
"Zaid wants to be transported to beirut"
In 143a none of the two thematic arguments of ?akl
(eating) is lexically realized. However, no ambiguity
results from this situation. 143a is identical in
meaning to its English infinitival counterpart: "I
encouraged John to eat". Our argument is that for this
to be possible, the controlled theta role must have been
singled out as the external theta role. Thus, It must
be the case that the agent theta-role in this case is
assigned the status "external” and is therefore
externally assigned, given the rest of the analysis
developed so far ( a syntactic configuration involving a
VP which functions as a predicate etc. ) , and
uncontrovertial assumptions about the obligatory
assignment of (at least, external) theta-roles. Thus,
it may be safely concluded that the subject theta-role in
the control structures in question is assigned and that
it is assigned externally. Given this conclusion, the
fact remains that no accusative Case can be assigned to
the object NP in these cases and, therefore, given that
the verbs in question are, in principle, capable of
247
assigning accusative Case, it couldn't be the case that
Burzio's generalization is here at work. We must,
therefore, look for another explanation.
3.3.4 The Structure of Subjectless Action
Nominalizations
3. 3. 4 . 1 Bound Forms and Genitive Case Assignment
we consider first standard cases such as the
following.
144- a) (H) axilat ha-tapuax
eating the apple
The underlying structure of 144a, b is the following.
248
145- DP
/ \
D NP
I / \
POSS N*
/ \
N VP
I / \
NOM V NP
I I
eat the apple
Our assumptions about the properties of the nominal bound
morpheme NOM were presented in section 3. 2. 1.2. We may
repeat them here briefly.
146- a) Argument structure of NOM: (Ri , Rj )
b) Subcategorization frame of NOM: [ VP]
In the active Action Nominalization constructions
considered in section 3. 2. 1.2 NOM acted as a mediator for
the purposes of theta-role assignment. In those cases
the external theta role of VP was assigned (linked) to
the internal R-role of NOM which was further linked to
the external R role of the subject NP. In such cases the
relation between the internal R-role of NOM and the R-
249
role of the subject NP is like the one which holds
between an NP used predicatively and the NP subject of a
sentence. However, we assume that in the absence of a
subject NP, NOM may simply retain the external theta-role
of VP. This is the most crucial property of NOM. The
reference of this theta-role may either be construed as
arbitrary, or generic, or else it may enter the control
relationship as will be discussed later. Thus, in cases
in which no overt subject is present, the external theta-
role of VP is "inherited" by NOM. In such a state of
affairs NOM may be said to be an argument, that is, i t
fulfills the function of a missing NP. I would like now
to propose the following definition of a "nominal" bound
morpheme .
147- A bound morpheme is nominal if and only if
i t is an argument.
As was just seen, NOM has the ability to change status in
this respect depending on whether a subject NP is present
or not. If a subject NP is absent then NOM functions as
an argument and is therefore "nominal". This is the
situation in 145 in which the distribution of theta-roles
may be illustrated as follows.
250
148“ NP k
/ \
N’k
/ \
N k VPi
I / \
NOM V NPj
R k ,Ri i, j
My suggestion is that, if the configuration in 148 can be
characterized in these terms, then it becomes possible to
view some of its properties, in particular, the
impossibility of accusative Case assignment, as part of a
wider phenomenon having to do with the general nature of
accusative Case assignment.
Larson (1988) suggests the following assumption
about structural objective Case.
149- V Assigns objective Case in the configuration
[in fl INFL [vp V.... ]]
If it is assumed that only structural objective

s
(accusative) Case is available in general , then it is
clear that 149, as it is, does not exhaust all instances
of structural Objective Case assignment. To begin with,
it certainly does not apply to accusative Case assignment
251
in active Action Nominalizat ions . But note also that,
given the analysis of causative constructions of the
double accusative variant outlined in section 3. 2. 2. 3, an
analysis which seems to be fairly well motivated, then
the Case marking of the second accusative NP remains
uncovered by 149.
150- IP
/ \
NPi I'
/ \
I VP
/ I \
V NP 2 VP
I / \
CAUSE V NPa
Thus, in 150 only the verbal morpheme CAUSE may be viewed
as satisfying the condition in 149 and the assignment of
accusative Case to NPa remains unaccounted for. Thus, if
149 is to have any validity, it should be reformulated in

9
a more general fashion . Given the three cases
mentioned above, it may be assumed that there exists a
class of elements having the capacity to license
accusative Case marking of an object by a governing verb.
It is, of course, an interesting question what property
252
it is which is shared by all these elements and by virtue
of which they may be taken to form a class. As a first
approximation we may suggest that the most typical of
this class are elements occupying the INFL node such as
the cluster of inflectional features which constitute
the content of a finite INFL and, probably, the element
occupying the INFL node in an infinitival clause, namely,
the preposition le (to) in Hebrew and to in English. At
this preliminary stage it is at least clear that all of
the elements which may be reasonably argued to fulfil
such function are elements which belong to what may be
referred to as the A* system, that is, elements of which
the distribution is limited to what are standardly known
as A* positions. Here we are making use of the well
established distinction between A positions (argument
positions) and A* positions (nonargument positions). A
positions are positions which bear a grammatical function
such as subject of, object of etc. and it is to elements
occupying such positions that Case features and thematic
roles are assigned. Note now that at least this minimal
condition is satisfied by both NOM and CAUSE and as will
be argued in chapter 4, some elements occupying the head
of CP position may also be assumed to fulfil the same
function. In what follows I will assume that these
elements share the abstract feature +F and that it is
253
by virtue of this feature that an element can license
the assignment of accusative Case.
It must now be clarified what structural
relationship must hold between a +F element and a verb
(or a VP) in order for the assignment of Accusative Case
to be possible. For this purpose, let us first define
the relation of being in the domain of as follows,
151- For a maximal projection and a head
o/ is in the domain of iff p c-commands
and there is no head such that 0 is in
the domain of ft and c-commands TT
151 may be relativized to +F elements as follows ,
152- For a maximal projection and a +F head
is in the domain of (3 iff p c-commands
and there is no +F head V such that is i n
the domain of X and J? c-commands *fl* .
Next we may reformulate Larson's (1988) suggestion as
follows ,
153- V assigns Accusative Case in the configuration
[x • X [vp V.... ]]
Where X is +F and VP is in the domain of X.
254
153 seems to take care of the three cases considered if
INFL, NOM and CAUSE are assumed to be +F elements.
However, configurations such as 148 are an exception to
153 since the structural condition it specifies is
satisfied and the verb in question may be capable of
assigning Case as well. Note now that, given the
discussion of 148 above, there is one obvious way in
which it constitutes an exception, namely, NOM in 148 has
the status of an argument. Given the definition of a
"nominal** bound morpheme in 147, we may reformulate 153
as follows.
153- V assigns accusative Case in the configuration
[x« X [vp V.... ]] Iff.
i) X is +F
ii) VP is in the domain of X.
iii) X is non-"nominal"
As can now be seen, 148, a subjectless Action
Nominalization configuration, is a case in which the
assignment of accusative Case is not allowed. This seems
to be a natural result since, as just observed, the most
fundamental characteristic of +F elements is their
function as A* elements. Therefore, by assuming the
function of an argument NOM clearly loses the main
255
property by virtue o f which i t forms part of this class
of elements.
Thus, in the configuration 1 4 8 the o b j e c t NP cannot
be assigned accusative C a se and therefore some other
strategy o f Case assignment must be invoked. The only
other strategy available is for NP to move to the

specifier position and be assigned Case by the head noun
in a bound genitive configuration.
This is then the account proposed here for the
"Burzio's generalization effect" observed in s u b j e c t l e s s
Action Nominalizations. Obviously, it remains to be seen
whether this explanation has any general applicability.

As will be shown in chapter 5 , the same explanation may
be invoked for the occurrence of the same phenomenon in
Agent Nominalizations. The surface structure
corresponding to 1 4 8 , that is, the structure of 144 a,b

would be the following.
155- DP
/ \
D NP
/ \ / \
Nj D NPi N*
I II / \
eating POSS the N VP
apple I / \
ej V et
25 6
Note, in light of the above discussion, that the
relation between the moved NP and its trace in 155
couldn't be an A/A relation. This is because the
specifier position in this particular case is not an
argument position. This status is now assumed by NOM.
Therefore, this relation may be viewed as an A* /A
relation. In particular, the relation between the moved
NP and its trace in this case is different from the one
holding between a derived NP subject and its trace in,
say, Raising constructions. In such constructions the
derived subject is assigned a theta-role by the upward
propagation of a thematic index from a trace in a lower
position to its sister node as illustrated in Chapter 1.
Such a thematic relation would be impossible in 155.
Rather, the NP occupies the VP internal position at NP-
structure and moves to the specifier position only in
order to satisfy Case requirements.
The binding facts in the subjectless bound genitive
variant ( c f . section 3. 3. 2. 3) are also accounted for by
this analysis.
156- a) (H) reiyat acmo
seeing himself
b) (H) qatlu nafs- i -hi
killing self-Gen-him
"killing himself"
257
In these cases the external theta role assigned to NOM
and the internal theta role assigned to the anaphor are
coarguments. The binding relation here is the same as
the one which holds between a subject and an object in an
active finite sentence with the same asymmetries.
The structure of the corresponding double genitive
in Hebrew is identical in all relevant respects except,
of course, for the Sel-phrase which I assume to be
adjoined as in the following configuration. Otherwise,
the only difference between the bound and the double
genitive is the usual one, namely, that in the second the
object NP is a pronoun rather than a full NP.
157- A subjectless double genitive
DP
/ \
D NP
/ \ / \
Ni D NPj N*
II / \
POSS pron. N’ VP
/ \ / \
ei Sel-NP V ej
258
3. 3. 4 . 2 Control
Given our assumptions concerning obligatory control
as presented in chapter 1, the implications with respect
to control of Action Nominalizations i n Arabic may now
become clear. As has already been observed in this
chapter, control of subjectless Action Nominalization
constructions in Arabic is control of a "position".
Typical pairs which demonstrate this property are the
following .
158-a) (A) Haawala Zaid-un ?ixraag-a l-?awlaad-i
tried Zaid-Nom taking out-Acc the kids-Gen
miin 1- bayt-i
from the house-Dat
"Zaid tried to take the kids out of the house"
b ) (A) Haawwala Zaid-un al-xuruug miin 1-
tried Zaid-Nom the going out from the
bayt -i
house-Dat
"Zaid tried to go out of the house"
259
159-a) (A) yuriidu Zaid-un naql-a 1-
wants Zaid-Nom transportation-Acc the
kitaab-i ?ila bayruut
book -Dat to beirut
"Zaid wants to transport the book to beirut"
b) (A) yuriidu Zaid-un al-?intiqaal-a ?ila
want Zaid-Nom the transportation to
bayruut
beirut
"Zaid wants to be transported to beirut”
Whereas in the a examples above it is an agent theta role
which is controlled, in the b examples it is a theme,
even though it is the same verb which governs the
controlled constituent in both examples. This relation
is thus insensitive to thematic role distinctions and is
clearly an instance of the general phenomenon of
obligatory control. In this case, however, unlike the
case in controlled infinitives, the lexically established
control relationship holds between an argument of the
governing verb and the argument structure of NOM. Thus,
the lexical entry of the control verb "Haawala" (try)
would look roughly as follows.
260
160- Haawala (try) : Agi , Xj
< NOMi >
As is the case with the element COMP in lexical entries
which encode the control relation with infinitival
complements (see Chap. 1) , the lexical representation
above encodes a requirement that a thematic index
associated with NOM be coindexed with an argument of the
governing verb (the external argument in this case) . I
assume that no difficulties arise from a lexical
specification which makes reference to the lexical head
NOM rather than to its maximal projection DP. The
overall DP structure of the constituent in question
follows, presumably, from general principles of phrase
structure.
- In many of the control structures attested in Arabic
the controlled constituent is governed by a lexically
specified and, most likely, semantically empty
preposition. This does not raise any difficulty for the
present analysis since these prepositions do not disturb
the direct relationship between the governing verb and
NOM and may be straightforwardly accommodated into
lexical representations as shown by the examples below.
(A, B , C , etc. in the lexical entries below stand for
thematic role names such as "Agent" "Theme" etc.)
261
161- a) Tamara (order): Aj , Bi , Ck
I I
< NP> < bi NOMi >
b) ( A ) Tamarat hind-un zaid-an bi- 1 -xuruuj-i
ordered hind-Nom zaid-Acc with the leaving-Dat
"Hind ordered zaid to leave”
162- a) ? arRama (force): Aj , Bi , Ck
I I
< NP> < 9ala NOMi >
b) (A) ? ar Ram a zaid-un hind-an 9ala tark-i
forced zaid-Nom hind-Acc on leaving-Dat
1- bilaad -i
the country-Dat
"Zaid forced hind to leave the country”
163- a) tamakkana (be able): Ai , Bj
< min NOMi >
b) (A) lam yatamakkan zaid-un min qiraa?at-i
not be able (jus) zaid-Nom from reading -Dat
jjariidat -i
the newspaper-Gen
"Zaid was not able to read the newspaper”
262
As is the case with control of infinitives, the
requirement encoded by a lexical entry such as the one in
160 can only be satisfied if it is the internal argument
of NOM which is coindex d with the controlling NP. A
coindexing of the external argument of NOM with the
controlling NP will result in the following
configuration .
.164- IP
/ \
NPi i I•
/ \
VP
/ \
V NPa 1
try / \
N*i
/ \
Ni VPj
I / \
NOM Vj NPa k
Ri , Rj eat
l,k
However, with NPi and NPa coindexed, 164 is ruled out as
a simple violation of principle C of the binding theory.
263
Only coindexation with the internal theta role of NOM can
avoid such a situation and therefore no ambiguity in this
respect can ever arise.
On this view of control the variation between Hebrew
and Arabic derives from the fact that Hebrew does not
have lexical entries in which control relations are
specified in terms of NOM but, rather, only in terms of
COMP. Presumably, this variation involves a parametric
choice in such a way that one option is taken to the
exclusion of the other. That is, we should not expect to
find mixed cases of a language which has both control of
COMP and control of NOM. Viewed in this perspective,
Hebrew is an interesting case since it has structures
involving both COMP and NOM but has obligatory control

10
only with COMP .
Finally, as one would expect, given the general
analysis of Action Nominalizations advocated here,
nominalized control verbs should maintain control
properties since in the underlying structure of such
cases it would be the control verb itself which occurs.
We should therefore expect to find cases of a controlled
nominalization which is embedded under a nominalized
control verb. The following example with the nominalized
form of the control verb Haawala (try) confirms this

1 1
expectation .
264
165- (A) 9ind muHaawalat- i -him 1-faraar-a ?ila
with attempt Dat their the £light-Acc to
1-Rarb
the west
"upon their attempt to flee to the west"
265
Footnotes to Chapter III
1
This conception of NOM as an abstract rather than a
concrete morphological object having an actual
morphological form is designed to avoid possible
conflicts which may arise between the material in VP and
the nominalizing morpheme. It seems that the analysis of
Action Nominalization constructions suggested in Borer
(forthcoming) commits exactly this error. Thus, adopting
(with some modifications) the proposals of Hazout (1988) ,
an earlier version of the present analysis, Borer
(forthcoming) proposes the following configuration for an
Action Nominalization construction.
NPi
/ \
N'i
/ \
N VP
I / \
haCCaCat NPa VP
I /
the army (SPEC)
\
NPa
I
the city
Concentrating just on the morphological issue, the

element figuring as head of the NP in i is the actual
morphological form of a noun, haCCaCa. The particular
phonological shape of the C(onsonant) positions in this
word is to be filled in by a movement of a verb stem as
indicated by the arrow. Note that the form figuring as
head in i is a form particular to a certain Binyan
pattern, the hiCCiC pattern. However, not every verb has
(either a verbal or nominal) hiCCiC form. Thus, the verb
haras (destroy) has a CaCaC form, of which the
corresponding derived nominal form is C (a) CiCa, and lacks
a form in the hi-CCiC binyan. Now, if the verb stem of
haras (destroy) is to be inserted in an underlying
structure such as i a mismatch between the two elements,
the head of the construction and the verb is going to
arise.
The same problem of a possible conflict arises with
respect to argument structure and the deployment of
arguments in a tree structure such as i. This may be
illustrated with causative verbs in Arabic (see section
3. 2. 2. 4 for the analysis of Arabic causative
constructions). Thus, a causative verb, or its nominal
counterpart, cooccurs with a subject and two accusative
objects as in the following examples.
266
ii) (A) Sarraba zaid-un l-?awlaad-a 1-Haliib-a
made drink Zaid-Nom the | kids -Acc the milk-Acc
"Zaid made the kids drink the milk"
iii) (A) taSriib zaid-en l-?awlaad-a 1-Haliib-a

making-drink Zaid-Gen the kids -Acc the milk-Acc
"Zaid's making the kids < rink the milk"
Note now that under Borer's suggestion there is nothing

to avoid a situation in which the head of the
construction is a causative form| but where the VP
governed by it contains only a subject and a single
object, that is, the argument structure corresponding to
a simple transitive verb such aS| Saraba (drink).
iv) NPi
/ \
N'i
/ \
N VP
1 / \
taCCiiC NP2 VP
1 / 1 \
Zaid (SPEC) 1 V
I \
V NPa
I I
drink the milk
In fact, an underlying structure such as (iv) does not

necessarily result in an ill-formed Surface form.
Indeed, (iv) may give rise to the following perfectly
well formed sequence.
(v) (A) taSriib Zaid-en aH-Haliib-a

making-drink Zaid-Gen the| milk -Acc
"making Zaid drink the milk"
The problem is that an analysis like| (iv) makes the claim

that in this structure only two theta roles are assigned,
namely, the agent and theme of drink . The agent of
CAUSE, the "causer", is in this structure non-existent.
It is, however, a simple matter {to show that the "causer"
is there and must be there. Thus, if (v) is embedded
under a control verb as in (vi) below, it is necessarily
the "causer" which is understood) to be controlled.
267
(vi) (A) Haawala muHammad taSriib -a zaid-en
tried muhammed making-drink-Acc Zaid-Gen
al-Haliib-a
the milk -Acc
"Muhammed tried to make Zaid drink the milk"
Thus, an analysis like the one proposed by Borer

(forthcoming) allows for the occurrence of mismatch
between argument structure and surface morphology.
Argument structure is a function of verb root and a
Binyan pattern combined, rather than of the verb root
alone. A structure such as (i) above makes it impossible
to ensure a correlation between the morphological form of
the verb (and the corresponding noun) and the argument
structure as it is realized within VP. The present
analysis, as well as the original one in Hazout (1988),
provides a simple way out of this problem.
2
In fact, one may think of one more, rather trivial,
possible way to establish a relationship between the head
and its arguments, namely, by linear adjacency. But,
obviously, here too no one to one mapping is possible.
Thus, in an active form the subject must be adjacent to
the head. But, in the absence of a subject, the NP
adjacent to the head is interpreted as an object and not
as a subject.
3
The same problem arises for Doron's analysis if one
assumes a "theta-theoretic" version of the binding
theory. This is because both arguments in 44 a,b are
considered by Doron to be internal and therefore, again,
symmetrically related.
With respect to this assumption we might point out
another, rather striking, inconsistency in Doron's
analysis. It is assumed in that work that Burzio's
generalization is an active principle of the grammar
"which could be interpreted to state that any lexical
item assigning accusative Case must link its external
argument” (Pg. 12) . This principle is invoked as an
explanation for the ungrammaticality of the following
example in which a subject is missing.
*i) ha-harisa et ha-ir

the destruction OM the city
One might wonder at this point what explains the

grammaticality of 44a. Given Doron's assumption that
both arguments in that example are internal, one might
expect it to be impossible for accusative Case to be
268
assigned. However, in that example, accusative Case not
only can be assigned but, in fact, must be assigned.
*ii) harisat ha-oyev ha-ir

destruction the enemy the city
The same kind of inconsistency arises in Doron’s analysis

of the following example.
iii) harisat ha-ir Sei ha-oyev

destruction the city Sei the enemy
"The enemy's destruction of the city"
This example is assigned the following structure in which

"the enemy" in the Sel-phrase is said to be an external
argument (see section 3. 2. 2. 3 on the analysis of these
cases) .
iv) N"
/ \
N' NP
/ \ Sei the enemy
N NP
Given, again, Burzio's generalization, one might expect

it at least to be possible for accusative Case to be
assigned since the external theta-role is, in this case,
linked and the "strong" head noun harisa (destruction) is
certainly capable of assigning accusative Case to its
object. Obviously, none of this is the case. The
following example in which the object NP is preceded by
the object marker et is completely impossible.
*v) ha-harisa et ha-ir Sei ha-oyev

the destruction OM the city Sei the enemy .
4
The account presented in Borer (1984) (Pg. 80-88) is
a variant of the analysis criticized and rejected in this
section. In her account Borer gives the following
paradigm and assigns
ia the structure in ii below.
i) a) (H) ktivat Dan et ha-maamar

writing Dan OM the article
"Dan's writing of the article"
b) (H) ktivat ha-maamar Sei Dan
writing the article Sei Dan
269
c) (H) ktivat Dan Sel ha-maamar
writing Dan Sel the article
ii) N ’'i
I \
N"i N"2
/ / \
N'i et N"a
I \ I
Ni N* *3 the article
I I
writing Dan
In this configuration Dan , the (logical) subject, is the

genitive complement of the deverbal head noun "writing".
It is assumed that, since in this state of affairs no
genitive Case can be assigned to the (logical) object
"the article", "the accusative dummy Case marker" et is
inserted to assign Case to NPa . The same situation is
assumed to hold in ib with a reversal of positions
between (logical) subject and (logical) object. Here it
is the "dummy" Case marker Sei which is inserted. Thus,
Borer assumes ib to have the following structure.
iii) N"i
I \
N"i N”a
/ / \
N’i Sei N"a
/ \ I
Ni N**3 Dan
I I
writing the articles
Note first that, given the discussion in section 3. 2. 2. 3 ,

it is clear that ia and ib involve two different
configurations. Therefore, either ii or iii or both
must be the wrong structure. Next, it is assumed by
Borer that the particles Sei and et, being "dummy" Case
markers, are required to be present only for the needs of
Case marking. This implies that the Object Marker et is
a Case assigner rather then the manifestation of Case
assignment by something else, say, a verb. I assume that
this is simply wrong, that is, et is the manifestation of
government by a verb and not a Case assigner itself.
There is no need to go here into a detailed discussion of
this point. It suffices to note that the corresponding
Arabic construction, identical to the Hebrew construction
in all relevant respects, shows morphological accusative
Case marking of the object, the same Case marking that
270
would appear on the object of a verb in a regular
sentence .
iv) a) ( A ) ?akala muHammad-un ttufaaHat-a

ate muhammed-Nom the apple-Acc
“Muhammed ate the apple"
b) ( A ) ?aklu muHammad-en ttufaaHat-a
eating muhammed-Gen the apple-Acc
"Muhammed's eating of the apple"
Coming now to the more fundamental issue, note that from

the paradigm in i one important member is missing,
namely, the following, in which the intended reading is
one in which Dan is the agent of "writing" and "the
article" is the theme.
*v) (H) ktivat ha-maamar et Dan

writing the article OM Dan
Thus, the discussion in Borer (1984) implies that the

subject and the object in this construction may freely
change positions with Case markers blindly inserted to
satisfy Case requirements. From this perspective, of the
four logically possible combinations y is the missing
variant .
What is then wrong with y ? Intuitively, et is an
object marker and therefore should not precede a
"logical" subject (or agent) as it does in y. The
crucial difficulty however is the following: What counts
as an object in the configurations ii and iii ? That
is, given configurations ii and iii, in what sense is
"the article” in ia an object and Dan in y not. It is
hard to see in what sense of the notion "object" can the
NP "the article" in ii be said to be an object. It is
obviously not an object in the configurational sense,
that is, an NP governed by a verb and immediately
dominated by VP ( [NP,VP] ) which is the standard
definition of this notion in the government and binding
framework, and, in any case, at least in Borer's account,
the subject NP in ib (configuration iii) occupies exactly
the same position. Thus, in mapping a Case marker onto a
certain NP, it couldn't be any configurational
sensitivities which would avoid the case marker et from
being inserted in the way it is inserted in y. Besides,
it is far from being clear what kind of position the
(logical) object in ii occupies. Is it a complement (of
what?) or an adjunct?
We are forced then, again, to the assumption that
the insertion of et must be sensitive to logical
("subject", "object" taken as primitive notions) or
thematic distinctions rather then to configurational
ones. Here, however, we run into the same problems
271
indicated above for Doron's analysis, namely, there is
no unique, unambiguous mapping from logical or thematic
roles to Case markings. The only restriction is that
accusative Case may not be assigned in the absence of the
subject, but this is not the situation in y. Thus, a
claim that accusative Case marking is mapped onto a theme
or a (logical) object would raise the expectation that
the NP identified as such in, say, ib could be preceded
by et. This is however false.
*vi) (H) ha-ktiva et ha-maamar Sei Dan

the writng OM the article Sei Dan
At least under Borer's approach, the only difference

between ib and vi on the one hand and ia on the other is
one of order of constituents, that is, of surface
structure and not of logical and thematic structure and
therefore vi would be expected to be grammatical.
We see then that the analyses of both Borer (1984)
and Doron (1989) fail to see the crucial point that the
occurrence of et is the outcome of a certain
configurational relationship between an NP and a
governing head rather then some other type of
relationship. Neither the analysis proposed by Borer nor
the one suggested by Doron provide a configuration in
which one NP can be unambiguously singled out as an
"object” .
Given this basic shortcoming certain other facts
remain unexplained under Borer's (1984) lexicalist
approach. First, as noticed by Borer, clitic doubling is
possible only with Sei (figure iii) and impossible with
et.
vii) a) (H) ktivat -oi Sei ha-maamar i

writing-his Sei the article
"the writing of the article"
*b) (H) ktivat -oi et ha-maamari
writing-his OM the article
Given the identity of structure assigned to the a and b

examples above, there is no explanation for this contrast
in grammaticality. Obviously, it couldn't have anything
to do with the "dummy" Case markers Sei and et. This
contrast is naturally accounted for within the approach
suggested in this thesis.
Another major shortcoming of Borer's analysis has to
do with the issue of anaphoric relationships in these
constructions. Here the relevant examples are the
following.
272
viii) a) (H) reiyat a cm a Sel ha-mora
seeing herself Sel the teacher
"The teacher's seeing of herself"
*b) (H) reiyat acma et ha-mora
seeing herself OM the teacher
According to Borer, Sei has the option of being inserted

in the phonological component of the grammar. Thus, a t
the relevant level of representation (viii) a is a flat
structure which makes it possible for "the teacher" to c-
command and be the antecedent of "herself". On the other
hand, et, the object marker, is present at all levels of
representation and therefore makes it impossible for an
NP governed by it to be the antecedent of an NP outside
its own phrase. However, as noted by Borer herself (Pg.
150) , examples in which an object governed by e t serves
as an antecedent are rather common.
ix) (H) Dan hera et Dina le acma

Dan showed OM Dina to herself
Thus, again, the contrast between (viii)b and (ix)

remains unexplained in Borer's approach. It does have,
however, a fairly simple explanation under the present
approach, namely, (viii)b is an instance of the asymmetry
between a subject and a (VP internal) object, (viii) a
being a different construction. In (ix) , on the other
hand, both "Dina" and "herself" are internal arguments
and can therefore be anaphorically related with "Dina"
as antecedent. Obviously, et plays no role in this
relationship.
Thus, to conclude, Borer's analysis faces much of
the same problems faced by Doron's analysis, namely, the
distinction between a subject and an object in these
constructions must be configurational. Both analyses,
however, simply do not provide the right configuration
for this distinction to be made possible.
It should be noted a t this point that much of the
problems faced by Borer's analysis result from her
assumption of the complement approach to bound genitives.
Looking back at configuration (ii) , we may observe that,
since in this structure Dan, the subject, is assumed to
be a complement, the object "the article" must be
relegated to an adjoined position of unclear status. In
the same way, the same configuration (iii) must be
assumed for (i)b with the subject Dan in an adjoined
position. The difficulties pointed out above are thus a
direct result of this set of assumptions.
Is the complement approach to bound genitives
compatible with any variant of a syntactic approach to
Action Nominalizations? Most probably not. In a
syntactic analysis of (i)a the only element that must
273
clearly be the complement of anything is the object Dan,
which is assumed here to be the complement of a verb and
dominated by VP. Now, obviously, a VP in these
constructions must be assumed to be governed by a
nominal bound morpheme functioning as head of the
construction. Thus, in the resulting configuration there
is nowhere in which the subject could fit as the
complement of anything. It may therefore be concluded
that a complement approach to bound genitives is
incompatible with a syntactic analysis of Action
Nominalizations . In view of the striking inadequacies of
lexical approaches to Action Nominalizations and the
clear advantages of a syntactic approach, it may be
concluded that a complement approach to bound genitives
is most probably wrong.
* The account provided here relies heavily on the data

and discussion of Arabic causatives in Mushaweh (1986)
and Hoyt (1989). The analysis proposed here is, however,
radically different from those suggested in these two
works. The data provided in Cowell (1964) has also been
of great importance.
6
Note that this analysis of causative constructions,
in particular, the assumed difference of structure
between the double accusative and the dative variants is
in sharp contrast to recent well known and quite
influential treatments of causative constructions. Here
I have in mind mainly Baker's (1985,1988) work on
causatives. Baker's work clearly, and explicitly,
excludes the possibility of the two morphemes CAUSE and V
being combined in the lexicon. This position is forced
upon Baker by his assumption of UTAH, a principle from
which it follows that NP's which are assigned the same
theta-role in both structures should be hierarchically
related to each other in the same way in the underlying
configurations of both structures. However, the
evidence from Arabic concerning the differences between
the two causative variants speaks clearly against Baker’s
general approach to causatives as well as against UTAH.
For a detailed argumentation, see below my discussion of
Hoyt (1989) which is an attempt to analyze Arabic
causative constructions in the spirit of Baker's and
Larson's recent proposals.
274
7
By making here the distinction between processes at
the syntactic as opposed to the lexical level, I do not
wish to imply that the two options are freely available.
Thus, Hebrew has causatives as well but these seem to be
lexically rather then syntactically derived. Causative
verbs in Hebrew are mostly of the hi-CCiC form, examples
of which are: hi-XTiV (dictate), hi-LBiS (dress
someone) , he-?eXiL (feed) hi-STiK (make quite) , and
rather rarely of the CiCeC form as in LiMeD (teach) .
These causative verbs are clearly lexical given two main
criteria. First, it is mostly impossible to have a
causative verb followed by two accusative NP's.
*i) Dina hilbiSa et Dan et ha-sveder

Dina dressed OM Dan OM the sweater
??ii) Dina heexila et Dan et ha-tapuax
Dina fed OM Dan OM the apple
Glinert (1989) (Pg. 163) enumerates only three verbs that

may be followed by two accusative NP's, limed (teach) ,
Saal (ask a question) and heevir (take.... across...).
Second, Hebrew causative verbs may freely assume
derived and idiomatic meanings. For example, hilbiS
("dress somebody" as opposed to "make someone put
something on"), he' exil (feed), hixtiv (dictate), etc.
8
Larson (1988) assumes, in fact, that in any instance
of government of a direct object by a verb two accusative
Case markings are assigned, structural and inherent. For
the ideas suggested here to hold it must be assumed that
this is not so, but rather, only structural accusative
Case is assigned. Note that the assumption of two types
of accusative case forces Larson to the conclusion that
in Passivization the two Cases are suppressed whereas in
Dative-shift only the inherent Case is suppressed. On
the other hand, no evidence is brought by Larson to show
that the assumption of two types of accusative Case is
necessary for constructions other then Dative-shift.
8
There seems to be growing cross linguistic evidence
in favour of the idea that accusative Case assignment in
sentential constructions is a process which takes place
in a configuration involving both VP and I and which
depends, most likely, on the particular specification of
I. Emonds (1989) proposes an account along these lines
275
for the phenomenon o£ nominative objects in finnish, old
russian and languages of the baltic group (lithuanian and
latvian) . The phenomenon is that under certain
conditions the morphological nominative instead of
accusative marking is used for direct objects as well as
cognate objects and certain adverbial NPs. Relying on
work by Timberlake, Emonds notes that a necessary
condition for the application of this process in all the
languages examined is that the verb be non-finite. Thus,
the assignment of accusative Case to the object seems to
be directly dependent on distinctions such as
finite/non-f inite which are specified on I. In Emonds
words, a non-finite I is "too weak" to allow for
accusative Case to be assigned. In this case Nominative
is assigned as the "default Case".
10
Interestingly, one finds cases of "control" of an
Action Nominalisation in subject position in Hebrew.
i) reiyat acmo ba-mara xaSuva le- Dan

seeing himself in the mirror is important for Dan
In such cases "Dan" must be interpreted as the antecedent

of "acmo" (himself) in the sub jet. However, given that
Hebrew, unlike Arabic, does not have obligatory control
of Action Nominalisations figuring as complements, there
are good reasons to suspect that the two control
phenomena are of a different nature.
11
Taken from Al-Hayaat October 3 1990.
276
CHAPTER IV
INFINITIVES AND GERUNDS
4.1 Introduction
This chapter will be concerned mainly with two
construction types known within the terminology currently
used in the literature on Hebrew grammar as the Verbal
Gerund and the Nominal Gerund. It will also include some
discussion of standard infinitival clauses which seem to
be closely related to the verbal gerund. Since gerundive
and infinitival constructions of the type discussed here
are non existent in Arabic, this chapter will be
exclusively concerned with Hebrew. 1 a,b below are
typical examples of the verbal and the nominal gerund
respectively .
1- a) bi- reot Dan et Dina
with seeing Dan OM Dina
"With Dan seeing Dina"
b) Suv ha-poalim me- ha-avoda
return the workers from the work
"The return of the workers from work"
277
Introducing the verbal gerund, let us take a brief look
at some of its main characteristics. First, verbal
gerunds are most typically used as temporal adverbs as in
2.
2- bi-reot Dan et Dina hi yac?a me- ha-xeder
with seeing Dan OM Dina she left from the room
"With Dan seeing Dina she left the room"
Thus, a verbal gerundive clause may never figure as an
argument (subject, object etc.). With respect to the
elements occurring in its surface form, the initial
element in a verbal gerund is always a preposition,
interpreted temporally (before, after, "at the same time"
etc.). Immediately following the initial preposition is
an infinitival verb form, morphologically identical to
the verb form which appears in a standard infinitive
(except, of course, to the initial preposition which is
different for infinitives) . Immediately following and
adjacent to the verb is a subject ("Dan" in la) . The
subject, in turn, may be followed by any complement which
happens to be required by the verb. Thus, the surface
sequence of elements in a verbal gerund is as follows,
where XP is any complement (possibly more then one) which
is lexically required by the verb.
278
3- Verbal Gerund: P - V i m - NP - (XP)
Going next to the nominal gerund, forms such as in lb are
NP's and have the regular distribution of NP's. Thus,
the NP in lb may occur in the following sentence.
4- anaxnu mexakim le-Suv ha-poalim la- avoda
we waiting to return the workers to the work
"We are waiting for the return of the workers to
work"
Given its distributional as well as other properties
(e.g. modification by adjectives, interaction with
genitive constructions) , the initial element in lb seems
to be a noun. It is followed by an NP which may be said
to be the logical subject of the related verb and by
other, possible, prepositional complements of that verb.
Thus, the surface sequence of elements in the nominal
gerund is the following, where N stands for the
gerundive form of the verb.
5- Nominal Gerund: N - NP - (PP)
Unlike the verbal gerund in which almost any verb may
occur (the exceptions will be discussed and explained in
section 4.2), in the nominal gerund, it is only a small
279
subset of verbs of which the gerundive form may occur.
Morphologically, however, the form of the head of the
nominal gerund construction is identical to its
counterpart in the verbal gerund. In fact, and, again,
abstracting away from the initial preposition, it is the
same as the verb form appearing in a standard infinitive.
This is illustrated in the following table.
6- Infinitive Verbal Ger. Nominal Ger.
le-header be-header header
to-be absent with-being absent being absent
la-Suv im-Suv Suv
to-return with-return return
le-healem be-healem healem
to-disappear with-disappearance disappearance
1
Traditional grammarians grouped together the verbal
gerund and the infinitive under the term "Infinitive
Construct". I assume that the main motivation for this
classification has been the fact that in both cases the
infinitival verb form is "in construction" with a
preposition, this, in addition to a certain phonological
280
difference between this form and the absolute form, an
archaic biblical form which will not be discussed in this
work. More recent work (e.g. Berman (1978), Glinert
(1989)) tends to group the verbal and the nominal gerunds
together, while taking the infinitive to be a separate
form. This is obviously reflected in the terminology.
Whatever its theoretical motivations, it seems to me
that the traditional classification is basically correct.
As I will attempt to show, both the standard infinitive
and the verbal gerund involve a clausal structure. This
is, of course, the standard view of infinitives, at least
within the broad outlines of the Government and Binding
framework assumed here. It will therefore be the verbal
gerund which will enjoy more detailed attention in this
chapter. The discussion in section 4.2 will be devoted
to the verbal gerund. The nominal gerund, on the other
hand, does not involve a clausal structure. Rather, it
involves an underlying structure similar to the one
argued for in chapter 3 for Action Nominalisation
constructions and a syntactic process of nominalisation.
That is, a process of head movement applying to a verb
and attaching it to a nominal bound morpheme, GER, which
governs the verb and the VP of which it is the head. The
crucial properties of the nominal gerund may be explained
by making certain, rather minimal, assumptions about the
281
properties of GER as will become clear in the discussion
in section 4.3.
Notwithstanding my view of the correct
classification of these constructions, as just stated, I
will adhere throughout this chapter to modern terminology
which has now become standard in reference grammars
(Glinert (1989)) and descriptively oriented work (e.g.
Berman (1978)). It is obvious, however, that the issue
of correct classification derives mainly from the
(mostly) identical verbal (or deverbal) form appearing in

2
all three constructions . Thus, in all three
constructions it is the same combination of verb root and
vowel pattern which occurs. This raises the question of
whether a unified approach to all three constructions is
conceivable in which a verb form would combine with the
same underlying morpheme in all three cases. This,
however, is not necessary. Rather, what seems necessary,
and certainly plausible, is to assume that the derivation
of the nominal gerund involves a " nominal izat ion- type"
of process applying to a nominal bound morpheme and a
verb form, whereas the infinitive and the verbal gerund,
involve a clausal structure in which a verb form combines
with a non finite verbal INFL to produce an infinitival
verb form. Thus, we will assume that the morphological
entity which combines with a verb form to produce the
particular infinitival shape is doubly lexically listed
282
as either a noun or a verbal inflectional element. The
lexically listed nominal element GER may be attributed
certain properties related to its thematic, or argument
structure, properties which are generally typical of
nouns. As will be seen, a similar situation holds with
respect to participial constructions. In chapter 5,
which is devoted to participial constructions, a more
detailed discussion of this type of problem is
undertaken.
The situation just described is similar to the one
existing in English, where one finds a variety of forms
and constructions involving the morpheme -ING. While
some ING- forms are clearly nominal (7) ,
7- a) The killing of a prisoner of war is a crime
b) Loud singing disturbs John
other ING- forms/constructions are of intermediary
verbal /nominal nature (8) ,
8- a ) Their killing a prisoner of war was a crime
b) His singing loudly disturbs John
and others, such as the following progressive form, are
simply verbal.
283
9- Mary is running
It seems to me rather implausible to assume that a
unified analysis of all these constructions is possible
in which a unique -ING morpheme ( a nominal morpheme or
possibly, a realization olf INFL) combines with a verb to
produce all existing Verb+ING forms. Rather, it could
be that the same approach would be appropriate in this
case as the one suggested! here for Hebrew. The
discussion in this and the next chapter will attempt to
demonstrate the advantages of such an approach in dealing
with the Hebrew and Arabic data
4.2 Verbal Gerunds
4.2.1 The Structure of Infinitival Clauses
The main claim about the structure of the verbal
gerund has already been presented in the previous
section. The claim is, to repeat, that the verbal gerund
has a clausal structure just like a standard infinitive
In particular, just like lan infinitive, its verb form is
derived by a movement operation adjoining a verb to a
non-finite INFL. I will adopt here a standard view of
284
the finite/non-finite distinction. A finite INFL is one
which is specified as [+TENSE] . Such a specification of
INFL is made possible by assuming the following rule,
10- INFL -----> [ [+/- TENSE] , (AGR) ]
A non-finite INFL is thus an INFL node specified as [-
TENSE] . The claim is then that both an infinitive and a
verbal gerund involve a [-TENSE] INFL. Given the basic
assumption that both structures are clausal, the obvious
question to raise is where the differences between the
two constructions come from. The two main differences
between the two constructions are the following.
The first concerns their distribution. An
infinitival clause can function as an argument and occupy
any argument position in the sentence. Thus, it can be a
complement to a verb (11) , an adjective (12) or a
preposition (13) as well as function as the subject of a

3
sentence (14) .
11- Dan roce [le-sayem et ha-avoda]
Dan wants to finish OM the work
12- naim [li-hyot ba -xeder]
pleasant to be in the room
"It is pleasant to be in the room"
285
13- ze yoter xaSuv mi [li-lmod ivrit]
it (is) more important than to learn Hebrew
14- [li-lmod anglit] ze xaSuv
to learn English it important
"To learn English is important"
A verbal gerundive clause may never be an argument.
Rather, its most typical use, but, possibly, not the only
one, is as a time adverbial.
15- [im hikanes Dan la- xeder] hitpateax vikuax
with entering Dan to the room developed a dispute
"With Dan's entering the room a dispute developed"
A second major difference between the two constructions
concerns the occurrence of a subject. In an infinitival
clause no lexically realized subject may appear (16) ,
whereas in a verbal gerund exactly the opposite situation
obtains, namely, there must be a lexical subject (17) .
16- *a) Dan li-•r'ot et Dina
Dan to see OM Dina
*b) li-r'ot Dan et Dina
to see Dan OM Dina
*17- bi-r ' ot et Dina
with seeing OM Dina
286
In the account proposed in this chapter these differences
derive from one single factor, namely, the nature of the
element occupying the position of head of CP in each
case. In an infinitival clause it is the abstract
functional head COMP, whereas in the verbal gerund it is
a preposition, one out of a small class of prepositions.
Our assumptions about the structure of infinitival
clauses were presented in chapter 1. Here we will just
repeat them in their main outlines. The structure of an
infinitival clause is assumed to be as follows.
18- CP
/ \
/ \
COMP IPt
(Rj , Ri ) / \
NP I'i
I / \
e le-I VPi
In an infinitival clause, the external theta-role of VP
is vertically assigned first to I * . It is assumed that,
in general, in infinitival clauses, COMP, or rather, its
internal theta-role inherits the external theta-role of
VP. In structures involving obligatory control this
287
provides the input needed to satisfy the requirements of
a higher control verb, control being a relation between a
lexically designated argument of a control verb and a
governed COMP. Thus, the subject position of IP may
only be occupied by an NP-trace, an element which must be
assigned but cannot satisfy a theta role. Of the other
empty elements which may be seriously considered
(assuming the non-existence of PRO) , a small pro would
require to be identified by inflectional morphology which
is absent in this case, and an expletive pro, as will be
argued later in this chapter, must be assigned nominative
Case, which is not available to this position. Thus,
COMP inherits the external theta role of VP and has also
its own external referential theta-role R by virtue of
which it can figure in argument positions in the
sentence .
18 shows the preposition "le" (to) attached to INFL.
This preposition plays no role in the thematic structure
of the clause and it is absent from gerundive clauses.
Following a suggestion of R. Higgins I will adopt the
idea that the preposition le, situated in the INFL
position, fulfills the function of a non "nominal" +F
element in the sense of this notion suggested in section
3. 3. 4.1 of the previous chapter. Thus, it provides the
environment for the assignment of accusative Case. Its
presence is not required in the verbal gerund since in
288
this construction this function is fulfilled by a
prepositional element occupying the COMP position as will
become clear soon.
Thus, the two main facts illustrated in 11-16 are
explained for infinitives. Given its external theta-role
R , COMP and the CP that it heads can be the argument of a
predicate (11-14) . An NP in the subject position of an
infinitival clause can be assigned Case neither by INFL
nor by COMP (or a governor from outside the clause) and
is therefore ruled out.
4.2.2 The Structure of Verbal Gerunds
4 .2 . 2.1 A Sentential Analysis of Verbal
Gerunds (I )
The underlying structure of a verbal gerund, as in
example 19a, is represented in 19b,
19- a) be- exol Dan et ha-tapuax
with eating Dan OM the apple
"With Dan eating the apple”
289
b) CP
/ \
C
/ \
P IP
I / \
be NP I'
with 1 / \
Dan I VP
[-TENSE] / \
V NP
eat the apple
In this structure the specifier position of IP is
occupied by a lexical NP. The head of IP is a non-
finite INFL. As just suggested above, the absence of the
"dummy” preposition "le" (to) in verbal gerunds may be
explained as having to do with the fact that its function
in an infinitival clause as a licensing element for the
assignment of accusative Case is now fulfilled by the
preposition occupying the COMP position as head of the
entire construction. Given that no +F element occupies
the I position, VP in 19b above is in the domain of P , in
accordance with the definition of this notion in 150
(Chap 3) .
The surface structure of a verbal gerund is derived
290
by two successive head movement operations. First, the
verb moves and adjoins to I.
20- CP
/ \
/ \
P IP
/ \
NP I’
/ \
I VP
/ \ / \
I V V NP
Then, I moves and adjoins to P in the head position o£
CP.
291
21- CP
/ \
C*
/ \
P IP
/ \ / \
P I NP I'
/ \ / \
IV I VP
I / \
e V NP
The outcome o f these operations is a complex preposition

((P+I+V) 4 which i s in a position to govern and a s s i g n
Case to an NP in the subject position of I P .
Just l i k e any other NP governed and assigned Case by
a lexical head (N, V, P) , i f the su bj e c t NP in 2 1 happens
to be a pronoun i t is realized as a c l i t i c on the head
governing it.
22- a ) b i- reot -i et ha-seret

wit h seeing-me OM the fi lm
"Upon my seeing the film"
292
b) leaxar saym-o et ha-limudim
after f inishing-his OM the studies
"After his finishing the studies"
The facts about the distribution of the verbal gerund may
be explained as a direct outcome of the properties of the
prepositions that figure as the head of this
construction. Unlike the COMP of standard infinitives,
such prepositions do not have an external referential
theta-role and, therefore, the construction of which they
are the head may not be the argument of a verb. Rather,
they function just like any regular temporal PP. The
most typical function of such PP's is as time adverbials
as in the following example.
23- axare ha-bxirot tukam memSala
after the elections will be formed a government
"After the elections a government will be formed"
24 is the verbal gerundive paraphrase of 23.
24- leaxar hearex ha-bxirot tukam
after being held the elections will be formed a
memSala
government
"After elections are held a government will be formed”
293
Note, however, that temporally interpreted PP's may also
figure as predicates. This is not a very common use of
such PP's but it is possible if the NP of which they are
predicated can be construed as referring to a certain
point or interval in time rather than to an object or a
non temporal entity.
25- a) ha-maaraxa ha-Sniya hay ta leaxar ha-hafsaka
the act the second was after the intermission
"The second act was after the intermission"
b) ha-mikre hay a axare ha-diyun
the incident was after the discussion
This predicative use is also available for verbal
gerundive clauses.
26- a) ha-maaraxa ha-Sniya hay ta miyad im
the act the second was immediately with
histayem ha-hafsaka
terminating the intermission
"The second act was immediately after the
termination of the intermission”
294
b) ha-mikre baya miyad im higamer
the incident was immediately with finishing
ha-diyun
the discussion
"The incident happened right after the discussion
ended"
4.2.2 .2 A Non-Sentential Analysis of Verbal Gerunds
Given the similarity in use and distribution between
verbal gerunds and simple temporal PP's, as just
demonstrated above, the possibility may be raised that
the structure of a verbal gerund is, in fact, not the one
suggested in the previous section but, rather, one
corresponding to the PP's in 25a, b. That is, a verbal
gerund, it may be suggested, is a PP which has the
structure P-NP, where the NP in this structure may have a
more complex "nominalization-like" structure and
derivation but in which the gerundive head is the head
noun. The arguments against such a possibility are
simple and straightforward as follows.
The first argument has to do with distribution. The
material following the initial preposition simply does
not have the distribution of an NP, that is, it cannot
appear independently of the preposition in regular NP
295
positions. The following examples show a verbal
gerundive as subject, direct object and object of a
preposition respectively. Such positions are perfectly
natural for Action Nominalization constructions
discussed in chapter 3.
27- *a) reot Dan et Dina haya maftia
seeing Dan OM Dina wasi surprising
*b) hiklat-nu et daber Dan el ha-kahal
recorded-we OM speaking Dan to the audience
*c) Sama-nu al reot Dan et Dina
heared-we about seeing Dan OM Dina
Next, a verbal gerundive head cannot be modified by an
adjective,
*28- be- haros ha-iriya ha-axzari et ha
with destruction the municipality the cruel OM the
bay it
house
Here, the intended interpretation would be one in which
the destruction (of the house) is described as cruel.
Note that the inacceptability of 28 may not be attributed
to the violation of some possible adjacency requirement
imposed on the relationship between an adjective and the
296
noun that i t modifies. Such a restriction simply does
not exist as the following example shows in which the
adjective may only be understood as modifying the non
adjacent head noun.
29- beyt ha-mora ha-gadol
house the teacher (fm) the big (ms)
“The big house of the teacher"
Thus, being unable to be modified by an adjective, it
would be implausible to assume that the verbal gerundive
head is a noun.
In an analysis in which the gerundive head is taken
to be a noun, the most natural, and the only available,
assumption with respect to the way it is related to the
subject following it would be that it assigns genitive
Case to it. On such an assumption, the head and the
subject would be related to each other as two terms in a
bound genitive construction. This, however, raises the
expectation that the free and the double genitive
variants would also be possible, but this is clearly not
the case.
30- *a) bi-reot Sel Dan et ha-seret
with seeing Sei Dan OM the film
297
*b) bi-reot -o Sel Dan et ha-seret
with seeing-his Sel Dan OM the film
This argument can actually be stated in a simpler way,
namely, if a gerundive head is a noun then i t should be
possible for i t to figure in genitive constructions of
all types, but this, again, is not the case.
A last argument against this proposal has to do with

0
the possibilities of coordination . In general, it is
perfectly possible for a preposition to govern two or
more coordinated NP’s,
31- a) axare aruxat ha-erev ve ha-seret ba-
after dinner and the film on
televizia
television
b) im Dan ve Dina
with Dan and Dina
If the alternative analysis considered in this section
was correct, we would expect to find two verbal gerundive
"NP's" coordinated under a single preposition. This can
be shown to be wrong.
298
*32- bi-Smoa Dan et ha-muzika ve
with hearing Dan OM the music and
reot Dina et ha-seret
seeing Dina OM the movie
These are, then, the arguments against the P-NP analysis
of verbal gerunds. In view of these arguments i t may be
concluded that this analysis is fairly implausible.
Since this is, however, the only conceivable alternative
to the clausal structure proposed in the previous
section, these arguments make the clausal analysis, or
some version of it, all the more plausible. We may turn
now to a further elaboration of that analysis.
4. 2.2.3 A Sentential Analysis of Verbal Gerunds (II)
As was suggested in section 4. 2. 2.1, the verbal
gerund has a sentential structure and may be viewed as a
type of infinitival clause in that its INFL head is
specified as [-TENSE] (and also does not carry agreement
features) . Its surface structure is derived by two
successive applications of head movement, first moving
the verb to INFL and then INFL+V to COMP. The resulting
tree structure is repeated below for convenience.
299
33- CP
/ \
/ \
P IP
/ \ / \
P I NP I'
/ \ / \
I V I VP
I / \
e V XP
Given this structure, all of the facts enumerated in the
previous section as arguments against the "P-NP" analysis
are directly accounted for. Beginning with the
distribution facts in 27, a substructure in which the
initial preposition is missing is simply not a
constituent and therefore it shouldn't be expected to be
able to appear independently. Moreover, if it is correct
to assume, as we do here, that the subject is assigned
Case by the initial preposition, then the underlined
sequences in 27, and therefore the sentences as a whole,
are also ruled out by the Case filter.
As to adjectival modification (example 28), given
300
the proposed structure, the gerundive head is not a noun
but, rather, a non-f ini te form of the verb and there is,
therefore, no reason to assume that it should be possible
for it to be modified by an adjective. The same
explanation holds with respect to the facts in 30. The
distribution of Sel-phrases is limited to NP's and
therefore they may not occur in a configuration like 32,
related as genitive arguments to the gerundive head.
Note, on the other hand, that adverbs are perfectly
admissible in verbal gerunds as is to be expected, given
the proposed analysis.
34- [ be-nos'e -nu leat ] garam-nu li-pkake tnua
with driving-we slowly caused-we to jams traffic
"By driving slowly we caused traffic jams"
Turning next to coordination possibilities in gerundive
clauses, an ungrammatical example such as 32 would have,
presumably, the following "pre- head movement" structure.
301
35- CP
/ \
C
/ \
p IP
1 / \
be IP IP
1 / \ / \
with NP I’ NP I'
1 / \ 1 / \
Dan I VP Dina I VP
/ \ /
V NP V NP
Il II
hear the music see the movie
I assume that 3 2 is ruled out simply because there is no
adjunction site for the verb in the second clause (see)
which would give rise to a surface word order such as the
one in 32, that is, one in which reot (see) follows the
first IP and precedes the second IP.
One does, however, find cases of conjunction which
may be attributed a configuration identical to the one in
35. One such example is 36a which I assume to have the
structure in 36b.
302
36- a ) be-hagia ha-banim be-Saa Seva ve- ha-
with-arriving the boys at hour seven and the
banot be-Saa Smone
girls at hour eight
"With the boys arriving at seven o’clock and the
girls at eight o'clock"
b) CP
/ \
P IP
/ \ / 1
p V IP and \
1 1 / \ IP
with arriv. NP I’ / \
1 / \ NP I*
the I VP the / \
boys 1 / \ girls I VP
e V Adv. 1 / \
1 1 e V Adv.
e at 7 1 1
e at 8
36 is an instance of an across-the-board movement. Given
the constituent structure and derivation suggested here
303
for verbal gerunds, such cases are naturally accounted
for. Note that the occurrence of a temporal adverb in
each one of the conjuncts serves as an indication that
there are in this structure two VP's and therefore two
sentential (IP) constituents (otherwise, the sequence
"The girls at 8 o'clock" would have to be taken for an
NP, which is rather implausible)
Further support for the clausal structure proposed
here is provided by the facts concerning binding
relationships in these constructions. Here the facts are
rather straightforward, namely, the same subject object
asymmetries that hold in a finite clause hold in a verbal
gerund. Thus, the contrast between the two following
examples is exactly what we expect.
37- a) bi-reot Dan et acmo
with seeing Dan OM himself
"With Dan seeing himself"
*b) bi-reot acmo et Dan
with seeing himself OM Dan
Equally, the passive variant of 37a is bad (38a) just
like its sentential counterpart (38b) .
38- *a) be- heraot acmo
with being seen himself
304
*b) acmo nira
himself was seen
On the basis of the analysis proposed here there is no
reason to expect 38a to be good. In this, verbal gerunds
stand in an interesting contrast to Action
Nominalizations such as in the following example
(discussed in chapter 3) .
39- reiyat acmo
seeing himself
"The seeing of himself"
We may turn now to some facts concerning the initial
preposition. As was claimed and argued for, the initial
preposition in a verbal gerund is not the head of a PP
but rather occupies the COMP position as the head of CP.
As was also mentioned earlier, it is only a small class
of prepositions that may appear as the initial element in
a verbal gerund. Thus, one does find prepositions for
which the most typical, or only, use is as temporal
prepositions but which may not occur in a verbal gerund.
40- a) me az ha-mahapexa
since the revolution
305
*b) meaz reot Dan et ha-seret
since seeing Dan OM the film
41- a) be-meSex ha-hafgana
during the demonstration
*b) be-meSex exol Dan et ha-tapuax
during eating Dan OM the apple
42- a) ad yom riSon
until Sunday
*b) ad Smoa Dan et ha-sipur
until hearing Dan OM the story
These facts, again, can be naturally accounted for on the
basis of the present analysis. A preposition may simply
be lexically specified as being able or not to figure as
the head of CP. We may assume that this is simply
achieved by assigning a preposition a certain feature,
say, [+COMP] . Conversely, one finds also instances of
prepositions that may appear in a verbal gerund and
receive a temporal interpretation but may not be used as
temporal prepositions when occurring as heads of a PP.
Consider the following examples.
306
43- a) lifne
before
b ) xaxare ha-teuna
after the accident
*c) im
with
*d) be
with
44- a ) lifne
before
b ) axare hitraxaSut ha-teuna
after occurrence the accident
??c) im "The occurrence of the accident”
with
*d) be
with
45- a ) lifne
before
b) axare hitraxeS ha-teuna
after occurring the accident
c ) im "the occurrence of the accident
with
d ) be
with
307
The UP ha-teuna (the accident) in 43 and the Action
Nominal izat ion the occurrence of the accident in 44 both
denote an event. Therefore, the most natural, or only,
way to interpret the governing preposition in 43-44 is as
a temporal preposition. This is, however, impossible for
the prepositions im and be. These two prepositions,
however, are perfectly fine with a verbal gerund (45) and
are, in fact, the most common and typical for this
construction. This, again, may be simply accounted for
by attributing the feature [+COMP] to these two
prepositions. Thus, a preposition specified as [+COMP]
may only occur as the head of CP. If this is so then we
must assume that the prepositions lifne (before) and
axare (after) have actually two separate variants, one
which is specified as [+COMP] and one which is not. Both
variants, however, will be semantically temporal. This
may be objected to at first glance as an unnecessary
complication but may be supported by the fact that the
same distinction needs to be made with respect to the
prepositions be and im. These two have an identical [-
COMP] counterpart which is, however, not semantically
temporal and may occur in PP's such as the following.
46- a) be amerika
in america
308
b) im Dina
with Dina
Thus, the distinction between prepositions which may
occur in a verbal gerund, as complementizers, and those
which don't must be made and given the present analysis
such a distinction may be achieved naturally by assuming
the existence of a certain, limited, class of
prepositions marked as [+COMP] of which the distribution
is strictly limited to the COMP position. Obviously,
such a distinction wouldn't be available on the basis of
a "P-NP" analysis of verbal gerunds. This distinction
turns out to be independently necessary for the two
identical variants of be and im which happen to differ
also in their meaning, the [+COMP] variants being
semantically temporal while the
[-COMP] ones are not.
This result is, in fact, desirable and is to be
expected given the theta-theoretic approach assumed in
this work. This is so for the following reason. In
regular PP's such as the ones in 46, the NP may be said
to be the argument of the preposition. It is assigned,
or satisfies, an internal theta-role of the preposition.
On the other hand, the theta-theoretic relation which
holds between a preposition situated in the COMP position
and the IP it governs must be different. This is because
309
IP in these constructions has no external theta-role and
is therefore incapable of being an argument. For this to
be the case, however, the governing preposition must be
one which has no internal argument. We are thus led,
again, to a distinction such as the one suggested above
between two separate classes of prepositions, the [-COMP]
(argument taking) ones and the [+COMP] (non argument-
taking) ones. The semantic difference between the
prepositions be and im which are marked [+COMP] and their
[-COMP] counterparts provides a certain amount of support
for this result since, given the reasoning above, they
must be lexically separate items on independent grounds.
In 4.2.1 two major differences between verbal
gerunds and infinitives were pointed out, the first
concerning distribution and the second concerning the
occurrence of a subject. It was claimed that both
differences are related to the occurrence of a
preposition in the COMP position of the verbal gerund and
its absence from the COMP position of infinitives. The
distribution facts were shown to follow naturally from
this assumption and it now remains to explain the fact
concerning the subject. The fact is, to repeat, that in
a verbal gerund a lexically realized subject must occur.
Given the clausal analysis of gerunds proposed here, a
rather natural explanation suggests itself if i t is
assumed that any sentential structure has a subject
310
position which must be occupied by some element. This,
it seems, is far from being a non-standard assumption.
Rather, it is an assumption which seems to have a fairly
general justification and the following discussion will
provide further support for it. The question as to the
obligatory presence of a lexical subject is then reduced
to the one concerning which elements are allowed to
occur in the subject position of this particular
sentential structure. In the case of the verbal gerund,
as I wish to argue, it is only a lexical NP which may
occur. In the following I will simply go through the
different possibilities in an attempt to show that a
lexical NP is the only one available, to the exclusion of
any other NP types.
Let us first repeat for convenience the proposed
configuration and note that its most important feature
for the present discussion is the fact that the COMP
position in this configuration is occupied by a complex
preposition derived by head movement.
311
47- CP
/ \
/ \
P IP
/ \ / \
P I NP I’
/ \ / \
I V I VP
I / \
e V XP
Given this characterization of the COMP position in the
configuration above, it follows that the subject position
in this configuration must be occupied by an element
capable of being assigned and satisfying a theta-role.
This would be part of the explanation for the following
contrast.
48- *a) bi-Smoa et ha-yedia
with hearing OM the message
b) bi-Smoa Dina et ha-yedia
with hearing Dina OM the message
"With Dina hearing the message"
312
Thus, as was concluded in the above discussion and unlike
the abstract element COMP which occurs in standard
infinitives, the preposition which occupies this position
in the verbal gerund does not have the appropriate
argument structure and is, therefore, incapable of being
assigned the external theta-role of VP (which would
"percolate" first to IP) . Therefore an NP-trace is
excluded from the subject position of 48a * . Thus, from
the variety of empty elements available the only
remaining option which may seriously be considered is
"small" pro.
Considering first "theta-role-assigned" pro, to be
distinguished from "expletive" pro, this element is
generally assumed to be licensed by morphological
agreement features, available in a governing INFL, and
those are not available in the case of 48a. An
infinitival INFL (at least in Hebrew) lacks agreement
features and it is for this reason that it is also
incapable of assigning Nominative to a lexical NP
subject.
Note that if the subject NP in 47 happens to be a
pronoun then it is realized as a clitic attached to the
gerundive head, as in the following example.
313
49- be-yode -nu et ha-tSuva
with knowing-we OM the answer
"With us knowing the answer"
Under certain approaches, it would be assumed that in
such cases the subject position in 47 is, in tact,
occupied by pro which is said to be "identified" by the
cluster of morphological features which are the clitic.
Alternatively, and what seems to be a fairly reasonable
approach, cliticization may be viewed as a process which
belongs into the morphological component of the grammar
so that at the relevant syntactic level of representation
the subject position of 49 is simply occupied by a
pronoun. In any case, however, such constructions may be
said to have a subject and are, therefore, irrelevant.
Thus, in cases such as 48 a,b, in which the verb in
a verbal gerundive clause has an external theta-role to
assign, only a lexical NP may occur.
The only remaining cases for which the question
arises are those in which the main predicate is one which
does not assign an external theta-role. Such cases
involve, for example, predicates which take only an
internal sentential argument (50a) or impersonal passive
(50b) .
314
50- a) (haya) barur Se Dan xole
(was) clear that Dan sick
"It was clear that Dan is sick"
b) nimsar al ha-teuna
was reported about the accident
"It was reported about the accident"
I assume that the subject position of such sentences is
occupied by an expletive pro and that the condition for
the occurrence of an expletive pro is that it be assigned
nominative Case. Thus, I am assuming that identification
by inflectional morphology is not the relevant factor
here. This is supported by cases such as 50a in which
the main predicate is an adjective. Adjectives, and,
more generally, third person morphology, do not allow for
pro-drop in Hebrew, that is, the occurrence of pro in
subject position. Note now that the verbal gerunds
corresponding to 50a, b are ungrammatical.
51- *a) be-heyot barur Se Dan xole
with being clear that Dan (is) sick
*b) be-himaser al ha-teuna
with being reported about the accident
The claim is then that the requirement of nominative Case
assignment to an expletive element occupying the subject
315
position of these gerundive clauses is not met and
consequently that the only remaining option for a non-
lexical subject in a verbal gerund, namely, expletive
"pro", is also excluded.
That this explanation is on the right track, that
is, that the ungrammaticality of these cases (51a,b) must
have something to do with conditions imposed on an
element occupying the sentential subject position is
strongly supported by the following examples in which the
infinitival counterparts of 50a, b occur.
52-* a ) lihyot barur Se Dan xole ze acuv
to be clear that Dan (is) sick it (is) sad
*b) lehimaser al ha teuna ze lo
to be reported about the accident it (is) not
naim
pleasant
Note that the only difference between the grammatical 50
a , b and the ungrammatical 52 a,b is that the former are
finite whereas the latter are not. To narrow the issue
further down let us note that corresponding infinitival
clauses in which an external theta role is assigned to
the subject are perfectly fine.
316
53- a) liSmoa Se Dan xole ze acuv
to hear that Dan sick it sad
"to hear that Dan is sick is sad"
b) limsor al ha-teuna ze lo naim
to report about the accident it not pleasant
"to report about the accident is not pleasant"
Considering first the contrast between 53a, b and 52a, b we
may note that the only difference between these two cases
has to do with the thematic status of the subject
position, namely, in one case (53) it is assigned a theta
role and in the other (52) it is not. Looking next at
the contrast between 50a, b and 52a, b we note that here
the minimal difference is that the first are finite while
the latter are not. Thus, the ungrammatical cases are
only those in which no theta role is assigned to the
subject and INFL is specified as [-FINITE] . We may
therefore conclude that the ungrammaticality of 52a, b may
only be explained on the basis of some assumption about
the way an INFL node in these sentences is related to
something else in the construction. This "something
else", however, could only be the subject position since
this is the only position with which INFL may be
plausibly assumed to be related in a government
relationship and, otherwise, why would the availability
or non-availability of an external theta role make any
317
difference (Note that, in any case, no other NP is
present in the construction which could be affected by
the finite/non~finite distinction) . The explanation
suggested here is that the expletive pro occupying this
position is not in the environment of the nominative Case
assignment as would be necessary for its licensing. It
follows that the ungrammaticality of the gerundives 51
a,b has, most probably, something to do with restrictions
applying to the subject position too. The
ungrammaticality of the 51a, b is thus not an isolated
case but rather a particular instance of a general
phenomenon having to do with the general conditions that
are imposed on the subject position of a sentential
structure .
To conclude this section, it was argued that the
Hebrew verbal Gerund construction is a type of
infinitival clause. The most important detail in this
analysis is the assumption that the initial preposition
in this construction is a complementizer to which an
infinitival form of a verb is adjoined by head movement.
This assumption allowed us to explain the facts about the
distribution of the verbal Gerund as well as some
restrictions with respect to its subject position. The
obligatory occurrence of a subject in a verbal gerund was
explained on the basis of the general assumption that any
clause must have a subject. As was demonstrated above,
318
of the variety of NP elements that may come into
question, only a lexical NP is licensed in that position.
Obviously, such an explanation is only available if a
clausal structure is assumed. The parallelisms
demonstrated between the verbal gerund (51) , on the one
hand, and finite (50) and infinitival (52) clauses, on
the other, provide rather strong support for this

7
assumption .
4.2.3 The General Analysis of Infinitives:
Some Implications
It would be interesting at this point to briefly
consider some implications of the above discussion to the
general debate concerning the analysis of infinitival
constructions. As is well known, the assumption that
infinitives are clausal is standard within the Government
and Binding framework but is otherwise far from being
universally accepted. The relevance of the discussion
in the previous section derives from the fact that the
explanation of some of the data relies crucially on a
clausal analysis of both standard infinitives and verbal
gerunds. It would therefore be interesting to consider
to what extent nonsentential approaches to infinitives
can provide an account for these facts. Nonsentential
319
theories of infinitives are theories which take
infinitives to be VP's. For the sake of concreteness we
will consider here the particular semantic theory of
infinitives developed in Chierchia (1984). Chierchia's
approach may be briefly summarized as follows.
Generally, properties, the semantic value of VP's, have
two modes of being: as propositional functions and as
individual images of propositional functions. A finite
VP such as runs, as in John runs, is a propositional
function, an unsaturated structure that "looks for" an
argument to yield a proposition. Propositional
functions have individual correlates. The semantic
operation by which a propositional function such as runs
is mapped onto its individual correlate is achieved by
the nominalizing function ' '. The operation of
mapping a propositional function onto its individual
correlate is marked in the syntax by means of the
f inite/non-finite distinction. Thus, to run is an
expression referring to the individual projection of
runs . Being an individual projection of a propositional
function, to run cannot take an argument, from which
follows the ungrammaticality of,
*54- John to run
Concretely, the syntactic counterpart of the nominalizing
320
function • ' is assumed to be the infinitival morpheme
[-Agr]. That is, [-Agr] is semantically associated with
the nominalizing function * * and, in the same way that
’ ' operates on an entire VP meaning, so does [-Agr]
have an entire VP in its scope.
The contrasts that concern us now are the following,
55- a) nimsar al ha-teuna ba- radio
was reported about the accident on the radio
"It was reported about the accident on the radio”
*b) lehimaser al ha-teuna ba- radio ze
to be reported about the accident on the radio it
acuv
sad
c ) Dan masar al ha-teuna ba- radio
Dan reported about the accident on the radio
d ) limsor al ha-teuna ba- radio ze acuv
to report about the accident in the radio it sad
"To report about the accident on the radio is sad"
56- a) hitbarer Se Dan xole
became clear that Dan sick
"It became clear that Dan is sick"
*b) lehitbarer Se Dan xole ze acuv
to become clear that Dan sick i t sad
c) Dan hirer et ha-nose
Dan clarified OM the issue
321
d) levarer et ha-nose ze xaSuv
to clarify OM the issue it important
"To clarify the issue is important"
The b examples above show the ungrammatical infinitival
counterparts of the grammatical finite sentences in a.
The verb in the a and the b examples does not assign an
external theta role. The verb in the ç examples does
assign an external theta role and its infinitival
counterpart in the d examples is perfectly grammatical.
Thus, the availability or unavailability of an external
theta role seems to be the only factor in the contrast
between the b and the d examples. Now, a theory that
assumes infinitives to be VP's, rather than clauses, will
have to provide an explanation of this contrast only in
terms of the lexical properties of these verbs. An
explanation in terms of the sentential structure in which
such verbs may be embedded is not available for such a
theory. A theory like Chierchia's (1984) will have to
assume that for some reason the semantic operation of
"nominalizing” a VP does not apply in these cases.
Here, it seems, a fairly plausible hypothesis suggests
itself as follows. The VP's in 55a and 56a do not have
an external theta role. In Chierchia's terms, these VP's
are not 1-place propositional functions (They are
therefore not propositional functions a t all) and can
322
therefore serve as input neither to the nominal i zing
function * * nor to its syntactic correlate [-Agr] . We
would therefore not expect to find infinitival forms of
these verbs.
This, however, turns out to be a wrong prediction as
the following examples show (compare with 55b and 56b
respectively) .
57 -a) amur lehimaser al ha-teuna
supposed to be reported about the accident
ba- radio
on the radio
"It is supposed to be reported about the accident
on the radio"
b) alul lehitbarer Se Dan xole
may become clear that Dan sick
"It may become clear that Dan is sick"
In these examples the infinitival clauses of 55b and 56b
are embedded under the (modal) raising predicates amur
(supposed to) and alul (may) respectively. T at amur and
alul function as raising predicates is attested by the
ungrammaticality of the following examples.
323
58-*a) amur ledaber ba- radio
supposed to speak on the radio
*b) alul levarer et ha-nose
may clarify OM the issue
Here the verbs embedded under amur and alul have an
external theta role to assign but a subject NP is
missing. As the examples in 57 show the requirement for
a subject couldn *t originate from the predicates amur and
alul themselves.
Now, in view of the grammaticality of 57a, b the
syntactic/morphological operation performed by
Chierchia's infinitival [-Agr] morpheme certainly applies
to the verbs hitbarer (become clear) and nimsar (be
reported). Thus, Chierchia's approach can provide no
explanation for the contrast between 57a, b on the one
hand and the b examples in 55 and 56 on the other. On
the other hand, given the sentential approach to
infinitives adopted here, a fairly simple explanation is
available, namely, an explanation in terms of the
restrictions on the distribution of an expletive pro.
Thus, the minimal difference between 57a, b, on the one
hand, and 55b and 56b, on the other, is that in 57a, b the
matrix subject position is assigned nominative Case and
therefore an expletive pro is licensed in that position.
In 55b and 56b the subject of the embedded infinitive is
324
"Caseless" . Such an account is only available under a
clausal analysis of infinitives since in 57a, b it is the
higher subject position of the raising predicates which
makes the difference, rather than a (possible) lower
subject position of the infinitive.
Note that if the analysis of verbal gerunds suggested
here is correct, then it must be assumed that an
infinitival VP can function as a predicate. A verbal
gerund in this analysis involves an infinitival VP and an
obligatorily present subject NP. This, again, is in
conflict with Chierchia's (1984) approach to infinitives
as individual projections of propositional functions.
However, the view of infinitives as individual
correlates of propositional functions seems to run
independently into some problems. Consider the following
example .
59- The president is to cross the street tomorrow
morning
Here it seems as though the infinitival "to cross the
street..." is predicated of the subject rather than
being "referentially identified" with it. Chierchia
(1984) associates the copula be with the
"predicativizing" function ' ’ , the reflex of the
nominalizing function '. It is not obvious, however,
325
that it is the operation of this function which is
involved in the interpretation of 59. If that was the
case then the same would be expected to apply in the

8
following example .
60- The problem is to cross the street
It is clear, however, that the problem in this example is
not interpreted in such a way that cross the street is
predicated of it. Note also that in this example the
problem is clearly a (both underlying and surface)
subject. This is shown by the fact that it can undergo
Subject-Aux Inversion as follows,
61- is the problem (really) to cross the street?
Under a sentential analysis of infinitives these
contrasts receive a simple account. Assuming that the
copula be functions as some sort of "raising" predicate
(it may either be the head of a VP or else be dominated
by I , the choice here is not important for the present
issue) , 59 above can be analyzed as involving a post-
copul ar IP (62a below) whereas 60 may be assumed to
involve a post-copular CP (62b below) .
326
62- a) IP
/ \
NPi I ’i
the / \
president I VPi
/ \
V IPi
is / \
NP I '' i
1 / \
et I VPi
1 1
to cross the
str e et
327
b) IP
/ \
NPi I*i
the / \
problem I VPi
/ \
V CPi
I / \
is COMP IPj
Ri ,Rj / \
NP l'j
I / \
ej I VPj
I / \
to Vj NP
I I
cross the
street
In 62b above it is the external theta role o£ CP which is
assigned to the subject NP rather than the external theta

9
role of a lower verb as is the case in 62a . It seems
that a VP approach to infinitives such as the one
developed in Chierchia (1984) is incapable of making the
distinction which is reflected by these two structures.
328
4.3 Nominal Gerunds
4.3.1 Introduction
Unlike verbal gerunds, nominal gerunds are NP's and
have the regular distribution of NP's. Accordingly, the
account to be proposed in this section will involve a
nominalization-type of structure in which a nominal bound
morpheme, GER, which is assumed to be the underlying head
of the construction, governs a VP. The derivational
operations assumed will be similar to the ones proposed
for Action Nominalization constructions. The
differences between the two constructions and the set of
properties characteristic of the nominal gerund will be
shown to follow from one single assumption about the
nature of GER. We may first, however, proceed to a
detailed presentation of the facts.
The following examples show a nominal gerund as
subject, direct object, indirect object and as a member
of a bound genitive construction respectively.
63- a) Suv —o Sei ha-diktator garam le-behala
return-his Sei the dictator caused to panic
"The return of the dictator caused panic"
329
b) irgan -nu et Suv ha-poalim la-avoda
organized-we OM return the workers to work
"we organized the return of the workers to work"
c) ha-bxirot samu kec le heyote-nu miut
the elections put an end to being -us a minority
"The elections put an end to our being a minority"
d) uvdat headr -o Sel ha-nasi garma le
fact absence-his Sei the president caused to
behala
panic
"The fact of the president's absence caused panic"
In view of these distributional facts, i t is clear that
the structural analysis to be assigned to the nominal
gerund must be radically different from the one assigned
to the verbal gerund in which no head noun is assumed.
The probably most intriguing fact about the nominal
gerund concerns the kind of verbs of which the gerundive
form may occur in it. As was mentioned earlier, the
class of these verbs is rather limited and includes some
13 to 15 verbs in all. The following is a sample of some
of the verbs which are most typical of this construction,
in their gerundive form.
330
64- bo "coming"
Suv "return"
cet "leaving"
header "being absent"
healem "dis appearing "
himace "being present
heyot "being"
It is fairly clear from this sample that the property
common to these verbs has nothing to do with any
phonological or morphological characteristics. Rather,
as the results of linguistic research in recent years
have shown, verbs similar in meaning to those listed in
64 seem to be grouped together with respect to various
syntactic processes in a large number of different and
completely unrelated languages and have become known as
ergative (or unaccusative) verbs. Such verbs are
assumed to be lexically specified as having only an
internal argument and no external theta-role to assign.
In the following, I will assume that the class of verbs
which may occur in the nominal gerund construction in
Hebrew is exactly the class of ergative verbs in this
language. Thus, the main task of the analysis will be to
show what i t is about the nominal gerund which allows
only ergative verbs to occur in it. This will be shown
to correlate with a certain number of other
331
characteristic properties of the construction. In what
follows I will first enumerate the properties that seem
to be relevant, classifying them as either
characteristically nominal or verbal. I will then
proceed to a presentation of an analysis.
4.3.2 Nominal Properties of Nominal Gerunds
Among the nominal properties of the nominal gerund
the most important one has already been pointed out,
namely, its NP-like distribution as demonstrated in 63.
Next, unlike verbal gerunds, a nominal gerund
construction allows the occurrence of a Sei phrase as in
the following examples.
65- a) Suv -o Sei Dan
return-his Sei Dan
"The return of Dan"
b) headr -o Sei ha-nasi
absence-his Sei the president
"The president's absence”
If a nominal gerund is headed by a noun, as it is most
plausible to assume in view of the distributional facts,
then it is also plausible to assume that in examples
332
such as the following, the (logical) subject is assigned
genitive Case, that is, it is in the bound genitive
construction with the head noun.
66- a) Suv ha-poalim la-avoda
return the workers to work
"The return of the workers to work"
b) header ha-talmidim me ha-kita
absence the students from the class
"The absence of the students from class"
The examples in 65, on the other hand, are instances of
the double genitive construction. In such a
construction, the clitic pronoun attached to the head
may be said to be in the bound genitive construction with
it. In the analysis of double genitives suggested in
chapter 2, the Sel-phrase is assumed to be related to the
pronoun by a certain mechanism available at the level of
LF rather than being directly related to the head noun.
Given the availability of bound and double genitives in
nominal gerunds, one might expect to find also instances
of the free genitive in this construction. Here,
however, a rather interesting gap is revealed. As the
examples below show, the free genitive is impossible in
the nominal gerund.
333
67- *a) ha-Suv Sel ha-poalim la-avoda
the return Sel the workers to work
*b) ha-header Sei ha-talmidim me ha-kita
the absence Sei the students from the class
The analysis to be presented soon will provide a natural
explanation for this gap.
Finally, unlike the verbal gerund, a nominal
gerundive head may be modified by an adjective as in the
following examples.
68- a) Suve-nu ha-mukdam la-avoda
return-our the early to work
"Our early return to work"
b) headr -o ha-maftia Sei ha-nasi
absence-his the surprising Sei the president
"The president's surprising absence"
4.3.3 Verbal Properties of Nominal Gerunds
Here, two main facts are to be mentioned. First,
the head of a nominal gerund may never appear on its own.
It must always be accompanied by a subject.
334
69- *a) ha-Suv la-avoda
the return to work
*b) ha-cet me ha-kita
the leaving from the class
I assume that this property may count as "verbal" since
it is most likely related to the lexical requirements
which may be attributed to the related verb. Thus, the
deverbal nouns corresponding to the gerundive forms above
may appear unaccompanied by a subject.
70- a) ha-Siva la-avoda
the return to work
b) ha-yecia me ha-kita
the leaving from the class
"The leaving of the class"
Here, the lexical requirements of the related verb are
not a t work, presumably, because the nouns Siva (return)
and yeci'a (leaving, departure) may be lexically derived
and thereby lose the complementation requirements of the
corresponding verb.
One last relevant fact to be reported concerns the
occurrence of adverbs. As the following examples show,
adverbs are perfectly admissible in nominal gerunds.
335
71- a) cet -am Sel ha-poalim be-hekdem
leaving-their Sel the workers early
"The workers leaving (departing) early"
b) Suv —o Sei ha-nasi im redet ha-
return-his Sei the president with fall the
xaSexa
darkness
"The return of the president with the fall of
darkness"
71b shows a verbal gerund figuring as an adverb in a
nominal gerundive construction. In fact, it is possible
for both an adverb and an adjective to occur in a nominal
gerund.
72- a) cet -am ha-muskam Sei ha-poalim
leaving-their the agreed upon Sei the workers
be-hekdem
early
"The worker's agreed upon early leaving (departure)"
b) Suv -o ha-maftia Sei ha-nasi
return-his the surprising Sei the president
im redet ha-xaSexa
with fall the darkness
"The president's surprising return with the fall of
darkness”
336
Examples such as 72 a,b may sound a little overburdened
but they are certainly possible. Most crucially, if an
adverb and an adjective are ever to cooccur in a nominal
gerund, the adjective must precede the adverb.
73- a) Suve-nu ha-maftia
return-our the surprising
"Our surprising return"
a) Suve-nu im redet ha-xaSexa
return-our with fall the darkness
"Our return with the fall of darkness”
c) Suve-nu ha-maftia im redet ha-xaSexa
return-our the surprising with fall the darkness
"Our surprising return with the fall of darkness"
*d) Suve-nu im redet ha-xaSexa ha-maftia
return-our with fall the darkness the surprising
Thus, the distribution of adjectives and adverbs is most
likely conditioned by the syntactic configuration
underlying this construction, and the analysis to be
outlined in the following section will attempt to provide
an account for it.
We may, at this point, summarize the facts reported
in this and the last section as in the following table.
337
74- A- Nominal Properties
i) NP-like Distribution
ii) The occurrence of adjectives
iii) The admissibility of Sel-phrases
(but the impossibility of Free Genitives)
B- Verbal Properties
i ) Obligatoriness of an argument
ii) The admissibility of adverbs
4.3.4 The Structure of Nominal Gerunds
The proposed underlying structure of a nominal
gerund, as alluded to previously, is the following.
75- NP
/ \
N'
/ \
N VP
I / \
GER V NP
In this configuration the bound morpheme GER is the head,
which subcategorizes for a VP. This configuration is in
338
fact identical to the one underlying Action
Nominalization constructions and the only difference
between the two is in the element occupying the head
position in each. It therefore must be the case, if our
approach is correct, that the differences between the two
constructions are derived from the different properties
that may be attributed to NOM and GER respectively. NOM
was characterized as a head noun having an external
theta-role R and an internal theta role R to which a
genitive NP may be related in the identificatory
relation.
76- Argument structure of NOM: (Ri ,Rj )
It was assumed that the internal theta role of NOM is
assigned the external theta-role of the VP governed by
it. This is what it means for NOM to inherit the
external theta-role of VP. The internal R-role of NOM
may be further linked with the referential theta-role of
a genitive NP, the result of this linking being that the
genitive NP is interpreted as the subject of VP.
339
77- DP
/ \
D NP
I / \
POSS NP N*
I / \
Dan N VPj
Rj I / \
NOM V NPk
Ri,Rj I |
eat the
2,k apple
The assumption made here with respect to GER is simply
that it has no internal R-role and is therefore incapable
of being assigned the external theta-role of the VP
governed by it. Consequently, it is also incapable of
"transmitting" that theta-role to an NP occupying the
specifier position of the construction. The lexical
representation of GER is thus simply the following.
78- a) Argument structure of GER: (Ri , )
b) Subcategorization frame of GER: [ VP]
Being a noun, GER has its own external referential theta-
340
role which allows it and the configuration of which it is
the head to occupy argument positions in a sentence.
Let us now look at the thematic structure of a
nominal gerund.
79- DP
/ \
NPi
/ \
N'i
/ \
Ni VP
I / \
GER Vj NPj
Ri
In this configuration, the external theta-role of GER is
vertically assigned, first to the dominating N* and then
to NP. An NP governed by the verb may be assigned the
internal theta-role of that verb. If the verb heading
the VP in 79 had an external theta-role, that theta-role
would be vertically assigned to VP. At this point,
however, it wouldn't be possible for this theta-role to
be assigned to GER, given the argument structure of GER,
in particular, its lack of an internal R-role.
Consequently, GER cannot serve as a mediator for theta-
341
role assignment between VP and an NP occupying its
specifier position. On the other hand, i t would also be
impossible for the external theta-role of VP to be
vertically assigned to N* and consequently to an NP in
the specifier position. This is because N’ already has
an external theta-role, namely, the referential argument
of GER. Thus, there is no way in which an external
theta-role of VP may be satisfied in a configuration such
as 79. It follows that the only verbs that may occur in
such a configuration are those which have only an
internal theta-role to assign and no external theta-
role. The occurrence of a non-ergative verb in this
construction will always result in ungrammaticality due
to an impossibility to satisfy a theta role.
The derivational stages following the underlying
structure in 79 consist, first, of a head movement of the
verb to GER, followed by a movement of NP to SPEC in
which position it may be assigned genitive case in the
manner usual for bound genitives .
342
80 DP
D NP
POSS N*
N VP
N NP
GER
b) DP
D NP
POSS NP N
N VP
GER
343
c) DP
/ \
D NP
/ \ / \
N D NP N'
/ \ 1 / \
V N POSS N VP
1 1 / \
GER e V e
1
I
Given , this analysis, all of the facts described in
the two preceding sections and summarized in 74 can be
directly accounted for.
Beginning with the properties classified as nominal,
the distributional properties of nominal gerunds are
explained on the basis of the argument structure of GER
and the fact that it functions as the head of NP. The
occurrence of adjectives in nominal gerunds is clearly
natural within this analysis.
The most interesting among the "nominal" properties
of nominal gerunds are the facts concerning genitive
relations between the head noun and the subject NP. As
illustrated in 80 above, an NP moved to the specifier
position may be in the bound genitive construction with
344
the head. This is, in fact, the only way for the
internal-* argument-NP to be assigned Case since none of
the verbs occurring in this construction is a "Case-
assigner". However, if that NP happens to be a pronoun
(81a) , then it may be "doubled" by an NP in a Sel-phrase
(81b). In this the properties of GER play no role. The
Sel-NP, in this case, is licensed by being related to the
pronoun, which is assigned a theta-role independently.
This way of being licensed is, however, not available for
the Sel-NP in 81c. In the absence of a pronoun, the only
way for it to be integrated into the structure would be
by being related to a thematic index in the argument
structure of the head noun. But this, in the case of GER
(and unlike the situation in Action Nominalisations) is
not available.
81- a) Suv-o
return-his
"his return"
b) Suv-o Sei Dan
return-his Sei Dan
"Dan's return"
*c) ha-Suv Sei Dan
the return Sei Dan
Let us look at 81c more closely. Given that a Sel-phrase
345
is adjoined to a nominal projection, presumably N* , it
cannot occupy an argument position in VP but must be
situated outside of i t . Thus, "the closest" that 81c
could get to a grammatical configuration would be
something like the following,
82- NP
/ \
the N’
/ \
N' VPt
/ \ / \
N Sel NP Vi NPi
/ \ Il I
V N Dan e e
I
GER
In 8 2 the only theta-role of the verb is assigned
internally to the object position which is, presumably,
occupied by a trace. Given that a trace cannot satisfy a
theta-role, the theta-role is vertically assigned to VP
from which it would be assigned to an argument in the
argument structure of the head noun. A Sei- phrase would
have its referential theta-role "linked" (coindexed) to
this argument and in this way end up being coindexed with
346
the theta-role originating from VP. In this way it would
be licensed as an argument. However, since GER has no
internal (R-) argument, none of this is possible and the
structure is ruled out. Thus, 81c is doubly ruled out as
a case in which a theta-role of the verb cannot be
satisfied and in which a Sel-phrase cannot be licensed as
an argument.
As to the "verbal" properties listed under B in 74,
the obligatoriness of an argument follows from the
present analysis since the lexical requirements of the
verb, figuring in the underlying structure of a nominal
gerund, must be satisfied.
Equally to be expected on the basis of this analysis
is the admissibility of adverbs, in particular, the
respective distribution of adverbs and adjectives
demonstrated in 71-73.
To conclude, the discussion in this section has
demonstrated, I believe, that the properties of the
nominal gerund may be given a fairly exhaustive
explanation on the basis of certain general assumptions
about the nature and distribution of theta-roles,
augmented by certain assumptions about the syntax, and,
in particular, the argument structure of a certain class
of nominal bound morphemes, the morpheme relevant for
this case being GER.
347
Footnotes to Chapter IV
cf . Gesenius ' Hebrew Grammar
2
Berman (1978) notes some isolated cases of
phonological differences between infinitives and gerunds
for which I have no explanation. Thus, certain
consonants in certain verbal roots undergo spirantization
in the gerundive but not in the infinitival form, for
example :
Inf. - liSbor (to break) Ger. - bi-Svor
Inf. - lispor (to count) Ger. - bi-sfor
3
It is plausible that, as argued in Hazout (1989) , the
infinitival clause in 14 does not occupy the subject
position but is rather related, in some way or another,
to the pronoun Ze occupying that position. This, however,
does not affect the distinction made here between
infinitives and verbal gerunds. The infinitive in this
case is still an argument and is interpreted as the
"semantic** subject of the clause.
4
The fact that I+V (or V+I, the order is irrelevant)
adjoins to the right rather than to the left of P may be
taken care of in a simple manner by assuming that P is
lexically subcategorized for an element to its right,
that is, P is a prefix for the purpose of word formation
operations applying in the syntax.
This was pointed out to me by Edwin williams.
* Note that this discussion presupposes the assumption

about the non-existence of PRO. But, clearly, even if
one were to assume the availability of PRO, it would
still be excluded from the subject position of 47 since
this position is governed by the preposition in COMP.
348
7
One issue not touched upon in this analysis concerns
the possibility of negation in verbal gerunds. Negation
in the verbal gerund, as opposed to infinitives, is
impossible.
i) XaSuv lo lirot et ha-seret

important not to see OM the film
"It is important not to see the film"
*ii) be- lo reot Dan et ha-seret
with not seeing Dan OM the film
*iii) be- reot lo Dan et ha-seret
with seeing not Dan OM the film
It seems that the analysis proposed here makes possible a

simple account of this fact, and of the observed
differences between infinitives and the verbal gerund, in
terms of the movement operation which the derivation of
the verbal gerund is assumed to involve. This would
follow if we assume that the (verbal) negation particle
lo (not) heads its own maximal projection
Neg-P. Given this assumption, the following would be the
structure of a verbal Gerund following the movement of V
to I (irrelevant details are omitted) .
iv) CP
/ \
P Neg-P
/ \
Neg IP
I / \
lo I’
I \
I VP
/ \
V I
We may say that the head of Neg-P in this configuration

selects IP. It follows that movement of the derived head
V+I to P would result in a violation of the ECP. In the
following configuration Neg-P is a barrier between the
moved head and its trace.
349
v) CP
/ \
P Neg-P
/ \ / \
P Ii Neg IP
/ \ I / \
V I lo I’
/ \
I VP
I
et
Such a situation does not arise in the derivation of

infinitival clauses since no further movement from I to C
takes place in such cases.
* Williams (1983a) suggests that some types of copula

constructions involve a reversal of order between subject
and predicate, so that a surface subject is in fact an
underlying predicate. Subject-Aux inversion is one of
the tests used by Williams (1983a) to establish the
status of a constituent as a subject. Of-course, it is
crucial for the present argument that the problem in 60
is a subject rather than a predicate, but it seems to me
that the Subject-Aux-Inversion test is not so
absolutely crucial since one finds also examples such as
the following.
(i) a) to know John is to love him

b) to love is to rejoice
Here both subject and predicate are infinitives so the

question of reversal of order becomes irrelevant since
one of the two infinitives must be the "deep" predicate.
Assuming, for instance, that the surface predicate in
(i)a to love him is also the underlying one, it is clear
that the theta role assigned to the subject is not the
external theta role of love. Under the analysis proposed
here it would be the external theta role of CP, a
possibility which is not available to Chierchia (1984).
9
As pointed out to me by R. Higgins, cases which
involve an "identif icational” type of interpretation can
have a complementizer for or indirect questions of
various sorts as in the two following examples
respectively .
350
i) The problem is for you to cross the street in five
seconds
ii) The problem is what to do with John
On the other hand, the presence of a complementizer is

never possible with a raising type of interpretation.
These facts clearly support the claim that the two types
of interpretation involve two different structures.
351
CHAPTER 5
THE SYNTAX OF PARTICIPIAL CONSTRUCTIONS
5.1 Introduction
5.1.1 An overview
Hebrew and Arabic, like many other languages,
manifest an extremely large variety of participial
constructions. By this we mean constructions in which
the participial (Benoni in Hebrew) form of the verb
occurs as what seems to be the head of these
constructions. The situation here is similar to the one
observed in the previous chapter with respect to Hebrew
infinitival and gerundive constructions. It is the
same morphological form, a certain combination of verb
root and vowel pattern, which occurs in a vast array of
constructions, ranging over different category types
(nominal, verbal and adjectival) in both their function
and distribution.
The mixed nature of Benoni participles is manifest
at both the lexical and the phrasal levels. Thus, one
finds both in Hebrew and in Arabic participial forms that
352
have been lexicalized as nouns as well as forms which
have been lexicalized as adjectives.
To start with, Benoni agent nouns such as the nouns
occurring in the following examples are fairly common.
1- a) (H) ha-menahel siyem et avodat-o
the director finished OM work-his
"The director finished his work"
b) ( A ) ra?ay-tu 1-kaatib-a fi- 1-maktabat-i
saw -I the writer-Acc in-the-library -Dat
"I saw the writer in the library"
Unlike participial forms which occur in phrasal
structures, such nominal forms lose the "verbal force" of
the related verb, that is, mainly, complementation
requirements associated with a particular verb.
The same is true with respect to Benoni adjectives.
Thus, the adjectival Benoni forms in the following
examples are related to the strictly transitive verbs
Samala (include) and saxaf (sweep away) respectively.
2- a) ( A ) maSruu9 Saamil
project inclusive
"An inclusive project"
353
b) (H) had axa soxefet
success sweeping
"A sweeping success"
Obviously, Benoni forms can occur as "strict" verbs
at the head position of a VP in sentential structures
such as the following.
3- a) (H) Dan ohev et acmo
Dan likes OM himself
b) (A) zaid-un ?aakilu ttufaaHat-a
Zaid-Nom eating the apple-Acc
"Zaid is eating the apple"
Moving on to participial constructions at the
phrasal level, section 2 of this chapter will present a
discussion of a nominal construction to be referred to as
Agent Nominalization in which a Benoni form seems to
figure as head. The Benoni head of this construction,
although acting as a noun in that it relates to its
complement in the bound form of genitive construal and in
allowing the occurrence of an adjective, still preserves
some of its verbal force as attested by the fact that its
complement is obligatory and that the occurrence of an
adverb is permitted.
Finally, section 3 of this chapter will be concerned
354
with another type of participial construction, the Benoni
Relative, a construction which in its use and
distribution may be either adjectival or nominal. In its
internal structure, however, the Benoni Relative may be
said to be "more verbal" than the Agent Nominalization
construction since the object of a verb in this
construction is assigned accusative Case rather than
being in the bound genitive configuration with it.
Thus, the picture that emerges while observing the
variety of participial forms and constructions is quite
perplexing. Nevertheless, there is a clear sense in
which the forms and constructions just mentioned are felt
to be related and it is again a question whether it is
possible to provide a general account of participial
constructions which would reflect this basic intuition.
As stated above, it is clear why this vast range of
phenomena is felt to belong together, namely, it is the
same morphological form, the same combination of
consonantal root and vowel pattern, which occurs in all
these forms. As just observed, identical lexical forms
of different categorial identity seem to exist side by
side. It would therefore seem reasonable to assume that
such forms are separately listed in the lexicon with
their distinct categorial identities. This may be
assumed to be the case with the Hebrew menahel
(directing, Verb) and menahel (director, Noun) , as well
355
as the Arabie Saamil (including, Verb) and Saamil
(including, Adjective) . It is therefore also reasonable
to assume that the two morphological components of a
Benoni form, namely, a verb form (which is itself the
combination of a verb root and a Binyan pattern) and the
(discontinuous) affix of the Benoni are also available as
separate entities in the lexicon. We have already seen
that such an assumption must be made in some other cases.
One such case was discussed in chapter 3 (section
3. 2. 2. 3) in which it was demonstrated that there are
strong reasons to assume that the underlying morpheme
CAUSE as well as the entire range of verb forms are
independently available in deep structures of the
double Accusative variant of causative constructions. We
may therefore make the further step with respect to
Benoni forms of assuming that the lexical entity
represented by the af fixai form of the Benoni, the
present participle, is multiply listed in the lexicon as
either nominal, adjectival or verbal (INFLectional) .
A lexically listed form of a certain categorial type
may be assumed to have certain lexical properties typical
of the particular syntactic category of which it is a
member. For example, we may assume that a nominal bound
morpheme to which we will refer by the designation B
(from Benoni) has an argument structure of the type
generally assumed to be associated with nominal heads
356
and, in particular, nominal bound morphemes such as the
ones studied in the two previous chapters, namely, NOM
and GER. Equally, a participial bound morpheme,
identical in form, of adjectival type may also be assumed
to be available as the underlying head of certain
adjectival constructions.
Our strategy then will be to analyze a participial
form into its two morphological components and to allow
one of them, namely, the bound morpheme referred to here
as B , to assume different categorial identities and to
occur as head in different underlying configurations.
Such an approach would have a clear advantage in that i t
does not multiply the number of basic category types that
are generally assumed to exist. Rather, as will be seen
in the following sections, it attempts to explain the
mixed nature of participial constructions only on the
basis of some assumptions regarding the existence of
certain abstract elements and the underlying
configurations in which they may occur. This, in
addition to certain derivational strategies that are
available within the general theory of grammar, will make
it possible for us to explain a fairly large array of
phenomena. It is this strategy which was employed in our
analysis of Action Nominalizations and gerundive
constructions .
Before going into a detailed analysis of nominal and
357
adjectival constructions involving the Benoni, I will
present in the next subsection a brief discussion of
participial forms in their verbal use in an attempt to
show that this use of the Benoni participle lies outside
the proper domain of verbal nouns with which this
dissertation is concerned.
5.1.2 Participles As Verbs
The so far most thoroughly studied constructions
involving a verbal use of the present participle have
been present tense sentences such as the following (for
recent work on this cf. Doron (1983) and Rapoport (1987)
on Hebrew and Mushaweh (1986) on Arabic) .
4- a) (H) Dan ohev et Dina
Dan loves OM Dina
"Dan loves Dina"
b) (A) huwa daarib-un zayd-an
he beat -Norn Zayd-Acc
"He is beating Zayd"
The verbal characteristics of the participial verb form
in the examples above include, first, its ability to mark
the NP object following it with Accusative Case. This is
358
seen in both the Hebrew and the Arabic examples above.
Another reason for viewin? the examples in 4 as involving
a verbal form is that the position occupied by the
participial form in these sentences can be occupied by a
tensed, past or future, verb form.
5- a) (H) Dan yohav et Dina
Dan will love OM Dina
"Dan will love Dina"
b) ( A ) huwa daraba zaid-an
he beat Zayd-Acc
"He beat Zayd"
A present participle verb form may certainly be modified
by an adverb.
6- a) (H) Dan oxel bananot maher
Dan eats bananas quickly
b) ( A ) huwa ?aakilu 1- xubz-a bi-sur9a
he eating the bread-Acc quickly
"He is eating the bread quickly"
One of the more intriguing questions concerning the
verb form in 4 has traditionally been the one concerning
its temporal specification. Thinking in terms of
temporal distinctions such as past, present and future,
359
the question may be raised as to whether a sentence such
as the following, or the INFL node which may be assumed
to figure in its underlying representation, is specified
as Present or is simply unspecified for tense.
7- (H> Dan oxel bananot
Dan eats/is eating bananas
A sentence like 7 may be interpreted as either describing
an action taking place at the moment of utterance, or
else receive a generic type of interpretation by which it
would be describing an activity habitually repeated over
a long period of time. This indeterminacy in temporal
interpretation may be better appreciated when one
observes cases in which a sentence such as 7 occurs as a
subordinated clause.
8- (H) Dina taxSov Se Dan oxel bananot
Dina will think that Dan eats/is eating bananas
Here, the activity described by "Dan eats/is eating
bananas" may be understood as taking place at either the
future time of the higher clause or at the present
moment, that is, the moment of utterance, or simply
generically as a habitually repeated activity. This
seems to provide simple but fairly solid support in
360
favour of a claim (cf. Doron (1983) and Rapoport (1987))
that so called present tense sentences are in fact not
specified for tense. It is simply not clear what it
would mean to say that they are specified for tense given
that such a specification wouldn't seem to have any
effect on the interpretation of such sentences.
In the spirit of Doron (1983) and Rapoport (1987) we
may assume a classification of Hebrew (and Arabic)
clauses according to which the INFL node in present tense
sentences of the type in 7 is specified as having
AGR(eement) features but as lacking specification for
tense (which may be represented as [0 TENSE]). This, as
opposed to infinitival clauses of which the INFL node
lacks AGR(eement) features and is specified as [- TENSE]
and tensed, past and future, sentences of which the INFL
node has both AGR(eement) features as well as
specification as [+ TENSE] (with possibly a further
specification as [+/- PAST] or something of this sort) .
It is in this respect, that is, their lack of tense
specification, that present tense sentences are felt to
be more "nominal" than their past or future counterparts.
There is however a clear and strong difference between
sentences of the type in 7 and sentences involving a
strictly nominal form of the present participle. Such a
difference may be expressed in terms of aspectual
361
distinctions such as completion or non-completion of an
action. The following pair illustrates this point.
9- a) (A) Zayd-un kaatib-u 1-kitaab-i
Zayd-Nom writer-Nom the book -Gen
"Zayd is the writer of the book"
b) (A) Zayd kaatib-u 1-kitaab-a
Zayd writing-Nom the book -Acc
"Zayd is writing the book"
In 9a it is predicated of Zayd that he is "the writer of
the book”. In this case the participial form is a noun
and kitaab (book) is in the genitive Case. The only
available interpretation in this case is one in which the
action of writing the book has been completed. 9b
differs in its surface form from 9a only in the Case
marking of kitaab (book), accusative in this case.
However, the interpretation of 9b must be such that the
book is understood as being still in the process of
writing.
Thus, there is a clear difference between nominal
and verbal uses of the present participle in both their
syntax and interpretation and it is therefore perfectly
justified to assume the existence of a participial form
which is strictly verbal in category. As noted above,
the analysis of "present tense" sentences has enjoyed
362
much attention in recent years. This includes both
sentences of the type exemplified in 4 and 7 above as
well as sentences involving non-verbal predication.
However, in view of the above observations, this use of
the present participle lies outside the proper domain of
verbal nouns and therefore no further discussion of i t
will be undertaken in this chapter.
5.2 Agent Nominalizations
As was noted earlier, certain participial forms are
lexically listed as nouns. This is the case with agent
nouns such as Somer (guard) and menahel (director) in
Hebrew and kaatib (writer) and muttarjim (translator) in
Arabic. These nominal forms can clearly occur in a
sentence unaccompanied by any arguments that would
normally be required by the related verb.
10- a) (H) ha-Somrim sagru et ha-knisa
the guards closed OM the entrance
b) (H) ha-menahel lo hegia
the director not arrived
"The director didn't arrive”
363
11- a) ( A ) raja9a 1-kaatib-u min bayruut
returned the writer-Nom from Beirut
"The writer returned from Beirut"
b) ( A ) qutila 1-muttargim -u
was killed the translator-Nom
"The translator was killed"
Given our claim that such forms are lexically listed as
nouns rather than derived by some general syntactic
process, we might expect to find verbs for which a
lexically listed nominal form is missing. This
expectation is clearly borne out. Thus, the Hebrew verbs
Savar (break) , nasa (carry) or heenik (give) simply lack
an agent nominal counterpart. These missing nominal
forms would be the ones appearing in the following
ungrammatical examples.
12- *a) (H) rai-nu et ha-Sover
saw -we OM the breaker
*b) (H) Dan nifgaS im ha-nose
Dan met with the carrier
*c) (H) ha-maanik lo hegia la- pgiSa
the giver not arrived to the meeting
The same situation obtains in Arabic. Thus, the verbs
kassara (break) , ajjala (postpone) and ?akala (eat) lack
364
a nominal counterpart of which the predicted form would
be the one appearing in the following ungrammatical
examples .
13- *a) (A) ra?ay-na 1-mukassir-a
saw- we the breaker -Acc
*b) (A) daxala l-mu?ajjil -u ba9da
entered the ’'postponer"-Nom after
1-muHaadarat-i
the lecture -Dat
*c) ( A ) jalasa l-?aakil-u fi l-mat9am
set the eater-Nom in the restaurant
It might therefore come as a surprise that participial
forms such as the ones appearing in 12,13 may occur as
head nouns as, for example, in the subject NP's of the
following sentences.
14- a) (H) Sovre ha-xalon barxu
breakers the window ran away
"The breakers of the window ran away"
b) (H) nos?e ha-alunka avdu ba- derex
carriers the stretcher got lost on the way
"The carriers of the stretcher got lost on the way"
365
c) (H) maanike ha-trumot hegi-u la- kinus
givers the donations arrived to the meeting
"The givers of the donations arrived to the meeting"
15- a) ( A ) haraba mukassir-u SSubbaak-i
ran away breaker -Norn the window -Gen
"The person who broke the window ran away”
b) (A) ?insarafa mu?ajjil -u 1-muHaadarat-i
left "postponer"-Nom the lecture -Gen
"The person who postponed the lecture left"
c ) ( A ) jalasa ?aakil-u ttufaaHat-i fi
sat eater -Norn the apple-Gen in
l-mat9am -i
the restaurant-Dat
"The person who ate the apple sat in the restaurant"
The reason for this difference between 14,15 on the one
hand and 12,13 on the other seems to be fairly clear.
The ungrammatical cases in 12,13 are cases in which a
lexically required complement of the participial form is
missing. Such requirements are satisfied in 14,15 which
are therefore grammatical.
It is generally known that strict observance of
complementation requirements is a property more typical
of verbs than of nouns. In this sense we may at least
say that the participial noun forms in these examples
(14,15) preserve some of the verbal force of the related
366
verbs. This, obviously, does not exhaust the issue since
it is not clear what it would imply to say that these
nominal forms preserve, or inherit, some of the verbal
force of the related verb. Indeed, the question is
whether these verbal properties to which we are referring
loosely with the expression "verbal force" should be
attributed to the nouns themselves or, rather, to the
related verbs. Some support to the view that this
"verbal force" should be attributed to the related
"underlying" verbs rather than to the nouns themselves
comes from the fact that there are participial agent
nouns which are related to strictly (or strongly)
transitive verbs but which may appear unaccompanied by
any complement which would normally be required by the
corresponding verb. Thus, the verbs nihel (direct) ,
9allama (teach) and racax (murder) are all transitive as
shown by the following ungrammatical examples.
16-*a) (H) Dan nihel
Dan directed
*b) ( A ) 9 al lama muHammad
taught muhammed
*c) (H) ha-Soter racax
the policeman murdered
367
These should be compared with their grammatical
counterparts .
17-a) (H) Dan nihel et ha-yeSiva
Dan directed OM the meeting
b) ( A ) 9allama muHammad al-?awlaad-a ddars -a
taught muhammed the kids- Acc the lesson-Acc
"Muhammed taught the kids the lesson"
c ) (H) ha-Soter racax et ha-asir
the policeman murdered OM the prisoner
The agent nouns corresponding to the verbs in the
examples above do not seem to inherit the transitivity of
these verbs.
18- a) (H) ha-menahel hegia
the director arrived
b) ( A ) xaraja l-mu9allim-u mini 1-madrasat-i
went out the teacher -Norn from the school-Dat
"The teacher left the school"
c) (H) ha-roceax barax
the murderer escaped
If agent nouns are related to the corresponding verbs by
some general process, i t is hard to see why such a
process would affect only certain verbs in a certain way
368
to the exclusion of others. Thus, an attempt to
attribute a certain "verbal force" to the "transitive"
agent nouns of examples 12-15 will leave us without an
explanation for why in some cases a derived agent noun
may lose its "verbal force" whereas in other cases it
doesn't. It seems fairly reasonable to assume that the
"verbal force" manifested by the participial agent nouns
of examples 12-15 is indeed the verbal force, that is,
the complementation requirements that may be attributed
to the related verbs rather than to the noun itself.
This suggests that these constructions (14,15), to which
we will refer from now on as Agent Nominalizations,
involve an underlying verb and a certain derivational
process by which the surface nominal form of what seems
to function as the head of these constructions is
derived .
Another piece of evidence which seems to support
such a conclusion has to do with adverbial modification
in Agent Nominalization constructions. Di-Sciullo and
Williams (1987) (citing Tom Roeper) observe the following
English example.
19- The swimmer across the river
Their observation is that this example can only mean "the
swimmer who is across the river" parallel to "the man
369
across the river" . It cannot mean "the one who swims
1
across the river" . In Di-Sciullo and Williams' terms,
an adjunct modifier of a word cannot be interpreted as an
adjunct modifier of the non-head part of that word.
Thus, being the non-head of the morphologically complex
word swimmer , the verb swim is inaccessible to
modification. Consider now the following example of
Agent Nominalization in Hebrew.
20- (H) doxafe ha-sirot me-ever la- nahar
pushers the boats across to the river
In this example the most likely interpretation is "those
who push the boats across the river" rather than "the
pushers of the boats who are across the river". Here the
modifier seems to apply to the VP meaning "push the
boats". In particular, the modifier seems to have access
to the verb daxaf (push) which is the non-head part of
the noun doxef (pusher) . Thus, if the general rule is
that modifiers don't have access to non-head parts of
words then it must be concluded that the verb daxaf
(push) is available for modification as an independent
word at some syntactic level of representation. But this
necessarily implies that the word doxef (pusher) is in
these constructions the output of a syntactic, rather
than a morphological process. The fact that the noun
370
doxef (pusher) does not exist independently, that it may
only occur in an Agent Nominalisation construction
followed by a direct object, provides further support to
this conclusion.
We will, thus, assume that Agent Nominalisations are
derived from an underlying structure and in a
derivational manner similar to what was assumed for
Action Nominalisations and the Nominal Gerund which were
studied in previous chapters. Note that such an
assumption implies that the process involved in the
derivation of Agent Nominalisations is general and
should, in principle, apply to any verb. This seems to
be supported by the facts. As noted, among others, by
Berman (1978) with respect to Hebrew, all Benoni
(participial) forms can occur as head nouns of an Agent
Nominalisation construction. On the other hand, only a
small set of all Benoni forms may occur as independent
nouns .
Thus, lexically listed agent nouns are strict nouns
and therefore do not preserve the verbal force of the
related verb. Being lexically listed is, clearly, an
arbitrary property and there is no reason to expect such
a process to apply to all verbs. On the other hand, a
process such as nominalisation which applies to verbs
simply by virtue of being verbs is expected to be general
and this indeed seems to be the case. We are therefore
371
justified in assuming such a process and as will be seen
a bit later more evidence may be brought in favour of
this assumption, based on the analysis of Arabic
causative constructions developed in chapter 3.
We turn now to the actual analysis. We are assuming
that constructions such as 21-22 below, involve a
nominalization-type of underlying structure and
derivation.
21- a) (H) oxle basar
eaters meat
"meat eaters"
b) (H) noale sandalim
wearer sandals
"Those who wear sandals"
22- a) (A) daarib-u 1-walad-i
"beater "-Norn the boy -Gen
"The one who beat the boy"
b) ( A ) qaatil-u rrajul-i
killer-Nom the man-Gen
"The one who killed the man"
The configuration underlying 21,22 is the following,
372
23- DP
/ \
D NP
1 / \
POSS N*
/ \
N VP
1 / \
B V NP
23 is a typical Nominalization configuration in which a
nominal bound morpheme, B in this case, subcategorizes
for a VP. The surface form related to 23 is derived, just
like all other nominalization constructions studied so
far, in three successive steps. The first of these steps
is a head movement of the verb to B, resulting in the
following configuration in which the circled subtree is
the derived Benoni head noun.
373
24- DP
/ \
D NPi
I / \
POSS N*
/ \
N VP
Vt N NP 2
The following stages i n the derivation involve a movement
of the object NP to SPEC, a position in which it can be
assigned genitive Case and stand in the bound genitive
form with the head noun in the usual manner. NP2 in 24
above can not be assigned accusative Case. This is the
same phenomenon as observed for subjectless Action
Nominal! zat ions in Chapter 3, and it can be explained in
the same way as will be demonstrated below.
25 and 26 illustrate the two additional steps in the
derivation.
374
25- DP
/ \
D NP‘
1 / \
POSS NP2J N'
/ \
N VP
\ / \
Vi N V ej
1 1
B ei
26- DP
/ \
D NPi
/ \ / \
N D NP 2 N’
/ \ 1 / \
V N POSS N VP
1 1
B ei
It is now c l e a r t h a t , j u s t l i k e other n o m i n a l i z a t i o n
c o n s t r u c t i o n s , the main properties o f Agent
N o m i n a l i z a t i o n s may be explained m a i n l y on the b a s i s of
some assumptions concerning the p r o p e r t i e s o f the nominal
375
head B. This must be so given the identical underlying
structures and derivation assumed for all three
constructions. The most salient property of the Agent
Nominalization construction which distinguishes it from
both Action Nominalizations and Nominal Gerunds is that,
while these two are interpreted as referring to some
event (destruction, eating) or state (absence, being
somewhere) , an Agent Nominalization construction is
necessarily interpreted as referring to the agent of an
action, that is, the entity, or person, which would be
referred to by the subject of the corresponding sentence.
I assume that this is the main property of this
construction that needs to be accounted for. In fact, it
can be accounted for in a fairly simple fashion and at
the same time be correlated with other properties of the
construction to which we will turn later.
The lexical representation of B may be assumed to be
the following.
27 - a ) Argument structure of B: (Ri , Ri )
b) Subcategorization frame of B: [ VP]
Like any other noun and like NOM and GER, B is assumed to
have an external theta-role R , associated with an index
(i) . Moreover, like NOM and unlike GER, B has an
internal R-role. However, unlike NOM, in the case of B
376
the two R roles are coindexed. The thematic structure of
an Agent Nominalisation construction may thus be
illustrated as follows (ignoring the dominating DP node) .
28- NPi
/ \
N’t
/ \
Nt VPt
I / \
B V NP
Ri ,Ri i,k
In the above configuration the external theta-role R is
assigned vertically to N* and then to the maximal
projection NP. VP has an external theta-role which must
be satisfied and can only be satisfied. by being assigned
to the internal argument of B. This is so because it
cannot be vertically assigned to N* and become its
external theta-role since N* already has an external
theta-role. Given, now, that the two arguments in the
argument structure of B are coindexed, we achieve the
result that the NP as a whole is interpreted as referring
to the bearer of the external theta-role of VP. This
looks, and is certainly interpreted, as if the external
R-role is assigned the external theta-role of VP.
377
Clearly, the NP (or DP) as a whole is assigned a theta-
role from the outside by virtue of occupying a certain
position in a sentential construction. But this does not
result in any ill-formedness since the two coindexed
arguments are represented separately. This is a natural
possibility given the kind of representations suggested
here.
This is then the analysis proposed here for the
Agent Nominalization construction. It clearly accounts
for the main observation concerning the interpretation of
this construction and as will be demonstrated soon it
also accounts for its other properties. However, before
going into these particular issues, let us return briefly
to one of the more general issues.
One of the main issues regarding the analysis of
Action Nominalization constructions was related to the
question concerning a syntactic as opposed to a
lexicalist approach to these constructions. The approach
adopted in this work may be characterized as a strongly
syntactic approach. One of the main arguments, probably
the strongest, developed in Chapter 3 in favour of a
syntactic approach was based on the analysis of
causative constructions in Arabic. As the analysis of
these constructions in Chapter 3 (section 3. 2. 2. 4)
demonstrated, the double accusative variant of causative
constructions involves a syntactic process of head
378
movement of which a causative verb form is the output.
The dative variant, on the other hand, was shown to
involve a lexically inserted causative verb which is the
output of a lexical process. The argument developed in
Chapter 3 was that, since Action Nominalization
constructions allow for both the double accusative as
well as the dative variant to occur, there is a strong
reason to believe that the option of a syntactic
derivation of the type suggested in this work is, in
principle, open.
Given the syntactic analysis suggested here for
Agent Nominalizations and the fact that these
constructions occur both in Hebrew and in Arabic, we
should expect to find causative Agent Nominalizations in
both their double accusative and their dative variants.
If this turned out to be the case then the syntactic
analysis just illustrated would receive a fairly strong
support. The following examples show that this is indeed
so.
29- a) ( A ) mudarris-u l-?awlaad-i ddars -a
teacher -Norn the kids -Gen the lesson-Acc
"The one who teaches the kids the lesson"
b) ( A ) mudarris-u ddars -i li- l-?awlaad-i
teacher -Norn the lesson-Gen to the kids-Dat
"The one who teaches the lesson to the kids"
379
30- a) ( A ) muSarrib -u l-?awlaad-i
"make drink-er"-Nom the kids -Gen
1- Haliib-a
the milk -Acc
b) ( A ) muSarrib -u 1-Haliib-i li-
"make drink-er"-Nom the milk -Gen to
1- ?awlaad-i
the kids -Dat
The a examples above show the double accusative variant.
In these examples it is only the second NP which shows
accusative Case marking. This is due to the absence of
an overt subject NP. We will see soon how this is
accounted for along the lines of the general account of
this phenomenon suggested in Chapter 3. Crucially, a
reversal of order between the two NP's is completely
impossible, under the same interpretation.
31-*a) (A) mudarris-u ddars -i l-?awlaad-a
teacher -Norn the lesson-Gen the kids -Acc
*b) (A) muSarrib -u 1-Haliib-i l-?awlaad-a
"make drink-er"-Nom the milk -Gen the kids-Acc
380
The same word order restriction holds of the
corresponding sentential constructions and is readily
explained if the underlying structure of the a examples
in 29,30 is assumed to be the same except, of course, to
the B morpheme governing the causative substructure in
the constructions examined here. The underlying
structure of a double accusative Agent Nominalization is
thus the following in which each one of the underlying
verbal heads, CAUSE and V , takes one of the two NP's as
its complement.
32- DP
/ \
D NP
I / \
POSS N'
/ \
N VP
I / I \
B V NP VP
11/ \
CAUSE the V NP
kids I I
drink the milk
learn the lesson
381
As was argued with respect to Action Nominalisations in
Chapter 3, a lexicalist approach to these constructions
would be unable to account for these facts since it can
provide no way to establish an asymmetry between the two
NP's. A reversal of order between the two NP’s is only
possible in the dative variant as in the b examples in
29,30.
Some remaining issues need to be touched upon in
order for our analysis to be complete. First, in view of
the underlying structure suggested here for Agent
Nominalizations , the one illustrated in 28, one might
wonder whether an NP could occur in the specifier
position of this configuration, in the way that this is
possible in active Action Nominalizations such as the
following (33). Clearly, the specifier position is
available for an NP to occur in both Action and Agent
Nominalizations .
33- (H) axilat Dan et ha-tapuax
eating Dan OM the apple
"Dan ' s eating the apple"
The corresponding Agent Nominalization which might look
as either of the examples below is clearly impossible.
It is important to appreciate that this is not obvious
since, as we know, at least in principle a Nominalization
382
type of structure does allow the possibility for two
genitive NP's to occur. But this is clearly impossible
as the following examples show.
34-*a) (H) oxel Dan et ha-tapuax
eater Dan OM the apple
*b) (A) mukassir-u zayd-en SSubbak -a
breaker -Norn zayd-Gen the window-Acc
*c) ( H ) oxel Dan Sei ha-tapuax
eater Dan Sei the apple
Clearly, a subject NP must be missing in order for the
structure to be grammatical.
35- a) (H) oxel ha-tapuax
eater the apple
"The eater of the apple"
b) ( A ) mukassir-u SSubbak -i
breaker -Norn the window-Gen
"The breaker of the window"
To see why the examples in 34 are impossible we may look
a t an underlying structure of an Agent Nominalization in
which the specifier position is occupied by an NP.
383
36- DP
/ \
D NP11
I / \
POSS NP2 N’ i
Ri / \
Ni VPi
I / \
BV NPa j
Ri , Ri i, j
In this configuration the external t h et a - r o le o f VP i s

assigned to the internal argument of B. Now, the only
way f o r NP2 to be thematically lic en se d wi thi n this
configuration would be by having its external R-role
coindexed with the internal argument o f B a s represented
in 3 6 . However, this brings about a s i tu a t io n i n which
NPi and NP2 share the same r e f e r e n t i a l index. That i s ,

i n a configuration like 3 6 , the NP a s a whole i s
dependent for it reference on an NP contained in i t .
This is a violation o f the ” i within i condition" (cf.
Chomsky ( 1 9 8 6 ) ) which prohibits the coindexation of a

phrase with a phrase that contains it ( *
i[....Xi ]). obviously, th is explanation relies on

our assumptions concerning the thematic structure of
these constructions. The f a c t that these assumptions
384
make possible such an explanation may be seen as
providing a strong support in favour of them.
These assumptions receive further support when it
comes to accounting for the facts concerning genitive
relations between the head noun and the complement NP in
Agent Nominalisations.
The examples given so far all involve cases in which the
head noun and the complement NP stand in the bound
genitive construction to each other. The question now is
whether the other types of genitive construal, the free
and the double genitive, which occur only in Hebrew, are
also possible.
Considering first the double genitive, we might say
that it is marginal but not totally excluded. Thus, the
following examples are clearly possible.
37- a) (H) nose-ha Sei gufat he-harug
carriers-her Sei body the dead
"The carriers of the dead body"
b) (H) horg-av Sei ha-nasi
killers-his Sei the-president
"The killers of the president"
c ) (H) oxle-ha Sei ha-uga
eaters-her Sei the cake
"The eaters of the cake"
385
Crucially, the forms nose (carrier) , horeq (killer) and
oxel (eater) do not exist independently as agent nouns
and it is therefore clear that the examples in 37 involve
Agent Nominalizations rather then simple NP's. In many
cases, however, double genitive variants of Agent
Nominalizations are unacceptable and it is possible that
this has to do with the generic type of interpretation
associated with these constructions. Thus, unlike Action
Nominalizations and Nominal Gerunds which may refer to a
particular event, Agent Nominalizations are most
typically used to refer to an agent who is habitually
involved in a certain type of activity. It is possible
that this type of interpretation is incompatible with
the definiteness inherent to the meaning of pronouns. In
cases in which an agent may be construed as being part of
a particular, rather than generic, event the use of a
pronoun becomes possible as in the examples given in 37
above. The underlying structure of the examples in 37
would be the following.
386
37- DP
/ \
D NP
I / \
POSS N*
/ \
N VP
I / \
BV NP
Pronoun
The adjunction of a Sel-phrase at a l a t e r stage results
in the following configuration.
39- DP
/ \
D NP
I / \
POSS NP N'
I / \
Pronoun N’ VP
/ \ / \
N Sel NP V e
/ \ I
V N e
I
B
387
39 undergoes the additional movement operation generally
involved in the formation of bound genitives, as assumed
in this work. The NP in the Sel-phrase is not
thematically related to the morpheme B , but is rather
dependent on the pronoun for its licensing and
interpretation as is generally the case for clitic
doubling in genitive constructions.
The picture seems to be much clearer when it comes
to free genitives. These are clearly impossible, as
demonstrated by the following pairs.
40- a) (H) nos?e ha-alunka
carriers the stretcher
"The carriers of the stretcher"
*b) (H) ha-nos?im Sei ha-alunka
the carriers Sei the stretcher
41- a) (H) horge ha-xayot
killers the animals
"The killers of the animals"
*b) (H) ha-horgim Sei ha-xayot
the killers Sei the animals
42- a) (H) oxle ha-uga
eaters the cake
"The eaters of the cake"
*b) ( H ) ha-oxlim Sei ha-uga
the eaters Sei the cake
388
These contrasts receive a simple explanation within the
theta-theoretic approach suggested here. To see this we
might look at what would be the underlying structure of
the b examples above (ignoring the irrelevant DP node) .
43- NPi
/ \
(the) N’t
/ \
N’ VPi
/ \ / \
N Sel NP V e
I R (k ) i,l
B (i)
Rt ,Ri
In the configuration above Sel NP is adjoined at the N*
level at a "later" stage in the derivation. A t this
stage, the external theta-role of VP is already assigned
to (coindexed with) the internal argument of B . The
internal theta-role of the verb (1) may only be assigned

»
to a trace in object position, since no lexical NP
occupies this position, and is therefore unsatisfied.
It cannot be assigned vertically to VP and become its
external theta-role since VP already has an external
theta-role. Thus, the internal theta-role of the verb
389
remains unsatisfied. Furthermore, the Sel NP could only
be thematically licensed by being coindexed with the
internal argument of B. But this, first, is not the
intended interpretation since in this way the Sel-NP ends
up being interpreted as a subject (external argument)
instead of an object (internal argument) . Secondly, this
results in an ill-formed representation for the same
reason pointed out for 36 above, namely, the Sel-phrase
ends up being co indexed with the higher NP node in
violation of the "i within i condition". Otherwise, the
Sel-NP must have a different index which would leave it
simply without a theta role.
It now remains to account for the fact concerning
the impossibility of accusative Case marking of a direct
object in these constructions. The following examples
are clearly impossible as NP's. (They are well formed as
VP's, but this is irrelevant).
44- *a) (H) nos? in et ha-alunka
carriers OM the stretcher
*b) (H) horgim et ha-xayot
killers OM the animals
*c) (H) oxi im et ha-uga
eaters OM the cake
45- *a) ( A ) mukassir--u SSubbak -a
breaker ■
-Norn the window-Acc
390
*b) (A) mu?ajjil-u 1-muHaadarat-a
"postponer"-Nom the lecture -Acc
*c) (A) ?aakil-u ttufaaHat-a
eater -Nom the apple-Acc
Here the same explanation seems to be available as the
one suggested in Chapter 3 for the same phenomenon
observed in subjectless Action Nominalizations. Thus, in
a configuration such as the one repeated below, the bound
morpheme B plays the same role as the morpheme NOM in the
cases studied in Chapter 3.
46- DP
/ \
D NPi
/ \
N't
/ \
Nt VPi
I / \
B V NP
Rt ,Ri i,k
B in this configuration is "nominal" in the sense of the
definition of this notion in 145 of Chapter 3, that is,
B, or rather, its internal R-role is assigned the
391
external theta-role of VP and in this way it functions as
an argument. Thus, in both this case and the case of
subjectless Action Nominalizations the bound morpheme
heading the construction fulfills the function of a
missing subject. Given that B in 46 is "nominal", the
assignment of accusative Case to the object NP is not
licensed and the only way for it to be assigned Case
would be by moving to the specifier position in which it
can receive Case in the manner usual for bound genitives.
To conclude, I believe to have shown in this section
that the general approach to nominalization constructions
developed in the two previous chapters makes possible a
straightforward and fairly exhaustive treatment of the
Agent Nominalization construction. As demonstrated, by
maintaining an identical structural analysis for Agent
Nominalizations as the one assumed for Action
Nominalizations and nominal gerunds, while making
minimal assumptions about the argument structure of the
abstract nominal element B, we were able to account for
virtually all of the relevant facts relating to the Agent
Nominalization construction.
392
5.3 Benoni (participial) Relatives
By the term Benoni Relative we are referring to
constructions such as the underlined parts of the
examples below.
47- a) (H) ha-ovrim al ha-xok yeanSu
the violating on the law will be punished
"Those who violate the law will be punished”
b) (H) ha-talmidim ha-mesaymim et avodat-am
the students the finishing OM work their
"The students who finish their work”
48- a) ( A ) akrahu al-mufuuna 9uhuud-a -hum
I hate thé keeping promises-Acc-their
"I hate those who keep their promises'*
b) ( A ) ttalamiith -u 1 kaatib-u 1-
the students-Nom the writing-Nom the
?imtiHaan-a
test “Acc
"The students who are writing the test"
The a examples above show a Benoni Relative in its NP-
like function. It may occupy argument positions such as
subject (47a) or object (48a). The b examples, on the
other hand, show a Benoni Relative in its adjectival
function. It is from this type of use of this
393
construction that the term Benoni Relative derives.
Indeed, the Benoni Relative in these cases appears in
the same function as a relative clause and its internal
structure seems to be, at least in some respects,
sentence-like. It is also for this reason that the
initial definite article (ha/al) in this construction was
felt by a number of authors (e.g. Berman 1978) to be some
sort of complementizer. However, as I will attempt to
show in this section, such an assumption is not
necessary and an analysis in which the initial element in
these constructions is taken to be a standard definite
article is fairly plausible. An analysis along these
lines seems to be possible, as will be argued, for the
Benoni Relative in both its adjectival and its NP-like
function. This double function will be irrelevant and
therefore ignored in most of the discussion. The
surface form of a Benoni Relative may be schematically
represented as follows ,
49- ( ha/al - Benoni form - X ]
In this representation X stands for anything which might
follow a verb in a regular sentential structure, in
particular, any complement that may be required by a
particular verb. The examples below show benoni forms
followed by a direct object, an infinitival complement
394
and a prepositional complement respectively. 50a shows
also a post-verbal adverb.
50- a) (A) al-?akiluuna ta9aam-a l-?axariin
the eating food -Acc the others-Gen
daaiman
always
"Those who always eat the food of others"
b) (H) ha-mesarvim le-hibaxen
the refusing to be tested
"Those who refuse to be tested"
c ) (H) ha-mitanyenim ba -nose
the interested in the-topic
"Those who are interested in the topic”
In its interpretation, a structure such as 49 is similar
to that of Agent Nominalizations in that the construction
as a whole is interpreted as referring to the entity
which would be denoted by the subject of the
corresponding sentence. One clear difference between
this construction and the Agent Nominalization
construction is demonstrated by example 50a, namely, it
allows for accusative Case marking of the direct object.
Thus, the Agent Nominalization corresponding to 50a would
be 51a, to be contrasted with the ungrammatical 51b. (51b
would be fine as a VP but this is irrelevant) .
395
51- a) ( A ) ?aakilu t9aam-i 1 -Taxariin
eàter food-Gen the- others
"The eater of the food of others"
*b) ( A ) Taakilu t9aam-a l-?axariin
eater food-Acc the-others
Given our assumptions about the configurational
conditions for accusative Case assignment as developed so
far, we must assume that the direct object in 50a is
governed by an underlying verb and is in a
configurational context involving a VP in the domain of a
non- "nominal" +F element (in the sense of 145, chap. 3) .
In this sense the Benoni Relative may be assumed to
involve a nominalisation process a t a higher level (in
terms of the tree structure) than what was suggested for
Agent Nominalisations. Exactly what this means in terms
of the technical analysis will be presented soon.
However, given our claim that the underlying structure of
a Benoni Relative involves a VP in the domain of a +F
element of some sort, the question may be raised as to
how much of a sentential structure is indeed involved in
this construction. In this respect, the differences
between this construction and the standard relative
clause are quite instructive. The facts presented below
have already been noted in the literature (cf. Berman
(1978)) and they clearly indicate that Benoni Relatives
396
do not involve a sentential construction but, most
probably, only a VP.
In discussing a possible sentential analysis, one
point must be made clear. A question that naturally
arises if such an analysis of Benoni Relatives is assumed
is where would the definite article be located in such a
structure. On this point it has been common practice to
speculate (e.g. Berman (1978)) that the definite article
in these constructions is some sort of complementizer.
Such a view, although not technically elaborated
anywhere in the literature, would seem natural in view of
the interpretation and function of these constructions,
especially in their adjectival use. There is no need to
go into the exact details of such a possible analysis.
The only point relevant for our needs is that in any
reasonable sentential analysis of Benoni Relatives the
definite article would necessarily be located above, or
outside, the sentence (IP) level. Thus, a sentential
analysis of Benoni Relatives would look roughly as
follows .
397
52- CP/?
/ \
c/? IP
1 / \
ha/al I*
/ \
VP
/ \
The point is, however, that something like the schematic
structure above must be wrong. Here we may start with
some £acts concerning the distribution of adverbs.
53- (H) ha-anaSim Se (tamid) holxim (tamid) ba-
the people that (always) walk (always) in the
rxov
street
"The people who always walk in the street"
54-*a) (H) ha-anaSim ha tamid holxim ba- rxov
the people the always walk in the street
b) ( H ) ha-anaSim ha holxim tamid ba- rxov
the people the walk always in the street
"The people who- always walk in the street"
Given the standard assumption that the Hebrew relative
clause complementizer Se is the head of CP (the same can
398
be demonstrated with respect to the Arabic complementizer
?allad* i ) , we would not expect it to have any effects on
the distribution of adverbs within the domain dominated
by IP. As 53 shows, this is exactly so. The situation
with Benoni Relatives is however different. As 54a
shows, an adverb may not intervene between the initial
definite article and the benoni verb which would,
presumably, be located in the position I in 52 after the
application of head movement adjoining V to I. Rather,
an adverb must follow the verb as in 54b. These fact are
not to be expected if something like 52 was the correct
analysis.
A significant fact concerning the type of adverbs
allowed may also be pointed out. Thus, sentential
adverbs are perfectly fine in relative clauses (55a) but
impossible, or very odd, in Benoni Relatives (55b) .
55- a) (H) ha-anaSim Se kanir'e/lelo safek ovdim
the people that probably/certainly work
kaSe
hard
’’The people who probably/certainly work hard"
??b) (H) ha-anaSim ha-ovdim kanire/lelo safek
the people the work probably/certainly
kaSe
hard
399
This contrast would be unexpected if both standard
relatives and Benoni Relatives were assumed to involve a
sentence level structure. I
Contrasts between Benoni Relatives and standard
relatives may also be observed with respect to negation.
These contrasts may be observed with the regular negation
word lo (not) , as well as the negation particle eyn which
appears normally only in present tense Benoni sentences.
56- a) (H) ha-anaSim Se I lo ovdim kaSe
the people that not work hard
"The people who don't work hard"
*b) (H) ha-anaSim ha-| lo ovdim kaSe
the people the | not work hard
57- a) (H) ha-anaSim Se 1 eyn-am medabrim ivrit
the people that not-they speak Hebrew
"The people whoi don't speak Hebrew"
*b) (H) ha-anaSim ha | eyn-am medabrim ivrit
the people the| not they speak Hebrew
Given, again, the standard (assumption that the relative
complementizer Se (that) isl the head of CP and the
grammaticality of the a examples in 56 and 57, it must be
that the negative particles| lo and eyn are located at a
lower position in the tree (structure than the COMP
position. Thus, an assumption that the definite article
400
ha / al is the head of CP in the Benoni Relatives of the
B examples would leave us, again, without an explanation
for the apparent conflict between its occurrence and the
occurrence of a negation word.
We may, at this point, note the rather significant
fact that some form of negation is not altogether
excluded from the position preceding a Benoni form in
these constructions. In particular, the Arabic
adjectival negation morpheme Rayr (non-) is perfectly
acceptable in this position.
58- (A) annaas-u al Rayr ?aakiluuna t9aam-a 1-
the people the non eat food-Acc the
?axariin
others
"People who don't eat the food of others"
This fact would follow naturally if the Benoni Relative
was assumed to involve an adjectival, rather than
sentential, structure of some sort. We will see soon how
this may be done.
Another important fact which would seem to support a
"non-verbal" and therefore non sentential analysis, as
opposed to a hypothesis of the type represented in 52,
has to do with the verb form which is permitted to occur
in Benoni and standard relatives respectively. In a
401
Benoni Relative it is only the Benoni (participial) form
which may ever occur.
59- * a) (H) ha-anaSim ha- siymu/yesaymu et ha-
the people the f inished/will finish OM the
avoda
work
*b) ( A ) annaas -u al ?akalu/ya?kulu
the people-Nom the ate/will eat
ta9aam-a -hum
food -Acc-their
On the other hand, such a restriction is non-existent
when it comes to standard relative clauses.
60- a) (H) ha-anaSim Se siymu/yesyamu et ba-
the people that f inished/will finish OM the
avoda
work
"The people who finished/will finish the work"
b) ( A ) annaas-u llad’iina ?akalu/ya?kulu
the people who ate /will eat
ta9aam-a -hum
food -Acc their
"The people who ate/will eat their food"
402
In fact, it would not be implausible to assume that
certain cooccurrence restrictions could be invoked which
would establish a certain correlation between the type of
element functioning as head of C P , in a configuration
such as 52, and a particular (temporal) specification of
I. We will see, however, that a simple account of the
facts in 59 is possible which is also in line with the
standard function of ha/al as definite articles.
Finally, there seems to be a fundamental difference
between standard relatives and Benoni Relatives which has
to do with the general phenomenon of relativization. In
a standard relative clause any NP position may be
relativized. The examples below show relativization
affecting a subject and an object respectively.
61- a) ( A ) al- ?awlaad-u llad'iina ?akalu
the children-Nom who ate
ta9aam-a -hum
food -Acc-their
"The children who ate their food"
b) (H) ha-yeladim Se Dina raata
the children that Dina saw
Clearly, in Benoni Relatives it is only the subject which
may be "relativised". In this, Benoni Relatives and
standard relatives are fundamentally different and the
403
analogy commonly suggested between them loses much of its
force. Given the above considerations, it seems safe to
conclude that a view of the initial definite article in
Benoni Relatives as some sort of complementizer as well
as a sentential analysis of these constructions are most
probably wrong.
Rather, I will assume that the definite article in
these constructions occupies the same position that a
definite article occupies in any DP or DegP. The main
feature of the analysis proposed here will be based on
the general strategy adopted in this work. Thus, in the
manner that this was done for previously examined
constructions, we will assume that the Benoni,
participial, head is syntactically rather then lexically
derived .
A participial form, as observed earlier, is a
combination of two morphological units, a verb form and a
Benoni affix. Here, again, we will make use of the idea
that the lexical content represented by the Benoni affix
may be multiply lexically listed under different
categorial identities. This idea is fairly plausible in
view of the fact observed in section 5.1.1 that
identical Benoni words exist which belong to different
categories. Thus, in addition to the nominal morpheme B
which was postulated to figure as head of Agent
Nominalization constructions in the previous section, we
404
will postulate the existence of an adjectival morpheme
B . The adjectival B subcategorizes for a VP and figures
as the head of an AP in the underlying structure of

3
Benoni Relatives which we assume to look as follows .
62- DegP
/ \
Deg AP
I / \
ha/al A’
/ \
A VP
I / \
B V XP
The only step in the derivation of a surface form of
Benoni Relatives consists of a head movement adjoining V
to B. The output of this operation is a Benoni verb
form.
405
63- DegP
Deg AP
ha/al
VP
Vi XP
This is another instance of what has been characterized
in this work as a syntactic analysis. There is no need
here to go into a detailed discussion of this point. It
will suffice to point out that these constructions, just
like Action and Agent Nominal izat ions discussed earlier,
allow the occurrence of causative constructions and, in
particular, the double accusative variant of these
constructions as in the two examples below. The
assumption of a syntactic derivation is therefore well
supported .
406
64- a) ( A ) al-muSarribuuna ?awlaad-a-hum
the "make drink-er-s” kids Acc their
al Haliib-a
the milk -Acc
"Those who make their kids drink milk"
b) ( A ) al-mu9allimu ayya-hu assuluuk -a
the teaching OM-him the behavior-Acc
1-Hin
the good
"The one who teaches him good behavior"
The two most crucial details in this analysis
concern the theta-theoretic properties of the adjectival
morpheme B and the definite article. Beginning with B,
unlike its nominal counterpart it has no arguments.
Thus, it is devoid of any thematic content and is similar
in this respect to INFL, the head of IP. It follows
that, unlike its nominal counterpart, the adjectival B
does not block the upward propagation of the external
theta-role of VP. The adjectival B is thus incapable of
being assigned the external theta-role of VP and is
therefore non- "nominal" in the particular sense of this
notion employed in this work. Being a non- "nominal" +F
element, i t provides the environment for the assignment
of Accusative Case. Thus, the adjectival B is a
morphological "nominalizer" in that i t subcategorizes for
407
a VP complement and provides a landing site for head
movement of which a Benoni form is the output, but it
lacks the thematic properties of nominalizers such as the
elements assumed to figure as heads of the various
nominalization constructions studied earlier. Rather,
this thematic function is fulfilled by the definite
article which is situated at the head position of DegP
and it is in this sense that this construction may be
said to be a nominalization at a higher level.
Thus, the thematic properties of the construction,
which in an Agent Nominalization are derived from
certain assumptions about the argument structure of the
nominal morpheme B, are derived in a Benoni Relative from
similar assumptions with respect to the definite article
ha/al. We will assume that ha/al (the) has the same
argument structure as the nominal B , namely, i t has an
external and an internal R-roles which share the same
index .
65- Argument Structure of ha/al: (Ri , Ri )
The thematic structure of a Benoni Relative may be
illustrated as follows.
408
66- DegPi
/ \
Degi APi
I / \
ha/al A’i
Ri ,Ri / \
A VPi
I / \
B V XP
i,k
In this structure the external theta-role of VP is,
vertically assigned to A* and to AP and is then assigned
to the internal argument of ha/al. The referential
theta-role of the definite article is vertically assigned
to DegP and it is by virtue of this property that the
structure as a whole may have an NP-like distribution.
Given that the two arguments of the definite article are
coindexed, the construction as a whole is interpreted as
denoting to the bearer of the subject theta-role.
It is assumed that the same structure as the one
illustrated in 66 underlies the adjectival use of Benoni
Relatives in constructions such as the following.
409
67- (H) N p [ ha-anaSim d« 3 p[ ha-rocim le-hikanes]
the people the want to enter
"the people who want to enter"
Thus, structures of the type in 66 have a double
referential/ modificational use. It is interesting to
note that this is not an isolated case of such a double
function. Arabic relative clauses headed by the relative
pronoun ?allad* i (that) (or its feminine or plural
variants) seem to behave in exactly the same way. The
following pair of examples shows the same "?allad* i-
phrase" functioning in one case as the subject of a

4
sentence (68a) and in the other (68b) as a relative
clause modifying a head noun.
68- a) ( A ) ?allad *i ya jma9u-na huwwa tawajjuhu-
that unites -us it being confronted-
-na ma9an ?ila l-isti9maar
-our together to the imperialism
"what unites us is our common confrontation with
imperialism"
410
b) ( A ) ssiraa9 ?allad'i yajma9u-na huwwa
the struggle that unites -us it
ssiraa9 Deda l-isti9maar
the struggle against the imperialism
"the struggle which unites us is the struggle
against imperialism"
Likewise, the somewhat archaic Hebrew examples below show
a phrasal constituent headed by the particle aSer
figuring in one case in referential function as a direct
object (69a) and in the other (69b) as a modifying
relative clause.
69- a) (H) kvar Saxax-nu et aSer lamad-nu
already forgot-we OM what learned-we
"we already forgot what we learned”
b) (H) Saxax-nu et ha-Siur aSer lamad-nu
forgot-we OM the lesson what learned-we
"we forgot the lesson that we learned"
In fact, the most obvious case of an expression which can
occur in this double function are simple definite
adjectives as in the following pair.
411
70- a) (H) ha-qdolim mesoxaxim al politika
the big discuss about politics
"the "grown-ups" discuss politics"
b) (H) ha-yeladim ha-qdolim mesoxaxim al
the boys the big discuss about
politika
politics
"the big boys discuss politics"
I will assume that definite adjectival phrases have a
structure corresponding to Benoni Relatives as follows.
71- DegPi
/ \
Degi APi
I / \
ha/al A'i
Ri ,Ri I
At
Xi ,
Thus, the definite article under this view is a
" nominal iz er " although not in its morphological

8
functioning but only in its thematic functioning . In
both a simple definite AP as in 71 above, as well as in
the more complex Benoni Relative in 66, the definite
412
article takes a complement which, on its own, can only
be used as a predicate, and produces an overall structure
which may be used as a referential expression.
It follows from these considerations that, for a
Benoni Relative to have an NP-like distribution, i t must
have an initial definite article. It is the definite
article which provides the referential theta-role which
makes it possible for a constituent to function as an
argument. This is obviously true as demonstrated by
earlier examples and by the following pair.
72- a) (H) ha-rocim le-hikanes xayavim le-Salem asara
the want to enter must to pay ten
dolar
dolars
"those who want to enter must pay ten dolars”
*b) (H) rocim le-hikanes xayavim le-Salem asara
want to enter must to pay ten
dolar
dolars
On the other hand, a Benoni Relative in its adjectival
function does not need to be preceded by a definite
article. It only has to be preceded by a definite
article if the noun it modifies is definite.
413
73- a) (H) ha-anaSim ha-Sotfim et ha-ricpa halxu
the people the wash OM the floor went
habayta
home
"The people who wash the floor went home"
b) (H) anaSim Sotfim et ha-ricpa niru
people wash OM the floor were seen
derex ha-xalon
through the window
"People washing the floor were seen through the
window"
The underlined part of 73b in which a definite article is
missing would have the following structure in which the
external theta-role of VP becomes the external theta-role
of DegP by upward propagation. This is possible since
the head of DegP in this case is not occupied by an
element with an argument structure.
414
74- DegP’>i
/ \
DegP * i
/ \
Deg APi
/ \
A’i
/ \
A VPi
/ \
V NP
I I
wash the floor
îr j
To conclude, we may now see briefly how all of the facts
reviewed earlier as arguments against a relative-
clause-type analysis follow naturally from the present
analysis. First, as was observed, neither adverbs nor
negation words may intervene between the definite article
and the Benoni head. The relevant examples were the
following.
415
75-*a) ( A ) annaass-u ddaaiman ?aakiluuna
the people-Nom the always eat
ta9aam-a l-?axariin
food -Acc the others
*b) (H) ha-anaSim ha- lo/eyn-am holxim ba-
the people the not/not-they walk in the
rexov
street
Given the structure in 62/63, it is clear what is wrong
with 75 a,b, namely, adverbials and negation words
couldn't occur in a position following the head DegP and
preceding the head of an AP. In 75a, b the adverb and the
negation can only be construed as dominated by the AP
node outside of the VP domain. But this is impossible
since both adverbs and negation words of this type are
limited in their distribution to verbal, or sentential,
environments .
On the other hand, the head of an AP may be
morphologically complex and include a negation particle
such as the Arabic particle Rayr (non) as in 58, but such
cases are exempt from the configurational restrictions
that are a t work in 75. I repeat here example 58 for
convenience .
416
76) (=58) (A) nnaas-u 1-Rayr îaakiluuna
the people-Nom the not eat
ta9aam-a l-?axariin
food -Acc the others
"The people who don't eat the food of others"
The head of AP in this case would have a structure as
illustrated below. The rest of the structure is
identical to the cases observed earlier.
77- DegP
/ \
Deg AP
/ \
A'
/ \
A VP
/ \
Rayr B
non-
It is also clear why the following example, in which
tensed forms of the verb occur, is bad.
417
*78- (H) ha- anaSim ha-halxu /yelxu ba- rxov
the people the walked/will walk in the street
The adjectival morpheme B is simply not specified for
tense since adjectives in general do not carry temporal
information.
Finally, it is clear from the proposed analysis why
it is only the subject which may be "relativized" in
Benoni Relatives, namely, it is only the external theta-
role of VP which may be vertically assigned in the
proposed configuration and, in this way, become the
external theta-role of the AP as a whole.
418
Footnotes to Chapter V
1
As pointed out to me by R. Higgins, Roeper's claim
about the interpretation of 19 is wrong. Clearly, the
example is biased against an adverbial construal of
across the river given the presence of the initial
definite article. According to R. Higgins, examples like
the following are possible
(i) She is a fisher in muddy water
in which in muddy water modifies the fishing rather than

the fisher. Obviously, this observation does not raise
any difficulty to the analysis of Agent Nominalizations
suggested in this chapter but simply raises the
possibility that a similar approach may be necessary for
some corresponding constructions in English.
2
Note that if the analysis so far is accepted as
reasonably satisfactory then it becomes inconceivable to
claim that the impossibility of accusative Case
assignment is due to passivization. The fact that the
explanation of this phenomenon suggested here is
applicable to both Action and Agent Nominalizations
shows, in my view, its superiority over a passivization
analysis.
3
The DegP analysis of AP's is due mainly to Abney
(1987) who makes use of observations of Jackendoff
(1977) . See appendix to Chapter 2 for an application of
this idea to the analysis of adjectival bound genitives
in Hebrew and Arabic.
Example 68a is taken from Al-Ahraam Sep 27 1990.
■ Given this analysis of definite articles and in view

of the assumed argument structure of nouns, something
needs to be said about the way the two components, a
definite article and a noun, combine to produce the
419
argument structure of a DP headed by a definite article.
This can be accomplished in a simple fashion as
illustrated in the following structure.
(i) DPi
/ \
D'i
/ \
Di NPi
I I
ha/al N*i
Ri ,Ri I
Ni
R1
In this configuration the external theta-role of NP,

which percolates up from the head noun to NP, is linked
to the internal argument of ha/al. Given that the two
arguments of ha/al are coindexed, the external argument
of ha/al, and therefore DP, will always end up being
coindexed with the external theta-role of N.
420
APPENDIX TO CHAPTER 2
ADJECTIVAL BOUND GENITIVES
Bound genitive constructions may be formed also with
adjectival heads. It seems that the analysis of bound genitives
proposed in section 2.1.3 may be naturally extended to
adjectives. The following are typical examples.
1- a) (H) (naar) yefe eynayim
( a boy) pretty eyes
"a boy pretty of eyes”
b) (H) (erec) cfufat uxlusia
(country) crowded population
"a densely populated country”
2- a) (A) walad-un jamiil-u 1-wajh-i
boy -Norn pretty-Nom the face-Gen
"a boy pretty of face"
b) (A) rajul-un Taahir-u 1-qalb -i
man -Norn pure -Norn the heart -Gen
"a man pure of heart”
As example lb shows, adjectival bound genitives are not limited
to expressions of inalienable possession such as body parts and
the like. Rather, it seems that the construction is fairly
productive although ( a t least in Hebrew) clearly marked as
421
literary in style. Just like nominal bound genitives, adjectival
bound genitives may not be preceded by a definite article.
3- *a) (H) (ha-naar) ha-yefe eynayim
(the boy) the pretty eyes
*b) (H) (ha-erec) ha-cfufat uxlusia
(the country) the crowded population
The structure of 1,2 would be the following.
4- DegP DegP
/ \ / \
Deg' =====> Deg*
/ \ / \
Deg AP Degi AP
I / \ / \ / \
POSS DP A* Ai Deg DP A'
[Def] I I I
A POSS A
[Def] |
ei
Note that the assumption of POSS as head of DegP is not
necessarily problematic. The "nominalness" of adjectives is
something which has long been noticed by both traditional and
modern linguists. Presumably, it is the same position which will
422
be occupied by a definite article in a definite AP/DegP.
There is, however, an interesting variation between Hebrew
and Arabic concerning the distribution of the definite article in
adjectival bound genitives. It seems as though Arabic does
violate the restriction on the occurrence of an initial definite
article. Thus, the definite variants of 2a, b would be rendered
as follows .
5- a) (A) (al-walad-u) jjamiil -u 1-wajh-i
(the boy-Nom) the pretty-Nom the face-Gen
"the boy pretty of face”
b) ( A ) (rrajul-u) TTaahir-u al-qalb -i
(the man -Nom) the pure-Nom the heart-Gen
. "the man pure of heart"
I would like to suggest that 5 a , b are not counterexamples to the
general analysis of bound genitives suggested in this chapter.
Rather, the explanation of these cases would be based on two
points. First, they involve a special type of adjectival
Relative construction. The initial definite article in 5a, b (the
one preceding the adjective, not the head noun) would be the
definite article of a containing adjectival Relative construction
in which the adjectival bound genitive is embedded. Second, the
sequence Def. article-Noun in adjectival bound genitives is
interpreted differently in Hebrew and in Arabic. In Hebrew it is
interpreted as definite whereas in Arabic i t is interpreted as
423
non-defini te (or generic?) . Note that this claim about Arabie is
made with respect to "definite" NP's in the specifier position of
AP's alone. This does not necessarily hold for other syntactic
environments in Arabic and, indeed, in most cases it doesn't.
This semantic difference between Hebrew and Arabic may be easily
demonstrated as follows. It is generally known that predicative
phrases (6b, c , 7b, c ) , unlike modifying adjectives in NP's (6a, 7a)
may not be definite, (the impossibility of definite predicates
seems to be a general phenomenon, cf . Higginbotham (1986) , here
we are obviously abstracting away from cases in which identity of
reference is claimed as in: "John is the president").
6- a) (H) ha-yeled ha-gadol
the boy the big
"the big boy"
*b) (H) ha-yeled ha-gadol
the boy (is) the big
c ) (H) ha-yeled gadol
the boy big
"the boy is big"
7- a) ( A ) al-walad-u jjamiil-u
the boy -Norn the pretty-Nom
"the pretty boy"
*b) ( A ) al-walad-u jjamiil-u
the boy -Norn (is) the pretty-Nom
424
c ) ( A ) al-walad-u jamiil-un
the boy -Nom pretty-Nom
"the boy is pretty"
However, the following examples show adjectival bound genitives
figuring as predicates of a sentence. The contrast between the
Hebrew and the Arabic examples can only be explained if one
assumes that the predicate in the Arabic example is indefinite,
(the examples in 8 are intended to be sentences not NP's)
8- *a) (H) ha-yeled yefe ha-eynayim
the boy pretty the eyes
b) ( A ) al-walad-u jamiil-u 1-wajh-i
the boy -Norn pretty-Nom the face-Gen
"the boy is pretty the face"
Obviously, the restriction on definite predicates does apply in
Arabic as demonstrated by 7b. Therefore, it must be the case
that the adjectival predicate in 8b is perceived as indefinite.
The grammatical Hebrew counterpart of 8b must not have a
definite article in its predicate.
9- (H) ha-yeled yefe eynayim
the boy pretty eyes
"the boy is pretty of eyes"
425
The analysis of special adjectival relatives will be discussed in
detail in chapter 5 of this dissertation. Let us assume that
they have the following structure, in which a DegP or a VP may
figure as complement in an AP.
10- DegP
/ \
Deg*
/ \
Deg AP
I / \
ha/al A*
I \
A [DegP/VPl
Having a definite article as head of DegP, 10 is semantically
definite and may only modify a definite head. The lower DegP in
this configuration may have the usual internal structure of a
DegP, in particular, i t may be an adjectival bound genitive.
We may now look back at example 5. In this example the head noun
is clearly definite since it is preceded by a definite article
and is, of course, interpreted as definite. It may therefore be
modified only by a definite modifier. However, as demonstrated
in 8b, an adjectival bound genitive in Arabic which shows an
occurrence of a definite article and which is, by assumption,
headed by a POSS marked as [Def] is nevertheless indefinite.
426
Thus, the only way for an adjectival bound genitive to be
definite, and to be able t o modify a definite head, would be for
it to be embedded in an adjectival relative of the type
illustrated in 10. The structure of 5 a,b is the following.
427
11- DP
/ \
D NP
I / \
the N’
/ \
N’ DegP
I / \
N Deg AP
I I / \
boy the A*
/ \
A DegP
/ \
Degi AP
/ \ / \
Ai Deg DP A'
pretty POSS the A
pure face/ I
hurt ei
Thus, 5 a,b do not constitute a counterexample to the analysis of
bound genitives proposed here.
428
APPENDIX TO CHAPTER III
THE PARTICLE "ET” AND ACCUSATIVE CASE MARKING
>
There is some dispute in recent work concerning
accusative Case marking and the nature of the particle et
which occurs in Action Nominalization constructions. The
dispute is mainly about the question as to whether et
should necessarily be taken to indicate the assignment of
accusative Case by a verb or, alternatively, is itself a
Case assigner. Being a Case assigner, the occurrence of
et in nominal constructions would be necessary because of
the absence of another Case marking element, namely, a
verb. The question is obviously important since much of
the argument developed in this chapter depends on the
assumption that the occurrence of et, and accusative Case
in general, is an indication of the government of an NP
by a verb and, therefore, the occurrence of a verb in
some underlying representation needs to be postulated.
It is, of course, a significant fact, in my view,
that the corresponding construction in Arabic doesn * t
have anything which would be the equivalent of the Hebrew
et. Rather, Arabic Action Nominalizations manifest the
same morphological accusative Case marking which occurs
in simple sentences. Thus, the dispute which developed
with respect to Hebrew would never take place with
429
respect to Arabie for which an analysis of the type
suggested in this chapter would be the most reasonable
approach. Much of the dispute concerning the Hebrew data
is caused by the fact that Action Nominalisations with an
indefinite direct object are in many cases less
acceptable than their definite counterparts.
1- a) axilat Dan et ha-uga
eating Dan OM the cake
"Dan's eating of the cake"
??b) axilat Dan uga
eating Dan a cake
But it seems to me that the indefinite counterparts are
not completely ungrammatical. They are perfectly
acceptable if the object NP is "heavier" as in the
following examples.
2- a) axilat Dan tapuxe ec rabim
eating Dan apples many
"Dan's eating many apples"
b) harisat ha-oyev kfarim rabim ba-negev
destruction the enemy villages many in the negev
"The enemy's destruction of many villages in
the negev"
430
c) ktivat-o Sel Dan sefer exad asuya lehimaSex
writing-his Sel Dan book one may last
zman rav
time a lot
"The writing of one book by Dan may last a long time"
It is of course an interesting question why the object NP
needs to be "heavier", but this question is probably
independent from the question of Case marking.
We turn now to the suggestion that et is not the
marking of Case assignment by another governing element
but is rather a Case marker itself, that is, a
preposition of some sort. Here we may simply note some
facts showing that the distribution of et is radically
different from that of any other prepositional element
and that therefore any such analogy is wrong.
First there is the known fact that et occurs only
with a definite NP.
3- a) Dan axai et ha-uga
Dan ate OM the cake
"Dan ate the cake"
b) Dan axal uga
Dan ate a cake
431
Obviously, it is the verb axal (eat) which must be
assumed to be the Case assigner in 3b. There is no
analogous fact when one looks at prepositions. A Case
assigning preposition must always be present.
The possibility to drop e t is attested in other
construction types. One example is relative clauses.
4- a) ha-sef er Se Dina kar?a
the book that Dina read
b) ha-sefer Se Dina kar?a ot-o
the book that Dina read OM it
"The book that Dina read"
5- *a) ha-kesef Se Dan viter
the money that Dan gave up
b) ha-kesef Se Dan viter al-av
the money that Dan gave up on i t
"The money that Dan gave up”
In 4a the relativized element is a direct object and the
position following the verb is empty. Presumably, the
element occupying this position, a variable, must be Case
assigned and the Case feature, it seems, may only be
provided by the verb. In 4b et occurs with a resumptive
pronoun attached to it. As 5a, b show, nothing like this
is possible with a preposition, even a semantically empty
one such as the preposition al (on, about) required by
432
the verb viter (give up) . Here the resumptive pronoun
strategy must be employed.
Finally, pronominal direct objects may be cliticized
on the particle e t (6a) or, in the absence of et, on the
governing verb (6b) . This strategy is more commonly used
with infinitival than with finite verbs. Again, nothing
like this may ever happen with a preposition (7a, b).
6- a) Dan roce lirot ot-a
Dan wants to see OM her
"Dan wants to see her"
b) Dan roce lirot-a
Dan wants to see-her
"Dan wants to see her"
7- a) Dan mesarev levater ale-ha
Dan refuses to give up on her
"Dan refuses to give her up"
*b) Dan mesarev levatr -a
Dan refuses to give up-her
In view of this evidence one must conclude, at
least, that et is the only preposition of its kind. But
this obviously begs the question as to what kind of
preposition et is. Given that the NF's with which et
occurs are direct objects with respect to all other
criteria which generally identify an NP as such, i t seems
433
that the reasonable characterization of it is as an
object marker rather than a preposition.
Now, it might still be argued (cf. Siloni (1990))
that the occurrence of et in "derived nouns" involves a
different kind, rather than the "ordinary" kind, of
accusative Case marking. But such an approach tends to
ignore the striking parallelism between verbs and their
nominal counterparts. That is, it is only verbs which
assign accusative Case that have nominal counterparts
which also assign accusative Case. In an approach which
takes the two types of Case marking to be different this
correlation seems rather accidental.
434
REFERENCES
Abney, S. (1986) "Functional elements and licensing"

Ms. MIT, Cambridge, Mass.
Abney, S. (1987) The English Noun-Phrase in its

Sentential Aspect, Doctoral Dissertation, MIT,
Cambridge , Mass .
Aoun, J . (1978) "Structure interne du groupe nominale en

Arabe", Ms. Universite de PARIS VIII.
Aoun, J . (1979) "Parts of speech: A case of

redistribution", in Belletti A. et al (eds)
GLOW colloquium, Pisa, Italy.
Aoun, J . (1981) The Formal Nature of Anaphoric

Relationships , Doctoral Dissertation, MIT,
Cambridge , Mass .
Aoun, J . and Sportiche D. (1983) "On the formal theory of

Government", The Linguistic Review 2.
Aristotle D e Interpretatione, Oxford University Press,

(1928) .
Bach, E. (1979) "Control in Montague Grammar",

Linguistic Inquiry 10.4.
Bach, E. (1983) "On the relationship between word grammar

and phrase grammar", Natural Language and Linguistic
Theory 1.
Baker, M. (1985) Incorporation: A Theory of Grammatical

Function Changing, Doctoral Dissertation, MIT,
Cambridge , Mass .
Baker, M. (1988) Incorporation: A Theory of Grammatical

Function Changing, The University of Chicago Press.
Baker, M. Johnson K. and Roberts I. (1989) "Passive

arguments raised", Linguistic Inguiry, 20.2.
Bekkum, W. Jac. Van (1983) "The origins of the infinitive

in Rabbinical Hebrew" , Journal of Semitic Studies,
28.2.
Berman, R . (1978) Modern Hebrew Structure, Tel-Aviv,

University Publishing Projects.
435
Borer, H. (1984) Parametric Syntax, Dordrecht, Foris
Publications .
Borer, H. (1988) "On the morphological parallelism

between compounds and constructs", in Booi j , G . and
Van Marie (eds) , Yearbook of Morphology, Dordrecht,
Foris Publications.
Borer, H. (1989) "Anaphoric Agr", In Jaegli 0. and Safir

K. (eds), The Null Subject Parameter, Kluwer Academic
Publishers .
Borer, H. (Forthcoming) "Derived Nominals and the

Causative-Inchoative alternation: Two case studies
in parallel morphology", Ms. UC Irvine.
Burzio, L. (1981) Intransitive Verbs and Italian

Auxiliaries , Doctoral Dissertation, MIT, Cambridge,
Mass .
Carter R. (1984) "Sous-Categorisation et régularités

selectionelles" , Communications 40.
Chierchia, G. (1984) Topics in the Syntax and Semantics

of Infinitives and Gerunds, Doctoral Dissertation,
University of Massachusetts, Amherst.
Chierchia, G. (1985) "Formal Semantics and the grammar of

Predication", Linguistic Inquiry, 16.3.
Chomsky, N. (1986a) Knowledge of Language, New york,

Praeger .
Chomsky, N. (1986b) Barriers , MIT Press, Cambridge,

Massachusetts.
Cohen, T. (1990) "A Determiner Phrase approach to Hebrew

nominals", Ms. Bar Ilan University
Comrie, B. (1976) "The syntax of action nominals, a cross

linguistic study", Lingua 40.
Cowell, M. (1964) A reference grammar of Syrian Arabic,

Washington, D.C. Georgetown University Press.
Di-Sciullo, A.M. and Williams E. (1987) On the Definition

of Word, MIT Press, Cambridge, Massachusetts.
Demirdache, H. (1989) "Nominative NP's in modern Standard

Arabic", Ms. MIT, Cambridge, Mass.
436
Doron, E. (1983) Verbless predicates in Hebrew, Doctoral
Dissertation, University of Texas, Austin.
Doron, E. (1989) "Derived nominals in Hebrew" Ms. Hebrew

University of Jerusalem.
Dowty, D. (1989) "On the semantic content of the notion

of 'thematic role’" in Chierchia, G., Partee B.H.
and Turner R. (eds) (1989) Properties , Types and
Meaning Vol. II Kluwer Academic Publishers.
Driver, S.R. (1892) A Treatise on the use of tenses

in Hebrew, Oxford, Clarendon Press.
Emonds, J . (1985) A Unified Theory of Syntactic

Categories , Dordrecht, Foris Publications.
Emonds, J . (1989) "Timberlake's nominative objects in

Finnish and old Russian" , Linguistic Society of
America, Annual conference.
Fabb, N. (1984) Syntactic Affixation, Doctoral

Dissertation, MIT, Cambridge, Mass.
Fassi-Fehri, A-K (1982) Linguistigue Arabe: Forme e t

Interpretation , Rabat, Publications de la faculté
des lettres et sciences humaines.
Fassi-Fehri, A-k (1988) "Generalized IP structure, Case

and VS word order", Ms. Faculty of letters, Rabat
Finer, D. (1985) "The Syntax of Switch Reference",

Linguistic Inguiry 16.
Fukui, N. and Speas M. (1986) "Specifiers and

Projections", Ms. MIT, Cambridge, Mass.
Gazdar, G., Klein E. , Pullum G. and Sag I. (1985)

Generalized Phrase Structure Grammar, Cambridge ,
Harvard University Press.
Gesenius, W. (1920) Hebrew Grammar, Oxford University

Press.
Gil, D. (1982) "Case marking, phonological size and

linear order", in Syntax and Semantics, Vol. 16
Academic Press Inc.
Glinert, L. (1989) The Grammar of Modern Hebrew,

Cambridge, University Press.
437
Grimshaw, J . (1989) Argument Structure, Ms. Brandeis
University.
Haegeman, L. and Van Riemsdijk H. (1986) ’’Verb projection

raising, Scope and the typology of rules affecting
verbs ” , Linguistic Inquiry 17.3.
Hale, K. (1983) "Warlpiri and the grammar of non-

configur ational languages", Natural Language and
Linguistic Theory 1.
Haywood, J .A. and Nahmad H.M. (1965) A New Arabic

Grammar , Cambridge, Harvard University Press.
Hazout, I. (1988) "Nominalization constructions in Modern

Hebrew," Ms. University of Massachusetts, Amherst.
Hazout, I. (1989) "The Hebrew pronoun Ze and the syntax

of sentential subjects", Ms. University of
Massachusetts, Amherst.
Heim, I. (1989) "Articles and definiteness", Ms. UCLA.
Higginbotham, J . (1983) "Logical Form, Binding and

Nominals " , Linguistic Inquiry 14.3.
Higginbotham, J . (1985) "On Semantics", Linguistic

Inguiry 16.4.
Higginbotham, J . (1987) "Definiteness and Predication"

in Reuland E. and Ter Meulen A. (eds.)
The Representation of (In) definiteness, Cambridge ,
MIT Press.
Higginbotham, J. (1989) "Elucidations of meaning",

Linguistics and Philosophy 12.4.
Horn, G. (1975) ”0n the nonsen tential nature of the

Poss-ing construction", Linguistic Analysis 1.4.
Hoyt, K. (1989) "Verb raising in Lebanese Arabic", MIT

working papers in Linguistics 11.
Huang, J . (1982) "Complex predicates in generalized

control", Ms. Cornell University.
Jackendoff, R . (1977) X* -Syntax: a Study of Phrase

Structure, Cambridge, MIT Press.
Jackendoff, R. (1990) "On Larson's treatment of the

double object construction", Linguistic Inquiry
21.3.
438
Jacobson, P . (1990) "Raising as function composition",
Linguistics and Philosophy 13.4.
Jaeggli, O. (1986) "Passive”, Linguistic Inquiry 17.
Kitagawa, Y. (1986) Subjects in Japanese and English,

Doctoral Dissertation, University of Massachusetts,
Amherst.
Kuroda, S-Y (1968) "English relativization and certain

related problems", Language 44.
Larson, R. (1988) "On the double object construction",

Linguistic Inguiry 19.3.
Lebeaux, D. (1988) Language. Acquisition and the Form

of the Grammar, Doctoral Dissertation, University of
Lefebvre, C. and Muysken P. (1988) Mixed categories,

Nominalisations in Quechua, Kluwer Academic
Publishers
Levin, B. and Rappaport M. (1986) "The formation of

adjectival Passives", Linguistic Inquiry 17.
Manzini, M.R. (1983) "On control and control theory",

Milsark, G.L. (1988) "Single -Ing", Linguistic Inquiry,

19.4.
Mohanan, K.P. (1983) "Functional and anaphoric control".

Moortgat, M. (1987) "Compositionality and the Syntax of

words " .
Muhammad, M.A. (1987) "On the parallelism between IP and

DP", Proceedings of the West Coast Conference on
Formal Linguistics.
Mushaweh, L. (1986) De la Syntaxe des Petites

Propositions , These de Doctorat, Universite de PARIS
VIII.
Muysken, P. (1989) "Predication chains: Case and

argument status in Quechua and Turkish",
Owens, J . (1984) "The Noun Phrase in Arabic grammatical

theory", Al-9arabiyya 17.
439
Owens, J . (1988) The Foundations of Grammar: Introduction
to Medieval Arabic Grammatical Theory, Amsterdam ,
John Benjamins.
Pollard, C. and Sag I. (1987) Information based Syntax and

Semantics , CSLI.
Rappaport, M. (1983) "On the nature of derived nominals”,

in Levin, Rappaport and Zaenen (eds.) Papers in LFG,
Bloomington, Indiana.
Rapoport, T. (1987) Popular Nominal and Small Clauses:

A Study in Israeli Hebrew, Doctoral Dissertation,
MIT, Cambridge, Mass.
Raposo, E. (1987) "Case theory and INCL-to-COMP: the

Inflected Infinitive in European Portugese” ,
Linguiste Inquiry , 18.1.
Reinhart, T. (1983) Anaphora and Semantic Interpretation,

The University of Chicago Press.
Reuland, E. (1983) "Governing -Ing", Linguistic Inquiry

14.1.
Ritter, E. (1985) "What eyn't there", Ms. MIT,

Cambridge , Mass .
Ritter, E. (1986) "NSO Noun Phrases in Modern Hebrew", in

McDonough J. and Plunkett B. (eds.) Proceedings of
NELS 17.
Riemsdijk, H. van (1983) "The Case of German adjectives",

in Heny, F. and Richards B. (eds.) Linguistic
Categories : Auxiliaries and related puzzles,
Dordrecht, Reidel.
Riemsdijk, H. van and Williams E. (1981) "NP-Structure" ,

The Linguistic Review 1.
Ross, J.R. (1972) "The category squish: Endstation

Hauptwort", Proceedings of CLS.
Rothstein S. (1985) The Syntactic Forms of Predication,

Doctoral Dissertation, MIT, distributed by IULC.
Rouveret, A. (1980) "Sur la notion de proposition finie.

Gouvenement et Inversion", Language 60.
Rouveret, A. and Vergnaud J-R. (1980) "Specifying

reference to the subject", Linguistic Inguiry, 11.1.
440
Schachter, P. (1985) "Parts of Speech", in Shopen T. (ed)
Language typology and Syntactic description Vol . I ,
Cambridge Univesity Press.
Scheindlin, R . (1978) 201 Arabic verbs, New York,

Barrons .
Schwarzschild, R. (1989) "The status of the particle *Sr

in Biblical Hebrew", Ms. University of
Shlonsky, U. (1987) Null and displaced Subjects, Doctoral

Dissertation, MIT, Cambridge, Mass.
Shlonsky, U. (1990) "Construct-State nominals and

Universal Grammar", Ms. Haifa University.
Siloni, T. (1990) "Hebrew noun phrases: Generalized noun

raising", Ms. Universite de Geneve.
Srivasta, V. (1988) "Correlatives in a theory of Relative

Clauses", Ms. Cornell University.
Stern, N. (1979) "The et verbs in Israeli Hebrew",

Hebrew Computational Linguistics 15.
Stowell, T. (1981) The Origins of Phrase Structure,

Doctoral Dissertation, MIT, Cambridge, Mass.
Stowell, T. (1987) "Subjects, Specifiers and X-bar

theory", Ms. UCLA.
Stowell, T. (1989) "Raising in Irish and the Projection

Principle" , Natural Language and Linguistic Theory
7.3.
Toman, J . (1986) "A (word-) Syntax for participles",

Linguistische Berichte 105.
Travis , L . ( 1 9 8 4 ) Parameters and Effects of Word-Order

Variation, Doctoral Dissertation, MIT, Cambridge,
Mass .
Vergnaud, J-R. (1985) Dependences et Niveaux de

Representation en Syntaxe, Amsterdam, John
Benjamins.
Waltke, B. K. and 0* Connor M. (1990) An Introduction

to Biblical Hebrew Syntax, Indiana, Eisenbrauns.
441
Webelhuth, G. (1989) Syntactic Saturation Phenomena and
the Modern Germanic Languages, Doctoral ,
Dissertation, University of Massachusetts, Amherst.
Williams, E. (1980) "Predication”, Linguistic Inguiry

11.1.
Williams, E. (1981) "Argument Structure and Morphology",

The Linguistic Review 1.
Williams, E. (1982) "The NP cycle". Linguistic Inquiry

13.2.
Williams, E. (1983a) "Syntactic vs. semantic categories",

Linguistics and Philosophy 6.
Williams, E. (1983b) "Against small clauses", Linguistic

Inquiry 14.2.
Williams, E. (1985) "PRO and subject of NP", Natural

Language and Linguistic Theory 3.
Williams, E. (1986) "A reassignment of the functions of

LF", Linguistic Inquiry 17.
Williams, E. (1987a) "Implicit arguments, the Binding

Theory and Control", Natural Language and Linguistic
Theory 5.
Williams, E. (1987b) "NP Trace in Theta Theory",

Linguistics and Philosophy 10.
Williams, E. (1989) "The anaphoric nature of theta

roles " , Linguistic Inquiry 20.3.
Williams, E. (1990) "Clause structure and C-Command", a

talk given at the annual meeting of the Israeli
. Society for Formal Linguistics, Tel-Aviv, June 11
1990.
Wise, H. (1975) A tranformational Grammar of Spoken

Egyptian Arabic, Oxford, Basil Blackwell.
Woisetschlaeger E. (1983) "On the question of

definiteness in "an old man's book"", Linguistic
Inguiry 14.1.
Wright, W. (1988) A grammar of the Arabic language (Third

edition), Cambridge University Press.
Wunderlich, D. (1987) "Partizipien im Deutschen",

Linquistische Berichte 111.
442
Zubizarreta, M-L. (1985) "The relation between
morphophonology and mor phosyntax , the case of
Romance Causatives”, Linguistic Inquiry 16.
Zubizarreta, M-L. (1987) Levels of Representation in the

Lexicon and in the Syntax, Dordrecht , Foris
Publications.
443

Hazout 1991 Thèse

Uploaded by

Copyright:

Available Formats

Hazout 1991 Thèse

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Hazout 1991 Thèse

Uploaded by

Copyright:

Available Formats

VERBAL NOUNS: THETA THEORETIC STUDIES IN

HEBREW AND ARABIC

Submitted to the Graduate School of the

HEBREW AND ARABIC

Approved as to style and content by:

Elizabeth Selkirk, Chair

Shmuel Bolozky, Member

Barbara H. Partee, Department Head

I would like to thank the members of my dissertation

Selkirk. Roger and Lisa read earlier drafts of this

extremely detailed and helpful and led to many

Thanks to Edwin Williams under whose direction most

be recocnized throughout this dissertation. The elegance

from the very beginning. His sympathy and encouragement

were important for the development of this work.

Gingi Fu, Greg Lamontaigne, Reine Pinsonneault ,

dissertation would bring some joy into my sister,

moment. This dissertation is dedicated to her memory.

VERBAL NOUNS: THETA THEORETIC STUDIES IN

HEBREW AND ARABIC

ILAN HAZOUT, LICENCE, UNIVERSITE DE PARIS VIII

MAITRISE, UNIVERSITE DE PARIS VIII

Ph.d. UNIVERSITY OF MASSACHUSETTS

Directed by: Professor Elizabeth Selkirk

This dissertation is a study of a variety of

constructions in Modern Hebrew and Standard Arabic which

involve nominalization processes. Such constructions

manifest a certain mixture of verbal and nominal

properties and are analyzed as involving a verbal

subconstituent, a VP, governed by an underlying nominal

head, a nominalizer . The surface form of the deverbal

head of such constructions is . the output of a head

movement operation adjoining a verb to the nominalizer

which governs it. The properties and the differences

between the different types of nominalization

constructions are explained on the basis of ceratin

assumptions about the thematic properties, the argument

structure, of the different nominalizers that are

postulated. The heads of nominalization constructions

are morphological as well as thematic nominalizers in

that they provide, in addition to a particular •-

properties of these costructions are derived from the

This approach to nominalizations is embedded within

structure combined with theoretical techniques developed

in recent work within the Government and Binding theory,

Chapter 1 presents the main theoretical assumptions

control. Chapter 2 is a comprehensive study of genitive

constructions in Hebrew and Arabic. Chapter 3 is a study

detailed argumentation in a favour of a non-lexicalist

standard infinitives and the verbal gerund, a

construction which is particular to Modern Hebrew.

functions as a morphological nominalizer.

1.1 Head Movement, Morphology

1.2.1 Argument Structure, Theta

1.3 Levels of Representation ................... 43

Footnotes to Chapter I................. 47

II THE SYNTAX OF GENITIVE CONSTRUCTIONS ......... ..49

2.1 Bound Genitives (The Construct State) ..... 51

2.1.1 The Data............................... 51

2.2 The Interpretation of Genitive