7 RF.

SUMES
ED 016 965 AL 000 976
CONSTRAINTS ON VARIABLES IN SYNTAX.
BY- ROSS: JOHN ROBERT
MASSACHUSETTS INST. OF TECH., CAMBRIDGE
PUB DATE SEP 67
EDRS PRICE MF-$2.66 HC- $21.00 523F.

DESCRIPTORS- *TRANSFORMATION GENERATIVE GRAMMAR, CONTEXT FREE

GRAMMAR: *DEEP STRUCTURE, *SURFACE STRUCTURE, TRANSFORMATIONS
(LANGUAGE) ,

IN ATTEMPTING TO DEFINE "SYNTACTIC VARIABLE," THE AUTHOR

BASES HIS DISCUSSION ON THE ASSUMPTION THAT SYNTACTIC .FACTS
ARE A COLLECTION OF TWO TYPES OF RULES- -CONTEXT -FREE PHRASE
STRUCTURE RULES (GENERATING UNDERLYING OR DEEP PHRASE
MARKERS) AND GRAMMATICAL TRANSFORMATIONS, WHICH MAP
UNDERLYING PHRASE MARKERS ONTO SUPERFICIAL (OR SURFACE)
PHRASE MARKERS. THE THESIS PRESENTS A SET OF CONSTRAINTS ON
VARIABLES--UNIVERSAL AND LANGUAGE-PARTICULAR--AND DISCUSSES
HOW THEY AFFECT SYNTACTIC RULES. HE POINTS OUT THAT CHOMSKY'S
A-OVER-A PRINCIPLE IS BOTH TOO STRONG AND TOO WEAK, BECAUSE
THE PRINCIPLE CANNOT BE EXTENDED TO ACCOUNT FOR THE PHENOMENA
WHICH THE AUTHOR CALLS "SYNTACTIC ISLANDS." THE ENTIRE
DISCUSSION CENTERS AROUND RULES AND CONSTRAINTS: WHICH ARE
ACTUALLY LIMITS ON THE POWER OF VARIABLES THAT CAN APPEAR IN
CERTAIN TYPES OF RULES. IT IS SHOWN THAT CONSTRAINTS,
ESPECIALLY WITH THE NOTION OF "COMMAND," DIVIDE P:'11ASE
MARKERS INTO "ISLANDS," SUGGESTING THAT THESE "ISLANDS"
BEHAVE LIKE PSYCHOLINGUISTIC ENTITIES. THE AUTHOR PREFERS TO
MAKE THE RELATIONSHIP BETW:EN GRAMMATICALITY AND
ACCEPTABILITY MORE ABSTRACT THAN PREVIOUSLY ASSUMED BY
TRANSFORMATIONALISTS. THIS THESIS WAS PREPARED AS PARTIAL
FULFILLMENT OF THE REQUIREMENTS FOR NE DEGREE OF DOCTOR OF
PHILOSOPHY AT THE MASSACHUSETTS INSTITUTE OF TECHNOLOGY. (FB)
 ."0

CONSTRAINTS ON VARIABLES IN SYNTAX

11." by
%4C)

III
w
r-4 J0414 ROBERT ROSS

U)
B.A., Yale University (1960)

M.A., University of Pennsylvania (1964)

U.S. DEPARTMENT OF HEALTH, EDUCATION & WELFARE

OFFICE OF EDUCATION

THIS DOCUMENT HAS BEEN REPRODUCED EXACTLY AS RECEIVED FROM THE

PERSON OR ORGANIZATION ORIGINATIRG IT. POINTS OF VIEW OR OPINIONS

STATED DO NOT NECESSARILY REPRESENT OFFICIAL OFFICE OF EDUCATION

POSITION OR POLICY.

SUBMITTED IN PARTIAL FULFILLMENT

OF THE REQUIREMENTS FOR THE
DEGREE OF DOCTOR OF
PHILOSOPHY

at the

a .....
"PERMISSION TO REPRODUCE THIS
delnatito MATERIAL HAS BEEN GRANTED MASSACHUSETTS INSTITUTE OF
BY 1/1-4411.4226_ TECHNOLOGY

TO ERIC AND ORGANIZATIONS OPERATING

UNDER AGREEMENTS WITH THE U.S. OFFICE OF
EDUCATION. FURTHER REPRODUCTION OUTSIDE
September, 1967
THE ERIC SYSTEM REQUIRES PERMISSION OF
THE °WNW

Signature of Author
Department of Modern Languages, August 21, 1967

Certified by
Thesis Supervisor

hairraan, Departmental Committee

Grliiduate students
 -W1 _

ii

DEDICATION

To four of my teachers.

Bernard Bloch, Zellig Harris, Noam Chomsky and Morris Halle

who have awoken in me, and intensified

by their ever-deeper insights, the desire
to understand Nan through an unraveling
of the mysteries of his language; and

to aly mother,

Eleanor Campbell Mott Ross,

who, although she does not understand

how anyone could want to study language,
has spared no effort to let me study
where, what, and how 1 want to,

I dedicate this thesis.

At,
 iii

CONSTRAINTS ON VARIABLES IN SYNTAX

by

John Robert Ross

submitted to the Department of Modern Languages and Linguistics on

August 21, 1967, in partial fulfillment of the requirements for the
degree of Doctor of Philosophy.

ABSTRACT

This thesis attempts a definition of the notion syntactic

variable, a notion which is of crucial importance if the central
fact of syntax, that there are unbounded syntactic processes, is
to be accounted for. A set of constraints on variables, some
universal, some lelgiage-particular, is presented, and the question
of what types of syntactic rules they affect is raised. It is
shown that these constraints, in conjunction with the notion of
command, partition phrase markers into islands -- the maximal
domains of applicability of all rules of a specified type.

Thesis Supervisor: Noam Chomsky

Title: Professor of Linguistics
 iv

FRAGESTELLUNG

The following anecdote is told of William James. I have

beer', unable to find any published reference to it, so it may be that

I have attributed it to the wrong man, or that it is apocryphal. Bc

that as it may, becaUse of its bull's-eye relevance to the study

of syntax, I have retold it here.

After a lecture on cosmology and the structure of the

solar system, James was accosted by a little old lady.

"Your theory that the sun is the center of the solar

1
system, and that the earth is a ball which rotates around it, has a

very convincing ring to it, Mr. James, butit's wrong. I've got a

better theory," said the little old lady.

"And what, is that, madam?" inquired James politely.

"That we lire on a crust of earth which is on the back

of a giant turtle."

Not wishing to demolish this absurd little theory by

bringing to bear the masses of scientific evidence he had at his

command, James decided to gently dissuade his opponent by making her

see some of the inadequacies of her position.

"If your theory is correct, madam," he asked, 'Nwhat

does this turtle stand on?"

"You're a very clever man, Mr. James, and that's a

very good question," replied the little old lady, "but I have an
answer to it. And it's this: the first turtle stands on the back of

a second, far larger, turtle, who stands directly under him."

."But what does this second. turtle stand on?" persisted

James patiently.

To this, the little old lady crowed' triumphantly,

"It's no:use; Mr. James --It's turtles all the way

down."

5
 vi

ACKNOWLEDGEMENTS

This thesis ends an overly long career as a professional

atflapnte a carper which has been a by to me, but a trial to the

many teachers and administrators who have gritted their teeth and

forgiven the lateness of papers and assignments (seVeral are still

late), the frequency with which I asked ill thought-out questions;

and my chronic unpunctuality. They have put up with all this in

t e hope, that something might become of me someday. To these

i

friends I can only offer this thesis, in the hope that it. will

in part repay their confidence in me.

For aid and support beyond the call of duty, I must

single out the following for special gratitude:

The Deutscher Akademische Austauschdienst, who

gave me a chance to dabble around at three

German universities, apparently learning nothing) -

but in reality finally realizing that it was time

to stop dabbling and study seriously;

The Woodrow Wilson Foundation, who gave me one

last chance that 1 had no right to, sending me

to the University of Pennsylvania) where at last

I did stop dabbling;

,
 vii

Don Walker, of the MITRE Corporation, who allowed

me to spend the summer of 1963 at MITRE, a summer

in which I produced nothing, but learned more

through reading and talking than in. any comparable

period of my life;

Naomi Sager, for whom I worked for almost a

year as a research assistant on the String Analysis

Project of the University of Pennsylvania, and

who never complained about the extent to which I

neglected my job;

Hu Matthews, who I worked for when I first came

to MIT, again giving precious little to show for

it;

John Olney, of the Systems Development Corporation,

who supported my work during a pleasant California

summer in 1965;

The National Institutes of Health, for a Pre-

Doctoral Fellowship during the year 1965-1966; and

 viii

Susumu Kuno, of the Harvard Computation Laboratory,

who supported my work during 1965-1966, and

eliminated many oversights and inadequacies in my

work with insightful counterexamples.

The typing of a thesis this size is a job of Herculean

proportions, and proofreading it can be almost as bad. It is there-

fure with great pleasure that I thank Ellie Dunn, Patricia Wanner,

and, because she did the bulk of it with a speed.and industry which

were incredible, especially Lorna Howell. The care and accuracy with

which these girls prepared the manuscript made proofreading as

enjoyable as I have ever known it.

I would also like to expresi my thanks to Dwight Bolinger,

of Harvard, for the care that he has devoted to reading, and commenting

on, various papers of mine, some related closely to the thesis, some

not, and for the many deep insights into syntax that his comments

afford.

To Roman Jakobson, I owe a special debt: not only has

he always given me freely of his time, for discussion of a wide range

of problems, but he loaned me his office in Boylston Hall, so that I

could break out of the becalmed state I had gotten into. Without

his generosity, the thesis would not have been finished this summer.

Each member of the BIT Linguistics Department: -..as helped

me overcome some obstacle in my work.' Hu Matthews helped me to see

 ix

Subject Constraint (cf. 6 4.4) more

the nature of the Sentential
insight that
clearly, and to formulate it.) Ed Klima's fundamental
of Chapter 4 (cf. § 4.1.1)
pruning interacts with the constraints
of the thesis. And Paul
has been the indirect source of most
behave the same with respect
Kiparsky's insight that factive clauses

rules (cf. 5 6.4) leads

to feature-changing rules and reordering
important concepts
directly to whatI regard as one of the most

developed below --the concept of islands.

is less direct,
My debt to the remaining three members

It was from Paul Postal's

but no less important, for'all that.
of a highly abstract,
lectures in 1964 and 1965 that the conception
contained only nouns
but probably universal, deep structure, which

and verbs, emerged. It is 'to the end of establishing the correctness

and my work, including this

of this conception that most of Lakoff's

thesis, has been directed.

Morris Halle, in addition to running a department which

discovery, has somehow

contains an atmosphere uniquely conducive to
distinction between
been able to get across to me the all-important

solutions to problems (i.e., devices that work, but...) and

in science.
explanations for phenomena, the most crucial distinction

What I owe to Noam Chomsky is incalculable. Unless he

2), it is doubtful
had formulated the A-over-A principle (cf. Chapter
this thesis is
whether I would have even noticed the problems which

devoted to solving. I disagree with him on many 'particular points of

analysis, but since it was really from his work that I learned how to

construct an argument for or against a proposed analysis, my ability to

disagree also derives from him. I am deeply grateful to him and to

Halle for helping me to understand what it is that a theory is.

It is impossible to thank all my friends individually for

their contributions, so I will select three. David Perlmutter, aside

from the great amount I have learned from his work, has also taught me

a lot about my own, through serving as a backboard for my new ideas

and pointing out unclarities and inconsistencies. He has also helped

proofread the thesis, for all of which I thank him. Bruce Fraser

has helped in every kind of way 7- linguistically, technically,

financially. I cannot thank him sufficiently.

This thesis is an integral part of a larger theory of

grammar which George Lakoff and I have been collaborating on for

the past several years. Since there is close interaction between

the theory of variables reported here and almost all facets of the

larger theory, it is impossible to guess what kind of thesis I would

have written on this topic had we not worked together in delving

down into deeper and deeper layers of turtles. Where I can remember,

I have tried to give him credit for particular ideas of his. I

ask him to accept this general word of thanks for all the places I

have forgotten.

Finally, I come to my family. Since in my view cats

are as necessary as air or water, I thank our cats Krishna and Aristotle

,,Mor.r777.-nw7,77.7. c,,,07:77.774777,77"--
 +

xi

for deigning to stay with us and seasoning our existence To my

new son Daniel. Erik I owe the added impetus that pushed me to finish

the thesis this summer. The ease with which this three-month-old

ti
child dislodged the completion of the thesis from its central position

in the universe, to assume this position himself, made use realize

that once he became ambulatory, thesis writing of all sorts would

cease.

I have no words with which to thank my wife Elke-Edda.

The writing of this thesis has been as much of an ordeal for her as

it has for me, for which I beg her forgiveness. For making my life

as easy as it could be, under the sword of Damocles, I thank her with

my heart.
 xii

Table of Contents,

.Dedications

Abstract

Fragestellune OOOOO OOOOOO ..iv

Acknowledgements.'Vi
10 OOO 1 Introduction

12
O

11.4bOOc OOOOOO .7

12.. OOOOOOO . 10

Footnotes . OOOOOOOOOO OOOOOOOOOOOO .........11

The Ar.over-A Principle,

20 OOOOO
2.1......4.0..... OOOOO .... OOOOO OOOOOO .14
:
2.2.. OOOOO 00000006 OOOOO OOOOO .16

OOOOOOOOOOOOOOO 24
dko.0;46000040 OOOOO OOOOO

OOOOO ........ 25

25
OOOOOO OOOOOOOOO 0000000600400000000000008 OOOOO 9050 OOOOOOO

0. OOOOOOOOOOOOOOOOO OOOOO 26
2.4.1 OOOO OO OOOOOO

2.4.2. OOOOOOOOOOOOOOOO * OOOOO . 29

OOOOO OOOOO I OOOOO 000000040!t;000400004000 OOOOOOOOO 34

Footnotes OO OOOOOOOO OOOOOOOOOOOOOOOOOOO 37

Tree Pruning

OOOOOOO OOOOO O

3.0.0.. OOOOO OOOOO OOOOO OOOOO ....................o......41

ft 4

Nft a2:4 :e ".0 ^

 30001600000000000000011000004000041.00060000.0041,40000000000006.00644

31100 OO
s

30002O.000646066.00046000000011410004000600006600606**040000,0400060M0645

301006.1.06000000040000.000040004.04600041444.40000047
OO 47

3oldele o e "1/
o ..
3,11.3. ................51
3.1.1.2. o 49

3113. .51
e ._.0

31.1.3.2 . . . e62
301010303.00000$0,000000000000.006.0600000.000000000066606.001T00,69

3,191.4,.......,..0.0.....73
3.1.3..-,00o80
3.1.4.,0 OO .:.88
316e*00494
OO

31.7.. 460 OOOOO 96

32. 6 6100:
3i108404100000400000,000000.004.00.1100000006000000000000006.98
3.1.90..e......a........99

4 00117
Footnotes....

41, 60118 .The IJmplex NP Constraint

.

Constraints on Reordering Transformations .

102

c4}.`
4'74 1,, .0
-..=11111111
 , ..

ATX

811O0001000080900000090110000110000001001100006600106011000.000.0..0041016,1

mpoom55olo1100550,WV110V orroosiposesesso4ripooloo,000m000mibo5
. . .

9n60400000600000000000.0000006.0000000000000401,004100000411000001010V
OOOOOOOOOOOOOOOO e1.

050111.0. 6E10
',Iry ,-,,........... 00,1000114b04111141100001004,94
. ..
'

7-......_S
L

coe.o.e,59Tv 89,T.....0..

sw..............99.......................,....................L.T.v
...

suo3 ..........4uTea4-- OOOOO KT...

aztfuTpa000'biu aanlonals

1p Z 8cr0000mIll0000mee011ooewoo0500055560)010o0e000041000100WITI

Niy 1 T9T600800006011011100000080000000008,111C000040006006100000000060,041z.
P
1.918,111004,1i000101000000950000000000000000000000011100000666W.500E0z0",

itr0000rnowillowomoosogoome1wor000romose00050000***00eVozo
- f

1711000110000000000001100000050000000.0006011606606660606066.4poitvorly
.

811
O0000,0#000000000000000050000000000)0000.01100000000000000801116z1,

98.1154,000000060601119000110410000000006000000051100000000000000eSSOCIOVV

.r.z.v .......0 .....5 ..........,................0 06T.......9.......0

s6Teeet0940004,e00000006*,000e0WO*110410900441o6011wooso*04100rnson

au_ plc/ 2UTdTa "*"1111.10T3MAUOD ** 118.4"."01"4"*"" 96T

961
666664060606011141666004004)0041001111101,0601000WISi90411410411111411006.000TOCO

90V00000000011100,00.00410111000540004190.0091114)00000000000000041TOCO

o5114114ioefornIs09140**0.00004,004oVies ** 0 ** 90z000s000rnott000mornee

1140000000000000TOZOC44.1 ***** LoV000060000001110000101,0960004100111000.6:1.,9

LI.c.6410000011101941000**90414,4)000000000410:0000004000410009,0044041006zizerill

EzzeiglowomeormoomIwo4,61504b000m04,414100,041Wern0400400Eorco

8ul,000041000410004,0000000411000000001100041004111.541501900000414,4144000v0Vril
Z £9
060411.04,000000110990111600400111,000900000004000000000000401100001110SOZOV

''..

,..4.--0' ,...,,..,
 t:
XV

436304100410414100000041000000414100000000000.41.411060000004044000006410600240

4.4. The Sentential Subject Constraint owes*,4,eowtboiso241

4.4.1...o....m....woesmoososooioessooyogoomeessoossisoo.o241
4.42most0000seees*Ioeoosoeosomooses0000000*44,.....essoseooses24.5
.

404.0300690000041004100410000041041000440410044004100041000,0.41410041004104100000252

405o64141041000004104100000000 0044000000000004100000410414,410416041000.4104141000255

50
Footnotes.... c0256

5; 1 Bounding,soeeasowase
..... .

Bounding,'Commandi. and Pronominalization,

.......... .

267

5.1.1.1,0268
501010000000000,04400410000000440041000410030000000000000004100040041o410.26 8

5.112e260
54296
5.113I04,...285
.11
50102090009000000000084100410414141410000944004100400,000000000000000410041,96298

5010201000000:*000004110000440041004100100000000000000414144000000000000.0298

5010202041004100041000.440000000.0441410000410000041000000410041000000041990.301

50102.3000040,4000.00000.000.444400004141000440.44444404141041444144440040041.440000306

5.103.6eolosomos000m0000o0.6o0ososoos**secom00000s00000soo..*6312
it

5.1.3.20:318
5010.3.1000000004100041410004141410410,000410410410004100000000000600000000312

5.1.321,318
5.1.3.2.2:322 .

47:77 ,77i77,77017g.""77""
, '4- .

71.
rzecTes
47zvrc
IIsozorrs
9zrts
lerc

izs,
0Trs

.
'.
te puvutta03
e too

0111114
IP,

50
*
so oo

... . .. G££..

.3 .
I

.IIII
000 f.41194
.

vcc
.0 05
I,'

.
35

8CE
sEr
BEE
TAX

SZEO

6zt
Icy

Evc.zzes
*Terris, II O sEt
oe
I II ice
ezerrig .. 0,Ke

*'
crs
Tcrs
zErs
. . ..
GOO

0. f, 1
ItC
L9E
zst

0Es
I .
. I.
uotwenlympuouola:

.
.. . vsE
ipsE

meg
.04, I

rCS .
.
crs . . .
IS II II
. I
fist

4
40 sce.
e 4p. . . .. C9C
00Erg O. G I II . C9C
'Eci. 0 4
.... . 479E0

OzEEs, I o .. g9C
.
.Ecrs
fetrs
. . . . ..
II
. L9E .
89E0(10,11
 xvii

5.3040080000,00.0000.040000000000000000,000004100000000.4,000,0000094369
.
.
..
4 ........ ,
,

5040,10i 00040W,00000000008000010004.0011,0000.i00000000004.110.00.000.372
1

Footnotes.0.00 WWWWWWW Wwww..174

On the Notion "Reorderinz Transformation"

6.00,0401*&.dowerneeeetroee,00,382
6.1. Some Rules'Obeying the Constraints6. 46.383

6103c4...................383

6.1.1.1...i...........e.o.......60.64........................385a

64,1,1,30

6.1.1.4
ee oP; :..388
6112....1..serne*,emernee0480386

OOOOO
i
391
64,1415 Ofieoesolo0000e 391
'itdi
60101666 .......4,394
64,1411.7.6........6..396
6.1.2.64100006011'/,.398
6120ee0398
6.102.10.00041.00000.4000008.0000000000398
6.1.2.2 ..06400
6.1.2 3

6.1.2.5...5......m....)0..406
6.1.2.6 .406
6.1.207.0.0ornowolleow.sorno.rneforneofoloos0000sossomoves..408

%loge ...N... a pi. aaa YM %N1.

641.2.8.641.41 ir, ...409 .
.

613:41,044:444,:e40,44, so.410
6136046a:4 .414 .

641113010000eie.. c
.4.411

.14415

6.1.3.3...........................................................416

620.:
6.2.

6.2.0....:
Chopping Rules...

IP *II IP IP IP It IP IP

6820300110000416000000000,0041.00000.00418,000000006000000000000000432
.4... .422

IP IP IP .428

6.2.4. 44441,40:orne,44434
6.20402..... .w437
...... ..........434
602.500.041100,.0,011WOMo40060411000Oplio*O0o000MOSOoOloo6000.41000.442

6.3. Reordering over Variables.... ....................... .442

6114061100
6.4. Islands.... ..........................................449

3,...O0.0..00.450
6.4010.
o .4..6449

6444,10. 4e414,006646,54iiloorn450
. ,

t"; ;;.,,
 &
.4!

461
xix

6.4.1.1 ..06451 . .
.,

6040164.6CO*
6.44.36.0m 006

6.4.2.
4.
.6.464

6.4.3 .6.469

6.4.3.1. .469

.477

6.5. SummarY.......0 3
Footnotes***6e6***64 .481
=10. Conclusion....

Bibliography0604 .495
Biography* .500 . ot
 ,...,..

Chapter 1

INTRODUCTION

The past decade of vaapAreh on tI.Anafnrmelt4nn.1 grammar

has substantiated amply, to my mind, the claim that the optimal frame-

work for the description of syntactic facts is a set' of rules, of two

types: context-free phrase structure rules., which generate an infinite

set of highly abstract formal objects, underlying (or deep) phrase

markers; and grammatical transformations, which map underlying phrase

markers onto an infinite set of objects of roughly the same formal .

1
character, superficial (or surface) phrase mar_ kers. Within this

framework, an evaluation measure is provided which must select, from .

a set of observationally adequate grammars of some language -- i.e.,

grammars which all generate the observed set of grammatical sentences

of the language -- the descriptively atklatre grammar -- the grammar

which makes correct predictions about strings of words not yet observed,

and can thus be said to reflect linguistic knowledge of speakers of the

2
language. Such knowledge includes intuitions about the immediate

constituents of sentences, about similarity, among constituents, and

about relatedness between sentences. For instance.; a descriptively

adequate grammar of English would have to predict the following facts

about sentence (1.1):

(1.1) A gun which I had cleaned went off.

a) The main constituent break occurs between

cleaned and went; I is a constituent; which I

.0

is not; etc..

A ,

-..tRi4
 2

b) The constituent gun which I had cleaned

is a constituent of ate same kind as the

constituent I. Similarly, went off is the

same type of constituent as had cleaned, and

neither is of the same type as I, a, or off.

c). Sentence (1.1) is related to sentence (1.2).

(1.2) A gun went off which I had cleaned. ry

Within a transformational grammar, intuitions of relatedness

between sentences are reconstructed by deriViag sets of related sentences

from the same or highly similar underlying phrase markers by means of

slightly differing sets of transformations. As a first approximation,

we could postulate a rule like (1.3) to convert the structure under-,

lying (1.1) to the one underlying (1.2)3 (here and elsewhere I will give

rules and tree diagrams in a simplified formes long as it makes no

difference for the point under discussion):'

(1.3) [NP S] VP,

VP OPT

2 3

1 0 3 4- 2

where the phrase marker (P-Marker) associated with (1.1) can be

4
represented as a tree diagram of roughly the following form :

A r .
 3

(1. 1')

NP

NP went. off

,Fun VP NP
I
which I h d cleaned

Rule (1.3) would convert (1.1') into thelerived*P-Marker (1.2')

n.
a gun went off NP . NP VP

whichI had cleaned

It is fairly easy to demonstrate that the present evaluation

measure gives a higher rating to a grammar which has (1.1') as an under-.

lyingP-Marker and derives (1.2') from it by using (1.3), than to one

which assumes (1.2') is.basic; but I will not undertake such a demon-

stration here, since the point at issue is more general, and these rules

I propose are only supposed to illustrate it, not to constitute a'

complete analysis.

Nbw consider the sentences (1.4) and (1.5).

(1.4) 1 gave a gun which I had cleaned to my ,rother.'

(1.5) I gal, a gun to my brother which I had cleaned.

 4

To relate (1.4) and (1.5) -- again, I omit the argument

wbich would prow that (1.5) must derive from (1.4) -- some rule like

(1.6) would be necessary.

(1.6) NP V [NP S] - PP
=nneS NP OPT

3. 2 3

1 0 3 + 2

By the provisions of the evaluation measure, we are forced

to collapse rules which are similar in certain ways, and (1.3) and,(1.6)
. ..
collapse to yield (1.7);
:
".

(1.7) [NP S]
NP PP
1
L"....see.....11 . OPT
2 3

1 0 3 + 2

Consideration of sentences like (1.8) and (19).

(1.8) He let the cats which were meowing out.

(1.9) He let the cats out which were meowing.

and similar sentences might lead one to reformulate (1,7) as an even

more general rule, (1.10), which I will call Extraposition from NP:

(1.10) Extraposition from NP

[NP SI Y
NP NP OPT
1611111......vaorISIMIIIIId

1 2 3

1 0 3+ 2 ..

.,,,,r.,
4,
 5

The symbols X and Y in (1.10) are variables which

range over all strings, including the null string. With them, the

rule as it stands is much too powerful.' For instance, (1.10) would

convert (1.11) into the ungrammatical (1.12).

(1.11) ,

NP
r
I

s
: :, ' 1 *,

t.... :

that NP VP

./'
NP S went
7\ off
1

.surprised -1. o one

.. .i .. "'

NP NP *-

- .-..-. ,.

which
1 :

I had
//"NN
cleanid .
. t

-. .

.:
.
(1.12) 1

NP
1

.,
.
.
4 LW

I
VP surprised : no one -which I cleaned

IL LA wtnt off

The fact is that an extraposed clause may never be moved

outside "the first sentence up," in the obvious interpretation of this

phrase, and there are a number of ways of incorporating this fact Into

.;

A-
 6
. " I. "
4.1
u yr or

a restriction on rule (1.10). One rather obvious way of blocking

sentences like (1.12) , which arise because of the great power which

variables in the structural index o: a 'transformation have, is simply

to eschew entirely the use of variables in the statement of the rule,

and to replaci (1.10) by an expanded version of (1.1), in which all

the nodes, or sequences of nodes, over which clauses may be extraposed

are merely listed disjunctively in the structural index of the rule.

Such a "solution" is feasible for this rule, but any linguist adopting

it will have merely postponed the day of reckoning when he will have

to find a more general way of constraining variables in structural

indices of transformations; for there are many rules whose statement'

requires variables, and these variables cannot be replaced, as far as

know, by disjunctive listings of nodes or sequences of nodes, as is

the case above, with respect to the rule of Extraposition from NP.

One example of a rule in which variables are essential

is the rule which forms WH-questions. It can be stated roughly as

follows (1 ignore many details which are irrelevant for the,purpose at

hand):

(1.13) X NP Y
OBLIG

1 2 3 =====> where 2 dominates WH +'some

2+1, 0 3

This rule produces sentences like those in (1.14), where

it is clear that the questioned element can be moved from sentences

which are indefinitely deeply embedded in a P-Marker:

" .,1111111110M
 1

t: 7

(1.14) What did Bill buy?

What did you force Bill to buy?

What did Harry say you had forced Bill to buj?

What was it obvious that Harry said you had

( forced Bill to buy?

A moment's reflection should convince anyone that it is

impossible to replace the variable X in (1.13) by some such disjunction

as that contained in (1.7): rule (1.13) is not stateable without

variables. And yet, just as was the case with rule (1.10), Extraposition

from NP, it is easy to see that (1.13) is far too strong, for it will .

generate infinitely many non-sentences, such as those in (1.15).

(1.15) * What did Bill buy potatoes and? .

* What did that Bill wore surprise everyone?

* What dl,d John fall asleep and Bill wear?

1.1. Sentences and non-sentences like those in (1.14) and (1.15)

show that some rules must contain variables but that somehow the power

of these variables must be restricted. It is the purpose of this thesis

to try to justify a set of constraints on variables, which I will

7.....?ose in detail in subsequent chapters. There are doubtless many

constraints on variables which are peculiar to individual languages, and

possibly some which are even peculiar to some rule in some particular
r
language, but I have by and large avoided detailed discussion of these

and have instead concentrated my research on constraints which I

suspect to be universal.
 8

It is obvious that the limited character of presently.'

available syntactic knowledge reduces drastically the chances of

4

survival of any universals which can be formulated today, for the

study of syntax is truly 'in its infancy. But it will be seen below

that the constraints on variables which I will propose are often of such

a complex nature that'to state them as constraints on rules in par-

ticular languages would greatly increase the power of transformational

,4

rules and of the kinds of operations on P-Markers they could perform.

But to assume more powerful apparatus in a theory than can be shown

/
to be necessary is contrary to basic tenets of the philosophy of
01 i

science, and so I will tentatively assume that many of the constraints

I have arrived at in my investigations of the few languages I am

familiar with are universal. It is easy to prove me mistaken in this

41

assumption: if languages can be found whose rules are not subject

to these constraints, then the apparatus in theory of generative . '

grammar which provides for the description of languageaaarticular

facts will have to be strengthened so that.rules like the question.

transformation in English, (1.13), for instance, can be stated and

correctly restricted to exclude ungrammatical sentences like those in

(1.15). But until such disconfirming evidence arises, the assumption

of a weaker theory for particular languages is dictated by principles

of the philosophy of science.

It is probably unnecessary to point out that it is common-

place to limit the power of the apparatus which is available for the

description of particular languages by "factoring out" of individual

1
 9

grammars) principles, conditions, conventions and concepts which are

necessary In all grammars: to factor out in this manner is to construct

a theory of language. So, for example, when the principle of operation

of the syntactic transformational cycle has been specified in

linguistic theory, it is unnecessary to include another description

of this principle-in-a 'grammar of French. And so it is also with such

well-known notions as free variation, 'grammatical sentence, constituent,,

coordinate structure, verb, and many others. The present work should

be looked upon as an attempt to add to this list a precise specification

of the notions actic variable. This notion is crucial for the

theory of synton for without it the most striking fact about syntactic
1,1 a.

,orocesses - the fact that they- may. operate over indefinitely large

domains - cannot be cap cored. And since almost all transformations

either are most generally stated, or can only be stated, with the

help of variables, no transformation which contains variables in

its structural index will work properly until syntactic theory has

provided variables which are neither too powerful nor too weak. It

is easy to construct counterexamples such as those in (1.15) for

almost every .transformation containing variables that has ever been

proposed in the literature on generative grammar. It is for this .

reason that attempts to constrain variables, like those which will

be discussed in Chapters 2, 4, and 5, are so important: without

the correct set of constraints,. it is impossible to formulate almost

all syntactic rules precisely, unless one is willing to so greatly

 10

increase the power of the descriptive apparatus that every variable

in every rule can be constrained individually. But one pursuing

this latter course will soon come to realize that many of the constraints

he imposes on individ.4a1 variables must be stated agaia and again; that

1

he is missing clear generalizations about language. the latter

course must be abandoned: the only possible course is to search for

universal constraints. This thesis is devoted to that search. .4.

1.2. The outline of this work is as follows. In Chapter 2,

I will discuss the only previous attempts to limit the power of

5
variables which I know of Chomsky's A- over -A principle and two conditions

subsequently proposed by him, and demonstrate that they are too strong

in some respects and too weak in others. In Chapter 3, I will discuss

a notion which will prove indispensable in stating the universal

constraints: the notion of node deletion, or tree pruning. In

Chapter 4, I state and discuss two putatively universal constraints on

variables, which overcome the inadequacies in the principles discussed

in Chapter 2, and several lea.a geacral constraints. The notion of

bounding, is introduced in Chapter 5. In Chapter 6, I discuss briefly

a number of rules and show that these rules are subject to the

constraints of Chapter 4, but that not all transformations : subject,f.,:,

to these constraints. The question is discussed as to w1 t formal

features of rules determine whether the variables in them are subject

to the constraints or not Chapter 7 is a brief recapitulation of. the

1.
.1.;

results of the thesis.

.", -,....4.44±kOda474
 11

Chapter 1

:. FOOTNOTES.

1. For an excellent introductory article on the difference between

underlying and 'superficial structure, cf. Postal (1964). A

more technical and far more complete exposition is given in

. ,
Chomsky (1965).

' :
2. For further discussion of the notions of observational and
.

descriptive adequacy, cf. ChrAsky (1964b).

3. My notation for transformations follows that of Rosenbaum (1965),

except where otherwise noted.

4. The assumption that relative clauses are introduced in the deep

structure by the rule NP 9 NP S will be justified in Lakoff

and Ross (in preparation b).

5. Except Langacker's notion of command (Langacker (1966)) and Klima's

notion in construction with (Klima (1964)), which will, be discussed

separately in i5 below, in connectio;A with the notion of bounding.

" t.

"°` kk:
V
- ,f 1.
 12

Chapter 2

THE iv-OVER-A PRINCIPLE

2.0. In a paper written for the 1962 Ninth In ternational Congress

of Lineuists, "Theflogical basis of linguistic theory" (Chomsky (1964a)),

on p. 930-931, while discussing the relative clause transformation and

the .question transformation, Chomsky makes the follow ing statement:

"The same point can be illustrated by an example of a

rather different sort. Consider the sentences

(6) (i) who(m) did Mary see walking toward the

. %
railroad station?
(ii) do you know the boy who(m) Mary saw
walking to the railroad station?

(7) Mary saw the boy walking toward the railroad

station.

(7) is multiply, ambiguous; in particular it can have

either the syntactic analysis (8i) or (8ii)

(8) (i) NP - Verb - NP - Complement

(ii) NP - Verb - NP

where the second NP in (8ii) consists of a NP

("the boy") with a restrictive relative clause.
The interpretation (8ii) is forced if we add "who
was" after "boy" in (7); the interpretation (8i)
is forced if we delete "ing" in (7). But (6i,61i)
are not subject to this ambiguity; the interpretation ..

(8ii) is ruled out, in these cases. Once again,

these are facts that a grammar would have to state
to achieve descriptive adequacy. (Notice that
there is a further ambiguity, where "Mary" is
1. taken as the subject of "walk ", but this is not
relevant to the present discussion.)

The problem of explanatory adequacy is, again,

. that of finding a principled basis for the factually
correct descriptio Consider how (6i) and (6ii)

W=m;imillmmilsWirmwW41,732,77.7wArsir4W4-707;,-;:-'
 03.

41 .
13

must be generated in a transformational grammar

for English. Each mus t be formed by transformation
from a terminal string S underlying (7). In
each case, a transformation ap lies to S which
selects the second NP, moves. It to the front of
1.5
the string S, and replaces it by a wh-form. [I
have not quoted footnote 15 here, for it does not
bear on the h-over-A principle -JRR] But in the case
of (7) with the structural description (8ii), thi
r
specification is ambiguous, since we must determine
.f whether the Second NP -- the one to be prefixed --
is "the boy" o=. "the boy walking to the railroad
station," each of which is an NP. Since trans-
formations must be unambiguous,, this matter must
be resolved in the general theory. The natural
way to resolve it is by a general requirement
that the dominating, rather than the dominated,
element mmstalways be selected in such a case.
This general condition, when appropriately formalized,
might then be proposed as a hypothetical linguistic
universal: What it asserts is that if tha phrase
X of category A is embedded within a larger
phrase ZXW which is also of category A, then
no rule applying to the category A applies to
X (but only to ZXW)."
' I.
It is the principle stated in this last sentence which I .

will refer to as the A-over-A principle. In terms of tree. diagram

(2.1), the principle asserts that all transformatiolis which refer

to A must apply to the topmost instance of A in (2.1), not the

dominated A, which I have circled.

I....
(2.1)
+ I.
s

)i 1 , I,. I .
. :
Nt ,7 '4,

1%. ,
1101 /
4 I
.
, *'

e_
t

.,-
41"4"".
 14

2.3 Chomsky, in the course of revising the paper quoted aboVe for

separate publication as the monograph Current Issues in Linguistic

Theory, (Chomsky 1964b), realized that the A-over-A principle was too

strong. On page 46,: in footnote 10, he gives the examples "who would

you approve of my seeing?", "what are you uncertain about giving to

John?", and "what would you be surprisedby'his reading?", where in

each case the question word, who or what, itself an NP, has been

moved out of another NP aNp Ey2191131121M1hilaj. [Np skkaison.

to John], [NP his jaadjagssataking])/. Other examples of this sort

are not difficult to construct, and there are even cases where the

relative clause transformation can move either a dominated NP or

any one of an unbounded number of NP's which dominate it.

(2.2) NP 1 .

.,,,./°"'""ft.
NP
,..-47'4,....
Det NP VP

the book '. I ':. Vi NP

.

. ,..
'' lost NP .
PP
... ..

Det N NP
r

tile"colver- arrows Det

'tLe book

9.
The relative clause rule", when applied to (2.2), will produce

either the book, the cover of which I lost, or the book which I lost
the cover of, the second of which would be ruled out by the A-over-A

principle. The example can be made more complicated b embedding

the NP in ever larger NP's, and as far as I know, this process

can be repeated without limit. Thus if the strucf.dre underlying (2.3)

(2.3) The government prescribes the height of the

lettering on the covers 4f the reports.

is embedded as a relative clause into an NP whose head noun is

,reports, the relative clause rule must produf& (at least) four

relative clauses: the'reports, the height, of the letteriagoa

the covers of which'the overnment -%rescribes- the'reptalsi.the

lettering on the covers'of which the overnment rescribes'the

height of: the reports the covers of which the government

t
It

prescribes the height of the letteriagau and the reports which .'.

the government the height of thelettering'on'tle

covers of. The problem of how to formulate the relative clause

rule so that it will produce all four of these is an important' --

and difficult one which I will discuss in SOM3 detail later

(cf. §4.3 below); but for the purposes of the present discussion

it is enough to note that the A- over-A principle would exclude

all but the first of these four clauses. Many other examples of

the same kind, whicL show that the principle as originally stated,

is too strong, can be found, so it would appear that it must

either be modified somehow, or abandoned and replaced by some

weaker principle. I have not been able to find any successful

s
 16
,

modification, and therefore, I have pursued the latter course.

2.2. Of course, it was not merely to handle certain

restrictions on question and relative clause formation that the

ki.over-A principle was proposed. And it is incumbent upon anyone

who wishes to modify or replace this principle to take into

consideration all cases which it dealt with satisfactorily. As

far as 'I know, the following is a complete list of all cases

which the principle handled convincingly. In all of these, I have

been able to construct an alternative explanation which still

allows the generation of such sentences as were demonstrated in --.:''.

to be improperly excluded by the A-over -A, principle. In

i 2.1 ,

all of the cases but one, I will not present here the alternative

I have found, but rather postpone the explanation until a more

For ease of referunce, :y

nazural time in the sequence of exposition.

I will repeat here several examples which I have already discussed,

so that all cases which seem to support the A-over-A principle are

grouped together.

Elements 1:41: relative clauses may not be

- questioned or relativized. Thus, the sentence ,

,
I chasedl.the boy'whothrewla snowball] at 'our .

NP NP
teacher.] can ziever be embedded as a relative

clause in an N? whose head noun is 'snowball:

,
4 r. ,

; ,, sentence (2.4) is ungrammatical...

r .

,:' :

'4 . 4 41 .

' 1 .-f .
. 1
, 4 , ; .,. i
,}C '

'
 17

It (2.4) * He is the snowball which I chased the

4.
boy who threw at our teacher.

It is easy to sae how the A-over-A principle

in the source sentence the NP- a

would exelude this:

snowball is embedded within a larger NP tjejLoyc./ho

..
'threwa 'snowball at our teacher, and the principle
nodes can
dictates that only dominating, not dominated,

be affected by the operation of a rule.

. .

of
Thiè restrtetion also applies to elements
f
.. 51
(i.e., those in which the
reduced relative clauses
3 the VP- bikinis , ..
initialvhich'is has been deleted ):
, .
, in the following
: 'is impossible to.question or relativik.s
. .,. .
.

sentence: she reported, r -,all'_the'Rirls wearing,

NP
Thus the following question
:-:

:,...
ibikiniili to the police.
.
.
,
.,

,
,
4

_ :
,
:
is impossible: . :

,f .! i ,,t .
IA
,
, r,, ' , ;
4 .4
1 1,., . I, 01 ..i . .0 '

' 0 :., I r {' ',. :: : the

(2.5) * Which bikinis did she report all,
,1

i
,. . ( ,,,....
..
I
,
. ,,,:,,.,,..,
....
.
,

st ,.:, ., . ; .. 7

girls wearing to the police? ,

B. Elements of sentences in apposition to such sen-

tential nouns as fatt,'idea; doubt,t's2essloa,

1184, cannot be questioned or relativized.

C the fact that

Thus tile sentence Tom mentioned
NP
E a'bi.32 cannot be 2mbedded
'she had worn,
NP
as a relative clause into an
NP whose head
 .

18

noun is bikini: sentence (2.6) is ungrammatical:

(2.6) * Where's the. bikini which Tom mentioned

the fact that Sue had worn ?,

Once again, it is easy to s'ee how the A-over-A

principle can be made use of in a~:-Audiri this sentence.

An extraposed clause may never be moved outside

"The first sentence up," as was discussed

. .4.

', briefly in S 1.0. Assuming that an approximately

correct formulation of the rule for'Extrasositiam,

"from 'NP Is the one which was given in (140),

:
;
. which I repeat here eonvemiance,

: $ '! *k (31010) Ixtrasositian'from'NP,

.$ ."
1
.
I
, X I. ENP 03 Y
NP OPT

1. .2 3

1 0 3+2

we see that unless it is somehow restricted, it

.4: will have two results when it is applied on the

'1
'topmost cycle of the structure shown in (2.7).
.

4 ,,
$.4 .

t .5
;
.r I°,
r
>Fa1

3 1r 17.; ry ; . ,
,
+1. t
4.
" r
t r 1,14I
t. .
4

C
.

. I

tk
It
. t. .

.,,,,
r-4". ..:"" 4'1414%<,-e,Nti4
 I
.-t

19

- (2.7)

'I

NP
N was 'given

(1)241,e.
"that NP
.
had been made
;

* t'

It ...f -the-e aim ..

1
17 VP
4
'that

John -had lied

Either S2 (the subscripts have no systematic r. .

'

: i
.
.,.... :.;:.

t significance and are merely inserted as an aid to exposi-

tion) could be moved to the end of Si, which rould yield

. .

.
... . .
....

,
the grammatical sentence (LS) ,
'

'0.' : .
I
.,
(2.8) A proof was given that the .claim that

John had lied had been made.

or S could be moved to the end of Si,. which would

3
t

result in the ungrammatical (209),*

t

(2.9) * A proof that the claim had been. made was

;

giver that John had lied. .

Sentences like (2.9) could be avoided if the

A-over-A principle' was strengthened somewhat so that

if a P -Marker had two proper analyses wish respect

.
. :. ;

e
I :1
I
;

k .
VI.'
,
.
4-

,:romossoklisem
 20

the structural index of some transformation4, Where

one proper analysis "dominated" the, other, in a sense

Which is intuitively fairly clear, but would probably

be difficult to state formally; then the trasforma-

tion in question would only perform the operations

specified in. its strucZural change5 with respect to

the "dominating" proper analysis. Begging the question

rt
of how these notions could be made precise, it should

be clear that the sequence of nodes 0 Sim? which .

is immediately dominated by NP, in (2.7) "dominates",

in the intended sense, the sequence of nodes [NP

'

which, is immediately dominated by N22; so'Extrapositimn

'from NP could not produce (2.9) from (2,7), if the

strengthened version of the A-over-A principle which

was sketched immediately above were adopted. . A

.. NP
In a relative clause structure, ie/Ss it is,
NP' S'

i.
not possible to question or rPlativize the

dominated NO. This is the case discussed

by Chomsky in the passage quoted in 5 2.0

".above. An example of the kind of sentence

that must be excluded is the following: it

, '

.*-: is not possible to question (2.10) by moving

a.

tT trj 4.
 -

1I

21

someone to the front of the sentence and

leaving the relative clause 'who I' was

Vith: 'behind.
It

41.4. IY.4 (2.10) He expected ([someone] who I was

NP
acquainted within, to show up.

Thus (2.11) is ungrammatical:

(2.11) * Who did he expect who I was

4. ,

t acquainted with to show up?

In (2.10), if the NP someone is to be questioned,

the whole NP which dominates it, someone who I was

:Aclaglpedvith, must be moved forward with it, yielding',

.(2.12), or, by later extraposition, (2.13)

'.(2.12) Who who I was acquainted with did he

expect to show up?

.14
.
".' (2.13) Who did he expect to show up who I

1
was acquainted with?

It should be obvious how the AfoverA principle

'would exclude (2.11).

4 ,,
t
. i
' ' :Cy
' E, , A NP which is exhaustively dominated6 by a
,.., . , . ..,

.,......0. . ...
.;' ' i ...
;
Determiner cannot be questioned or relativized
,
I.
":e
1141 :*' out of the NP which immediately' dominates that
,
. !..
S.

(t Determiner. Thus, from (2.14) it is impossible

to form :(2.15)i

;r/4(f -42? ..
f: . ,

22
' .

(2.14)

NP

found Det N
./' 1

NP "bodk
.
4. a

' Det N Pose - .

.%

;` some one ''s 4

1
(2.15) * Whose did you find book ?'

Only (2.16) is possible:

(2.16) Whose book did you find?

and the to.over-4 principle correctly makes this assertion.

, J..

P. Alt NP which is a, conjunct in a coordinate NP

' .

. :. structure cannot be questioned or relativized.

4 Thus, in (2.17), neither, of the conjoined NP's

. , . :
may be questioned (2.18) and (2.19) are both
." . .
." ,t4'; ,

impossible.

(2.17) Ile will put the chair betweeliplIm[pmme

,t,

table] And some sofajle]

,
NP
What sofa will he put the chair bet-

ween some table and?

(2.19) * What .table will he put the chair,...,

, ..,
f some.
between nail!' sofa? 1 .

4'4^ '4' !' .

 23

principle will exclude

The last example was suggested by'James McCauley

(cf. McCawley (1964)). He points out that if

the'Adlective Shift Rule, the r le which permutes

.

,. -
; ..,t.

. a reduced relative clause with the noun it modifies,

ective, and not

-A principle;

ed from

the

noun 'sate, instead of with the whole compound

t. ;
 24

Thus, without the stronger version of the

S 4
l.nverA principle Vitich wan discussed shrive
'1. .t
i
t ,
in connection with E2...ittapp
s it ion from NP
a
5/
rule (2.20), when applied to (2.21) would
4.;
t' yield the incorrect ** a'boolcbig case

instead of the desired a big book case.

2.3 As was stated above, z have been able to find alternative

explanations for all seven of the cases discussed in 6 2.2 above. '

Cases A, 110 and C will be accounted for by the Complex NP Constraint, .

which will be discussed below, in 5 4.1. In case D, ungrammatical

4'

?I 4
sentences like (2.11) will be shown to be excluded by either of two

independent conditions: the Complex NP Constraint of § 4.1, or the -

.

Pied Piping Convention, which will be discussed in 6 4.3, in connection

..

with relative clauses. The Pied Piping Convention will also be used

to exclude the ungrammatidal sentences which arose in case E. And.

case P will be accounted for by a special condition of great

generality which will be discussed ins I 4.2 0 the Coordinate

Structure Constraint.
;,
Case G remains to be explained without invoking the: ."

A-over-4 principle, and it seems to me that the most likely line

4
of explanation lies in rejecting the assumption that the correct

statement of the 'Adjective Shift1tule is the one given above in

(2,20). The rule of (2.20) must have many restrictions placed-on

 51

25

into the ungrammatical

for otherwise it will transform I painted it*red
7
* -ainted red it.10 and we showgd the children untranslatable passages,

* we showed the untranslatable children assa es etc. Clearly

into
e
adjectives
it ii necessary to restrict the operation of this rule to

which are part of the same 111, as the N over which the adjective

permutes, One simple wad to do this would be to modify.(2.20) so

that it is stated as shown in (2.21): .

(2.22) X rNp Det, N Adjlta; Y

1 2 3 4

1. 3 2 0 4

Although the formulation in (2.22) avoids the difficulty'

pointed out by EcCawley, recent work (cf. Lakoff and Ross (op.'cit.))

Indicates that it is still inadequate. I will not discuss this inadequacy,

examples:,
,1 here, for to do so would be unnecessary for my present purpose:

suff4ce to show.
of ungrammatical sentences like * I painted red it

is too strong
that MCCawley's formulation of the Adjective Shift*Rule

and must be replaced by some rule formulated along the general lines

of (2.22)0 Thus case G provides no support for the A,over-A principle.

2.4.

InCILLl'ueltELLtieor(Chomsky (1964b))*
2.4.0.

having realized that the A-over-A principle was too strong, Chomsky
rule.
proposed two other conditions on the relative clause and question

These need to be scrutinized carefully, so that it can be ascertained

to what extent they can replace the AmtAver-A, principle.- Admittedly,

k
. -.. . . .
, . 1

i
t 1.
1

. 4
v 4
e .1

1' , : i
1

. ' :

'1,0
 26

Chomsky at no time claims that these two conditions will have the

same coverage as the principle, but since the facts given in cases

A through P h.vn to ht. AnnotetIteA f.r Onyvny, lt is of interest

to see how far his two conditions can go towards this end.

In the quote that follows, '(6)' refers,, to the following

rule, which Chomsky states on p. 38, and which he asserts is the

basic rule in question and relative clause formation.

F (6) Y 1411 + Z +'X -Y Z.

.; 2.4.1. .
The first of the proposed conditions on this rule is

on pp. 43-44: .. ' 1 .. 4:

, . S,.4 i 1* . '
1

4
S
e , "Notice that although several noun
.,,
1
Phrases in a sentence may have A attached
:4
,,. to them, the operation (6) must be limited to .

: .1 a single application to each underlying terminal

,. ..'...,t. string. Thus we can have 'who saw what?', 'you
met the man who saw what?', 'you read the book
that who sawl','you saw the book that was next
to what?', etc., but not 'who what sawr, you
1 ,
saw the book which which was next to' (as a
. , - 6
declarative), and so on, as could arise from .1 .

multiple applications of this rule. These.

:

e t, ..,'.; ,.!,.
examples show that (6) cannot apply twice to
' 4 l'. ......
;
a given string as a Relativization and cannot .,

.4.., ....- apply twice as an Interrogative transformation, ;,.:,

11,4
. ,

.. 1..1q. ,

but it is equally true that it cannot apply to

et

,
,e ,

'!e,.. .:,. '' ', I, ei

'A:
'I
.
'..'° a given string once as a Relativization and -
o
'.
,

:'..1.;,6,,i;
,6
:;.1; :

.
, ,
once as an Interrogative transformation. Thus
if rule (6) has applied to form a string which
.

.
is embedded as a relative clause, it cannot
reapply to this embedded string, preposing one - ,..,
! ,..- of its Noun Phrases to the full sentence. Thus : :' ... '.

we can have the interrogative 'he saw the.man

..r read the book that was on what?' , but not 'what
.
did he see the man read the book that was on'; -
, ., Ind we can have 'he wondered where John put .

H
whatl', but not 'what did he wonder where John
pute'i 4tc."

. 66.......1

A./
 1,

27

My first objection to this condition, which I will refer

to as Condition 1, is that is seems to me, to be somewhat too strong.

That is, I find the sentences in (2.23) all more or less acceptale:
, .

,,,
(2.23) au., He told me about a book which I can't
, t,
1. 4( 'whether to buy or not
, "t, . 1 ,'"
figure out how to read.
' I

H 1. `
where to obtain.
I 1 .

what to do about. ,

I i
S. ....
1 ..

b. He told me al-out a book which I r ..

k$ 4
!

41. " . 1
why he read.
figure out ?whether I should read
??when I should read.
why
c. Which books did he tell you ?whether
?when
he wanted-to read?

For some reason that is obscure to me, I find sentences e7

.5 "'
8 .1
like those in (2.23a), where the embedded question consists of a r

wh-word followed by an infinitive, by and large more acceptable than '

corresponding sentences, like those in (2.23b), where the wh -word is

1.14....
followed by a clause with a finite verb. And yet there are many

sentences which differ in no way which I can d / sceru from those in

r ,

(2.23b-c) but which I find totally unacceptable. (Chomsky's example, ,

"* what did he wonder where John put?" is a good case in point). So,

for speakers who agree with me in finding at least some sentences like :

those in (2.23) acceptable, Condition 1 is too strong as it stands,

although examples like Chomsky's make it clear that it is partially!

true. This all, indicates that much more work needs to be dons on

this condition, so that a. weaker version of it may be found.

^'

.tv,
$
4:
6 .1

.
 28

......welIt

It is apparent that even a correct version of Condition 1

must be supplemented somehow by ethe. principles; for, of the six cases

which were discussed in 5 2.2, Condition_l can only account for case A.

And it should be noted that even in case A, it is not obvious how

10,
Condition 1 should be stated so that it will apply to embedded
.

questions, full relative _clauses end reduccd relative clauses. That

is, in (2024a) 'and (2.24b), it is easy to state formally that, in

Chomsky's terms, "operation (6)" has applied once, for there is a . ,

.
subetring which is headed by a wh-word.

(2.24) a. I know who is mad at John.

: , b. I know a boy who is mad at John. _ . ,

/

: But in (2.25)., which has been derived from 1/.24b) through ete operation
. '
I of the Relative Clause Reduction Rule, there is no longer any wh-word

in the sentPnce which could be used as an indication that Condition I

.

, 1 Qust be invoked. '." :

...:4 ; 1

":
..;
(2.25) I knott e boy mad at John.
.

The fact that NP's in the position of John in (2.25) ' s

cannot be relativized or questioned (cf. the ungrammaticality of

:4
.....E*119LASYoualcm
LILlosgagg10 would have to be stated in some other , 4

way than in terms of wh-words, possibly, for instance, as follows:

`*- .

(2.26) No element of a constituent of an NP which'

' ,
,:
modifies the head noun may be questioned or

relativized. ., .

1
But thin condition is strong enough to account for cases A and (with ;,

.t 1. ;
A .
1
.1. .7-

It
,
It.
 4

r
n
A 5"

29 ;:

.
suitable modification) B, of 6 2.2; an4 in fact, condition (2.26),
. .

when suitably formalized, is the cornerstone of whet I have called the

. ..-
,. -
,.
I;
,

Constraint, and will be discussed in detail in S 4.14 .

), . .
.... Complex NP
t-.
. , ,
.

,
It appears, therefore, that Condition 1 is of limited

utility, except insofar as it can be given in a weakened reformulation ., .

..t
.41,
which will allow some of the sentences in (2.23) to be generated, but
, .

,1 '
IA
;44.4
.1''0!','. will, exclude others, like ChQmsky's example of "* what did he wonder :lc
:
:::::

?..., , -, : :. ,..
,

where,John put?''. I should add that none of the conditions I will -.... .. ..
,
..:
i'..--.,,
.,..
i..,
4
,: ",
,.,

.
', .. t: 1/4.

,.
-- .
. , .. :,..
...,. :.'
.
.
i ' .. propose in Chapters 4 or 5,can be modified, in any way, that I know
\:. ...., .,..

.. ..,4.1 ; ,... . ... '

exclude this last example; so it is evident that some version
4. .
.
.

of, to
t e
.. . .

, .....,-

of Condition 1 must appear in the grammar of English, or, if this ' S,

i t

condition should prove to be univeisal, in linguistic theory. ,,

,wt;
, '4. a

The second condition which Chomsky proposes for his rule, -:.;.-._ ,

...:.4, .

....4....,0.,.!:;.;,...;..,,...., ..
(6) . is stated as follows: ,,.....
, ..

:-..... ..,,,.:...,.,..7"..,
I: ; r -, ,-,:

: , ,,,,:tt
s.,:.. I. In i,L.;;:..ni . .
4' ' t .. .
4

,.. f a

I. 1'
. *4 1
.. V,

1-
8 n

:
.'
t j,. i' 4 ''. ft
Finally, it is clear that the first . .,
:4Vi:IA
, ,

-4-'',
,
,. ..
,..
: segment Y of the structural condition of rule ..pi .! i ,, .
.,
. .
,

,, :

A I
. , ,
. .;' : N 1 et:
i : .',.;
!,..

. (6) must be suitably restricted. Thus we cannot .1;". -

.,,,,-,...,, .-,.....
:: '

. ..
..f..
,.- .. have such interrogatives as 'what presumably
. . 0 "' ) ' i ' t. . '
tl" did Bill see' from 'presumably Bill saw something 9 ...--eu -
.;.s.,I.,,.:. 14 ,.t. t ., . : . ,... .
, .."
, -' , '-4 and so on. This suggests that we restrict Y .
,.4.41,-....;

..1',..--4.*:',.....
,....2:, ,..-:

.. -. .
' -"Y.i.o in (6) to the form NP + ..c. With this further
condition, we also succeed in excluding such
,

non-sentences as 'what for me to understand

, " er

, ,
.
,,
;-,,,t' ,'. would be difficult?', although the perfectly
,, .,..;:

.. ,..,!i\:
., .
correct form 'what would it be difficult for .

.
., ..

I ,. me to understand?' is still permitted. Thus

91: .
this condition would account for a distinction .:,..-:5'.. ''
.
bgtween the occurrences of 'for me to understand
_ ,...;-:
.....- something' in the:contexts '---- would be
,..., .- difficult' 'ard lit would be difficult ---ftlt- .,..'2......- ti.', .. *; .'''
os 4. ,. ',:.. . t
,, ,i ' ;
.

*-...

:, '' : ' ':....,i ' O .;,, %. I,' .. :

,
IN
t
4
I / Z $ 4
%, ,I .Y :. ..
:ii 4
;',
A
1
%
4..
1,

, . , 4., % k

I i .
. . % 1

., d ''.... : .4 ". '

'is
.,
,.. ....
. .. '.; ..
V 0. # r . . ',
V

.,
.
1
.
1
o s' , t e.
, ...
, ,

. 4 i . .
. ri-
,

*;.

- !,*--
 I.

, , Ai
.. I, 4. II'
* .4.
"i t 0' ' ', 1.
1

. .k 0 a . , , i !
; '
.i,
ie 7P4 . ,
.
t '
.. k 1...'', '.
' 1 ' y
L

r
.,,,,, r
?, ...?..,--

'No,..N
:1
o
: l ', ', '' ''. 30
Vf

,:t-:1`.

4 ,7,z; Aa
it.
...44*. '1
a . i
'..,.., ,

Bo far as applicability of (6) is concerned.""

I - *
i ,q
_ ..
,,
: f

.- I

4.. :"'.r,
.
(op. cit. pp. 45-46) [I do not quote footnote
_
,
? ; lit i,-, i!..;
. .;,, 1
.,
10 here, becatise its content has been discussed
..
,,; ,,.,:l: .::;th.:
in 5 2.3. above,-an4I it. is of no direct relevance
li
-,,i I, till:, ! i".' .; ; i!zi.,41 , tc the point at hand
3.,:,.:., ,

.,, 1,, :Of.. ii °I;V:.

H , ! .! This condition, which I will refer to as "Condition 2",
4,
- ;-,,..:.,;
.. . , .-.::

bears close scrutiny, even though it is clear that there is no overlap

,.;;:i .:,,: ....,1
-,:,
;:,i i''.-'
-.,
,

. at all between it and tht A-over-A principle none of the ungrgumaatical .

.. .
. .

1. ,... sentences discussed in cases A through P of 6 2.2 will be

a -
,, excluded by Condition 2. ye'
- 5- '
In the first place, the first example is not convincing. :4..
:.

The fact that Chomsky's example * what presumablyda Bill see? is

, ..,
, -,. ,
an 4 .
. i ,

ungrammatical has nothing to do with the fact that adverb star:a

...:. %,.-
.
,
..,::. b.
..
.
,
.....

.
, .: ,-,'
...,: ''..--,,,,:
,. :: , t, :. ,
the sentence; as was noted in footnote 8 above, questions are :c.i:1;-..,,. .

. et .7 i :' . r
i

, :, / ;:-?:,,441 .;';: .,:

incompatible with sentence adverbs in any position. Thus, neither
? ,::-.4. . .

.."
. . ..... J.
.r. :
,

4.
in Bill presumably saw something nor in Bill saw something., presumably

' ,
.

%
.

; can the word somethinst be questioned: * tahat did Bill ayes

;. ;
".-"

"'
'.1
"
and * what did Bill see, presumably are both probably to be -''
f `' 1.-
-,
,

- , , excluded. It may be that Condition 2 is correct anyway, but if it

. , is, all of the sentences in (2.27), (2.28), and (2.29) must be -

,
. .r . .

I
. .. 1 .. -
I

.,. t
4. explained away, for they appear to be counterexamples.
''' 'ti
,,r ! .,:,
.
s 3:1::.,.,... ''.,-; . ,7
;. '''''i..," (2.27) After maintaining that you were sick, why did ..,... ,,. ..
:::,:,
;.-;..,;.,.'
i'''': :
;,- ,. ';! 1:,:i. '';'.1.'. (. a

1
, .p.,
,..o ' , ; I .
:1....,,,.,,
. .,., ,., .';'`1" ., .
'..., s, i'.',
--ir"'.1.',
t'l., !,.,.1.: -.
,':. ,;.''$.:'
.
-:.!..
" J IA.. .(,.
i !.. you get out of bed?
,
..-.

I
s...,.i.., :"....%;,...
,, ,... .

,
.
.

:
i..
.
,i ,
,4-.'...1
,
':. 11. ; ',:,:':' 'it ,,`,,(: 4 .:,'I'.:)i , I '. 4 :i. :,:
.. ..... ., :-.
''': . , `:' d
..,14 f, ,'...
.. , : ...r
,f : .'''' '"Y,' '1'.'
I*
. . ., , ,,, ...., .1, .,... i!, ,. is n ...; ,,,. , i.. 01,,
..t,, .
'
:It 1
4.:

e .:. "I. . .- S'

1%.t.:I.
.: i Although you've never been in one, what would 2.:.-4...,1::::...,.`:..;.t.
' :,, ,,.
' i f. %.....,..
:. ...
i

41; , . ,, . ,,0.41' !, . ), y ,
... Is' -, . / . f, .
!,
. / 0 .

1, : :
s;., .. : t. ! . . .
..

.9

.. : 4.
: ,t ,;', .: !4. : ;,1,,,',
Itlii, . t,......,:.
. I

./ 1. i..
0,

.,' '7+1 ; "

I 4

i''' you do in a typhoon?

i'4'. ',
'-, ........... .q....
. 1 i1 % II ,,,..
:;,.... . - ,
0
,, '
,
' ,

.
1 ,.,
; '::. 1'''
:
.:.. .
:1'1'''''!:: ''
,..
:.7..!..2.1'''''
,....
,
. .1'; In light of
:,
longwill you ..',4';'
this promotion, how long
:: A
..,.....
' ,
..
14-. '4'44.
, ' . . ., ,v. , 61:- :c , .
;.
.' ,

:,
'''.
'',,
''' p
,
.
,
..,
-

.
;'- '
(T.
,
.. . at ' '. ...,
.
- . `
,, ;. ;. . .. ,

i 1,14 , .
:. , ...
..
4 a t . ;
;
_
44
-0 ' .. ' . . ,',.. : , .
*
I ..s ,

t"
*
'fri . 1 .).
. I
; I.,. :4 4-4...4 1 ,'' i'
, ,t .- ,* ' ; ,,..
't k ) i' ..
. 1 , .1 ' . i '',
.'
4 y .
.. q '. 4 .
. . .
.
. ' . § .. . , ' .1
4 : f
.
: ;' I ,V . . 1.

00,44.

6,4 40"
,- , 0. I '.4.4 .` 4' .- 4 481.1.1.1!
 \./

'

i! .. ' 1
.i
1.
I !...
"
.
-"to * 1 .
! ' I1 I.I
:4
, ..,
i .
, * ,t
4
4
4 . .;
t
4! *V ..) .' t. 1 ; .. .3, ; :,..
.; ' ' .). '''. " I,

.4. -
, .S,A....I. ,-..4. , 31
,
t ' . ,
. .

.
.t
. .

4
stay here?

.
-; 14: ,
" Furthermore, what prompted you to hit John?

4
,41
If it rains, will you finally give up and go 1

: .: t: ": l' -'="

..
. ..r.' t
.
; . .
f

,
,: ;, "t .,.. :it ,,'; ,.
., 't
home?
..;
,t
.; .: .,. '.5., ,.,

!
.

1 ,.:
gi
i
I
-
"
, ..., .

(2.28) A: Why, after maintaining that you were sick r, did 1'

..,11.:4!'
1.1; ,
' '
,

e ' "
"i':.. -;
,

; .4 you.get out of bed?

.
.
*".;
ft .
.1 !t
, I , ' ,
What, although you've never been in one, would
4.3. .:t "t
I I 4 6 4

.; , -1.-}5': '' 7od do in a typhoon?

13
1
4ti
t
t' ril
4 44.
I

-L,

I
. .
1",. :
4 , Now long, in light of this 1 11
Pttnaqiitifit WAAL& i
' i4 ; ' .
; .S. ,14
I

' ; ; ..; 0 0: .. , 1.
3 ":4 , . ;'' i .

4, t . `11
"."
you estay here? 1 ' .. ' .4 4. .4

-. ; n t., . ,, - .

, ....is, s. What, 'furthermore, prompted you to hit John? ;-..-.'n-:;--1.. ....'".::.;

...,...;:`.".i.,:.!:.*'*-z
:..,....... ;.t:.:..r.- ,..,
"'"r ' ,
/..' . .
What, presumably,,did Bill seat .... ,.,,,.... .

:
.
. -
t, :..... .-.
it, 4 (.,'; X ;

4
.( ft
4
."
a r
'
' 4' . '

.. ) 4,01 4n% p.4...

what can you do with the wounded?
E144
.
. ,
1, J .

The type of explanation which at first seems attractive ... :

,

is one involving rule ordering,. That is one might suggest that the
t 44;`

keLtim11111111311Majtule should apply first, and that the the adverbial.'

r t
4
elements which start the sentences in (2.27) should be moved to the
.

front of the sentence, past the'trwordo, to yield the sentences in a ,

=

i r

. (2.27). Subsequently, a second adverb movement rule might move the

i
preposed adverbs to the position immediately following the gh7word, . .1., ., A
I
; I`

and insert ;Os= markers on either side of them. To giv,. an example,

) ., ..,
.. ;-' .

the second sentence in (2.27) and (2.28) would be derived as followst

I. . .. ....

.
..
..
-
... ;.
.

.s
- -.

:%:-: f
;i
: f
 e
4.1 ,

'
1
*
32
8 i 48
t
".
.
. 1

'

'something in a typhoon, although you've

1

1 , ...,." " Bacei you would do wh

t ,,_.- *'

. .

v. 4 ., ` , . " '
1,', ' 4,'
s
never been in one. 11.? t+
` r
,ir?1 .
ss
.I: .
i. . 11
I.1 !
%I. , i
sh.
- '
'.
,

I
ir r

4 '1 s
a

sI
4 4
;
V,
I
question formation
A.
4' 4 ,4
IA .t"
,
IA
.
' '

, t it

,
"
'' . ...;*

..
, . what would you do in a typhoon, although you've never been ,'
4.:4*I

one? -'
. . y la '
,, 1st adverb movement
41 / 44
"
4
":1 v:
1"....
`' .....

'
e
s4
,,
s,.

I.
t4
i
v
s. -Z1
(2 .27)
.
*

*.
Although you've never been in one, what-would ".
:411; - you do in a typhoon ?f
4;:. ,'.4 t
4. .
2nd adverb movement
.1.;S.- .1
; "1':; 7-
.
,
i Xj
.
.
. 2.28) What, although you've never been in
. - .
.

7
r .; I a'
r-.11
Molt do 4" a typhoon?
St I I

Note that if this proposal is adopted, Condition 2 can S

:
e t.
be dispensed with anyway, for at the time at which the question rule - :
-

applies, no adverbs have yet been moved into sentence-initial position. .

0.

.
But there is still some doubt in my mind as to whether the rule- ,

; ordering, explanation is possible, because the sentences of (2.30) have .;...

, such low .siccaptability that I doubt they should be generated at all.9 r
r s s
,
,
(2.3 a.- wonder, after maintaining that you r

lase
,;' were sick why you got out of bed'..
.
,4 '' ' 1 ' -' , .
. - . 4 .
i 4
.'
7", ' 1 ''''' 1"' " 'A -1a., . 4 ...",. ....
.i t' I ,, .
1 .'
, . is,
IN .
*$
'
'

4. I'
,4,..
'4,;
1

i
',., A'- 's' f $
! " 't.' "t
/44 .
s,
,s nv

.''.., .
1 i.
4,4!,
.
'0 A";

". .,
tt t,',.4 ...:q
4

1P.
'
.,f ... 1
ts
r , 04' Ov.! tiail ....,..* '
,if.',4'

1"." e fi'''' lyric,.

V

:(
*0
T..:
6,:
'

.$ '1,'
4
i
6
%
:,`,4.

-.!
Y
.4,,
''

,
4' -t.i
4', 1,
r
.
,,', 7, ..,
"?
.

.
.,,: : ::3..,;'
.

..,.4: ,1 0,* !Pt

: .,
.. s
T! ..';,. .41 s 4:4
c
..4.'n
7.I. 4
''i,
''
: it .:...,
N
,.. r
4 ,,..
*I :
1
.'
..,
s
.4
' t. . .:
,......
`' ,, i `
, ".
s,
..)
.
, .: .e. ;
'

1 v.,:, ;- - ...,. ' '.` e '..:. .,,.- ,, .

I.
.. .' ir.. % ' ,-
, ,
, 1.
. ..

.,`
. ,
,;
1

,.-
,

.: . ....,, . ,.:. .t. ;. .

.

7 .;.' ;... # ' ,. ., .

,..;
, ,
.
# 7 ,
... : t ,' .
.
. . . .

t.

Aso
 4 . .

. 0 7
.

4 r

33
..! ,

, ,A . , : .. .
,: ..,..
,,,. ., -./"....,,...... .
. .
.

., .. -I 4 1.: .:-, .'3 ,l'; : ,

,, r, -I
1 it A

:
I
1 .4 t t .(41.t.' b. 1. ? Tom will ask you, although you've never
O
. e ':0 ?a y
. 4
,
1. ,

.;.,
y

' ,... , ,e-

i
... .
, : ,
. ...
been in one, what you would do in a typhoon.
, ,.,,,..4-
. .4 , ! C.. . ,4::".;......44 71. I ' "0 . `i

I 7
A"
I '" .
,. ?*/ wonder, if it rains, whether he will
*, I I

c'
; .1 - finally give up and go home.
' ; . . . 1 lo,
e. '
d. *It is not known,if it rains, whether he
. .

.,t ,,t., e
, , *41 ": ,
1
,?70.
, . 4
will finally give up and go hone.
. . t.
f
e. :. *She raised the question if it
e
. rains, whether he will finally give up
'55-
and go hone.
. ,
Since the sentences in (2.30) all contain embedded '. 3-
/ ,

questions, the first adverb movement rule, which produces the sentences.:
.
of (2.27), will also generate the ones in (2.30), unless it can be t

'
1`

1, '
restricted somehow, which seems doubtful to me. And if the first

adverb movement rule camnat be prevented from generating them, then

" "ie
..S :"

the second adverb movement rule, which Converts sentences like those

in (2.27) to ones like those in (2028), must somehow be made obligatory

I

I
:.%1' when it operates on embedded questions. It does not appear to me as
,.!.

.!,,4
. .if conditions of either of these kinds on the adverb movement rules.'
-4' e
t ;
,
cannot be stated, but it does begin to seem that the rule - ordering
n: 4 .4

mode of explanation may not be the optimal one.

, 4 s
11.

:.
If the correct explanation is not toibe found in the 5%44

may be necessary4.
;
r I
,g
'r
i 1;
ordering of the rules, then some version of Condition 2
i "

, I ft
I say some version", because it seems to me that the sentences in, 1

.;
I + (2.29) constitute clear (thou h rather trivial) counterexamples
5,0
I
r. I
.s
t
'

: '1" r 1
. 4,
4, .

rt. t.-4
4" ,
4.7; r. a. V .
. . ,
r

. r or oar.
,. 4.
. , 1
A

i ,5

.
 -V,11-

A. ".4
I ,
V''. 4,
4,41.

'4 , ti
4
a
: I

, .

34 4

Ft
444 N
r 4,

4 ;$

: to the condition as it was originally stated.

I would like to call particular attention to the last 4 "".

.11.

sentence of (2.28), what,,yresuambly; did' 8311 ? This sentence 'tr. I r

A '

:
='
04
seems perfectly acceptable, as long as heavy pauses separate
T.
,.. :
,!4,
;/. ;
.i,
' :',9:1 ' presumably, from the rest of the sentence.
(1
4

This fact is especially

' it " 11 ''.
;.,:. '.. 1 ' A

t 1 :..
;.,,.
t., P
;±k ','4,' baffling, since it seems that presumably can occur nowhere else in , ?

,
,,
'4
.ekt .., 4. .. ', ,
. : t .j .-,. 4 ..,
.
.
1

., .:,,.
..,
i
);i: :
w::..,
the questioned sentence, unless I was wrong in excluding the question 4
-

..

.,..

I e,: , . 4 41 which .has it occurring finally, preceded by a comma: ?*what did

4!i t

,Bill see, presumablz1 It is obvious that much more work will have ;

ity . ;: ;:;
to be done in this area before answers to many of the questions
I. 1

I
'
4 have raised can be attempted. "4
4
:f.,1
. re:
;` .
One last comment about Condition 2 should be made: 'I :`:: t-
:::, :. -
. - .
,' s:, , although it is strong enough to exclude Cnomsky's example, * wnat
, . :,

,
2 NI ) '
,- ,
s' :
....
, .

' l .i' I
1' '11 ' ., i "CA. ..1 14,
.

. for me to understand would be difficult?, I will show below in 4

.0.

6 4.4 that sentences like this can be excluded by a' latch more ,
44

.widely applicable condition than Condition 2, and one that is ,

f V, ..

independently motivated. So it appears that although Condition 2 :, ',':-. ''!" --'

:, ..., .44'
'rt;!'.1;, ' '
may be correct, the cnly support for it is to be found in the .. .,v''-!-,:
t:
''' it
..:.
' 1,4!):kZ
::: (..-
21'?'
:!,.s....,:....,.i.,-
: ;
k .
4

:,4:$14;:l.. i. ,
. . :'.. "7' .1-. .
::: .':
.:
confused mass of eases which have to do with the interrelationship '', .s.-1: '.-.
, ; ." . ti 410

. .
:,
of the tgo,adverb movement riles and the question formgtion rule. .... .
.

) , , ,
v.)
a
. ,;( 4,
.1. frw .4.3. . 44
:
'VC

,.
2.5. In summar, I have tried to demonstrate in this chapter.' e.1

that the three conditions on the relative clause and question formation
-

'c.
. .

k
; 4; . 41.
4";

. , ,
-.. ,i
`,
;...1 l' ' .4 . ,

..44i'' 1 ... .
.

' ,.
'., . .,',
"S '
.r
ft
. .
!:' . " i'

.
41
;'' ":"i

,4 .3
 rf

. I
t
/. 44. $

.1
CP:

4
f
4; 1.
4i "'
4"'

1.
4. I
.
1
4
35
^
'

4 t
0.4

rule which Chomsky has proposed all suffer from defects of various
,

kinds. The k-over-,A principle, while shown in S 2.1 to be too

., Ar

strong in a non-trivial way, still' is the most important of the three, . .0,1 1!%
1

, ..,
because of the wide range of cases it successfully accounts for. - , ,-
,

'.1-': %:.''
Condition 1 seems tb be somewhat too strong, in somevay which 1
AA:'.:...*Ii.i;!,'
cannot yet delimit precisely; but insofar as it is correct in the :.t.--,;4..,.n:,
, ..
.. ,..4.,.

. ., .;,,
: ;.

3 7 k ,
i.:..,
.ss!:,---1,.-;.-W, ;IY
4
t ' restrictions it imposes upon the reletivizing or questioning of .,. , ,-, .

.11 ;
-I.

- elements in embedded questions, it is valuable and should be added

4 .1.- :
s
either to the rules of English grammar or to the theory of grammar.'
" - ' ..", ,

' 4.
-
But it seems that this condltion, if it is to apply both to full and 'A

to reduced relative clauses, cannot be formulated in terms of Chomakep

notion of "single application of rule (6) to a string"; rather, It -

-,;, ,,.
.'
. must be formulated along the lines suggested in (2.26), and, as will .. "'
t. : .

s
be shown in 5 4.1, (2.26) contains in rough form, the central . , ,
:
..1,t
I notion of the Complex 1P Constraint, which has much independent s 4

; 4

"
.10 'S
,

motivation. In any case, Condition 1 fails to account for most

,,,; '
r.
=',1*

of six cases of 5 2.2. The status of Condition 2 is undecidedL.:',.'Y,i'--..

4.
4:.
. 4
.
A
' because of the present lack of knowledge about the complex syntactic.,
*. '7.
"4
t
e
"
4
,
. At

t
.,, 4

'
phenomena which way provide support for it. But whether it is

s
.
'
.
; ..
. . eventually adopted or not it can account for none of the six cases ,
t
1,.
't ;

e .-,, 1`.
I of 2:2. :'t
,

-1'
' 4
;.1
4. 4,'
I hope that in my criticisms of the three nonditions
;

proposed by Chomaky I have not given the' impression that I wish to

t
belittle them, merely because they can be proven to be wrong today;
4,. ..
-a .
r ' ';t 1 ...43,,,. i . . . a
r 8
, ..

.,
k 4.',
, .,
,,r
.1
4,
'el'
.. .:
'. -, -,
,
.. .,4 .
1,, s tt ,;4..., .' 14 e. 4 :-. ; ! . , , .4!.., - I ',
o
,
.0.1 kr
,- ' :,' 0 4.4 ,' '1,, 4 s
I, ,!: : 1. . .1'' '&`:. e 7 i. x. 4 ....
. ,
*4

" :, ; ' to . . `
I
7...
. /' '' .'1:''':... % ...i. ,... t : ".` k "' r
.
,

.
; .
4
o ..
.
1 . .
,...
.
. :
i , .
..,
;
, .
.1

3,
% ' . .
l ' ' ' .. .
. a . 1.
'

'81TV:es:Pt' .1,43.744e.t.y
-4es
...74 TT_8"

t.?; (1. , '

4.,=d,`ra..'51fe z. ; 4' 1
.
 -

,
. . a

& -F
: I ' t' ,
,. .. )
4 '
; ;
,

,
, I )4 :-tt1z 4*t 4pt,, ES
1

4 t ,
jJ4ta
:
t
, l w4
I - t is 4
k
4

., . : ;
-

., .,
1 : ..
: . .' .'.:

:; 'Y ,

- : :::':: .
$1, .#J14 r
, .-t
:
S i ,

;
,
' :
, . ;,: ?

j
,.

S.. for the cOfltaXy is trtie : these conditions , in particulsr the ' , . ,
:

.,, ,. ..
'
. ':
'
' ?
kover"A principle , provide the basis fo the peaent work. For :

10 choinakf reuiarked , :. .
lii
..
; *
34

r
'tPrecisely constructed models for
%
'tTh linguistic structure can play an iinportat . '

role, t,oth negative and positive, in the process

,. .,

.;
1

1: '4t tj( ' , '

' 4' of discoVery itself. By pushing a precise but ' ,,;

;ì1
t

'I 3 inadequate foXiflulation to an unacceptable

4 '-
conclusion, we can often expose the exact source '

' %' of this inadequacy end, consequently, gain a , .' ,

I t. '
deeper understanding of the linguistic data." , "
r
.
t1: , (Choineky (1957), p.5)
l

'

The in task of this work i to provide a set of '- t : i

t, )
:. :'
:j
$-, r
:

constraints which Will avoid the detects pointed o*t in 2.1 '
:

:
: .
z,
.. .. i J t t).
'i'
iI

£4 and will account for all the casea in 2.2. Before tU can o '

.
;:: :
. ,
':'

e
$$
be attempted, in Chapter 4, one digres8ion must intervene: "- -. b'
I

..'
:

;,...r.:
I
thapter 3, itt which the notion of tree"pruning, which interacts in
:' .; ; . ' ; '::. .
.
1 : , . '

i p various ways with the conatrainte of haptrs 4 and 5, is discussed. :

,,
:, : .;
. -:;
. . * t , . . ' -
:
;
I
.
1 . . rt
.

4
: ?
c .
I , i : '
'q :
%
?
I
, 47 ' 9 41 t
, ' t , '' 'fl I
I ,? ' I :

1' 4 A
'.
I
? t
T.S 4 Ø T f .
,:
I

I ' tp;
:, ";t i
It :i 4 :; ::
1;
,
4$
4
: _ : : .
f
I -.i

' $
i4t:, "'
, l ' :
:
'i ': : : :
4 : ,

:; ;

l_ 1
' s?d. (
h ' lt h1 J, I
: LI .
! ' I - I ; .c
ii
h i 1

1
l3 1; ; (% ;;; : 44

I
b :. :
4 i:: SI
3 . ,I4 '1' -. . ' .
9'. '' .
74'J) 1.4 4 '.
. i
.
r* 4,
4
3
1
,1 . '. "
4. 7
f4f'3 13 1
4'
4'
43,
';' ,' ' :
.4 -
, I
i '4 3
4
1".
444 1 ., .
S It3 t 4
I
4

.4
4 4 41
j 4 4' ,

I
. 't k t 1.
I
.4

.., .1 t .. . -.
'. ',4' 41.-J - .4'.r ., . '.
I '.4 1.... 4. ',
5 4
-'
I S #4 3' 4 4
.4

4 .4
41
1' 1 4
5 4 * '.4 $

4 $ / .4
-

1 :" :.

: /
4 I ..
1. \
"
S.
S
4 Il
;'t .
4

4
1
S

1
4.
5
,
i
'-

4 5 44 - ..f'r 4'.
4, -
I 4

4
?/j .4
I.;., 44 I 4
4 1' 4
,

444 4 ,4,'4 ' 4 74 1'

4
?' " ' 4 *
'.4.

'
.4

:
117

:'''
 , ,

I
4
-I ' . -
J 1
v

i
tiø-
tt,4 ;f 4' ,
I
: ;
;

;: .'
:1..: 1 '.

¶ ;
:
9 : ' ,4 '' 1
'
k
i
"? t
;tS : .. . /t'

:: .

:
: 'i 31 '?
I . . . , ', p ' ' , e '
,, - .
p , ,
: ' '
. '

44i ,;
: J .
1 .*
', u"' ::-
4 1 4
\t pit k 4 4
,

4 ' /.#._/ j , '

:- , 4 ? "' f i

I):Iv.
.

chapter 2 , , ., '_

) I ,
+t 'a

f
' 1
( ,c ' -

1
t:
POOTNOTES ' f ' ' .
;

: .r.
; ,

r4
t't I
I.. f.,. ,
.' .''. 3 a
I '
4 * ', .. , ,,

'
% I
1',
':. ,
'
:

br a justification of the as8ignwent o NP statua to theec '

-. ..,' L
L& :'
1
I _ ..

I
f

e±edded eentence, cf. Roaenbaunt (.96S). ' ''.

I
L
) : t':t.
S :: :

a:t
I4'

2, juetification for the c]im that the rule NP NP S .

., ' "

:. ' "
,
; a -, j ' .'
r

,
ig the correct deep etruture of relative c1aueo , a c1im i
4\
:-

j_
'; !; ,
1I 4
I

% 4 :t . et
which is implicit in Choky' earlier discuaiica of relativi : . . ' t
# '-.' '

' _3 h -- ' .

"t$I
t, ''a t I
4
V 7
;$i c1aute (cf. Chon*ky (1964a) , p. 9O bottom, and p. 33 top) :
F -'
&g,;
; -
* 44
ta

' 4 :rt4 '

p

tt: cf Lakoff and Rosø (in preparation b) .

. . .:' k
'
N :
. '
c ' : ' ' '"

: :

: £ r: : :; :
r;
I
y
, 'I
3j Por a discussion of tie relative c1ase reduction rule, '
I

'a l ;I $,
'i I

'a

(
I
I :: c :

_
cf. Sin:Lth (1961) . ;, ':

i;,,
: : .

'i '! ;'' 1

q
4r :
S ' ' ' : r ,
i' I.
'U .1

t( ''.
:t
'

iI
,
I

1
;l ,t3' !' 'i
I 'j

.it !!
4. The mos t coinp].ete diacuacion of the notions P-Marker , ' / '
ci ; : "
, , ,
;d '.4'
I
1 ? ,
;g
j 1 j , ;

;t
( proDerna1ysis and a triietural index l.a contained Lu : . '
'
'
* ;
I

1 tk :
''
'ft'atl : ; 1

(1955) A Ahotter account is given in Fraser (1963) .

$ a. a

I
a
a
?
;, j
i
j

a'
;
) I

, S. For an explanation of the term "6tructural change" cf, the

I
t a -
1 i

references of-fn-4,-or Choins?y (1957),or Lees (1960). ,'' Y

;
-
l'' \ % tt '.
'a
1_r ,, , a ." V'
a.
4
. ''.

'
'

a a

':
\ % t
! . g P t i
a
. a
'

t
e a - '

:.
,.:

:' a ',a / . ,

-- : . -;.'
' '4.
I I -,
a,
'a -'a
I I

a '' / a
a

"
a a .
a '-

'1
a a
,
a

. :-.
1

1:
: .

a '
a

I
' I$' a
a

a l

Ut$IHSl..aI_ -

'I

£
I ':

. -
 38

6. The relation exhaustively dominates is the converse of the

converse of the ISA relation (cf. Frasers(1963)). I use the

term (weakly) dominate as follows: if A (weakly) dominates

B, then A exhaustively dominates XBY, where X and Y

are (possible null) variables and B is a single symbol or

a string of symbols.- A immediately dominates B if and

only if A damiinates B and there is no Z such that A.

dominates Z and Z dominates B.

7. Sentences like I sainted red- all the houses which had white

doors are derived by a different rule which moves "complex"

NP (for an attempted partial explanation of this term,

cf. § 3.1.1.3.2. below) to the end of the first S above

them. Some results of this rule are the sentences I would

consider unwise an attem t to visit her now, Pete attributed

to Masaccio a beautiful old fresco which Joan swooned over,

They elected president a'man who had never run for public

office before,- etc.

8. There are two .acts about such sentences as those in (2.23)

which indicate that the clauses in them that start with a

wh-word axe in fact questions, and not the type of clause

which has been called "the free relative clause," such as

the wh-word clauses in I eat what she cooks or I live where

he lives.
 39

1. Questions exclude sentence adverbs, like perhaps,

possiblx, etc., as was pointed out by

Katz and Postal (cf. Katz and Postal (1964), p. 87-88).

Thus the following sentences are impossible:

* Did John probably hurt himself?

* What will she perhaps wear?

* Where did you possibly find this?

The same restriction, however it is to be stated,

which is far from being clear, obtains after such

verbs as ask and wonder,

* I wonder whether to probably leave,

* Tom asked where he should possibly put the car.

although after ask there are contexts where these

adverbs can occur; e.g., Tom asked where Jane

probably put the car. There is still much to be

explained here.

2. The word else can appear after the wh-word in questions

What else did he say?

Where else did you stop?

Why else would he have come?

and after the wh-word in clauses after wonder, ask,

kaow, find out, determine, guess, etc.

 40

I wonder what else he said.

Tom asked where else I stopped.

? I Know why else he would have come.

but it cannot appear after the wh-word of a free

relative clause

* I ate what else she cooked.

* I live where else he lives.

9. I will occasionally wish to designate more than two degrees

of acceptability; when I do so I assert that I find that

sentences prefixed with an asterisk are completely unacceptable;,

those prefixed with a question mark followed by an asterisk

are only barely acceptable, if at all; those prefixed with a

question mark are not quite fully acceptable; and those with

no prefix are completely acceptable.

 41

Chapter 3

TREE PRUNING1

3.0.

3 nin A fairly serious failinR of the present theory of

generative grammar is that it assigns to many senteaceq. derived

constituent structures which seem intuitively to be overly complex.'

For instance, sentence (3.1) would probably be assigned some such

structure as the one given in (3.2):

(3.1) John is taller than Dill.

(3 . 2 )

VP

is Av
Adj er 'than 1i).

tall NP

Bill

At present, I am not interested in the question of

what the node over the constituent than Bill (if indeed it is a

constituent at all) should be labeled, so I have avoided the issue

by labeling it with a question mark. What concerns me at present

is only the question of whdther the NP"Bill should be immediately

 42

iominated by the circled node S. It seems intuitively abhorrent

to assert that, in sentence (3.1), the single word Bill has the

same status as a constituent as the whole sentence, and yet that

is precisely the assertion that the labeled bracketing in (3,2)

makes. And yet 'in sentence (3.3), from which (3.1) is derived

by the deletion of the second occurrence of the word is, it

seems more reasonable that the phrase Bill is should be called

1
a sentence,

(3.3) John is taller than Bill is.

for there is every reason to believe that the underlying structure

contained the sentence Bill is tall. Transformational grammarians

since Harris (cf. Harris (1957), p. 166) have agreed that sentences

containing comparatives derive from sources containing at least

two sentences, and in more complex comparative sentences, like

those in (3.4)

fq LI This sofa is longer than the room is wide.

Tom is smarter than anyone thought he would

prove himself to be.

Bannister ran a little faster than it was

necessary for him to run.

there is no intuitive difficulty in labeling as sentences the

phrases which follow than. But the phrase Bill is in (3.3),

which it seems correct to call a sentence, ceases to be felt to

be one when the word is is deleted.

 43

Similarly, it seems counter-intuitive to claim, with

the present theory, that the correct structure to assign to a NP

like his yellow cat, is one roughly like the one shown in (3.5).

(3.5) NP

Dec

Art Postart "c t

NP VP

N Adj

Poss 'yellow,

he

Once again, recent research in syntax has called into

question many facets of the analysis implicit in (3.5) (cf. Postal

4

(196a) and Lakoff and Ross (in preparation, b)), but at present

I am only interested in the fact that it seems incorrect to claim

2
that the words his and are sentences. In the present

theory, an NP like the one diagrammed in (3.5) would, correctly

I think, be derived from an underlying NP with two relative

clauses: the cat which I have which is,....Lea:)%7. The motivation

for deriving possessi'vzs and prenominal adjectives from relative

clauses is well-known enough not to need recapitulation here

 44

(cf., e.g., Harris (1957)), although several real'problems

remain (cf. Winter (1965)). But it seems to me that the analysis

is well-established enough to make the appearance of the two

circled S nodes in (3.5) more than a pseudo-problem.

3.0.1. To overcome the inadequacies of the present theory,

which I have just discussed, I propose that the following principle

be added to the theory of derived constituent structure:

(3.6) "S'- Pruning: delete any embedded node S

which does not branch (i.e,, which does not

immediately dominate at least two nodes).

This principle should not be thought of as a rule which

is stated as one of the ordered rules of any grammary but rather as a

condition upon the well-formedness of trees, which is stated once in

linguistic theory, and applies to delete any non-branching S nodes

which occur in any derivations of sentences of any language. The

condition that (3.6) only affect embedded S nodes, which was

suggested to me by asloge Lakoff, is necessary to prevent the node

S which should dominate imperative sentences like go homer iron

deleting when the subject,Isu, is deleted."'

It is easy to see that (3.6) will operate on the

circled instances a the node S which were pointed out to be

intuitively incorrect in diagrams (3.2) and (3.5), but the only

evidence I have given so far for adopting (3.6) is that without

er......r-ropmrreroorbrpr....."'"`1,
 45

it, counter-intuitive derived structures would be produced. This

is already a sufficient reason for incorporating (3.6) or some-

thing like it into the theory, but it might be objected that (3.6)

could be replaced by some other convention which would do as well

for the two cases I have discussed. Below, however, in § 3.1, I

will discuss eight cases which I know of,whose correct analysis

seems to me to depend upon occurrences of S being pruned out

either by the principle stated in (3.6) or by some more general

principle which subsumes It. These cases constitute even stronger

evidence for (3.6), for in each case the rules which would be

required in order to describe the facts accurately without the

principle are far more complex than the rules which can be

formulated if the principle is adopted. In most cases,'ad hoc

conditions would have to be placed upon the latter rules, but in

some cases extra rules would have to be added, and in one case,

which is discussed in § 3.1.4, the facts seem to me to resist

description completely, unless one allows the Complex NP Constraint

(cf. § 4.1), which is applicable elsewhere in English and which

I believe to be universal, to be avoided somehow for just these

cases.

3.0.2. Before I start in on a detailed analysis of the eight

cases, I would like to add one final prefatory comment, which was

suggested to me by James Thorne, in a recent letter. Traditional

 46

grammarians distinguished between'phrases,and clauses, and while

a considerable effort has been made, both in structuralist

linguistics and in generative grammar, to reconstruct the former

notion (the resulting theoretical entities have been called

(immediate) constituents, Iagmemes, or trees), little attention

has been focussed on the latter notion, to the best of my know-

ledge, in any recent theoretical work. In the framework of

generative grammar, it would seem that the most natural

reconstruction for the traditional-notion:of'clause of a

sentence would he "any subpart (not necessarily proper) of the

terminal string of the final derived phrase marker of a sentence

which is dominated by the node S." But' without some notion of

tree-pruning, the cases discussed above, (3.2) and (3.5), are

counter-examples to this reconstruction, for no traditional

grammarian would designate as clauses the words Bill, his, or

yellow. However, with principle (3.6), these words are no

longer dominated by S in the derived phrase marker, so the

definition just proposed is again in lane with the traditional

notion. It might be thought that the distinction between clause

and phrase is a minor one, but I feel that the contrary is the

case. Many rules can only be stated if the notion of clause is

available (three of these -- the Latin word order rule, the

Serbo-Croatian clitic placement rule, and the English reflexive

rule -- will be discussed in the next section), and I think it

 47

is fair to say that the fundamental idea of transformational

grammar -- Harris's insight that complex sentences can be thought

of as being in some way "composed" of more elementary sentences,

whieh may only appear in A defnrmeA gbape in the romplelc cientenee

can be traced back to the realization that what might be called

"clauses of the underlying structure" may differ from the things

which have traditionally been called simply "clauses," but which

it might be more accurate to call "clauses of the superficial

structure." And the failure of traditional grammarians to

recognize that the clauses' a2 and I'shave myself underl=3

the phrases 'ILsa and'shaving myself in (3.7)

(3.7) I want to go.

Shaving myself is difficult for me.

may derive in part from the fact that such principles as (3.6)

were not available to them.

The first of the eight cases I will discuss 'ilas to do .

with the interaction of the Particle Movement Rule and "colplex"

NP. Verb particles in English are a subset of the English

prepositions which occur in such two-word idiomatic verbs as

eke out think'overicallm, show off, etc.4 Since there is a

 48

close lexical connection between verb and particleAbruit, for

instance, only occurs in,English in construction with the

particle about ), in previous transformational accounts it has

been assumed that the structure underlying (3.8i.) is basic and

that .(3.8b) is derived from it by a rule roughly like the one

given in (3.9) (cf. Chomsky (1962), p. 228).

(3.8) a. The shock touched off the explosion.

(3.8) b. The shock touched the explosion off.'

(3.9) Particle Movement

X V Pt-t: - NP - Y
OPT OBLIG if 3 is a pronot
1 2 4 BLOCKS if 3 is "compl(

1 0 3+2 4

The condition that (3.9) be obligatory if the object

NP is a pronoun has been imposed in order to exclude sentences

like * I called ua him. But it is the second condition on (3.9)

which I am primarily interested in)in connection with the problem

of node deletion. Chomsky notes (cf. Chomsky (1961), fn. 18)

That whatever "complex" in the second condition on (3.9) may mean,

it cannot be equated with "long", for he finds (3.10a) though

far longer, far more acceptable than (3.10b).

(3.10) a, I called almost all of the men from

Boston up.
 0

49

b. * I called the man you met up.

I agree with his intuitions, but I must point out that

there are people who find (3.10b) perfectly acceptable, and there

may even be people who find :ct better than (3.10a) . The whole

problem area of what NP are felt to be "heavy" or "complex"

borders on questions of style, and there seems to be a baffling

array of dialectal, or possibly even idiolectal, variations here.

Since I have not made a systematic study of this variation, I

can have no hope of finding examples whose acceptability will be

agreed or by all readers, if indeed such examples exist. Instead

I must resort to describing the facts of my own speech, insofar as

they can be ascertained with any consistency, for this area is really

a grnunatical shadowland, and I fear my own judgments may change

from time to time. I can only hope that most readers will share

my judgments, at least in part.

3.1.1.2. With this caveat, I would like to propose the following

definition as a partial explication of the notion of "complex" NP.

(3.11) A noun phrase is complex if it dominates the

the node S.

Used in conjunction with the principle for S-pruning, (3.6),

definition (3.11) explains why sentence (3.10b) is less acceptable

than sentence (3.10a): in the d.c.s. or the former, the node S

will dominate the relative clause'you'met, so the object NP,

 50

definition (3.11); but in

the man you met, is complex, under

is
(3.10a), although the postnominal modifier'from Boston

derivedfromarelativeaause-who are from Boston, the node

will have
S which dominates this clause in the deep structure

Rule5
been pruned by (3.6) when the Relative Clause Reduction

deletes the subject NP* who and the copula are.

in
A similar explanation holds for the sentences

(3.12), (3.13), and (3.14). The b version of each of these

which dominate
sentences is more acceptable, because the nodes S

deleted after the who is

the relative clauses of the a versions are

has been dropped by thelelative Clause Reduction

Rule

(3.12) a. * I ran a man who was old down.

b. I ran an old man down.

(3.13) a. * I'm going to call somebody who is

strong up.

b. ? I'm going to call somebody strong up..

(3.14) A. * I polished the vase which was from

India up.

b. ? I polished the vase from India up.

I find sentences (3.13b) and (3.14b) somewhat worse than

(3.12b), although none_of.them are complex according to definition

(3.11). It is thus clear that (3.11) cannot be strengthened to a

for an NP to dominate the node S is a sufficient,

biconditional:
A
but not a necessary, condition for diminished acceptability.
 51

possible explanation for the less than full acceptability of

(3.13b) arid (3.14b) will be suggested below, in S 3.1.1.3.

Nevertheless, despite the fact that principle (3.6) cannot explain

the variations in acceptability among the b sentences, the fact

that it and definition (3.11) can predict the difference between

the a sentences and the b sentences is an indication of the

correctness of (3.6).

3.1.1.3.

3.1.1.3.1. I will now discuss what I consider to be an inadequacy

of the previous analysis of particles, or of any analysis which

includes conditions like these on (3.9). The second condition on

(3.9), it will be remembered, was one which prohibited' Particle

Movement, from moving a particle over a complex NP. I wish to

argue that to state this as a condition on Particle Movement

alone is to miss a very general fact about complex NP in English.

In sentences (3.15) to (3.19) below, the a-sentences, in which the

direct object immediately follows the verb, are basic, as is

demonstrated by thelmaxeptability of the b-sentences, in which

the direct object has been moved to the end of the verb phrase.

(3.15)-W: He attributed the fire to a short circuit.

b. *He attributed to a short circuit the fire.

c. He attributed to a short circuit the fire which

destroyed:4ost of my factory.
 52

(3.16) a. He threw the letter into the wastebasket.

b. * He threw into the wastebasket the letter.

c. He threw into the wastebasket the letter

which he had not decoded.

(3.17 a. We elected my father president.

b. * .We elected president my father.

c. We elected president my father, who had

just turned 60.

(3.18) a. They dismissed the proposal as too costly.

b. * They dismissed as to costly the proposal. .

c. They dismissed as too costly the proposal ,

for the State to build a sidewalk from

Dartmouth 'to Smith.

(3.19) a. I consider the problem unsolvable.

b. * I_consider unsolvable the problem.

c. I consider unsolvable the problem of

keeping the house warm in winter.

The grammaticality of the c-sentences can be explained by

a rule which optionally moves a complex NP to the end of the first

sentence up. As the non-sentences in (3.20) show, however, this rule

must be restricted in some way,

iforced
(3.20) a. * I to eat hot soup all the children
lwanted
who were swimming.
 53

b. * 'I told that we were in trouble a an

who had a kind face©

c. * I watched talk(ing) all the children who

had never seen the sea.

d. * He restrained from attempting to bend the

bars a cellmate he had know= the outside.

for all of them'are the result of moving a complex NP to the end

of the S which contains it. It might be proposed that the rule

should be restricted so that a complex NP can move to the end of

its S only if it does not pass over a VP in moving there. Such a

condition would be sufficient to exclude the ungrammatical examples

in (3.20), but unfortunately it would also exclude (3.18c) and

(3.19c) , since I see no reason why tha cosh and

unsolvable should not be considered to be verb phrases. Furthermore,

the sentences in (3.21), which show that one complex NP can be

moved over another, provide additional evidence against the proposiaa

condition, for the second complex NP, over which the one being

moved permutes, will of course_contain a VP. (I have underlined

these VP's in (3.21)0

(3.21) a. He attributed to a short circuit which

was caused by an overloaded transducer

the fire which destroyed most of my factory.

b. He threw into, the wastebasket which stood

by his desk a letter which he had not

decoded.
 54

c. They dismissed as too costly to people

who'live in*thesuburbs the proposal for

the State_tobuild a sidewalk from Dartmouth

to Smith.

Clearly the condition must be weakened somewhat, but

before this is attempted,.one-further.class of constructions must be

....

taken.into consideration.

(3.22) a, found to be delicious some fruit which

I picked up on the way home.

b. I found delicious some fruit which I

picked up on the way home.

(3.23) a. ?* The mayor regarded as being absurd the

proposal to build a sidewalk from

Dartmouth to Smith.

b. The mayor regarded as absurd the proposal

to build a sidewalk from Dartmouth to

Smith.

(3.24) a. * I consider to be a foci the senator

who made the opening speech.

b. ? I consider a fool the senator who made

the opening speech.

For me, at least, the a-sentences above are considerably

worse than the b-sentences, although some speakers may find the

distinction not to be as ciearcut as I have indicki., This then

 55

indicates that the rule which moves complex NP must be made

sensitive to the presence of the copulalbe, for the.a and b-

va, 4 U. 4
A444^.44
44G101 CM/ W V 4 %I J.i .16. G4 411.4:y
4,
4.44
444.44.
44444 W
^.44.4e,
e 444 a %.1,
44.4
ai.AM
14.4 wnmm^y4n,n1 Pt Q

and does not appear in the ones which are grammatical. Under previous

generative analyses of adjectives, such as the one found in Chomsky

(1965), on p. 102, in which be is not treated as a verb, but rather

as a terminal element of the base component, no simple statement of

the restriction on the complex NP rule is possible, as far as I

can see. However, under a new analysis of adjectives, which I have

proposed in some detail elsewhere (cf. Ross (1966c)), the restriction

is easily stated. In this new analysis, which is independently

motivated by a number of constructions, be is treated as a real verb

which takes a sentential object. Using the feature [4. Adj:16, the

underlying structure of John is has as shown in (3.25).

(3.25) S

NP VP

John Au(*(MV
I

Pres V NP
I
///N.N..
r +v 1 it S
-,Adi
-,'

1 NP
be
John Aux MV

? [ +V

'happy
 56

(I have used a question mark for the auxiliary of the embedded

sentence to indicate my uncertainty as to whether it should appear'

at all there, and if so what node it should dominate)

Under the analysis which is implicit in (3.25), the

restriction which is necessary to exclude the sentences in (3.20),

(3.22a), (3.23a), and (3.24a), while allowing (3.18c), (3.19c), (3.21)

(3.22b), (3.23b), and (3.24b), can be stated as follows: a complex

NP may permute to the end of the first sentence up, providing

1 +v 1
it permutes over no true verb (i.e., ))unless that verb is
LAdj Ji
dominated by an NP. More formally, the rule is

(3.26) awlex NP Shift

X NP Y
OPT
1 2 3

1 0 3 + 2

Condition 1: 2 dominates S

2: BLOCKS if 3 = X
1
4.+V
Adj
+ X
2

where there exists no NP which

+V 1 7
dominates [-Ad .

J-12.

Notice that (3.26) will generate (3.20b) * I'told

that we were it trouble a man who had a kind face? It might seem

that this sentence could be excluded on the basis of the very general

output condition on performance, which is stated in (3.27):

 0

57

(3.27) Grammatical sentences containing an internal

NP which exhaustively dominates S are

8
unacceptable.

(3.27) would explain why (3.20b) is unacceptable -- it

contains an internal NP which exhaustively dominates the sentence

that we were in trouble, Some condition like (3.27) seems to be

necessary in any case: note that (3.27) also explains why the

a-sentences of (3.28) to (3.33) are worse than the corresponding

b- or c-sentences.

(3.28) a. * Did that John showed up please you?

b. Did the fact that John showed up please

you?

c. Did it please you that John showed up?

(3.29) a.?* That that John showed up pleased her

was obvious.

b. ? That the fact that John showed up pleased

her was obvious.

c. That it pleased h3r that John showed up

was obvious.

(3.30) a. ? *For whether she died to remain unclear

would spoil the play. 1

ib. ? For the question as t° whether she died

of

to remain unclear would spoil the play.

c. For it to remain unclear (as to) whether

she died would spoil the play.

 58

(3.31) a. 2* I want that Bill left to remain a

secret.

b. I want ale fact that Bill left to

remain a secret.

c. I want it to remain a secret that Bill

left.

(3.32) a. * What what I ate cost almost broke me.

b. What the thing which I ate cost almost

broke me.

c. What the thing cost which I ate almost

broke me.

(3.33) a. * I went out with a girl who that John

showed up pleased.

b. ? I went out with a girl who the fact

that John showed up pleased.

c. I went out with a girl who it pleased

that John showed up.'

In each of the a-sentences, (3.27) applies and explains

their unacceptability. In the b-sentences, (3.27) does not apply,

because a head noun (fact, suestion,orlAn0 has been added to

the internal sentence that produced the unacceptability in the

a-sentences, so that they are no longer exhaustively dominated by

NP. And in the c-sentences, extraposition has applied, and the

offending sentences_are no longer exhaustively dominated by NP.

 59

But although (3.27) will explain why the a-sentences

as a class are worse than the b- or c-sentences, it will not explain

why (3,29a); (3,30a); and (3.31a) are slightly better than the

others, which means it is not sufficient. And although (3.27) seems

9
to be right, in many cases, I do not think it can explain the

ungrammaticality of (3.20b), which I find to be absolute word salad.

Sentences (3.28) to (3.33), while ponderous and taxing to read, are

still,decipherablet but (3.20b) is baffling. This means that some

other condition must be placed on (3.26); what I believe to be the

correct one is given in (3.34) (bat cf. 6,3, 3 Lis 4),

(3.34) Condition 3: (3.26) BLOCKS if Y NP where

3
lej 0 [E. ÷
he
(3.34) seems to produce the right results in many cases:

it allows (3.15c) and (3.16c), but excludes (3.20b). Furthermore, it

correctly prevents (3.35a) from becoming (3.35b), and (3.36a) from

. becoming (3.36b).

(3.35) a. I loaned a man who was watching the

race my binoculars.

b. * I loaned my binoculars a. man who was

watching the race.

(3.36) a. She asked a man who was near the window

whether it looked like rain.

b. * She asked whether it looked like rain a

man who was near the window.

 60

However, Condition 3 also incorrectly excludes (3.17c) -- We

c s elected_presideraLmfatherj who had just turned 60, for president
.
Is an VIP. At present I see no way arovad'this wrong result.

Nevertheless, it seems beyond dispute that a rule like

(3.26) must appear in the grammar so that complex NP can be

displaced from their underlying positions. This rule will be optional,

and it must be supplemented by some output condition which will

stipulate that if a sentence contains an un-permuted complex NP

near the end" of its .VP, the acceptability of the sentence is

lowered. Thus, for instance, the sentences of (3.37) must all be

designated to be unacceptable in varying degrees.

(3.37) a. * We called my father, who had just

turned 60, up.

b. ?* We elected my father, who had just

turned 60, president.

c. ? All those speeches made my father,

wilt had just turned 60, mad.

d. * They gave my father, who had just

turned 60, it.

However, there are.many more sentence types than those

in (3.37) which must be taken into account before this output

condition can be stated in its fullest generality. Some of these

follow:
 61.

(3.38) a. He figured it out.

b. * He figured out ,it.

He f4gwrollrhnt cint

d. * He figured out that.

e. He figured Ann out.

f. ?* He figured out Ann.

g. He figured something out.

h. ? He figured out something.

i. He figured the answer out.

j. He figured out the answer.

(3.39) a. * I sent him it.

b. I sent him that.

c. ? I sent him Andy.

d. I sent him something.

(3.40) a. ?* We elected the man who he had brought

with him president.

b. ? We made the reports which he had brought

with him available.

c. They gave the reports which he had

brought with him to me.

Once again, I must emphasize that these judgments, which

are not sharply defined in any case, may only hold for my own speech.

Nevertheless, I, would expect similar phenomena to exist in most dialects.

ti

62

3.1.1.3.2. It seams to me that such facts of acceptability as

those indicated in (3.37) - (3.40) can most readily be accounted

for by 'a theory constructed along the following lines. First of

all, all the sentences in (3.37) - (3.40) should be generated by

the grammar and designated as being fully grammatical. With the

exception of Complex NP. Shift, (3.26), no conditions having to

do with complexity will be imposed on any rule, and the same

thing applies to conditions having to do with pronouns. This

means that neither of the conditions on Particle Movement, (3.9),

will appear, and both (3.37a) and (3.38b) will be generated.

Similarly, the Dative Rule will not be restricted so as not to

apply if the direct object is a pronoun: (3.37d) and (3.39a)

10
will also be generated.

Instead of restricting the operation of particular

V)h.\I, rules, I propose that an output condition, much like (3.27), be

stated, which imposes an ordering upon the constituents which

follow the verb of the sentence which contains them, and lowers

the acceptability of sentences whose constituents are not

arranged in accordance with this condition. It will be remembered

that (3.27) had a cimilar effect: it rendered unacceptable

perfectly grammatical sentences which contained an NP which

exhaustively dominated the node. S.

The output condition which I propose ia (3.41) is

highly tentative, for I have not done much research on this exti:cnely
 63

difficult problem. (The lower the number before a constituent

in. (3.41), the closer it must be to the verb.)

(3.41) Out ut Condition on PostVerbal Constituents

1. Direct object pronouns

2. a. Indirect object pronouns.

sb. Demonstrative pronouns and integers

used as pronouns'(give me two)

3. Proper names

4. a. Particles (ue in call up)

b. NP with no postnominal modifiers

11
5. Reduced directional phrases (out in let out)

6. NP like president in elect him president

7. Single adjectives like available in make

thLreport.L available

8. Indirect object phrases and directional

phrases

9. Non-complex NP with postnominal modifiers

10. Complex NP

11. company in. kelkac22111EL

The ordering in (3.41) is doubtless wrong in many

particulars, but it incorporates some generalizations which cannot

be expressed if conditions on rules, such. as the ones stated on (3.9),

are used instead of it. For instance, to say that direct object

pronouns occupy the first place in such an ordering as (3.41) is to

kruirrONINfflimmInwvirimPonnerwr
 64

simultaneously exclude both (3.38b) and (3.39a); but in a system

which makes use of conditions on rules, one condition would be

needed to exclude each. Furthermore, in this latter system, there

is no way to indicate that both of the sentences to be excluded are

unacceptable for the same reason, but (3.41) does make this claim,

which I believe to be a true one.

I will now attempt to justify (3.41), insofar as that

is possible in my present state of ignorance. In many cases,

particularly in the higher numbers of (3.41), I have put one

constituent before another on the basis of very scant evidence.

Firstly, (3.41) is only a partial ordering, and a

number in it which is followed by the letters a and b indicates

that for me, there seems to be no preferred ordering of the

a-constituents with respect to the b- constituents. This is the

case in two instances: I find no difference in acceptability

between I called an old friend up and I called old friend

(these are the two constituent types mentioned in 4 of (3.41)),

nor between the sentences give me that! and give'that to me!

(2 of (3.41)).

Secondly, (3.41) makes the prediction that violations

of the hierarchy which arises from permutations of constituent

types which are close to one another in terms of (3.41) will

lead to smaller losses of acceptability than permutations of

constituent types wh:l.n are far apart in (3.41), and this

 111

65

prediction seems to be borne out in a number of cases. For instance,

the sentencelsoliatin (3 follows 4) is better

than I tried to figure out that (2 follows 4). I also find Le_ t the

dogs which are barking out (5 follows 10) somewhat better than

Knock the dogs which are barking out (4 follows 10). These two

sentences provide the motivation for distinguishing in (3.41)

between the reduced directional adverbs discussed in footnote 11

and true particles. In addition, I find that while constituent

types 4a and 4b are equally acceptable in either order, constituents

of type 5 are more comfortable to the right of constituents of

type 4b than to the left of them. So knock out'the sentry! is as

natural as knock the sentry out!, whereas let'out is

somewhat less natural than let'the sentr,- out

only motivation for ordering constituents of types

6, 7 and 8 as I have is that it seems to me that complex NP

A

(type 10) can precede 8 more readily than it can precede 7, and

7 more readily than 6. This is exemplified in (3.40): (3.40a),

which is the least acceptable for me, has the order 10-6; (3.40b),

which is slightly better, has the order 10-7; and (3.40c), which

is'almost, if not totally acceptable, has the order 10-8.

Constituents of type 9, for example, the NP amebo4k

slam, are ordered closer to the verb than complex NP like

somebody.Fho is strong. This explains why (3.13b), which has the

order 9-4, is better than (3.13a), which has the order 10-4. The

same explanation can be given for the difference in acceptability

 66

between (3.14a) and (3.14b).

Finally, I have included in type 11 such words as

company in 1=2_211mm, through in see (someonc)..shrought.to in

a122.L(sol__neone)to and 'on in zLi...ieoneLsa_utsoti, because for me

these words must always end their VP, unless a relative clause

has been extraposed around. them. In the sentences below, the

a-sentences are the least acceptable, the b-sentences, in which

a complex NP precedesa constituent of type 11, are somewhat

more acceptablP* and the c- sentences, in which Extraposition

from NP has applied, are the most acceptable of all, although.

12
they are still awkward.

(3.42) a. * He kept company some girls who had

been injured in the wreck.

b. ?* He kept some girls who had been injured

in the wreck company.

c. ? He kept some girls company who had.

been injured in the wreck.

(3.43) a. * I insist on seeing through all the

students who started out the term in

13
my class.

b. ?* I insist in seeing all the students who

started out the term in my class through.

c. I insist on seeing all the student' through

who started out the term in my calss.

 67

(3.44) a. * The doctor brought to the passengers

who had passed cut from the fumes.

* Tha Anntnie rickaa..46u.A.a wuy

had passed out from the fumes to.

c. ? The doctor brought the passengers to

who had passed out from the fumes.

(3.45) a. * He tries to put on everyone who he

doesn't like.

b. ?* He tries to put everyone who he doesn't

like on.

c. ? He tries to put everyone on who he

doesn't like.

These sentences raise many problems I cannot deal with..

Firstly, I cannot explain why (3.43c) should seem more acceptable

than the other c-sentences, or why (3.44b) should seem less

acceptable than the other b-sentences. Secondly, it may be the

case that the ?-sentences are so bad that they should not be

generated at all -- this would entail restricting (3.26) so that

complex NP immediately to the left of such words as company,,

through, etc. could not undergo the Complex NP Shift Rule. More

damaging is the fact that the hierarchy in (3.41) predicts that all

the b-sentences should be the most acceptable of all, in fact

perfectly acceptable, but in no case are they anything better than

barely acceptable. This means that the hierarchy must either be

 supplementary output
modified or that it must be supplemented by some

containing
condition which lowers the acceptability of any sentence

a complex NP near its end, even though the ordering in (3.41) is

NP
adhered to. So, for example, in (3.46), even though the object

(3.26), Complex
of the verb watch is complex and very lengthy, rule

Shift,, cannot move it over the VP talk because of Condition 2

NP

on (3.26).

(3.46) * I watched the Indians who the man who had

been my advisor in my freshman year had

advised me to study when I got to Utah talk.

Notice also that the unacceptability of such sentences

(3.45) can be reduced

as (3.46) and of the b- sentences in (3.42)

by adding material to the end of the sentence:

? I watched the Indians who the man

who had
(3.46)

been my advisor in my freshman year had

advised me to study when I got to Utah talk,

because I was fascinated by the way their

view of the world seemed to be constrained

by the structure of their language.

(3.42b") 2 He kept some girls who had been injured in

the wreck company, and meanwhile I scouted

around to see if I could find a phone.

 69

(3.43b') ? I insisted on seeing all the students who

started out the term in my class through,

after they had all chipped in to buy me a

going-away present.

(3.44b') ?* The doctor brought the passengers who had

passed out from the fumes to, but many of

them suffered relapses at various times

during the night.

(3.45b'). ? He tries to put everyone who he doesn't like

on, by pretending to be deaf.

These sentences show that it will be very hard to state

in formal terms just what "near the end of an S" means, for it secns

that the acceptability of sentences like the b-sentences and sentence

(3.46) must be assigned by a quasi-continuous function of the length

and complexity of the_object NP and the length auj,cpmplexity of

what follows. And (3.41) is at best a. first approximation of such

a function.

3.1.1.3.3. One final important question which must be raised is

the following: what is the theoretical status of such output

conditions as (3.27) and (3.41)? In the case of the former, it

seems that although it has not yet been formulated adequately, it

is not being overly optimistic to hope that a more adequate version

of (3.27) may turn out to be universal. But it is out of the

 70

question that the particular content of a condition such as (3.41)

could be universal, for in (3.41), the constituent types are

definedwithreferencetoconstituentslikeParticle_Reduced

Directional Phrase, comDany in he:to.alma, etc., all of which

are peculiar to English. One might wish, therefore, to make a

theoretical distinction between (3.27) and (3.41), referring to

universal conditions as "performance filters," and to all

languageparticular phenomena, such as those discussed in

connection with (3.41), as ordinary_rules_of particular grammars.

In my opinion, it is correct to draw such a distinction, but I

would like to emphasize that if (3.41) is to be added to the

grammar of English, it will be a rule of a type wl-tch is completely

different from other transformational tiles, First of all, where

other rules change one. PMarker to another, (3.41) does not: it

merely changes the acceptability index of PMarkers. Secondly,

"violations" of (3.41) do not produce total unacceptability (except

in extreme cases), but rather a partial loss of acceptability, with

the amount of loss a function of the input tree and the structure

of the rule. It is easy to see that other rules are entirely

different in this respect: if an ordinary rule applies to a tree

it should not have applied to, or does not apply when it should

have, it is rher the case that an unintelligible string is

produced 02.1QE111.115.1.1011L1:1221111.1.1.SLD, or if
intelligible (though ungrammatical); the strings produced do not
 71

vary in amount'of deviance according to the input structure'(that

is, the forced me for me to wash myself is as deviant as I forced

vou for you to wash the veQetables.)

These considerations suggest that if (3.41) is to be

put into the grammar of English, it should be segregated from the

normal type of transformational rules, to whose output it applies,

and placed in a component by itself, a component which I tentatively

propose to call the stylistic component. Of course, (3.41) will

not be the only rule in this component, but at my present state of

knowledge, I can only suggest two other rules that seem to be likely

candidates for inclusion in it. The first is the Scrambling ...Ill

in Latin and other "free word order" languages, which will be

discussed separately in § 3.1.2 below, and the second is the

condition which must be imposed on prenominal adjectives with

respect to their closeness to the noun they modify. In the case

of the latter problem, if adjective sequences were to be constrained

in deep structure, an entirely new system of selectional restrictions

would have to be created, and this system would only be used to

generate the permissible sequences of adjectives, as far as I know.

In other words, to attempt to account for order-of-adjectives

phenomena in deep structure would require setting up an elaborate and

totally ad hoc mechanism, which would greatly increase the class

of languages characterized by the theory of generative grammar, but

unnecessarily, for the extra descriptive power would be used to

 72

solve only one problem. On the other hand, if another output

condition, highly similar to (3.41), were to be added to the

stylistic component, which the discussion above has demonstrated

is likely to be necessary in any event, then the theory would not

be weakened at all. Furthermore, it seems to me that the type

of phenomena which the two. conditions would account for are

phenomena of the same type. That is, in both cases, we have to

do with constituents which occur in a 4.rlierred order. It is

not that'let out John! and'a spotted young; &m are to be

categorically ruled out, but rather that let John out!. and a

14
young spotted doR are more natura1. So it seems to me that it

would be wise to separate into disjoint parts of the grammar rules

which must produce constituents in an order from which any deviations

15
produce ungrammaticality, from rules which produce constituents in

an order which, within limits, is variable. The only possible reason

that I know of to question the decision to relegate constraints on the

order of adjectives to the stylistic component is the possibility that

NP with different orders of adjectives may not be synonymous, in which

case, of course, order constraints would have to be stated in the

base. It has been suggested by Quine (cf. Quine (1960) p. 138) that

the NP abiLlumplcalL3nlazdesignates a butterfly that is

\both European and big, while the NP 2..12E2211111IITtalfly may

designate a butterfly hich is in fact small, but is big for

European standards. I am not sure of the validity of this example,

 73

and I have not studied the problem closely enough to be able to

say whether such examples are sufficient to refute my proposal to

pbonnrqpnn in thA ntyliatic component; or

not. I mention the problem here only to call it to the attention

of the reader.

3.1.1.4. To summariLe briefly what I have touched on in this

digression, I have suggested that to put two conditions on the

previously proposed Particle Movement Rule , was to miss

the generalization that both conditions were merely extreme cases

of a rule relating the length and complexity of constituents of

verb phrases to their ordering after the verb. To capture this

generalization, have proposed adding a stylistic component

to the set of components of a generative grammar, and stating in

it language-particular output conditions, such as (3.41), which

capture the notion of preferred order, and reduce the acceptability

iI

of sentences whose constituents are in an order other than the

specified by the stylistic rules. It was in.the ordering given

in (3.41) that the notion of node deletion, the main topic of S 3,

played a rale, for the constituent types 9 and 10 were shown to

function differently with respect to the other constituent types

of (3.41), and these two types can be conveniently distinguished

in constituent structure terms if the principle of S-pruning

which was stated in (3.6) is r 'e use of.

 74

3.1.2. The second case which seams to require some notion of

node deletion has to do with Latin word order. In Latin, as in

languages like Russian, Czech, etc. the order of major elements

within a clause is free, within certain limits. Thus the subject

NP may precede or follow the VP, the object NP may precede or

follow the V, etc. In Latin poetry, it was even possible for

adjectives to be separated from the nouns they modified. Robin

Lakoff has kindly provided me with the following example from

Horace (Carmine (Odes I), 5)

(3.47) Quis mut.: grad to puer in rose

What may a slender you boy on rose

r
perfusus liquidis urget
I
odoribus

drenched liquid makes love to (with) scents

grato, Pyrrha, sub antro?

delightful Pyrrha in a cave

'What slender boy, drenched with perfumes

Is making love to you, Pyrrha,

On a heap of roses, in a delightful cave?'

 75

Words in (3.47) joined by lines are discontinuous

constituents which have been derived from contiguous constituents

in a slightly deeper structure by a rule of roughly the following

form:

(3.48) **Scrambling

NP
1
X V:
N
V
VP
1t'
N
V
Y
,

Adj Adj
Adv
`Adv
J
OPT
1 2 3 4

1 3 2 4

Condition :16 S dominates 2 if and only if

i

i dominates 3.
S

Rule (3.48) scrambles major constituents,*sublectIl

the restriction that the be in the same clause. For instance, (3.48)

will convert (3.49a) into (3.49b),

(3.49) a. Hom3 bonus amat aminam pulchram.

b. Pulchram hom3 amat aminam bonus.

'The good man loves the beautiful woman.'

because for the purposes of scrambling, adnominal adjectives behave

as if they were in the same clause as the nouns they modify. But

note that this fact entails that node deletion has occurred, for in

the underlying structure, adnominal modifiers are not in the same

 76

clause as the noun they modify. The deep structure for (3.49) is

that shown in (3.50). The latter is converted into the former by

a rule of Relative Clause Reduction cognate with the one proposed

in Smith (1961).

(3.50)17

P
i

homB NP
....**"-- 1

'amat N7' 0
all Adj Camin am NP VP
1 1
I.
1 rNN
est bonus tla V Adj
. ..
I 1

est pulr,hra

The Relative Clause Reduction Rule will delete altest

and ma./ est from the embedded relative clauses in (3.50). If the

S-pruning principle of (3.6) were not in the theory of grammar, the

circled S-nodes in (3.50) would not be deleted, and Scrambling

would not be able to apply to the adjectives bonus and 2ulchram to

permute them with the elements of the main clause of t3.50), for

the adjectives would be in clauses of their own. But the fact that

(3.49b) is grammatical indicates that'Scramblim must affect them,

and thus this fact constitutes further evidence for the correctness

of principle (3.6).
 77

For my present purposes, I am not t:,,fix-sy concerned the.;

(3.48) is too strong, for the problems involved in specifying exactly

the correct subset of the strings which will be generated by (3.48)

are far too complicated for me to even mention them here, let alone

come to grips with them. In § 3.1.1.3 above, I suggested that rules

like (3.48) be placed in the stylistic component, because they are

formally so unlike other transformational rules. In the first place,

since '(3.48) can apply an indefinite number of times to its own

output, every sentence will have an infinite number of derivations.

It seems tmong to use normal rules of derived constituent structure

to assign trees to the output of this rule, for the number of trees

that will be assigned to any sentence, although it will be bounded,

very large, and there will be no correlation between the

number of derived trees and perceived ambiguities, as there is in

happier circumstances. In short, it is clear that rules like (3.48)

are so different from other syntactic rules that have been studied

in generative grammar that any attempt to make them superficially

resemble other transformations is misguided and misleading. They

are fo.-ally so different from previously encountered rules that

the theory of language must be changed somehow so that Scrambling_

can be placed in a different component from other syntactic rules,

thereby formally reflecting the differences I have been discussing.

It is possible that'Scrambling should be effected in

the stylistic component, as I suggested in § 3.1.1.3.3, but it
 78

should be emphasized that. there are as many formal differences between

Scramblinaand output conditions like (3.41), which I also suggested

should be stylistic rules, as there are between Scrambling and trans-

formational rules like Extraposition from NP. But it does seem, in

some ill-defined sense, that 121.412221naand output conditions like

(3.41) both have to do with such low-level matters as taste or

idiolect, which have often been grouped under the heading of stylistics;

so that it may yet be appropriate to assert that they both belong in

the same component of a grammar. But at present, our knowledge of

constraints on Scrambling, or on conditions like (3.41), or in fact

on any stylistic problems whatsoever, is so limited that nothing

but speculation is appropriate.

One final point should be made with reference to

Scrambltalg It may be possible to formulate this rule in a partially

universal way, so that it is only.necessary to specify in a particular

grammar whether it applies or not. This suggestion must be modified

somewhat, for it appears that languages with "free word order" may

differ among themselves as to the contents of the second and third

terms of the .........MScraklina2111. Thus although it appears that in Latin,

adjectives can be permuted away from the noun they modify, this

possibility either does not exist at all in Russian or is severely

limited there. This suggests that the theory of language must be

constructed in such a way, that universal:skeleton rules can be stated.

AM.
 79

that the
The skeleton for the universal scrambling rule would state

subject NP can precede or follow the VP, that the. VP can have

its constituents arranged in any order, and possibly a few other

universal conditions. In the grammar of any "free word order"

language, it would then only be necessary to state that the

scrambling skeleton rule could be applied, and to list any language.

particular additions to the skeleton. For example, in both Latin

and Russian, it would be necessary to note that scrambling could

apply, and in Latin, it would be necessary to specify in addition

that adjectives can be scrambled.

I should point out that such important traditional

concepts as "free word-order language" can only be reconstructed

by introducing some such notion as that of skeleton rule into

linguistic theory, for, as I pointed out, the grammars of languages

which exhibit "free" word-order do not all contain the same rule --

the rules in each which effect the scrambling are slightly different.

Therefore, it is necessary to factor out that part of the various

scrambling rules which is language-independent and to state this

skeleton once in linguistic theory. Then the notion "free word-order

language" can be equated with the notion "language having a grammar

making use of the Scrambling skeleton."

All the points discussed in this section are highly

conjectural, but they do not materially affect the point at hand,

.1
 80

which is that in order to state the version of the Scrambling Rule,

no matter in what component it appears, nor how much of it can.be

factored out and put into a universal skeleton rule, some net4c-

of tree - pruning must be in the theory.

3.1.3. A closely related phenomenon provides an additional

piece of evidence.for (3.6): the phenomenon of case-marking. In

Latin,. as in many other languages, noun phrases must be marked for

case in various contexts. The exact number of cases which are

distinguished in any particular language is not my concern here: the

important thing is that when an NP is marked with some case, say

accusative, then all markable elements of that NP must have the

feature [4.' Accusative] added to them. In Latin, determiners,

adjectives, possessive .adjectives, participles, some numerals, and

the head noun of the--NP are markable, and nothing else is. In

particular, elements of clauses contained in an NP are not

markable. Thus if the Relative Clause Reduction Rule does not

apply to the rightmost circled S of (3.50) above, the adjective

pulchra cannot be marked (+ Accusative]: sentence (3.51), which

would be the result of such a marking, is ungrammatical.

(3.51) * homE qui est bonus amat aminam quae est ,J.chram.

However, as sentence (3.49a) shows, once the Relp,,ive Clause

Reduction Rule has applied,'2u2slabecomes markable, and the

accusative form nulchram is produced. Once again, these facts can be

 81

accounted for simply if some principle of node deletion is invoked.

The case-marking rule, which distributes the case feature with which

the whole NP is marked onto all markable elements dominated by

it, must be constrained so that no elements are marked which are

dominated by an S which is in turn dominated by the NP in

question, as the ungramraticality of (3.51) clearly shows. Therefore,

in order for puichra to become markable, after the alse est of the

rightmost relative clause is (3.50) has been deleted, and the

circled node S no longer branches, some S-pruning principle

must delete it. Facts corresponding to these can also be found

in Germanic, Slavic, and Balto-Finnic, so it is inaly that the

solution to the Latin case-marking problem is at least partially

universal.

might remark in passing, however, that there are

many unsolved problems which have to do with the case-marking rule.

Consider, for example, sentence (3.52) and its approximate labeled

bracketing, (3.53):

(3.52) Puer amat puellau quae est similis deae.

'The boy loves 'a girl who is similar to a

goddess.'
 82

(3.53) S

NP VP

V NP

puer amat

puellam NP VP

lae Adj NP
1

est similis N

dae

If the Relative Clause Reduction Rule applies to (3.53) ,

to delete the alae est of the relative clause, principle (3.6) will

delete the circled node S, as was the case with the P-marker (3.50),

and the adjective sim, no longer contained in a clause dominated

by the object NP of 0.53), will become similem, as in (3.54).

(3.54) Puer amat puellam similem deae.

The problem is to specify how the case marking rule is

to be constrained so that deaelgoddess' (dative singular) will not

become deam 'goddess' (accusative singular); for if this occurs, the

sentence will no longer be grammatical (cf. (3.54')).

(3.54') * Puer amat puellam similem deam.

It might be proposed that the case-marking rule should

not only be restricted from marking elements in clauses which are

dominated by the NP being marked, but also from marking elements

in NP which are dominated by the NP being marked. This, then,

 83

would be a kind of A- over -A restriction which only applies to the

case-marking rule. It can easily be seen how this condition will

pi.GI,G4416 uc.c.4%.
cq1
vro.r.r, 4,-,nnrrtIntitt converted t'A'AtInM

even if Relative Clause Reduction applies, and it can also be

used to prevent (3.55a) from being converted into (3.55b)

(3.55) a. puella amat mad. frarem.

'The girl loves a friend's brother.'

b. * puella amat amIcum frarem.

because at the time the case-marking rule would apply, the sentence

(3.55a) would have approximately the structure shown in (3.56),

(3.56)

puella
N
NP
i

amat Det

NP
'%, N
1

fr"iter
.1

amYcl

and since amIcl 'a friend (gen.)' is an NP dominated by an NP, the

A-over-A restriction.on the case-marking rule would prevent it from

being changed to am. Once again, the same facts obtain in

Germanic, Slavic, and Balto-Finnic.

 84

However, it seems that this limited A-over-A

restriction is both too strong and too weak. It is too strong in

that it would exclude (3.57) below

(3.57) ,puella amat meum fratrem..

'The girl loves my brother.'

unless meum 'my'.had somehow ceased to be dominated by NP, for

otherwise the structure of (3.57) at the time ease- marking applies

would be exactly that shown in (3.56), except that*meus would

appear in the place of amici. In traditional grammar, words like

meus are called "possessive adjectives," a term which aptly

characterizes their behavior under case-marking rules, but which

provides no explanation as to how they have come to behave differently

from NP in the genitive case, like amIcl. I have no explanation

for the facts at present, but Postal has suggested a promising new

analysis of pronouns which may provide a key to the answer (Postal

(1966)). Postal argues convincingly that personal pronouns such

as I, you, he, etc., should be treated as underlying articles

(actually, in the deepest structure, these articles, as well as

words like the, a, some, etc., which have been traditionally

categorized as articles, would all be represented as features on

the noun they modify) which modify the pronoun one, and that they

acquire their derived status as nouns because of a rule which deletes

one and leaves its article (i.e., he, she;ve etc.) as the only
 0.

node still dominated by the node N which dominated one in the

deep structure. I will not recapitulate here the various arguments

Postal advances in support of this analysis: for my purpos.:s, it

is sufficient to assume their correctness. For if Postal's analysis

is correct, and pronouns are articles at some stage it their

derivational history, it may be-possible_to_save the A-over-A

condition on case-marking from being too strong. L § 3.2 below

I will discuss briefly the possibility of there be; ng a principle.

similar to (3.6) which would delete the node NP under certain

conditions. At present there is only weak evidpte for NP

deletion, and I do not know how the principle effecting it should

be formulated, if indeed such a principle should be added to the

theory of grammar at all. But it seems to me that it may be

possible to formulate it in such a way that if the structure

underlying a pronoun is assigned the case feature Genitive],

somehow this structure is changed to meet the conditions for NP

pruning, and the NP dominating it is deleted. The A-over-A

restriction on the 'ase- marking rule could then be kept. Thus, if

the NP amid' frater 'a friend's brother' were marked Accusative],

hater would change to frame, but amI would not change to amIcum

for amid' would be dominated by NP, and the A-over-A condition

on case-marking would be in effect. On the other hand, if'meus'frater

'my brother' is marked (.1.. Accusative], the rule distributing one case

which is assigned to the whole NP to the markable elements dominated

 86

by the NP will affect both 'mews and fraer, for neither is a

NP, and the correct form, meum frarem will result. This proposal

is highly programmatic at present, for it depends crucially on an

exact formulation .of the NP pruning principle, and such a

formulation is not at present available.18

Although it does not seem possible at present to

formulate a case-marking rule which is generally adequate, it seems

to be true that in all languages which mark for case, elements in

clauses dominated by the noun phrases being marked are not markable.

I do not know whether in all case languages with a rule for reducing

relative clauses, the unmarkable elements of.the full clauses become

markable after the clauses have been reduced, as is the case in

Latin, Slavic, Germanic, and Balto-Finnic, but I suspect this to

be true too.

Notice that if the former hypothesis is correct, another

rule whose statement would require quantifiers (cf. fn. 7 above) caa

be relegated to linguistic theory. For if the hypothesis does not

hold universally, then the case-markag rules for languages where it

does hold would look roughly like this:

(3.58) [2 X - Y - Z [ flasylipi
1
OBLIG
1 2 3 4

1 [ 2 3 4
+casej

Condition: It is not the case that NP > S and S > 2.

k k
 87

Here I have assumed that an earlier rule, which assigns a case to

a whole NP on the basis of its syntactic function, has adjoined

the node (+case ] (this is a variable ranging over (+ Accusative],

[4. Dative], etc.) to the entire NT, but nothing depends on this

assumption. The'important fact to notice is that subscripts, which

are logically equivalent to quantifiers, must be used to state the

condition. This is not to say that it is necessarily true that

rules like (3.58) are not language-specific, but rather that if

Unitier-5:, ty
my hypothesis that elements of clausLs areeot markable proves to

be wrong, it will be necessary to abandon at least in part the

restriction that transformations must be stated without making

use of quantifiers over P-markers( vj. f6.4..g.1

Xn summary, whether or not it turns out to be true

that in all case-marking languages, full and reduced relative

clauses behave differentially with respect to the case-marking

transfc'lationp the fact that it is true of Lztin, Slavic, Germanic

and Balto-Finnic supports the hypothesis that a principle for

S-pruning must be in the theory of grammar, for the case-marking

facts in these languages can be most economically explained on the

basis of the differences in constituent structure which such a

principle, ould produce.

 '777""'"7"r2,"-nrrillff

88

3.1.4. The fourth example in 70ihiCh node deletion plays a

role, which has,to do.with the placement ofclitics in SerboCroatian,

was diseoveted by Wayles Browae (e f. Btovitle (1960). A5 Brovrae

points out, there exists a rule in SerboCroatian which moves to the

second position in their sentence all of the clitics (these are a

number of short words like pronouns the copula, a morpheme indicating

the coniitional, etc. -- an exhaustive listing of these words is not

necessary here.) The clitics occur in a certain order there, but

what this order is is not relevant here. For example, wince the

words le 'it' (acc.) and mi 'I' (dat.) are clitics, if no prior

rules were applied to sentence (3.59), which has approximately

the structure shown in (3.60), a rule of Clitic Placement would

convert (3.60) to the structure underlying (3.61).

V
(3.59) Ivan tell. da Ivan ata :le mi.

Ivan wanted that Ivan read it to me.

'Ivan wanted Iv= to read it to me.'

(3.60)
 89

(3.61) Ivan Zeli da je Ivan Cita.

'Ivan wanted Ivan to rad it to me.

However, when the subject NP of the embedded sentence

19 of the matrix sentence (just which NP

is identical to some NP

is not relevant for this example), a rule which I will refer to as

zaial NP Deletion optionally deletes the subject of the embedded

sentence, simultaneously deleting the complemetizer da 'that' and

1. ,V
converting the main verb (cita) into an infinitive kcitatf). But

if this occurs, as Browne points out the clitics 4e and mi must be

moved to the position immediately preceding :ieli 'wanted', for if

1gui NP Deletion has applied, the sentence which must be produced

is (3.62).

(3.62) Ivan mi je deli vcitati.

It will be observed that the position of the clitics

le and mi before the main verb of (3.62), ieli provides compelling

motivation for Spruning, for if the circled occurrence of the

node S in (3.60) is not deleted by (3.6) after the operation of

Egui NP Deletion has caused it to cease to branch, Clitic Placement

will apply vacuously to (3.60), for je and mi will already occupy

second poE lion in the most deeply embedded S. Thus unless node

deletion applies, they will not move at and.(3.62) will not

be generated.

The clitics must be moved so that they become'the

second element of the first sentence above them. (Actually, they

 90

are adjoined to the right side of the first element of this sentence,

and are phonolo2ically in the same word as this elements Thus, in

(3.62) 11.2121..je is a phonological word.) It is of theoretical

interest that, given the presently available theoretical conventions,

it is only possible to specify formally that the clitics may not

be moved out of the first sentence above them by using subscripts

on rule conditions (or, equivalently, quantifiers on P-markers), as

in (3..63) below.

(3.63) -Clitic Placement2°

X -[Y Z (-r ciitic] W

Si Si
OPT
1 2 3 4 5 6=====-4>.

1 2+4 3 0 5 6

Conditions: (1) 2 is a single node

(2) If Sj > 4, it is not the case

that Si > Sj.

It would of course be absurd to hope that such a rule

as (3.63) could be universal, so the question is, must the

restriction that conditions on transformational rules be Boolean

conditions on analyzeability be given up? And if so, must all

possible combinations of subscripts in conditions be countenanced?

I believe the correct answers to these questions to be a qualified

yes and a definite no, respectively. I will argue be/ow, in

 91

discussing the notion of 1)oundint, that a new convention must be

introduced into the theory of grammar: it must be made possible

to refer to the right and left boundaries of the first sentence

up or of the first sentence down from any term of the structural

index of a transformation. If this convention is made available,

I think that the unlimited'power of quantificational conditions

on rules need not be countenanced. However, I cannot argue these

claims-at this point in the exposition. I will return to them

in § 5.

It should be obvious, however, that whether or not

my proposed convention is or is not strong enough to obviate the

need for quantificational conditions, and whether the rule for

Clitic Placement should be stated as in (3.63), or in a new

formulation which makes use of my proposed convention, the

argument for S-pruning, which is my main concern here, remains

valid. Unless principle (3.6) applies to delete the circled S

in (3.60), after Ecui NP Deletion has deleted da and Ivan, it will

be necessary to add an ad hoc rule to derive sentence (3.62). This

fact constitutes confirming evidence of the strongest kind that

principle (3.6) must be in the theory of grammar.

3.1.5. The fifth example involving S-pruning has to do with

sentences containing'as

- Tom drives as that man drives.

b. Tom drives 'as that man does.

 !77-7r!!!!Z7111.1

92

c. Tom drives like that man.

I wish to argue that (3.64b) is derived from (3.64a)

by the deletion under identity of the verb in the as-clauseo and

furthermore, that (3.64c) is derived from (3.64b) by the deletion

under identity of the auxiliary in the as-clause. If only an NP

follows'asv it is obligatorily converted to like. There are, of

course, dialects in which (3.64a) and (3.64b) are impossible unless

like has been substituted for'as there too. Poi me, in casual

speech, (3.644 and (3.64b) are only possible with like, although

I believe the as-versions are the ones sanctioned for more formal

purposes.

Note there is a difference in relativizability between

the first two sentences and the last one. That is, relative clauses

on the noun man cannot be formed from (3.64a) or (3.64b), although

this in possible in the case of (3.64c).

(3.65) a.. * I know a man who Tom drives as drives.

b. * I know a man who Tom drives as does.

c. I know a man who Tom drives like.

I think the ungrammaticality of the first two sentences

of (3.65) can be explained on very general grounds if the structure

shown in (3.66) is postulated to be the approximate underlying

structure for sentence (3.64a) (and thus; derivatively, for the

other two sentences of (3.64) too),

 ...16,6,1K .. 6,00*

93

VP

Tom

drives

the .tat
Art N V NP

1
thatman'drives
I
P ;ATP

in Art

some Ira,

After the relative clause rule and a. rule deleting the

preposition in have applied to (3.66), sentence (3.67) results:

(3.67) Tom drives the way that that man drives.

A later rule will have to convert the way that to as

or like, depending on what follows, and if this rule can be ordered

late, the fact that'that man in (3.64a) and (3.64b) is not

relativizable can be reduced to the fact that'that'man is not

relativizable in (3.67). And this latter fact follows from a very

general condition, which was stated in approximate form in (2.26)

§ 2.4.1, and which will be. gone intoin'greater detail in § 4.1,

of

the Complex NP Constraint. It prevents the relativization of any

element contained in a relative clause. This condition is met even

 94.

if the verb drive in the relative clause of (3.67) is deleted, under

identity with the verb in the main clause, .yielding (3.68), a

structure which may later be converted into (3.64b).

(3.68) Toms drives the way that that man does.

But if the deletion proceeds further, and even the

word does of (3.6P' is erased, then the circled node S in (3.66)

will cease to branch and will be deleted by principle (3.6). With

this deletion, the condition ceases to be met, and the NPthat man

becomes relativizable.

Although the details of this explanation of the differences

among the sentences of (3.65) will not become clear until the condition

I have made use of is given final formulation in 4.1, I think that

enough has been said here to prove the point at hand -- that the

explanation depends in a crucial way upon the notion of node deletion.

Assuming that I am correct in supposing all the sentences in (3.64)

should be derived from the same underlying structure, the fact that

(3.64c) behaves differently than (3.64a) and (3.64b) with respect

to the relative clause transformation suggests that the former

sentence differs from the latter two in constituent structure.

Principle (3.6), if adopted, would provide such a difference, so

(3.6) is supported by the facts of (3.65).

3.1.6. The final three sets of facts which support (3.6) come

from areas of grammar which I understand so poorly that I will not

 95

even speculate as to what the full analyses in each case are, but

merely suggest that when full analyses are.available, they will make

use of an S-pruning principle like (3.6).

The first of these sets of facts has to do with

comparatives, and bears a strong resemblance to the case discussed

immediately above, in § '3.1.5. Although boil of the sentences

in (3.69) are grammatical, as the sentences in (3.70) show, the

NP that man is only relativizable in (3.69b), which has been

derived from (3.69a) by deleting'is.

(3.69) a. John is taller than that man is.

b. is taller than that man.

(3.70) a. * I-know-a-man-who John is taller than is.

b. I know a man who John is taller than.

Facts parallel to these in all respects can also bs

shown to hold for the comparison of equality.

(3.71) a, John is as tall as that man is.

b. John is as tall as that man.

. (3.72) a. * I know a man who John is as tall as is.

b. I know a man who John is as tall as.

Although more efforts have been expended on the comparative

than on any other construction, and although there exist a wide variety

of proposed analyses to choose from (cf., e.g. Smith (1961), Lees (1961),

Hale (1965), Hale (to appear), Lakoff (1965),-Ross (1965) and

Qualls (to appear)), it seems to me that no satisfactory deep structure

 96

has been arrived at, although the range and complexity of examples

that have been taken into consideration is extremely wide. I cannot,

therefore, explain in detail why it is that (3.70a) and (3.72a)

are ungrammatical, while (3.70b) and (3.72b) are not, but it does

seem likely that the eventual explanation of this fact will hinge

on the fact that the node .S which dominates the phrase that man 'is,

in (3.69a) and (3.71a) will have been deleted by (3.6) when the

word is is deleted by the transformation which converts (3.69a) and

(3.71a) to (3.69b) and (3.71b) respectively.

3.1.7. The second set of facts which seems to depend on

S-pruning also has to do with comparatives and with the way they

interact with the rule which permutes an adjective from a reduced

relative clause to prenominal position (this rule was discussed and

given a preliminary formulation in § 2.3 above), Assuming that

the adjectives in (3.73) - (3.75) are all derived from the same

underlying structure, which is ems moot point,

(3.73) a. Mary has never kissed a man who is

taller than John is.

b. Mary'has never kissed a man who is

taller than John.

(3.74) a. Mary has never kissed a man taller

than John is.

b. Mary has never kissed a Man taller

than John.

(3.75) a. * Mary has never kissed a man taller than John is.
 97

b. Mary has never kissed a man taller than John.

the ungrammaticality of (3.75a) is presumably to be explained by

constraining the rule which accomplishes the shift of the adjective

to prpnominal positi-.1 go that enmparpd adiativag pay only undergo

this rule if the than-clause does not contain a sentence. Principle

(3.6) asserts that this is not the case for (3.74b), although it

is the case for (3.74a), and thus provides a basis for explaining

the difference in grammaticality of (3.75a) and (3.75b).

I believe the facts of the comparison of equality to

parallel these facts (cf. the sentences in (3.76)),

(3.76) a. 7* Mary has never kissed as tall a man

as John is.

b. Mary has never kissed as tall a man

as John.

but for some obscure reason, (3.76a) does not seem to me to be as

clearly ungrammatical as (3.75a).

These constructions raise many interesting problems

which cannot be gone into here, and so little is known about them

that it may turn out that the explanation which I have proposed

for the differences between (3.75a) and (3.751%) and between (3.76a)

and (3.76b) is incorrect; but at the present state of knowledge,

Wiese differences seem to be connected with S-pruning in some way, and

thus to provide weak support for principle (3.6).

3.1.8. The last case which seems to require S-pruning has to do

with contrastive stress in Hungarian, Kiefer has noted (cf. Kiefer

(1966)) that there exist adverbs in Hungarian which cannot be

contrastively stressed. At present, this fact is totally isolated,

unexplained, and, as a matter of fact, not statable within the

present theory of grammar.- Not7eiloinh-is-now known about these

adverbs for it to be possible to predict how the theory will have

to be changed to accommodate this fact, but there is one indication

that S-pruning will figure into the solution. IA J Jr

La Asia 4.1(

Kiefer notes that the adverb-dllandodn is

one of those which cannot bear contrastive stress*in normal circumstances.

That is, in the Hungarian equivalent of a sentence such as (3.77),

plandofin could not be contrastively stressed.

Valoiki dllaadolm drveket hozott fel.

(3.77)

Somebody constantly arguments brought up.

'Somebody constantly brought up arguments.'

Lut it is also a fact that if an NP in Hungarian is

contrastively stressed, the first lexical element of that NP is the

phonological carrier of the contrastive stress for the entire NP. And

if the structure underlying (3.77) is embedded as a relative clause

on the noun*ervet 'argument', reduced, and shifted to prenomnal

position, as in (3.78);*dilandodn can become the first lexical

element of an NP and, if that NP is contrastively stressed,

dllandogn will bear that stress,

 99

(3.78) Az dilandoan felhozott drvek rosszik

The constantly up brought arguments wrong were.

"'he COft5Lalitay brought up arguments ware wrongs

It seems reasonable to me that whatever the precise

constituent structure reconstruction of the phrase "in normal circumstances",

which funderlined above, may turn out to be, it will depend to some

extent on whether the tdVerb to be stressed is immediately dominated

by the,node S or not, or possibly it will depend on the number

of nodes intervening between the adverb in question and the "first

sentence up." If either of these conjectures proves correct, then

it will probably prove useful to invoke some principle of S-pruning

like (3.6), so that-the reduced relative clause gllandogn felhozott

'repeatedly brought up' will no longer .be dominated by the node S in

(3.78). But here again, as in the case of the examples discussed

in Si 3.1.6. and 3.1.7, there are so many unsolved problems that

it is impossible to be certain that S-pruning is involved.

3.1.9. To summarize briefly, in §i 3.1.1. - 3.1.8, I have

discussed eight cases which all support, some more strongly than

others, the hypothesis advanced in § 3.0 -- that principle (3.6)

should be added to the theory of grammar. There is an additional

class of cases having to cl4 with conjunction)which space limitations

forbid me to go into here, but which will be discussed at length

in Lakoff and Ross (in preparation b). The analysis of Cont. on

...,Tyrlrrirry-11ffr,f17111111.07471,4116
 V--

100

22
Reduction which we propose there depends crucially on pruning

rules, in particular on a rule fir pruning non- branching S, which

thus constitutes further evidence for (3.6) . Therefore, I feel

that it is safe to conclude that pruning rules must appear in

the theory of grammar, at least for the node S. The fragmentary

evidence which suggests that rules which prune NP and VP may

be necessary is discussed imediately below in § 3,2.

3.2. At present I know of no reasons other than intuitive

ones for arguing that the node NP must be deleted; and the only

argument except for intuition for deleting VP which I know of is

connected, in a minor way, with the analysis of theContjaictim

Reduction Rule which will be presented in Lakoff and Ross (op. cit.),

but which cannot be gone into here. Yuki Xuroda first suggested
23
the possibility that other constituears than t might be deleted.

His idea was that if the head of a phrase (the Ilead of NP is N,

of VP, V) is deleted, the phrase should be deleted with it. This

idea seems to be a promising approach to the problem of establishing

some constituent structure difference between mews and'amici (cf. § 3.1.3

above), so that the case of the first can be changed, but not that of

the second, but there are problems with it aside from those mentioned

in fn. 18. Thus, presumably phrases like the brave the dead, thelust,

keep their status as an NP) even though the underlying head noun,
 101

ones, has been deleted. I have no argument for this other than

intuition, but it does seem strongly counter -intuitive to claim,

as Kuroda's principle would seem to force us to, that the phrase

the brave in (3.79) is not dominated by NP.

(3.79) The brave are not afraid to die.

The intuition that the brave is a constituent of some kind in (3.79)

is strong, and if it is not an NP, what is it? In 1-esearch on

conjunction conducted by Lakoff and me, it has seemed to us that

a necessary, though not a sufficient, ^onAitiva for node deletion

is that the node not branch. So if Kuroda's principle is supplemented

by the general condition that only non branching nodes delete, the

difficulty connected with (3.79) can be avoided,.

But there still remain problems Which Kuroda's principle

is not strong enough to handle adequately. Thus, in footnote 2

above, it was pointed out that it may eeem counter-intuitive to call

the word yellow in the NP his'yellow cat a VP. But if my

proposed analysis of predicate adjectives is correct (cf. (3325) above),

then yellow will be the head of a VP in the deep structure, so by what

rule can this VP be pruned?

In short, while there is strong evidence that a principle

of S-pruning is needed in the theory of grammar, and even evidence

that supports the formulation of this principle which was given in

(3.6), the evidence that NP and VP must be deleted is weak, and

no adequate formulation has been found of principles by which their

deletion might be effected.

 102

Chapter 3

FOOTNOTES

1. I would like to acknowledge here my indebtedness to several

of my friends and colleagues, whose ideas and counter-

examples have greatly infliiide-ed-the-formulation of the

principles in this chapter Paul Postal, in a lecture for

a course he conducted in the spring of 1965, first brought

to my attention the counter-intuitiveness of much of the

derived constituent structure (1.c.s.) which was assigned

by the then current theOry. This counter-intuitiveness,

which. is discussed in § 3.0, provided the original impetus

for constructing a systematic theory of node deletion. To

Yuki Kuroda I owe the important idea that node deletion

might not be restricted to the node .S, as I had originally

proposed, but should rather be generalized to affect all

branching nodes. His proposal will be discussed briefly

in § 3.2 below, in connection with the problem of deletion

of the node NP. I have profited from my discussions with

Susumu Kuno about the problems of case-marking, and especially

from many long conversations with George Lakoff about the

consequences for :Jaciple) of node deletion of an analysis

of conjunction which will be presented in Lakoff and Ross

(in preparation b).

 103

2. It may also seem counter-intuitive to label the word

a VP, although this intuition is not so clearcut, to me,

at least.

3. For some discussion of this analysis of imperatives, cf.

Katz and Postal (1964). An important critique of this

analysis, containing a large class of constructions that have

hitherto not been taken into account is given in (Bolinger

CL94

4. For a detailed discussion of many problems in verb-particle

constructions and references to earlier work on particles,

cf. Fraser (1965).

5. For some discussion of this rule, cf. Smith (1961).

6. Postal and Lakoff have pointed out that words which traditionally

categorized as verbs and ad1ectives are better considered to

be subcategories of the same lexical category, Predicate which,

follo /ing Lakoff (cf. Lakoff (1965)), I will designate with the

feature ( +V]. What were traditionally called adjectives are

designated with the feature bundle (Add], and what were

traditionally called 'verbs are designated by [lib

 104

7. It should be emphasized that the use of a subscript on 0/

-Adj 1

in Condition 2 conceals a hornet's nest of problems. Ir the

first place, there is only one other rule which I know of

which can only be stated by using subscripts: the rule which

scrambles major constituents in a clause in so-called "free

word -order languages" like Latin, Serbo-Croation, Russian, etc.

This rule will be discussed in i 3.1.2. Secondly, it is

evident that the subscripts in the condition on (3.26) are

used in a way which is logically equivalent to using quantifiers.

That is, Condition 2 has the following logical structure:

(for all ( ±ldjii) 11-v + X2) if and only if

(Y X1 + 6 Ar i alr

(there is an NP j) NPi dominates L-Adj

manin

Aside from these two rules, it has previously been thought possible

to restrict conditions on transformational rules to Boolean

conditions on analyzability (cf.:Chomsky (1965), p. 144).

George Lakoff and I will argue in our forthcoming monograph

(Lakoff and Ross (op. cit.), that it must also be possible to

state conditions in terms of immediate domination, a notion

which can only be defined log.;.c.u.ly with quantifiers, if the

only primitive notion in the theory is domination (cf. 6 2, fn. 6

above). That is, to say that A immediately dominates B is

to say that there euists no node Z. such that A dominates Z

and Z dominates B. However, I would be opposed to the

 letoluNIMPliec,..* Ali, e

105

suggestion that the restriction to Boolean conditions on

zItalyzability be dropped entirely, for to drop it would be

to greatly increase the set of possible rules and thereby

to weaken the theory. It may be possible to restrict

quantifiers to conditions on very late transformational

rules, which is much to be preferred to all^wing such

restrictions on any rule whatsoever. It seems likely that

bath (3.26) and theanyjakaliacan come very late in the

ordering, but too little is known about.this at present.

8. I here make use of the distinction between grammaticality

and acceptability discussed by Chomsky (1965), § 1.2. By

"nternal", I mean "embedded", in the technical sense defined

in Chomsky (1961) -- that is, an NP is internal to a sentence

if it.is both preceded and followed by non-null parts of that

sentence. I have used the word "internal" here because it

seems to me that in recent work, the word "embedded" has

been used in a sense different from Chomsky's original one --

a sense which must be excluded for the purposes of (3.27), O

For example, it is often said that C\e sentence Bill was sick

is shmbedded",,in the sentence kreryone t1,.21.athlLIzt Bill was

sick, even though it*is not internal to it (in my sense).
 106

9. Sentences like the following, which (3.27) would predict to

be unacceptable,"but which are in fact far more acceptable

than (3.28a) (3.33a),

Bill said (that)for her to enlist would ue impossible.

Jack thinks (that) what he's eating is scrambled eggs.

constitute counterevidence to (3.27). At present, I do not

see how to modify it so that these sentences will not be

produced with as low an acceptability index as is assigned

to (3.28a) - (3.33a).

10. The'Dative'Rule relates sentences like Lsata4=12Docac

andimye a book to Mau.. It is thoroughly discussed in

Fillmore (1965).

Emmon Bach has recently pointed out (cf. his note "Problom-

inalizatioe University of Texas mimeograph, 1967) that ..ertain

facts about the Dative Rule and'Pronominalization in German

lead to an ordering paradox. The same holds true of English,

which I will discuss here.

It has been usual to make the Dative Rule obligatory if the

direct object is a pronoun, thus excluding (3.37d) and (3.39a).

(Here I have assumed that sentences Mar

are basic and that sentences with'to are derived from them, but

nothing depends on this assumption.) This presupposes the

ordering below:

PPWITT'AVIIVIROMI, grarfolfprwanrIt
 107

'Pronominalization

Dative

But there are sentences whidl suggest that the reverse

ordering is necessary:

I gave Mollyi her book.

* I gave heri M011y'si book.

I gave Molly'si book to heri.

* I gave heri book to Mollyi.

It will be seen that the pronoun always follows the noun it

refers to in these sentences. This means that the ordering

or the rules must be,

"Dative

Pronominalization

for if the reverse order obtained, the first of the four,

sentences could be converted into the fourth. But if Dative

is optional and precedes Pronominalization, how can the

following derivation be prevented?

BASE: I gave the girl who. - .wanted _the book the booki

(: Da_ tive optionally

does not apply

Pronominalization

applies
(
* I gave the girl who wanted the booki its.
 108

The only solution I can find within the current theory it.to

postulate a second'Dative Rule which applies only when the

direct object has become a pronoun. Obvinusly howevel., the

current theory is wrong and must be modified. The modification

I propose is taken up immediately below.

11. Fraser (op. cit.) made the interesting discovery that a

subclass of what had previously been thought to be verb-

particle combinations, verbs like le_ t out,'take in, loa_ d on

...-elbaw off, etc., should really not be treated as verb-particles

at all. Rather, verbs like these should be considered to be

derived from verb phrases like_1et'Utl.outlaLsom2211111,

sal....L.1;toscm,...11(eitixiethin, 'load (it) on (to'somethinal

elbowitt....g.iin), etc., where the prepositional

phrase in parentheses is deleted bythe rule which converts

John smokes something, to John smokes, ancrl an-rove

of

somethi tol.uprove, a rule which seems to be required in

been studied
P. wide variety of cases, but which has never

intensively. Fraser points out several facts about these

verbs which show clearly their diffarenceT from ordinary

verb-particle combinations:

1) The prepositions of these verbs will conjoin

'(she took boxes in and'out), particles will not

( *1 'showed her u and off)

 43

Similarly, it seems counter- intuitive to claim, with

the present theory, that the correct structure to assign to a NP

like his yellow cat, is one roughly like the one shown in (3.5).

(3.5) NP

Det

Art Postart

()
NP VP

Adj

N Poss yellow

he

Once again, recent research in syntax has called into

question many facets of the analysis implicit in (3.5) (cf. Postal

(1965a) and Lakoff and Ross (in preparation b)), but at present

I am only interested in the fact that it seems incorrect to claim

2
that the words his and ze12.2z are sentences. In the present

theory, an NP like the one diagrammed in (3.5) would, correctly

I think, be derived from an underlying NP with two relative

clauses: the cat which I'have which'isx22122. The motivation

for deriving possessi...cs and prenominal adjectives from relative

clauses is well-known enough not to need recapitulation here

 109

2) These verbs do occur in action nominalizations,

while verb-particles do not *(his brinit/aol.

the trays into but t 00±).

1%
41 OVWG 111 pasi.aormr.A.a.aw. J.LIA,A.0 Livua

out of the window,may always occur with these

verbs (hei.et her out into the garden, they were

"loading them on from the warehouse, he elbowed

-.1.t off into the well they took it in up the

-stairway`, but there are verb-particle constructions

which exclude them (*I burned it up from Boston)

*I'showed her` asp out of the 'window, *Shei?a

"whiled the morning away into the well).

4) If a verb stem occurs with one of these prepositions

from reduced directional phrases, it will occur

with many more. Thus,.since throw out is one of

these verbs, it is to be expected that other

directional prepositions will also occur with

'throw (e.g., over, under, down,'220'offiacross,

on, in,lway, around). The same is true of verbs

like 112210., take, send, Shoot,liand, etc., but no

such prediction is possible with true verb particles.

Thus, although figure out exists, there is no

:fijure'off.i'Dgure*in,0 etc.
 110

Alter the unspecified NP and second preposition nave been

deleted from a VP like'let'the cat autiaLpomethingL the

remaining prepositionlout, is optionally moved to the left,

around the object NP, and adjoined to the verb.

12. Sentences like (3.42); (3.44), and (3.45) point up a very

interesting fact: there are well-formed deep structures which

no sequence of rules can convert into fully acceptable surface

structures. Trivial examples of this kind have been known for

same time.- one such example is any well-formed deep structure

which would result in a surface structure so long that it

could not be scanned in one lifetime -- but to the best of my

knowledge, it has not been noted previouSly that short sentences

which have this property also exist. Such sentences provide

evidence of the strongest-kind_for_output conditions like (3.41)

for without such conditions, a grammar would have to claim that

one of the versions of (3.42), (3.44) and (3.45) is fully

acceptable, a claim which is simply,:not true.

13. Sentence (3.43a) is acceptable, of course, if the main verb

see through is taken to mean (approximately) "not be fooled by",

but not if it means "continue to support until some specified

end point."
IA. The most detailed treatment of this problem which I know of it

given ..ta a paper by Zeno Vendler, "The order of Adjectives,"

Transformations and niscourse Analysis Project, paper number 31,

1
University of Pennsylvania mimeograph.

Mark Liberman has recently pointed out that the word one is

ambiguous in the sentence James bou ht a wonderful old brick

-house and/1 bought a wooden one. One can mean simply 'house,

but it can also meauvouderful old house. Since it is desirable

to restrict pronominalization to constituents, this suggests

that the input structure of.the above sentence, when one has the

latter meaning, must be the one underlying the unacceptable

string *James bou ht a brick wonderful old'house and I bou ht

a wooden wonderful old house. The rule which inserts the

pronoun one matches the double - underlined phrases and optionally

replaces the right-hand phrase with one. Ifone is not inserted,

some rule which scrambles prenominal adjectives optionally

applies to the adjectives in both of the conjoined sentences,

and some output condition will then evaluate the acceptability

of the output string, Liberman's observation seems to me to

provide extremely strong evidence for modifying the theory

of grammar so that it contains some kind of stylistic component,

for I can see no way of accounting for it within the present

theory.
 112

15. As a case in point, consider preverbal pronouns in Prench.

I1 ,x' a des autres is grammatical, whereas**1.1

'des autres is totally ungrammatical:

16. On the theoretical implications of using subscripts in

conditions on rules,s.cf. fn. 7 above.

17. In diagram (3.50), I have, for expository purposes only,

-Aot given what T. believe is the correct labeled bracketing.

In Latin,as in Engliah,there is reason to think that the

underlying structure of sentences containing predicate

adjectives is roughly that shows in (3.25),

18. Unfortunately, there are facts in Latin and Russian which will

remain unaccounted for, even if some principle for NP pruning

can be worked out. For in these two languages, third person

pronouns in the genitive case do not become "possessive

adjectives"- (i.e., their case is not changed by the case-

marking rule). Thus, while mews frater 'my brother' becomes

meum fratrem in the accusative case, eius hater 'his brother'

becomes eius fratrem, not the parallel *eum fratrem. But in

German, third person genitive pronouns*do inflect like adjectives,

so it is clear that while many features of the case-marking rule

may be universal, these interact with language-particular features

in a way that is at present inexplicable.

 113

19. It has been realized for a fairly long time that the notion

of identity which is required in the, theory of grammar must

include identity of reference (hints of this are present in

Chomsky (1962), p. 238, and a specific proposal for formally

indicating coreferentiality is made in Chomsky (1965) p. 145-

147). In addition,.as Lees-pointed-out (cf. Lees (1960),

p. 75), identity of strings of words is not sufficient; rather

the requisite notion must be defined as identity of constituent

structure. The example Lees uses to .point out this interesting

fact is the following. Since both sentences a and b below occur,

a. Drowning cats are hard to rescue.

b. Drowning cats is against the law.

if string identity were sufficient to correctly predict what

non-restrictive relative clauses can be formed, it should be

possible to embed sentence b into sentence a, for both share

the string drowning cats. But the ungrammaticality of c shows

that the stronger type of identity which was proposed by Lees

must be adopted.

c. *Drowning cats, which is against the law, are hard

to rescue.

In fact, there are examples which show that an even stronger

notion of identity is necessary: a constituent which is to be

pronominalize:, by virtue of its identity to some other constituent.

 114

must be identical in deep structure to that constituent. Examples

which illustrate this point involve syntactically ambiguous

sentences which are derived from different deep structures but

have the same d.c.s. Several such sentences are given below.

d. I know a taller man than John.

e. When did Bill-promise-to-call me?

f. The shooting of the pTisoners shocked me.

In d, one reading derives from a deepstructure containing the

deep structure of John knows a tall. man, the other from one

containiag the deep structure of John is tall. In e, when can

-s
modify promise or call, and in f, prisoners can have been

derived from an underlying subject (the prisoners shot something)

or from an underlying object (someone shot the prisoners). If

any of the sentences in d, e, or f is pronominalized as in 20 h,

or i,

g. He told Peter that I know a taller man than John, but

Peter didn't believe it.

h. I divulged when Bill promised to call me, but I did

so reluctantly.

i. I'll talk to John on Friday about the report that

the shooting of the prisoners shocked me, and to

.his wife on Saturday.

it is clear that reference has.been made to the deep structures

of d, e, and f, for the sentences in g, h, and i are only ambiguous

 1_15

in two ways, not four.

The problems that deep structure ideAtity raise for linguistic

theory are extremely complex. They will be taken up in detail

in Lakoff and Ross (op. cit.). (1(.044 g 6', a, 3.'1.640.

20. At present, rule (3.63) is not stated correctly, for according

to the specification of elementaries given in the structural

change there, the clitics are adjoined to the first element of

the first sentence above- them_as_ sisters. Thus they will not,

without some special provision for the introduction of word

boundaries, be part of the first word of the sentence. What

seems to be necessary is that the clitics be adjoined to the.

first element of the sentence by a new type of adjunction:

daughter adjunction. What must happen is that the leftmost

branch of (3.60), which I have reproduced here and labeled

a, must be converted into either b or c depending on how the

word boundary rules are formulated.

a. b. c. S

NP NP

Ivan Ivan

This rule is the only one I know of where daughter adjunction

is required, and I am reluctant to argue, on the basis of this

rule alone, for a change in the number of kinds of elementary

 116

operations which the theory of grammar provides. At present I

can see no other course to follow, but I will postpone proposing

such a radical change in the theory until more is known about

Clitic Placement or until other rules are found whose statement

requires daughter adjunction.

21. The reasons for arguing that manner adverbs are not constituents

of VP, as was proposed in Chomsky (1965), but rather of S,

are presented in Lakoff and Ross (1966).

22. This is the rule which reduces such sentences as John knows

the answer and Bill knows the answer to John and Bill know the
answer, and Otto sells Buicks and Otto sells Fords to Otto

sells Buicks and Fords, etc. (C( 2,111 I) 3-4") 6.1.2.3).

23. In an unpublished, untitled paper written in the fall of 1965.

''''' 1444u
 117

Chapter 4

CONSTRAINTS ON REORDERING TRANSFORMATIONS

4.0. In this chapter and the next one, I will propose a set

of constraints, some universal, some lairm6ge-particular, which I

will show to have roughly the same effect as the te-aver-A, principle.

That is, I will show that with these constraints, it is possible to

account for the six constructions in § 2.2 which Constitute evidence

for the principle, while avoiding the counter-examples of § 2.1.

The A-over-A, principle was postulated to be a constraint on trans-

formaticual operations of all kinds, but I will attempt to show, in

Chapter 6, that the_constraints of Chapters 4 and 5 (and hence, the

principle as well) should only apply to transformations which exhibit

certain well-defined formal properties. The constraints of Chapter 4

only affect what I will refer to informally as reordering transformations --

transformations which have the effect of moving one or more terms of the

structural description around some other terms of it. (The precise

definition of this notion will, not be given until Chapter 6.) Two

examples of reordering transformations are the Question Rule and the

Relative Clause Formation Rule, which are stated very schematically

in (4.1) and (4.2) below.

_#PRITAWRerAttelliMPIRmIRFAMwwvirmrsownersolegmen~...
. -
WHIRMORIMONITIPPRIMIMPIWIRIPWpmspiawwwww
.
,
 118

(4.1). Question
Q- X - NP Y .

OBLIG
1 2 3 4

1 1+2 n 4

Condition: 3 dominates WH + some

(4.2) Relative Clause Formation

W NP (s X - NP 11 -
? JNP
OBLI
1 2 3 4 5 6

1 2 4+3 0 5 6

Condition: 2 a 4

I will use ungrammatical questions and relative clauses

to illustrate the effects that the constraints of this chapter have on

all reordering transformations. In Chapter 6)1 will present a list

of all the other reordering transformations I know of, and show that

they obey the same constraints.

4.1. The Com-,lex NP Constraint

4.1.1. It is to Edward S. Klima that the essential insight

underlying my formulation of this constraint is due. Noticing that

the NP that man could be questioned in (4.3b) , but not in (4.3a)

(cf. (4.4)), Klima proposed the constraint stated in (4.5):

 119

(4.3) a. I read a statement which was about that man.

b. I read a statement about that man.

(4.4) a. * The man who I read a statement which was

about is sick.

b. The man who I read a statement about is

sick.

(4:5) Elements dominated by a sentence which is

dominated by a noun phrase cannot be

questioned or relativized.

If Klima's constraint is used in coninaction with the

principle for S-deletion stated in (3.6), it can explain the difference

in grammaticality between (4.4a) and (4.4b), for it is only in (4.3a)

that the NP that man is contained in a sentence which is itsel:

contained in an NP: when (4.3e) is converted into (4.4b) by the

Relative Clause Reduction Rule, the node S which dominates the clause

which was about that man in (4.3a) is pruned by (3.6) .

Although I do not believe it is possible to maintain

(4.5), for reasons I will present immediately below, it will be

seen that my final formulation of the Complex NP Constraint makes

crucial use of the central idea in Klima's formulation: the idea

that node deletion affects the potential of constituents to undergo

reordering transformations. This hypothesis may seem obvious, at the

present stage*of eevelopmeLt of the theory of grammar, but when Klima

first suggested it, when the theory of tree-pruning was much less
 120

well-developed than it is at present, it was far from being obvious.

In fact, this idea is really the cornerstone of my research on variables.

4.1.2. As I intimated above, however, I find that (4.5) must

be rejected, in its present form. For consider the NP that man,

in (4.6) : as (4.7) shows,, it is relativizable,

(4.6) I read (sthat the police were going to

Li P

interrogate that manjs .

1
NP

(4.7) the man who I read that the police were going

to interrogate

and yet the that-clause which contains it would seem to be a noun

phrase, as I have indicated in the bracketing of (4.6). Presumably,

the approximate deep structure of (4.6) is that shown in (4.8),

(4.8)

V NP

%lad

it

thec.1_.cewereoitointerrogate that man

and unless some way.is found of pruning the circled node S or the
 121

boxed node NP in (4.8), condition (4.5) will prevent the relativization

of that man. There is abundant evidence that the first alternative

is not feasible:

(4.9) ,a. I read that Bill had seen me.

b. * I read that Bill had seen myself.

(4.10) a. Evidence that he was drunk will be presented.

b. Evidence will be presented that he was drunk.

1
(4.11).a.ThatUll.;,ms unpopular distressed himi.

b. That he was unpopular distressed Bill..

i

The Reflexivization Rule does not "go down into" sentences

(cf. Lees and Klima (1963), Postal (1966b)); thus the fact that (4.9a)

is grammatical, while (4.9b) is not, is evie:nce that that-clauses are

dominated by S at the time that reflexivization takes place.

Similarly, the fact that that-clauses may be extraposed, as is the case

in (4.10b) , indicates that they are dominated by the node S at the

time that this rule applies. Finally, the fact that backward
2
pronominalization into that-clauses is possible (cf. (4.11a)) also

argues that they must be dominated by the node S. So it seems

implausible that the circled node S should be deleted by some principle

which supplements (3.6); and there is no independent support for such

an additional pruning principle in any case. Therefore, the only other

way to save (4.5) is to claim that the boxed node NP must be deleted

in the process of converting (4.8) into the surface structure which

underlies (4.6).
 122

Can the node NP be deleted? In § 3.2 above, I

discussed briefly Kuroda's proposal to generalize the notion of tree-

pruning in such a way that arty non-branching node whose head had

been deleted world be pruned. Jihile it is possible to propose such

a generalized version of (3.6), there is as yet no syntactic evidence

which indicates that node deletion must prune out occurrences of

NP or VP. The complex problems involving case - marking with respect

-
to amici and eius on the one hand and meus on tLe other, which I

discussed in 5 3.1.3 above, might be solvable if use were made of

some principle, of NP deletion, but this has yet to be worked out

in detail; and unless some other evidence can be found for NP

pruning, invoking it to delete the boxed NP in (4.8) is merely ad hoc.

For there are many pieces of evidence which show that that-clauses are

dominated by NP at some point in their derivation.

(4.12) a. That the defendant had been rude was stoutly

denied by his lawyer.

b. What I said was that she was lying.

c. Bill told me something awful: that ice won't

sink.

d. Muriel said nothing else than that she had

been insulted.

That-clauses passivize (4.12a) , occur

in pseudo-cleft sentences (4.12b), after the colon in equative sentences

(4.12c), and after than in sentences like (4.12d): in all of these

 123

contexts, phrases can occur which are unquestionably noun phrases

(e.g., Little Willy, potatoes, flyingplane's, etc.), and Lakoff and

. Ajt
I argue that the syntactic environments defined by (4.12) can on 11.'"P

be filled with noun phrases (cf. Lakoff and Ross (in preparation a)).

If our arguments are correct, then that-clauses must be dominated

by NP at some stage of their derivation. But it might be claimed

3
that the late rule of It Deletion , which deletes the abstract pronoun

it when it immediately precedes a sentence, could change phrase-

markers in such a way that the NP node which dominated it S would

undergo pruning before Question and Relative Clause Formation had

applied. Not enough is known about rule ordering at present for this

possibility to be excluded, but it should be noted that even if it

should prove to be possible to order It Deletion before all reordering

transformations, thereby accounting for the grammaticality of (4.7)

by providing for the deletion of the boxed NP of (4.8), it would

still be necessary to explain wh-7 there is no difference in grammaticality

between (4.13a) and (4.13b),

(4.13) a. This is a hat which I'm going to see to it

lee
ti/frit that my wife buys.

b. This is a hat which I'm going to see that

my wife buys.

After the verb se_; e... ...koL, the deletion of it is optional

(in my dialect), and therefore, by the previous argument, while the

 124

that-clause in (4.13b) might not be dominated by NP, the that-clause

in (4.13a) still would be. So unless some additional convention for

NP pruning could be devised for this case too, (4.5) would not allow

the generation of (4.13a). Again, I must reiterate that there is no

known evidence for pruning NP under any other circumstances, so the

ad hoc character of the explanation which is necessitated if (4.5).

is adopted is readily apparent.

But there is an even more compelling reason to reject

(4.5) than the ones above: as I pointed out in § 2.4.1 above, it

is in general the case that elements of reduced relative clauses

and elements of full relative clauses behave exactly the same with

respect to reordering transformations. This can be seen from the

following examples: NP which are in the same position as Maxime

ia the sentences of (4.14) cannot be questioned (cf. the

ungrammaticality of (4.15)),

(4.14) a. Phineas knows a girl who is jealous of

Maxim.

b. Phineas knows a girl who is behind Maxime.

c. Phi:leas knows a girl who is working with

Maxime.

(4.15) a. * Who does Phineas know a girl who is jealous of? .

b. * Who does Phineas know a girl who is behind?

c. * Who does Phineas k--,ow a girl who is working

with?
 125

nor can they be questioned,even after the relative clauses of (4.14)

have been reduced (this is evidenced by the ungrammaticality of

(4.16)).

(4.16) a. * Who does Phineas know a girl jealous of?

* Who does Phineas know a girl behind?

c. * Who does Phineas know a girl working with?

It was facts like these which motivated the condition

stated in (2.26) above, which I repeat for convenience here.

(2.26) No element of a constituent of an NP which modifies

the head noun may be questioned or relativized.

In the light of the facts of (4.15) , and (4.16), it

would appear that it is the grammaticality of (4.4b) which is

problematic, not the ungrammaticality of the sentences in (4.16).

And there are parallel facts which have to do with Reflexivization,

which I will present in § 4.1.6 below, which also support this

interpretation. So condition (4.5), which takes the differences

between the sentences in (4.4) to be typical, would seem to to 126 0-

projection to an incorrect general conclusion from a case where

special circumstances obtain. In the next section, I will give some

evidence which allows the formulation of a broader-based generalization.

 126

4.1.3. The sentences of (4,17), which only differ in that the

NP object of believe has a lexical head noun in the first, but not

*in ?i,. caonnA Aiffor nn rn rolnrivixnhilityi nn the cnrrenpondine

sentences of (4.18) show.

(4.17) a. I believed the claim that Otto was wearing

this hat.

b. I believed that Otto was wearing this hat.

(4.18) a. * The hat which I believed the claim that

Otto was wearing is red.

b. The hat which I believed that Otto was

wearing is red.

If the analysis proposed by Lakoff and me (op. cit.) is

correct, the d.c.s. of (4.17a) will be roughly that shown in (4.19):

(4.19)

V NP
. I

believed NP .

the claim th51"EVP

Otto

was wearing_ this hat

` -=', - -
 127

Whether or not we can show it to be correct that abstract

nouns followed by sentential clauses in apposition to them have

exactly the same -(NP-SL,

/IL
structure that we argue relative clauses 1
have, it is clear that these constructions are highly similar.

Condition (4.20), the'Complex NP Constraint, is formulated in an effort

to exploit this similarity to explain the ungrammaticality of sentences

like (4.18a) and (4.15) on the same basis.

(4.20) The Complex NP Constraint

No element contained in a sentence dominated by

a noun phrase with a lexical head noun may be

moved out of that noun phrase by a transformation.

To put it diagrammatically, (4.20) prevents any constituent

A from being reordered out of the S in constituents like the NP

shown in (4.21),

(4.21) NP

Li-N
+Lex]
 128

as the X's on the two arrows pointing left or right from A designate.

(Vote that (4.20) does not prohibit elements from reordering within

the .a..minatea emntannct and in

ram fact 1-1 c,1rr. are manly
0 riles which effect
such reorderings. Some will be discussed in § 5.1 below.)

I have assumed the existence of a feature, [4- Lex] , to

distinguish between lexical items like claim in (4.17a) or _girl in

(4.14) on the one hand, and the abstract pronoun it of (4.13a) on the

other. Since it is possible to move elements out of sentences in

construction with the third of these, as (4.13a) attests, but not

out of sentences in construction with the first two ((4.18a) and (4.15)

ere ungrammatical), it will be necessary for the theory of grammar

to keep them distinct somehow. The featurej+ Lexical] may not turn

out to be the correct one; I have chosen it not only on the basis of

the facts just cited but also with regard to the following parallel

case in Japanese.

4.1.4. In Japanese, and I believe in all other languages as

well, no elements of a relative clause may be relativized. Japanese

relative clauses invariably precede the noun they modify. Superficially,

they appear to be formed by simply deleting the occurrence of the

identical NP in the matrix sentence. Thus when the sentence (4.22)

is embedded as a modifier onto the NP sono sakana wa 'this fish',

which is the subject of (4.23), (4.24) results.

 129.

(4.22) kodomo ga4 sakana o tabete iru,

child fish eating is

'The child is eating the fish,'

(4.23) Sono sakana wa ookii.

That fish big

'What fish is big:

(4.24) Sono kodomo ga tabete iru sakana wa ookii.

That child eating is fish big

'That fish which the child is eating is big.'

The deep structure of (4.24) is that shown. in (4.25)5.

(4.25)

NP VP

sono NP V

NP ookii

NP

sakana

N E V

k. N
I

sakana tabete iru

 130

In the derivation of (4.24) from (4.25), when the

Relative Clause Formation Rule applies, the only apparent change that

occurs in (4.25) is that the boxed node NP disappears. It would

thus appear t6at the English version of the Relative Clause Formation

Rule, which was stated in (4.2), is fundamentally different from the

Japanese version, for in the former, the embedded identical NP is

reordered and placed at the front of the matrix sentence, while in

Japanese, the embedded NP is merely deleted.

But there are two facts which lead me to believe that

this dissimilarity is only superficial. First of all, the Japanese

Fo I V, 01,1441A
Relative Clause tztramm Rule is subject to the Complex NP Constraint

and also to the Coordinate Structure Constraint, which will be discussed

in § 4.2, and I will show, in Chapter 6, that simple deletion

transformations are not subject to these two conditions. Secondly,

in Japanese, as in all other languages I know of, the crossover

condition, which Postal has proposed, obtains.

This condit...on, as Postal Iririnally stated it, prevents

any transformation from interchanging two coreferential NP. Since the

Passive Rule effects such an interchange, reflexive sentences cannot

be passivized, as was noted by Lees and Klima (cf. Lees and Klima (1963)).

(4.26) a. Rutherford understands himself.

b. * Rutherford is understood. by himself.

c. *. Himself is understood by Rutherford.

 131

The condition can be generalized, however. Subjects

of sentences which appear as the object of say can normally be

relativized: that this is true of the NP pudding in (4.27a)

can be seen from the grammaticality of (4.27b):

(4.27) a. The man who ordered ice cream said the

puddingi would be tasty.

b. The pudding which the man who ordered

ice (-ream said would be tasty was a

horror show.

But if (4.27a) is changed so that the coreferential

NP the puddingi appears not only as the subject of would

but also as the deep object of ordered, ana if backward

pronominalization has applied, yielding (4.28),

(4.28) The man who ordered iti said the puddingi

would be.tasty.

then, for many speakers, the subject NP of the embedded sentence

is no-longer relativizable.

(4.29) * The puddingi which the man who ordered iti

said would be tasty was a horror show.

While (4.29) is an acceptable sentence if. the pronoun it refers to

some other NP, it is ungrammatical if it has the same referent as

the head noun of the subject of (4.29).

These facts can be explained by generalizing the cross-

over condition as shown in (4.30):

 132

(4.30) The Cross over Condition

No NP mentioned in the structural index

----of a transformation may be reordered by that rule

in such a way as to cross over a coreferential NP.

This condition is strong enough to excludc (4.29), for in

carrying out the Relative Clause Formation Rule to form (4.29), it

would have been necessary to move the subject of would be tasta,

leftwards over the coreferential pronoun it. This also explains why

the pronoun he in (4.31a) can refer to the same man as the head NP

the man but cannot do so in (4.31b) .

(4.31) a. The mans who said h el was tall.

* Theimml. who he. said was tall

However, (4.30-is too strong -- it would incorrectly

prevent (4.32a) from being passivized, and (4.32b) could not be

generated.

(4.32) a. The sheriffi denied that gangsters had

bribed him..
3.

b. That gangsters had bribed himi was denied

by the sheriff.

At present, I know of no way to weaken (4.30) to avoid this wrong

result.

The crossover condition also obtains in Japanese: the

Japanese version of the Passive Rule, which converts (4.33a) to (4.33b),

 133

(4.33) a. sono hito wa sakana o aratta

that man fish washed

'That man washed the fish.'

b. sakana wa sono hito ni arawareta

fish that man was washed

'The fish was washed by the man.'

cannot apply to reflexive sentences. (4.34a) cannot be passivized, as

the ungrammaticality of (4.34b) shows.

(4.34) a. sono hito wa zibun o aratta

that man

'That man washed himself.'

b. * zibun wa sono hito ni arawareta

I* That man was washed by himself.'

The crossover condition, by its very nature, applies only

to transformations which reorder constituents, so the fact that

grammatical and ungrammatical pairs of Japanese relative clauses

can be found which parallel those in (4.31) is a second indication

that the Japanese rule of Relative Clause Formation also involves

reordering, and not merely deletion.

(4.35)a.kare.ga nagai to itta hito..

he tall that said man

'The an who said he was tall.'

 134 .

b. * hito. ga nagai to itta hitoi II

1

man tell that said man

lic'Theman.whohe..said was tall '

1 1

The fact that the first occurrence of hits 'man' in

(4.35b) cannot have the same raZerent as the second one indicates that

the term 'cross over', which was used in the statement of (4.30), cannot

be taken sirply to refer to the linear order of words in the sentence,

for the underlying structure of (4.35a) is that shown in (4.36) .

(4.36).

As (4,35) shows, the boxed NP can be relativized, although

the circled NP cannot. If I am correct in attributing these facts to

the cross `over condition, which (4.34b) shows to be necessary in

t..../

Japanese in any case, then, if the rule of Relative Clause Formation

,,
 135

in Japanese operates is such a way as to move the identical NP in

the matrix sentence to the right end of the.embedded.sentence, in

L.w A.* 4. 4.. '0« I 4 ',Is

7
vuc vpposle w.recu.honA-2
from 4.4,a4. In wha.",,...--1-1 it moves A.Laki.i.4. L4

notion of "crossing over" must be defined in such a way as to take

into consideratiOn not only the one-dimensional linear ordering of

constituents, but also their two-dimensional hierarchical arrangement.

At any rate, whether or not my contention that the Japanese

version of Relative Clause Formation involves reordering is correct,

it is a fact that elements of relative clauses cannot be relativized.

For example, sentence (4.24), in which the NP kodom ga the child'

appears as the subject Of a relative clause, cannot be embedded as

a modifier of the subject NP of (4.37), as is shown by the

ungrammaticality of (4.38).

(4.37) kodomo ga byooki

child sick is

'The child is sick:

(4.38) * sono tabete iru sakana ga ookii k.odomo ga byooki da.

that eating is fish big child sick is.

1* The child who ED fish (he) is eating is big is sick.'

A
Furthermore, there are Japanese sentences)like (4.39))which

parallel those in (4.17); and, just as is the case in English, while

elements can be relativized from the object clause of (4.39b), which

corresponds to (4.17b) , this is not possible in (4.39a), which corresponds

to (4.17a). This can be seen from the ungramaaticality of (4.40a) and

the grammaticality of (4.40b).

 136

(4.39) a. Otto ga kono boosi o kabutte ita to iu syutyoo o watakusi wa sinzita.

Otto this hat wearing was that say claim I believed

'I believed the claim that Otto was wearing this hat.'

b. Otto ga kono boosi o kabutte ita koto o watakusi wa sinzita.

Otto this hat wearing was thing I believed

'I believed that Otto was gearing this hat.'

, 8 t
syatyoo o watakusi ga sinzita boosi wa aKaie
(4.40). a. *Otto ga kabutte ita to

wearing was that say claim I believed hat red

Otto

'*The hat which I believed the claim that Otto was wearing is red.'

b. Otto ga kabutte ita koto o watakusi ga sinzita boosi wa akai.

wearing was thing I believed hat red

Otto

'The hat which I believed that Otto was wearing is red.'

The underlying structure for (4.40b) is roughly that shown

in (4.41).
 137

(4.41)

NP
rNP VP
Ar VP

V
1

akai

N Nrf..-/.4." V boost

watakusi sinzita

VP ri.N
Lex
NP V
koto

kabutte ita
boosi

Although it is not clear to me what the deep structure

for sentences like (4.39a) should be, it seems reasonable to assume

that at the time the Relative Clause Formation Rule applies, the

major difference between this structure and the structure which results

from the deep structure of (4.39b) (the deep structure which appears

in (4.41) as a relative clause on boosi 'hat') would be that the

lexical noun syutyoo 'claim', would appear in place of the non-

lexical. noun koto 'thing'. Thus the circled NP boosi 'hat' in

(4.41) is relativizable, because the Complex NP Constraint only

prohibits elements which are contained in a sentence dominated by a

 138

NP with a lexical head noun from reordering, and the Japanese

nouns koto, mono, and no (if this last should be analysed as a

noun at all), which all mean roughly 'thing', are presumably

non-lexical. But nouns like mutr.22 'claim' are lexical, and

therefore the Complex NP Constraint must prevent elements of

sentences in apposition to them from reordering out of these

sentences, as the ungrammaticality of (4.40a) shows.

To summarize briefly, what I am proposing is that the

facts presented as evidence for the..A.Tover-A principle, in Cases A

and B of § 2,2 - namely that elements of relative clauses cannot

be relativized or questioned, and that in general, elements of

clause' in apposition to sentential nouns.also cannot -- should both

be accounted for by (4,20) -- the Complex NP Constraint. The

fact that elements of clauses in construction with "empty" nouns

like it (cf. (4.13a)) and koto 'thing' (cf. (4.40b)) can be

relativized, whereas this is,not possible in clauses in construction

with nouns like _girl (cf. (4.15)), claim (cf. (4.18a)), kodomo 'child'

(cf. (4.38)), and svutvoo 'claim' (cf. (4.40a)), necessitates that

the constraint be stated with reference to some such feature as

[± Lexical]. Ibelieve the Complex NP Constraint to be universal

(but cf. fn. 8), although there are problems with it even in English.

These will be taken up in the two sections immediately following,:

 139

4.1.5. The first difficulty with (4.20) concerns sentences

like those in (4.42),

(449) n, T am making_the c1 aim that the company

squandered the money.

I am discussing the claim that the company

squandered the.money.

Most speakers find NP in the position of the money not

to be relativizable in (4.42b), but to be so, or at least more nearly

so, in the. case of (4.42a).

.(4.43) a. ? The money which I am making the claim that

the company squandered amounts to $400,000.

b. * The money which I am discussing the claim

that the company squandered amounts to

$400,000.

Sentence (4.43b) can be made even more ungrammatical by prefixing the

noun claim with some possessive modifier,

(4.44) ** The ropey which I am discussing Sarah's

claim that the company squandered amounts

to $400,000.

and many speakers feel that while (4.43a) nay not be fully grammatical,

sentences like those in (4.45),.whose only significant difference from

(4.43a) lies in the definiteness of che article on the sentential

noun, are completely grammatical.

 140

hopes
(4.45) a. The money which I have that the
a feeling
company will squander amounts to $400,000.

b. The money which I will have a chance Lo

squander amounts to $400,000.

c. TIT money which I w 11 make a proposal

or us to squande amounts to $400,000.

that we squander

If any of these sentences are grammatical, either

condition (4.20) must be modified or abandoned, or the two sentences

in (4.42) must derive from quite different sources. As it stands, (4.20)

O

will block the generation of all the sentences in (4.43) (4.45): in

each case, the NP being relativized is contained in a sentence in

apposition to a lexical head noun.

There is some evidence that the second alternative may

be correct, i.e., that (4.20) can be preserved as is. I have not yet

been able to solve various problems of rule ordering that arise in

connection with this alternative, and it is only in the hope that the

following incomplete analysii may suggest a correct way of distinguishing

between (4.43a) and (4.43b) that I present it here.

Harris has proposed (cf. Harris (1957)) that sentences

like those in (4.46) be directly transformed into the corresponding

sentences ti (4.47), by a rule which he calls the modal transformation.

 141

(4.46) a. I snoozed.

b. Sam progressed.

c. Bill gave me $40.

L Max shoved the car.

e. I feel that Arch will show up.

(4.47) a. I' took a snooze

Sam made progress.

c. Bill made a gift to me of $40.

d. Maxgave-the car a shove.

e. I have a feeling that Arch will show up.

Since the surface structures of (4.46a) and (4.47a)

seem to be those shown in (4.48c) and (4.48b), respectively (the

situation is similar with respect to the other sentences of (4.46)

and (4.47)),

(4.48) a,3 S

VP
'
I V
1

snoozed

b.

NP VP

I V NP

took a

snooze
 142

Harris' rule cannot be stated within the currently available

theoretical framework, for at present, only transformations which

decrease structure can be formulated. The P-marker in (4.48a)

contains only one NP, but the one in (4.48b) contains two, so

the present theory would not allow a direct transformational

relation which converted the former into the latter (the opposite

direction would be possible, of course). So, at present, in the

theory of generative grammar, one could only claim (a) that the

sentences are only semantically related, or (b) that (4.48b) is

converted into (4.48a), or (c) that the deep structure of (4.48a)

is contained in the deep structure of (4.48b), as shown in (4.49):

(4.49)

IP
I NP

took S

'it IP
YP
I V
I

snooze

Proponents of this last approach would presumably argue

that after the embe-'ded subject in (4.49), 1, had been deleted by

Equi-NP Deletion, the verb'snooze TIould be substituted for the

 143

9
abstract pronoun, it, and the indefinite article would be segmentalized ,

yielding the structure in (4.48b).

do not know whether any of the above analyses is

correct, or whether structure-building transformations, which could

convert (4.48a) directly ii to (4.48b), should be countenanced within

the theory. But whatever analysis is adopted for the sentences in

(4.47), it should also be adopted for expressions like make the claim

that S, have hopes that S, have a chance to VP, etc., which were

used in (4.42) and (4.45) above. If analysis (a) is correct, then

both sentences in (4.42) would come from roughly the same deep

structure, (4.50).

(4.50)

r
NfP

I am

making NP
discussing

the N

andered
 ,
-....... 4-4,......-.

But the fact that the NP the money is relativizable

in (4.42a) but not in (4.42b) seems to argue against this analysis,

for how can this difference be accounted for, if both sentences have

roughly the same deep structure? Furthermore, thete is another fact

about the sentences in (4.42a) and (4.45a) which sets them off from

other sentences containing sentential nouns with clauses in apposition

to them. George Lakoff has pointed out to me that the rule which

optionally deletes the complementizer that in clauses which follow

a verb cannot apply if the verb has been substantivized. So, while

both (4.51a) and (4.51b) are grammatical, only the a-version of

(4.52) is possible.

(4.51)___a. Kleene proved that this set is recursive.

b. Kleene proved this set is recursive.

(4.52) a. The proof that this set is recursive is

difficult.

b. * The proof this set is recursive is difficult.

It seems to be the case that it is only in modal

constructions like make the claim that S, have hopes that S, etc.

that the complementizer that can be deleted after a sentential noun.

(4.53) a. ? I am making the claim the company squandered

the money.

b. I have hopes the company will squander the

money.
 145

c. I have a feeling the company will squander

the money.

---d. * I made a proposal we squander the money.

As (4.53d) shows, it does not seem to be the case that

that can be deleted in all modal constructions -- what the restrictions

are I do not know at present but...the...fact that it generally can be

deleted in these Constructions is another piece of evidence that

argues they should be analyzed differently than such sentences as

One final fact deserves mention here: to the best of

my knowledge, it is only in modal constructions that sentential nouns

which are related to transitive verbs cannot occur with a full range

or possessive modifiers. In sentences like those in (4.54), where

the main verb of the sentence containing 'claim is not make, any

possessive NP can modify claim.

-
Your
(4.54) a. Dick's claim that semantics is generative
etc.

is preposterous.

Myron's
b. We are discussing their. claim that
etc.

flying saucers are real.

 146

But after the verb make, and only after it, the

possessive modifier must refer back to the subject of make, if it

is possible to have such a modifier_at_all:

(4.55) Myron is making

9 =
* Suzie's claim that dead
* Dr. No's
etc.
{
is better than red.

The same is true of all modals, as the sentences in (4.56)

demonstrate.

(4.56) a. * I have Tom's feeling that the company will

squander the money.

b. * Myra took Betty's snooze.

c. * Bill made Sarah's gal to me of $40.

d. * Max gave the. car Levi's shove.

These three facts -- that the Complex NP Constraint is

not operative in modal constructions; that the complementizer that is

generally deletable there, and the fact that.possessive modifiers

must refer back to the subject of the modal verb -- indicate clearly

that sentential nouns like claim. hooe, etc. which occur in these

constructions must be derived differently in modal constructions than

they are elsewhere.

It is tempting to propose changing the theory so that

(4.48a) could be directly converted into (4.48b) by a structure-building

 147

rule of Modalization. Then the fact that elements are relativizable

in complement sentences after make the claim, have hopes, etc. and

the fact that that can be deleted there could be handled by ordering

the rules as follows: Relative Clause Formation, That Deletion,

.11111111

Modalization.

Unfortunately, this solution will not work, for if there

is a rule of Modalization, Passive must follow it:

(4.57) The claim that plutonium would not float was

made by the freshman.

But if Passive follows Relative Clause Formation, such

sentences as (4.58) will not be derivable.

(4.58) The man who was arrested by Officer McNulty

went mad.

Furthermore, if Passive follows That Deletion, what is

to prevent derivations like that shown in (4.59)?

(4.59) a. Jack is claiming that you won't need it.

That Deletion

b. Jack is claiming you won't need it.

Modalization

c. Jack is making the claim you won't need

it.---H>Passive

d. * The claim you won't need it is being made;

by Jack.
 148

These difficulties, which I have not been able to overcome,

have kept me from reaching a solution to the problem posed by the

modal construction for the Complex NP Constraint. But since it

seems clear that the complex sentential NP which occur in modal

constructions must be derived from some other source than the sentential

NP in other constructions, I have hopes that it will be possible to

preserve the Complex NP Constraint in the way it was stated in (4.20).

At any rate, I will not settle for merely an ad hoc rider on (4.20)

until the grammar of modal constructions is considerably better

understood than it is at present.

4.1.6. The second difficulty concerning (4.20) arises in

connection, with the sentences in (4.3) and (4.4), which I will

repeat below for convenience..

(4.3) a I read a statement which was about that man.

b. I read a statement about that man.'

(4.4) a. * The man who I read a statement which was

about is sick.

b. The man who I read a statement about is sick.

As I pointed out in § 4.1.2, it is not in general the

case that elements in reduced relative clauses can be relativized or

questioned: the fact that the sentences of (4.15) and (4.16) are

equally ungrammatical supports this contention. How then can it be

that the object of about in (4.3b) can be relativized, if (4.3b) derives

 149

from (4.3a) byway of the rule of Relative Clause Reduction?

The tentative answer to this question which I would

propose is that the re' zion between the sentences of (4.3) must be

much more complex than has hitherto been suspected. I suspect that

(4.3b) is nearer to being basic than (4.3a) is, and that in any case,

(4.3b) is not derived from (4.3a) by means of the rule of Relative

Clause Reduction. There are a number of peculiar facts about sentences

containing nouns like statement, some of which I will take up below,

which suggest the correctness of this idea.

First of all, such sentences behave uniquely under

reflexivization. As was shown in Lees and Klima (1963), the second

of two identical noun phrases is replaced by a reflexive pronoun,

subject to the condition that both NP's be in the same "simplex

sentence", to use their term. They do not state how this restriction

clear from
is to be expressed foimally, but their meaning will be

the following examples:

(4.60) a. You're going to hurt yourself one of these

days.

b. I spoke to Bill about himself.

(4,61) a; * That .Tom saw me surprised myself.

b. * He said that himself was hungry.

Reflexivization must be blocked in (4.61), for in both

cases, there is a node S which dominates one occurrenceof the two

NP's which does not dominate the other. Since this is not true of
 150
2

(4.60), Reflexivization must apply.

Consider now such sentences as those shown in (4.62)

(4.62)a.Ireadhim.1 a statement which was about

himi
?himself l
l

b. I read him, a statement about 1 ....

1 himseitj

I am not sure, but I believe (4.62a) is better, in my

own speech, with a non-reflexive pronoun than with a reflexive

pronoun. If there are dialects in which both of the sentences in

(4.62a) are fully grammatical, I can provide no explanation of such

facts, for in the overwhelming majority of cases, Reflexivization

cannot go down into relative clauses, and I would not know how to

characterize formally the relative clauses in sentences like

(4.62a) in such a way that Reflexivization could go down into

them, but not into clauses like the one shown in (4.63).

me
(4.63) I know a man who hates
myself

Therefore, for the purposes of this study, let us assume,

perhaps falsely, the existence of a dialect in whiw reflexive pronouns

are absolutely excluded in (4.62a) and are absolutely necess in

(4.62b). How could we explain such facts?

Given that a meta-rule of S-pruning like (3.6) must

be included in linguistic theory,. on the basis of the independent

evidence presented in § 3.1, it might be argued that the explanation

 151

must depend in some way on this meta-rule. That is, one could assume

that (4.62b) is derived from (4.62a) by the rule of Relative Clause

Reduction. Reflexivization would be blocked in (4.62a), because in

(4.64), which shows the approximate structure of (4,62a) , the circled

node S dominates the second occurrence of the NP he (1112), but

not the first, so the two NP's are not in the same simplex sentence.

VP

NP NP
I

read him NP

Art N NP

a statement

NP

about him

Then, of course, as in the cases discussed in §§ 3.1.1 -

3.1.3, when the Relative Clause Reduction Rule deletes which was in

(4.64), the circled S will no longer branch and will be pruned by

(3.6), thus bringing it about that the two occurrences of he (him) are

in CAe same simplex sentence, so that Reflexivization can convert the

sewnd one into himself.

 152

- d a or
- --,
or
0

417. ;
ood

at.,
- .
;A.
-7 ....or

...*.
--

"
...CT.
.
-

,
ow,
-.

T
-do.

L.
. /
.
no

anc cannot, as far as I. can see at present, be patched up to

overcome these inadequacies. The first difficulty arises in connection

with several facts which were first pointed out in two careful

studies of reflexives made by Florence Warshawsky (cf. Warshawsky

(1965a,b)). She pointed out that whether or not reflexivization

occurs in sentences like (4.62b) is correlated in some inexplicable

way with the type of determiner which precedes statement. In (4.65a),

where the determiners are indefinite, reflexivization seems to be

obligatory, in most dialects, whereas in (4.65b), where the determiners

are possessives, they do not occur (in most dialects). With the

definite articles the, this, that (4.65c), there seems to be great

dialectal variation. To my ear, the sentences sound odd with or

without reflexives.

(4.65) a. I read him two (several, some, no)

statements about himself.

b. * I read him Judy's statement about himself.

c.?* I read him the (this, that) statement

about himself.

Clearly, no principle like (3.6) can account for the facts in (4.65)

by itself -- additional conditions of some sort must be imposed on the

rule of Reflexivization (these sentences will be discussed again in

6.4) below). But, it might be argued, at least the principle of

 (1'

153

S-pruning makes it possible to state the Reflexivization Rule in such

a way that reflexives are excluded from (4.62a), while at least

some of them are allowed in sentences like (4:65a) and possibly (4.65c).

This argument seems appealing until it is realized that normally

Reflexivization does not go down into reduced relative clauses. For

example, if the relative clause in (4.66a) is reduced to the phrase

behind me, the NP me cannot be converted into a reflexive. The

same is true of the reduced clauses jealous of you and watching me

in (4.77b) and (4.78b).

(4.66) a. I know two men who are behind me.

b. I know two men behind me (*myself).

(4.67) a. You are too flip with people who are jealous

of you.

b.. You are too flip with people jealous of

you (*yourself).

(4.68) a. I screamed at some children who were

watching me.

b. I screamed at some children watching me

(*myself) .

In fact, excluding the problem as to whether reflexive

pronouns can appear in relative clauses of the type contained in

(4.62a) , I would hazard a guess that not only do rules of reflexivization

universally not go down into relative clauses, they also do not go c

into reduced relative clauses. For instance, in German, if the

 154

relative clause die'ihmaieb sind 'who are kind to him' in (4.69a)

is reduced to form (4.69b), the personal pronoun'ihm 'him' (dat.)

is aot converted to the reflexive pronoun'sich 'himself'.

(4.69) a. Hans verknallt sich nur in 112dchen, die

Hans falls only for girls, who

ihm lieb sind.

him kind are.

'Hans only falls for girls who are kind to him.'

b. Hans-verknallt_sich nur in ihm Liebe Madchea.

Hans falls only for him kind girls.

'Hans only falls for girls kind to him.'

If'sich is substituted for'ihm in (4.69b), as in

(4.70), the sentence produced has a different meaning and is unrelated

to the sentences in (4.69).

(4.70) Hans verknallt sich. nur in sich liebe Maddhen.

Hans falls only for themselves kind girls.

'Hans only falls for girls who are kind to themselves.

Thus, the most obvious explanation of the facts of

(4.62), an explanation making use of the rule ordering shown in (4.71)

(4.71)* Relative Clause Reduction

Reflexivization

and of some convention of S-pruning, would seem to be inadequate

for the same reason that (4.5) cannot adequately account for the

difference in grammaticality of the sentences in (4.4). Normally,

 155

Reflexivization does not go down into reduced relative clauses, so

the fact that reflexives can occur after about in (4.62 -) suggests

that the about-phrase is not clausal in origin.

Warshawsky (op. cit.) points out that Jaany of the nouns

which can appear in the blank in (4.72) are relarad to verbs.

, Gf himsel:.
(4.72) Nax showed me a
labout

A few of the verb-related nouns that occur in tais environment

are listed in (4.73a).; several for which no corresponding verb

exists are given in (4.73b). (Warshawsky gives much more extensive

lists of these nouns, which she calls "picture nouns".)

(4.73) a. description, statement, report, claim,

tale, drawing, painting, photograph,

etching, sketch

b. story, column, satire, book, letter, text,

article, sentence, paragraph, chapter,

picture

Warshawsky points out that the verbs associated with the nouns of

(4.73a) are all verbs of creation, and the nouns systematically

ambiguous with respect to whether they denote an abstract cx.ation

or some physical object upon which this creation is represented.

Further, she notes that certain of these verbs can occur only with

human subjects (cf. (4.74)),

 156

{painted
{ Michael the duck
(4 .74) sketched
*Michael's photograph
drew I

%.

pond.

but that others could have either human subjects or picture noun

subjects.

Michael

report
statement
description iltold of the conflict
(4.75) Michael's story described the country
article stated that we were at fault
book
?picture
I.

This list property is unlike any other grammatical fact I have

encountered, It is worth pointing out that it is not the case

that any abstract noun can serve as subject of these verbs -- only

picture nouns can, as is shown by the ungrammaticality of (4.76).

sthe space between my eyes

entencehood told of the conflict
(4.76)
f
* Harry's' civil rights described the country :

Marilyn's arrival 11
stated that we were at fault
,etc.

The fact that the deverbal nouns in (4.73a) behave the

same way as the apparently basic nouns in (4.73b) with respect to

relativization and questioning (cf. (4.4)), reflexivization (cf. (4.62))

and with respect to the curious selectional facts pointed .out in (4.75)
 157

provides strong evidence for treat!ng all picture nouns alike.

Warshawsky sug,:ests that verbs may be basic for picture nouns, and

that hypothetical verbs (cf. Lakoff (1c.w65)) such as to story,

to column, etc. be postulated as underlying the nouns of (4.73b).

This proposal seems quite reasonable, but in the absence of a

detailed analysis along these lines, little more can be. said

about it at present.
r-

In passing, it should be remarked that there are a

number of prepositional phrase adjuncts to noun phrases which exhibit

similar behavior r 'icture nouns. As (4.16b) shows, it is not

in general the case chat elements of postnominal prepositional

phrases can be questioned. But this is the case in tae sentences

of (4.77), as (4.78) shows.

(4.77) a. I gave Tom a key to that door.

fo

b. Harold has books by some young novelists.

c. Billy is looking for a road into the cavern.

(4.78) a. Which door did I give Tom a key to 2

for
b. Which novelists does Harold have boo by?

c. ? Which cavern is Billy looking for a road into?

Considerations of the same sort as were discussed above

would suggest that NP like a key to this door and a road into the

cavern should not be derived from ?a key which is to this door and

?a road which is into the cavern, which are at best of dubious

grammaticality in any -vent But what their deep structures might be

 158

is at present an unsolved problem.

4.1.7. To conclude this discussion, the constraint which I

stated in (4,20) correctly prevents elements of relative clauses

from being questioned or relativized. The remarks of footnote 8

and § 4.1.5 above indicate that this constraint is stated too

strongly at present, and the remarks in § 4.1.6 show that the

differences between the sentences of (4.4), although they appear to

fall within the scope of (4.20), are in fact much more complex than

has been realized. I know of no other counterexamples to the

Complex NP Constraint, and I therefore submit it for inclusion

in the list of putative linguistic universals, subject to whatever

modifications are necessary to avoid the extra strength pointed out

in footnote 8 and4.1.5.

4.2. The Coordinate Structure Constraint

4.2,1. In § 2.2, in Case F, it was pointed out that conjoined

XP cannot be questioned: this Was attested to by the ungrammaticality

of (2..18) and (2.19), which I repeat here for convenience.

(2.18) * What sofa will he put the chair between some

table and?

(2.19) * What table will he put the chair between and

some sofa?
The impossibility of questioning the circled NP nodes in diagram

(4.79) can be succeasfully accounted for by invoking the A-over-A

principle,

(4.79)

but this principle does not prevent the circled NP nodes in diagrams

(4.80) or (4.81) from being questioned or relativized.

(4.80)

He_ nr
(4.81)

S and

VP

the nurse the plumber

/4
Dolished her trombone

But all of the circled nodes must somehow be restricted from being

moved, as the ungrammatical sentences of (4.82) show.

(4.82) a. *.The lute which Henry plays and sings

madrigals is warped.

b. * The madrigals which Henry plays the lute

and sings sound lousy.

c. * The nurse who polished her trombone and

the plumber computed my tax was a blonde.

d. * Which trombone did the nurse polish and

the plumber computed my tax?

e. * The plumber who the nurse polished her

trombone and computed my tax was a hefty

fellow.

f. * Whose tax did the nurse polish her trombone

and the plumber compute?

 161

I. know of no principled way of excluding such 'structures

as those shown in.(4.80) and (4.81) from being introduced as relative

clauses, i.e., at the node S in (4.83),

(4.83) NP

NP

so it appears to be necessary to add the following constraint to the

meta-theory:

(4.84) The Coordinate Structure Constraint

In a coordinate structure, no conjunct may be

moved, nor may any element contained in a

conjunct be moved out of that conjunct.

4.2.2. I propose to define the notion coordinate structure

as eny structure conforadng to the schematic diagram in (4.85).

(4.85)

and
or
.
A A ....

{
 162

Of course, since (4.85) is intended to be a universal definition,

it must be understood as containing not the English morphemes

and and or_ but rather a mnre abstract_ language-independent

10
representation of these terms--. Furtheralore, the conjlinc.tion

should be understood as either preceding all its conjuncts, as in

English, French, or as following them, as in Japanese.

Coordinate structures contain at least two conjuncts, but may

contain any higher number of them.

As for the deep structure position of the conjunction

with respect to the conjuncts, there are many reasons for believing

that the structure of (4.86) is not that shown in (4.87), but

rather that shown in (4.88), where each occurrence of the conjunction

and forms a constituent with the following sentence instead of

being coordinate with it, as in (4.87).

(4.86) Irma washed the dishes, and Sally dried, and

Floyd loafed.

(4.87)

Irma washed the dishes a d Sally dried and Floyd loafed

1,!..PIORMIOPPTPPRVIITITRMRPOPWPWWIrfP.Prr..-,..Fgrrewns.frogro..-.
 163

(4.86)

Irma washed the dishes and Sally dried and Floyd loafed 1
One'syntactic reason is that if a conjoined sentence like

(4.89) is broken up into two sentences, as in (4.90), the conjunction

always goes with the second sentence, as in (4.90a), never with the

first, as in (4.90b).

(4.89) John left, andhe didn't even say goodbye.

(4.90) a. John left. And he didn't even say goodbye.

b. * John left and. He didn't even say goodbye.

A second syntactic reason is in that languages in which

coordinating conjunctions can become enclitics, which are then

inserted into one conjunct (this is the case with - que 'and' in

Latin, and with the word aber 'but' in German), these enclitics

are always associated with the following conjunct, never with the

preceding one. Thus (4.91) may be converted into (4.92a), but not

into (4.92b).

(4.91) Sie will tanzen, aber ich will nach Hause

gehen.

'She wants to dance, but I want to go home.'

 164

(4.92) a. Sie will tanzen; ich will aber nach Hause

gehen.

b. * Sie will aber tanzen; ich will nach Hause

11
gehen.

A third syntactic reason for, regarding (4.88) as the

correct structt is the following: since the Appositive Clause

Formation Rule must convert sentences like (4.93a) into (4.93b)C-tj.

(4.93) a. Even Harold failed, and he is the smartest

aoy in our class.

and he
b. Even Harold., is the smartest boy
who
in our class, failed..

there are very general theoretical grounds for arguing that the string

and he is the smartest boy in our class in (4.93a) is a constituent,

for except for this case, transformations can be constrained so that

only constituents may be adjoined.

Phonological evidence indicates strongly that the bracketing

of the subject NP of (4.94) must be that shown in (4.95a), and not

that shown in (4.95b) or (4.95c),

(4.94) Tom, and Dick, and Harry all love watermelon.

(4.95) a. ((Tom) (and Dick) (and Harry)) all love

watermelon.

b. ((Tom) (and) (Dick) (and) (Harry)) all

love watermelon.
 165

c. ((Tom and) (Dick and) (Harry)) all love

watermelon,

far intonational pauses come before coordiftatiag conjunctions, not

after them or equally on both sides of them.

So there is good evidence to indicate that the correct

structure of (4.86) must be that given in (4.88). But how does this

structure arise? Lakoff and l (op. cit.) propose that there be a

phrase structure rule schema like (4.96) in the base,

(and,
(4.96) S S , where n > 2
or

and that later the and or or which Is introduced by (4.96) be

12
copied and Chomsky-adjoined to each of the indefinitely many

S's that are introduced by (4.96) by a rule of Conjunction Copying.

So the deep structure of (4.86) would be approximately that shown in

(4.97), which the rule of Conjunction Copying will convert to (4.98).

(4.97)

an

VP NP VP VP
IP 111.3
Irma NP Sally V NP Floyd, V

was led the dishes dried something loafed

 166

(4.98)

and
and
Irma washed the dishes Sally dried something Floyd loaf

To derive (4.88) from (4.98), the first instance of and

is deleted by a general rule which I will not state here.. It is

deleted obligatorily if the conjuncts are sentences, as is the case

in (4.98), but it may optionally be converted into both if the car.juncts

are NP, VP, or V. The rules for conjunction with or are similar

in all respects, except that the initial or may be converted into

either in front of all conjuncts. Languages like French, where the

first conjunction does not have a suppletive alternant, provide

further motivation for this analysis:

(4.99) a. Et Jean et Pierre sont fatigu6s.

andJohn and Peter are tired.

'Both John and Peter are tired.'

b. Ou Jean ou Pierre doit le faire.

Or John or Peter must it do.

'Either John or Peter must do it.'

One final point in favor of this analysis should be

mentioned: the semantic interpretation of conjunctions, under this

analysis, is much more in line with the traditional logical analysis of

 167

conjunctions, which treats them as n-place predicates, than would be

the case if the previously accepted analyses were adopted. That is,

4C /A 071 4n nA, mA ec
4%. .0%0
4-1-tn
.--r ..... nc IL RAI
...--e, 1.11p. Pn-Milnet-inng

and and or are only different semantically from such two-place

relations as see, etc. in that the former can have an indefinitely

large number of arguments, while the latter is binary. But if some

such structure as (4.87) is postulated as the deep structure of (4.86),

quited:ssimilar projection rules will have to be constructed to

interpret (4.87) semantically, and the fact that and, or, and see

are semantically similar, in that all are relations, will not be

expressed formally.

4.2.3. Given the above definition of coordiLate structure, the

first clause of the Coordinate Structure Constraint will exclude

(2.18) and (2.19), while the second will exclude all the sentences

of (4.82). The latter sentences could neither be excluded by the

A-over-A principle nor by the Complex NP Constraint of § 4.1,

so it appears that condition (4.84) is necessary for reasons which

are independent of the problems raised by (2.18) and (2.19). Thus

(4.84) can be used to explain their ungrammaticality, just as the

A-over-A principle was.

It should be pointed out that there are instances of

the morpheme and which must be derived from different sources than

the two major sources discussed in Lakoff and Peters (1966). For
la 3
 168

instance, as (4.101) shows, there is a difference in relativizability

between (4.100a) and (4.100b), even though both sentences in (4.100)

appear rn contain gtructnrpg thnr are onnrclinnte_ by definition (4.85).

(4.100) a. I went to the store and bought some whisky.

b. I went to the store and Nike bought some

whisky.

(4.101) a. Here's the whisky which I went to the store

and bought.

b. ' Here's the whisky which I went to the store

and Mike bought.

However, as George Lakoff has pointed out to me, there

are clear syntactic indications that the relative clause in (4.101a)

is not an instance of ordinary sentence conjunction. First of all,

it is only with non-stative verbs as the main verb of the second

conjunct that sentences lik- (4.101a) can be constructed.

(4.102) a. Tony has a Fiat and yearns for a tall nurse.

b. * The tall nurse who Tony has a Fiat and

yearns for is cruel to him.

Secondly, the second conjunct cannot be negative:

(4.103) a. I went to the movies and didn't pick up

the shirts.

b. * The shirts which I went to the movies and

didn't pick up will cost us a lot of money.

 169

Thirdly, there are restrictions on the tenses that may

appear in such sentences as (4.101a) . Thus. (4.104a) parallels (4.100a)

hbtu.ease
111 eve:ytin - ) """
UW. MT)
4.4 the ryln4clev, is 'Int- T.P1ativi7Ab1e

as (4.104b) indicates.

(4.104) a. I went to the store and have bought some

excellent whisky.

b. * The excellent whidcy which I went to the

store and have bought was very costly.

The fact that (4.100a) , on one reading, is synonymous with (4:105a) ,

which contains a purpose clause, and the fact that the ungrammaticality

of (4.102b) , (4.103b) , and (4.104b) is matched by correspondingly

ungrammatical purposeclauses (cf. (4.105b), (4.105c), and (4.105n)

respectively) suggests that the reading of (4.100a) which allows the

formation of the relative clause'of (4.101a) be derived from whatever

the underlying structure is that underlies (4.105a). Note, by the

way, that relativization is also possible in (4.105a)\ as (4.106)

shows.

(4.105) a. I went to the store to buy some whisky.

b. * Tony has a Fiat to yearn for a tall nurse.

not to
c. * I went to the movies pick the
to not

shirts up.

d. * I went to the store to have bought some

whisky.

,.1r.rrrprovenvbe....
 170

(4,106) Here's the whisky which I went to the

store to buy.

limc= arc of.tvar .kubl.auceb aiLu wiL.Lcu a

similar 1:11e of argument suggests should not be derived from coordinate

nodes in deep structure, For example, consider the sentences in (4.107):

(4.107) a. She's gone and ruined her dress now.

b. I've got to try and find that screw.

c. Aunt Hattie wants you to be nice and kiss

your granny.

As I have no plausible analysis for these sentences, I will merely

point out, that they are not subject to (4.84):

(4.108).__a.___ Which dress has she gone and ruined now?

b. The screw which I've got to try and find

holds the frammis to the myolator.

c. Which granny does Aunt Hattie want me to

be nice and kiss?

The fact that the sentences of (4.108) and sentence (4.101a) are

grammatical might mean that (4.84) is simply wrong, but the facts

I presented in (4.102) - (4.106) suggest that this may not be so,

at least with regard to (4.101a) . Rather it may be tb case that none

of thesie sentences contain coordinate structures at the time when

questions, relative clauses, etc. are formed, but only are converted

into coordinate structures later, or that .they never contain coordinate

structures at all. In fact, I know of no other test for coordinate

 171

structure than the one (4.84) provides, and it therefore seems quite

reasonable to me to assume that one of the last two possibilities

mentioned above is correct.

It is perhaps worthwhile to show how (4.84) can provide

a test for coordinate structure. (4.109i) can be converted into

(4.1091) by the rule of Gamin/ (Ross 1967d)):

(4.109) a. The boy works in a skyscraper and the girl

works in a quonset hut.

b. The boy works in a skyscraper and the girl

in a quonset hut.

The structure underlying these sentences is that shown in (4.110).

(4.110)

and

NP

works works
1

in skyscraper

..1r.r.0.71.11.1F,T,,,,
 172

When Gappiag applies to (4.110), deleting the second

occurrence of the verb works, it might be proposed that either the

node VP which immediately dominates it or the circled node S

should be pruned, or both. There is no evidence which argues for

or against retention of the circled node VP, but if the circled

S were pruned, (4.110) would cease to be a coordinate structure, under

441e,
the definition given in (4.85), andlboxed NP),,, in (4.110) should

become movable. The fact that they do not (cf. (4.111))

(4.111) a. * Which boy works in a skyscraper and the

girl in a quonset hut?

b. * The skyscraper which the boy works in and

the girl in a quonset but belongs to

Uncle Sam.

c. * The girl who the by works in a skyscraper

and in a quonset but has a dimple on her

nose.

d. * Which quonset but does the boy work in a

skyscraper and the girl in?

is most simply accounted for by assuming that (4,110) retains its

coordinate structure even after appina. has applied, i.e., that the

putative convention which pruned the circled S was incorrect.

It can 'iso be shown that coordinate structure can

disappear, in the course of a derivation. So, for instance, Lakoff and

Peters (op. cit.) argue that (4.112) should be derived from (4.113) by
 173

a sequence of optional rules which convert an occurrence of and to

with and then adjoin the with-phrase to the main V? of the

CAnt'Arle.P, 13

Billy went to the, movies with ..Yy tmic.

(4.112)

check.

(4.113)

NP

and went to the movies

Bid a luscious check

The circled NP is not relativizable unless Conjunct

Movement has applied (cf. (4.114)):

(4.114) a. The luscious chick who Billy went to the

movies with will wed me ere the morn.

b. * The luscious chick who Billy and went to

the movies will wed me ere the morn.

Similarly, in the conjoined structure (4.115),

(4.115)

and

NP VP

Pietro Sofia

/4 \
:bought a Ferrari from me

The circled NP can only be relativized if the second conjoined

sentence has been inserted into the first as an appositive clause.

(4.116) a. * The Ferrari which Pietro bought from me

14
and Sofia adores him cost him a bundle.

b. The Ferrari which Pietro, who Sofia adores,

bought from me cost him a bundle.

These two facts illustrate a perhaps obvious point: whether or not

a constituent can be moved deends not on deep structure, but on

derived structure..

4.2.4.

4.2.4.1. There is an important class of rules to which (4.84) does

not apply. These are rule schemata which move a constituent out of all
 175

the conjuncts of a coordinate structure. In Lakoff and Ross

(in preparation b), an analysis of conjoined sentences is explored

which takes the process which converts such sentences as (4.117a)

into (4.117b) as being the fundamental process in conjunction.

(4.117) a. Sally might be pregnant, and everyone

believes Sheila definitely is pregnant.

b. Sally might be, and everyone believes

Sheila definitely is, pregnant.

We propose a rule of Conjunction Reduction which Chomsky-adjoins

to the right or left of the coordinate node a copy of some i.onstituent

which occurs in all conjuncts, on a right or left branch, respectively,

and then deletes the original nodes. Thus this rule converts (4.118),

whlch underlies (4.117), into (4.119). .

(4.118)

everyone V

be Neves

pregnant

re
 176

(4.119)

an pregnant

1
NP 'VP NP

Sal i.y mizht be everyone

1
ANP

believes it S

NAP VP

Sheila definitely is

It is important to note that Conjunction Reduction must

work "across the board" -- the element adjoined to the coordinate

node must occur in each conjunct. Thus (4.120a) can be converted to

(4.120b), but not (4.121a) to (4.121b) .

(4.120) a. Tom picked these grapes, and I washed

these grapes, and Suzie will prepare

these grapes.
 177

b. Tom picked, and I washed, and Suzie will

prepare, these grapes.

(4.121) a. Tom picked these grapes, and I washed

some turnips, and Suzie will prepare these

grapes.

b. * Tom picked, and I washed some turnips, and

Suzie will prepare, these grapes.

It appears that the rule of Relative Clause Formation

must also apply "across the board "; the relative clause in (4.122)

would seem to have to derive from a structure with an embedded

disjunction, as in (4.123),

(4.122) Students who fail the final exam or who

do not do the reading will be executed.

(4.123)

will be executed
NP

students or

VP NP VP

students fail the final exam students do not do the reading.

 178

rather than sentence (4.124), whose main clause is a disjunction,

because (4.124) is not synonymous with (4.122) .

(4.124) Students who fail the final exam will be

executed or students who do not do the

reading will be executed,

It is obvious that there are many rules which do not

necessarily apply across the board -- passives can be conjoined

with Actives (cf. (4.125a)) , and Particle Movement and Extrapositica,

may apply in some conjuncts but not in others (cf. (4.125b) and (4.125c)).

(4.125) a. John has been captured by the cops and I'm

of he'll talk.

b. I heated up the coffee and Sally wiped the

table off.

c. That Peter showed up is a miracle and it

is doubtful that he'll ever come again.

4.2.4.2. At present, since I only know of two rules which can

convincingly be argued to apply across the board, it is perhaps too

early to look for formal properties of rules which correlate with the

way the rules apply. Nonetheless, I find it significant that both

of the across-the-board rules operate in such a way as to remove

elements from conjuncts, while rules like Passive, Particle Movement,

Extraaosition, and many others like them which could be cited, me-ely
 179

rearrange items wit'ain a conjunct.

It is evident, even from the informal description of

Conjunction Reduction which was given above, that this rule moves

elements out of conjuncts, but it is not evident from the statement

of Relative Clause Formation which was givenin (4.2) that this rule

must also move elements out of conjuncts. Under the normal interpreta-

tion of the elementary operation of sister-adjunction, which is

symbolized by ' +' in the structural change of (4.2), when one term

is sister-adjoined to a variable and that variable is null for some

particular structure, nothing happens to that structure. That this

convention is necessary can be seen from the following consideration:

The rule of Extraposition-sister-adjoins the sentence

to a variable, as can be seen from the formal statement of this

rule in (4.126).

(4.126) Extraposition

X - [it Y
NP
OPT
1 2 3 4

1 2 0 4+3

With the abw7e condition on sister-adjunction, if

(4.126) were to applj to (4.127), no change would be effected:

the sentence in apposition to it would stay within its NP.

 180

NP VP

\NP
V

claimed

that Bob is a nut

Thu6. che next rule in the ordering, It Deletion,

could be formulated as shown in (4.128).

(4.128) It Deletion

X - [it - S] Y
NP
OBLIG
1 2 3 4

1 0 3 4

However, if the convention I have suggested were not

"15
in effect, "vacuous extraposition would be possible, and the

embedded sentence could be moved out of its NP and attached some-

where higher up the tree, as in (4.129) (just where it would attach

is not relevaim for my argument, and I have drawn two dotted lines

from the extraposed S in (4.129) to indicate two possibilities).

 181

(4.129)

eggs 11.,
.11=10

S
%\
I

claimed it that Bob was a nut

But if (4.127) can be converted into (4.129), then

(4.128) will have to be modified as shown in (4.130), for otherwise

this rule would not delete the it in (4.129), and the ungrammatical

(4.131) would result.

(4.130) X - it - S - Y

OBLIG
2 3 4
>
1 0 3 4

(4.131) * I claimed it that Bob was a nut.

But there are many sentences which show that (4.130) is

16
far too strong: it requires the deletion of it before any sentence

whatsoever, and it is easy to construct sentences where this extra

power leads to wrong results. In (4.132a), for instance, the it.

which ic; thc cLject of claim will 110 deleted, hicAr,_Auge it precedes

the clause [and I think so too) , and the ungrammatical (4.132b)

 182

will result.

(4.132) a. Although Bob may not be a nut, many people

have [and I think so too]s.

b. * Although Bob may not be a nut, many people

have claimed and I think so too.

To .avoid converting (4.132a) into (4.132b), while still

requiring the it in (4.131) to delete, some method would have to be

found of indicating that the sentence that Bob was a nut is somehow

"appropriate" as an environment for the deletion of the it of (4.131),

but that this is not the case with respect to the sentence and I think

so too in (4.132a) . In the absence of independent evidence for such a

convention of appropriateness, it seems more desirable to me to reject

the definition of sister-adjunction which gives rise to these difficulties

by allowing "vacuous" extraposition, and to impose the suggested

condition on this operation -- that if a term is sister-adjoined to a

null variable, no change in the d.c.s. will result.

Now let us return to the problem of the proper formulation

of the rule of Relative Clause Formationc Robin Lakoff has pointed

out to me that Ws in the position of the boy in (4.133) cannot be

relativized (cf. (4.134)).

(4.133) The. boy and the girl embraced.

(4.134) * The boy who and the girl embraced is my

neighbor.
 183

The fact that (4.134) is ungrammatical should be accounted for by

the Coordinate Structure Constraint, but since this constraint

only prevents constituents from being moved, it must be the case that

the formulation of the rule of Relative Clause Formation which was

given in (4.2) is wrong. (4.2) specifies that the identical NP

shall be sister-adjoined to a variable, and since this variable is

null in the case.of (4.133), by the argument given above, this NP

would not be moved by (4.2), and thus the constraint would not
17
be in effect. But if (4.2) is reformulated as in (4.135), the

identical NP will be moved, whether it is the first constituent

of the relative clause or not.

(4.135) Relative Clause Formation

,- NP
s
X - NP Yi
S NP
Z
-NP

OBLIG
1 2 3 4 5 6 >
1 2 4# [3 0 5] 6

Condition: 2 v: 4

The symbol '#' denotes the operation of Chomsky-adjunction, and the

brackets in the structural change indicate that the adjoined term is

not to be adjoined to term 3, but rather to the node which dominates

the sequence of terms enclosed in the brackets, in this case, the node
18
S. Thus (4.135) cony .7ts (4.136a) into (4.136b) .
 184

(4.136) a. NP

/
the 'oily:
,

NP
crn
V4

I V
/',ZN NP

saw the lay.

b. NP

NP

t'he 1.2sLy. NP

NP
"\\VP

I V

saw

And since (4.84) would prevent the circled NP node in (4.137) from

being raised and Chomsky-adjoined to the circled S by rule (4.135),

 185

sentences like (4.134) would be blocked.

(4.137) NP

the bob VP

and NT' embraced

Therefore, except for the possibility that the ungrammati-

cality of (4.134) can be explained by rule ordering, along the lines

suggested in footnote 17, it seems that it is necessary to formulate

such away that it becomes

the rule of Relative Clause Formation in

formally similar to the rule of Conjunction Reduction which Lakoff

and I have proposed. Both rules would have the effect of moving

and possibly
elements contained in conjuncts out of those conjuncts,

it is this formal property that the fact that they are both across-

the-board rules must be attributed to.

 186

4.2.4.3. There are other problems in grammar which are reminiscent

of the across-the-board application of the two rules just discussed.

These have to do with the necessity of excluding such sentences as

those in (4.139), while allowing those in (4.138).

(4.138) a. When did you get back and what did you

bring me?

b. (You) make yourself comfortable and I'll*

wash the dishes.

c. Did Mery show up 1::;.1 did you play chess?

(4 139) a. * Sally's sick and w at did you bring me?

b. * (You) make yourself comfortable and I

got sick.

c. * What are you eating or did you play chess ?19

At first glances, it might seem possible to distinguish

between (4.138a) and (4.139a) by claiming that the question Rule

must also be formulated in such a way as to Chansky-adjoin the

questioned element to the sentence which it is moved to the front

of. Support for such a proposal comes 7rom the fact that it is not

any more possible to question the NP the bov in (4.133) than it

was possible to relativize it.

(4.140) * Which boy and the girl embraced?

The facts of (4.134) and (4.140) are similar, and I

think that it is correct to maintain that the'Question'Rule must be

 187

reformulated in the same way as the rule of 'Relative Clause Formation

was reformulated in (4.135), so that the questioned element, too,

will be Chomsky-adjoined to the sentence. Also; since it seems

likely that yes-no questions should be derived from whether- clauses

whose initial element, after having been Chomski-adjoined, is later

deleted, sentence (4.141) 'could be excluded, while (4.138c) was

allowed.

(4.141) * I'm hungry an-, did you play chess?

orj

Promising though this approach seems, it is not capable

of being strengthened to account for a wide range of additional facts.

For instance, in Japanese questions, the questioned element is not

moved from its original position in the structure. Thus to question

the object of the verb mita 'saw' in (4.142),

(4.142) Zyoozyi wa sakana o mita.

George fish saw

'George saw a fish.'

it is sufficient to replace the word sakana 'fish' with the question

word nani 'what' and add the question, morpheme ka to the end of the

sentence, as in (4.143)

(4.143) Zyoozyi wa nani o mita Ica.

'What did George see?'

But the fact that (4.143) cannot be conjoined with a

declarative like (4,144), as the ungrammticality of (4.145) shows,

(4.144) neko ga nete iru.

cat sleeping is

'The cat is sleeping.

41
 188

mita ka (to)
(4,145) * Zyoozyi wa rani o mite neko
mi

ga nete iru.

loVhat did George see and the cat is sleeping.'

while two questions'can be conjoined (cf. (4.146)),

(4.146) Zyoozyiwa nani o mi aeko wa rani o tabetaka?

George what see cat what ate

'What did George see and what did the cat eat?'

indicates that the attempt to exclude sentences, some of whose conjuncts

are declaratives and others questions) by making the English rule of

Llpuestion an across-the-board rule cannot be a successful solution to

the problem in universal grammar of ensuring that only the "right kinds"

of sentences get conjoined. It would seem that the non-sentences of

(4.139) must therefore be excluded not by transformational constraints,

but rather by deep structural ones.

In fact, there is evidence within English which supports

this claim. Thus it seems that even questions like those in (4.147),

which contain more than one WH-word but presumably have no history of

reordering at all in their derivations, cannot be conjoined with

declaratives (cf. (4.148)), although they can be conjoined with normal

questions (cf. (4.149)):

(4.147) a, Who ate what?

b. What exploded when?

c. Who gave what to whom?

 Kwa WtrO 111411%.0

189

(4.148) a. Where did you go and who ate what?

b. What exploded when'md who was hurt?

c. How long did this fit of generosity last

and who gave what to whom?

(4.149) a. * I saw you there and who ate what?

b. * What exploded when and I warned you it

would?

c..* Who gave what to whom and I'm sickened at

this sentiment.

As far as I can see, cnly some kind of deep structure

constraint can be used to exclude (4.149). Moreover, the same is

true --ith respect to (4.138b). In one sense of this sentence, it is

synonymous with (4.150).

(4.150) If you make yourself comfortable, I'll

wash the dishes.

But there is another sense of (4.138b) which is a command, or a

suggestion; and if the word please is inserted into (4.138b), the

result has only this sense.

(4.151) (You) please make yourself comfortable

and I'll wash the dishes.

The fact that sentences like (4.139b) and (4.152) are ungrammatical

(4.152) *(You) please make yourself comfortable and

,

the cat is dead

I've studied Greek .

Jack left

. a ,
"I+ -
 190

cannot be accounted for by an appeal to some across-the-board rule

which has not applied to all conjuncts, because the only rule in

question, Imperative, only applies to the first conjunct to delete

the subject Ou. It therefore seems that only some deep structure

constraint on what tenses can be used in sentences which are

conjoined to commands can exclude (4.139b) and (4.152). Notice,

incidentally, that it is not in general the case that'if the first

sentence of a conjoined sentence is in the future tense all subsequent

conjuncts roust also be:

(4.153) Harry will be in the Marines next year

and Herman was drafted last night.

Exactly what the nature of deep structure constraints

on conjoined sentences is is an interesting topic which has been

studied far too little and which I can contribute nothing to at

present. Why, for instance should there be a difference between

(4.138c) and (4.139c)? Whatever the answer to this and similar

questions turns out to be, my basic point remains. valid: there are

both transformational and deep structural constraints which must

be formulated to apply to all conjuncts in a coordinate structure.

4.2.4.4. Sentences such as those in (4.154) raise problems which

may be related to across-the-board constraints.

 191

(4.154) a.. Sasha is gobbling down blintzes faster

than I can reheat them.

b. I want to peruse that contract before filing

it away.

c. Fred tore the curtain in roiling it up.

Although the sentences are so complex that positive

judgments are difficult to come by, I believe it to be the case that

when relative clauses are formed from the sentences in (4.154), both

the NP's blintzes, that contract and the curtain themselves and

their anaphoric pronouas may seem to be relativized at once, as is

the case in the sentences in (4.155).

(4.155) a.?? The blintzes which Sasha is gobbling down

faster than I can reheat are extremely

tasty, if I do say so.

b. ? I suspect thatthe contract which I wanted

to peruse before filing away may have

some loopholes.

c. The curtain which Fred tore in rolling up

was the kind gift of my maternal Aunt

Priscilla.

I believe it is theoretically possible to relativize

any number of NP's at once, although the resulting sentences are

somewhat less than felicitous: the a-sentences below have been

 192

converted into relative clauses in the corresponding bsentences.

(4.156) a. I want to peruse that contract before

damaging it while filing it .away.

b. ? The contra: which I want to peruse

before damaging while filing away is

written on Peruvian papyrus.

(4.157) a. ? I warlt to peruse that contract after

copying it by treating it in milk

while pressing it between two pieces

of marble in flattening it out.

b. ? *The contract which I want to peruse

after rlopying by treating in milk while

pressing between two pieces of marble

in flattening out is a beautiful piece

of art.

Whether or not such tortured constructions as this last

are to be accorded some degree of Englishness is not of great

importance for this study, since I ca101ot even propose a rule which
6
will generate less questionable examples, such as (4.155) and (4.156e).

What makes these sentences similar to the ones discussed in 5 4.2.4.2

above is the fact that not only does it seem possible to relativize

some NP simultaneously from a number of clauses, but it does not

seem possible to relativize an NP from only the second of these

clauses. Thus if the anaphoric pronouns of (4.154) are replaced by

---"""."4"..."ne"rynnrwrn......7-..7.-
 193

different NP, as in (4.158), these NP cannot be relativized, as

(4.159) shows.

(4.158) a. Sasha is gobbling down bliatzes faster

than I can reheat the fishballs.

b. I want to peruse that contract before

filing away the deed.

c. Fred tore the curtain in rolling up the

wallpaper.

(4.159) a. * I think Anita may have poisoned the

fishballs which Sasha is gobbling down

blintzes faster than I can reheat.

b. * The deed which I want to peruse that

contract before filing away is probably a

forgery.

c. ? *The wallpaper which Fred tore the curtain

in rolling up had a pleasing geometrical

pattern.

The sa.:t..1.1arity stops here, howe:er; for, bafflingly, it

is possible to relativize NP in just the first of these clauses

(cf. (4.160)):

(4.160) a. The blintzes which Sasha is gobbling down

faster than I can reheat the fishballs are

extremely tasty, if I dc say so.

 19'

b. I suspect that the contract which I want

to peruse before filing away the deed may

crspo InnphniA.q,

c. The curtain which Fred tore in rolling

the wallpaper up was the kind gift of

my maternal Aunt Priscilla.

Notice that it is similarly possible to relativize just

the NP's blintzes, that contract and the curtain in (4.154):

(4.161) a. The blintzes which Sasha is gobbling down

faster than I can reheat them are extremely

tasty, if I do say so.

b. ? I suspect that the contract which I

wanted to peruse before filing it away

may have some loopholes.

c. ? The curtain which Fred tore in rolling it

up was the kind gift of my maternal Aunt

Priscilla.

These facts suggest that it may be incorrect to attempt to derive

the sentences in (4.155) directly from (4.154) by some kind of

modified across-the-board rule. The sentences in (4.161) may be a

necessary first step in this derivation, with a rule of pronoun

deletion applying optionally to (4.161) to produce (44155). This idea

is given additional support by the fact that there are differences

in acceptability among the sentences of (4.155) which are exactly

reversed in the sentences of (4.161). That while (4.155a) is

far more awkward for me than (4.155b), which in turn is slightly

more awkward than the fully grammatical (4.155c), in (4.161), it

is the a-version which is fully grammatical, the b-- version which

is slightly doubtful, and the c-version which is the most dubious

of all. These differences can be accounted for if it is assumed

that the rule of pronoun deletion which transforms (4.161) into (4.155)

is obligatory in the case of (4.161c), optional in ahe case of

(4.161b), and not applicable in the case of (4.161a). This attempt at

explanation does not yet have much force, for I have no idea what

features of the environvent the optionality of this vile depends

upon, nor how to state the rule, but perhaps it is at least a correct

line of attack on this problem.

4.2.5, In summary, I have tried to show in the above sections

that Case F of §2.2 can be excluded by a constraint of geat

generality, the Coordinate Structure Constraint, which is needed

independently of the other constraints of this chapter. It is

more powerful than the A-over-A principle, which cannot exclude

sentences like (4.82). It can be used as a criterion for coordinate

structure, and on this basis, it was argued in § 4.2.3 that nodes

which are coordinate in deep structure may cease to be so in the course

 196

of a derivation and that nodes which appear to be coordinate in surface

structure may not be. The statement of the constraint in (4.84) was

...... Q.^ nron^evell. CINY 1410 canto. of thn !`'1 ACC

of across-the-board rules, which must operate in all conjuncts

simultaneously. A tentative hypothesis about the formal properties

of such across-the-board rules was advanced. At present, I know

of no rules which are not subject to the Coordinate Structure Constraint,

except for the rule of itive which I will discuss

in i 6.2.4 below, so I propose that this constraint be added to the

theory of grammar.

4.3. The Pied Piping Convention

4.3.1. In this section, I will suggest a constraint which can

successfully account for the evidence for the A-over-A principle which

was presented in case D and case E of § 2.2, and a convention which

will provide for the generation of all the relative clauses in the

sentences of (4.163). These must all be derived from (4.162), the

approximate structure of sentence (2.3), which I have repeated here, for

convenience.

(2.3) The government prescribes the height of

the lettering on the covers of the reports.

 ........."ofowrs,:-, ,"-----.7.1.--.- :=,:"«.-.7Z7 ".,..".="' -""'"'",

197

(4.162)

N1)

/IN
the government
1

rescr es NP

the heisia,

'the lettering,

"the covers

of. IILL929TU.

(4.163) a. Reports which the government prescribes

the height of the lettering on the covers

of are invariably boring.

,..4.7,7,...-..1.779.,171.74M+IRTF7717...1=

d'47+-27
 198

b. Reports the covers of which the government

prescribes the height of the lettering on

almost always put me to sleep.

c. Reports the lettering on the covers of.

which the government prescribes the height

of are a shocking waste of public funds.

d. Reports the height of the lettering on the

covers of which the government prescribes

should be abolished.

It can be seen that if the structure in (4.162) were

embedded as a relative clause modifier in a noun phrase whose head

noun is report, the rule of Relative Clause Formation, as it is

stated in (4.135), would only produce the relative clause in (4.163a).

If an attempt were made to modify the structural index of (4.135) in

such a way that the new rule would derive either (4.163a) or (4.163b)

from (4.162), the revised rule would be that.shown in (4.164):

0 - NP
(4.164) W NP - X - Y]S i Z
s NP
NP [NPNP P NP2NP
OBLIG
1 2 3 4 5 6

1 2[45]10 0 0 61 7

Condition: 2 m 5
 199

To derive the relative clause in (4.163c), the

further complication of the rule shown in (4.165) would be necessary,

1
r 1
0 - NP
(4.165) NP [ X-
S
[NpNP P - Y1 I
-S
Z
NP NP
NP P.
NP P NP11
NP NP
OBLIG

1 2 3 4 5 6

1 2[4 5]#[3 0 0 6] 7

Condition: 2 as 5

and deriving the clause in (4.163d) would entail adding a fourth line

to the disjunction inside the braces in (4.165). But since there is

no upper bound on the length of a branch consisting entirely of NP's,

like NP - NP in (4.162), in order to give a finite formulation

1 7

of this rule, which must be able to generate clauses like those of

(4.163) to any desired degree of complexity, either some abbreviatory

notation, under which the sequences of terms within the parentheses

of (4.164), (4.165), etc. can be collapsed, must be added to the

theory of grammar, or some special convention must be. Of these two,

the latter is weaker, for to add a new abbreviatory notation to the

theory is to make the claim that there are other cases, unrelated

to the case at hand, where rules must be collapsed according to the

new notation. No such cases exist, to my knowledge, so I propose the

 200

.convention given in (4.166) as a first approximation to an appropriate

universal convention.

(4.166) Any transformation which is stated in such

a way as to effect the reordering of some

specified node NP, where this node is

preceded and followed by variables, can

reorder this NP or any NP which dominates

20
it.

By the term "specified" in (4.166), X mean that node NP, in a branch

containing many NP nodes, which is singled out from all other nodes

on this branch by virtue of some added condition on the rule in

question, such as the condition on the rule of Relative Clause

Formation that the NP to be relativized be identical to the NP

which the clause modifies, or the condition on the rule of Question

that the questioned NP dominate WM-some. This convention, then,

provides that any reordering transformation which is stated as

operating on some NP singled out in some such way may instead operate

on any higher NP. Thus the formulation of Relative Clause Formation

which was given in (4.135), when supplemented by 0.166), will all^ti

for the adioining to the front of the sentence of the specified NP7,

the reports, or NP6, of the reports, or NP5, the covers of the reports,

etc., so that all of the clauses in (4.163) will be generated. That

(4.166) is too strong, in that it does not exclude the ungrammatical

sentences of (4.167) need not concern us here;

 201

(4.167) a. * Reports of which the government prescribes

the height of the Lettering on the covers

are invariably boring.

b. * Reports on the covers of which the

government prescribes the height of the

lettering almost alway put me to sleep.

c. * Reports of the lettering on the covers

of which the government prescribes the

height are shocking waste of public funds.

there seems to be a constraint, in my dialect at least, which prohibits

noun phrases which start with prepositions from being relativized and

questioned when these directly follow the NP they modify. Thus (4.168)

can be questioned to form (4.169a), but not (4.169b).

(4.168) He has books by several Greek authors.

(4.169) a. Which Greek authors does he have books by?

b. ?*By which Greek authors 2.oes he have books?

I will not attempt a more precise formulation of this restriction here:

instead, I will point out two further inadequacies in the formulation

of (4.166).

Firstly, if the structure shown in (4.170) were to be

embedded as a relativeclause on an NP whose head noun were'the boy,

 202

(4.170)

NP

watched

Bill 'the ha

the Coordinate Structure Constraint would not allow the formation

of (4.171)7

(4.171) * The boy who I watched Bill and was vain.

However, the circled node NP is dominated by the boxed node NP,

and convention (4.166) would allow this higher node to be preposed,

which would result in the ungrammatical (4.172).

(4.172) * The boy Bill and who(m) I watched was

vain.

The ungrammaticality of this sentence indicates the necessity of

revising (4.166) in such a way that if an NP dominating the specified

NP is coordinate, neither it nor any higher NP can be moved. I will

incorporate such a revision into the final version of the convention,

which will be stated in (4.180).

 203

The second inadequacy of (4.166) can be seen in connection

with P-marker (4.173).4.

(4.173)

VP

won't like the hat

while it is true that the circled node NP can be.relatiized as

(4.174) shows,

(4.174) They will give me a hat which I know

that 1 won't like.

once again, (4.166) would allow the preposing of the boxed node

NP, and the ungrammatical (4.175) would be produced.

 204

(4.175) * They will give me a hat that I won't

like which I know.

The modification of (4.166) that seems to be revired here is that

if a branch of a P-marker has an occurrence of the node S intervening

between two occurrences of the node NP, only the lower one can

be reordered. This restriction does not extend to the node VP,

however, as can be seen from the following example.

The approximate structure of the German sentence in

(4.176) is that shown in (4.177).

(4.176) Ich babe den Hund zu finden zu versuchen angefangen.

I have the dog to find to try begun

'I have begun to try to find the dog.'

(4.177)

ich

VP anciefanc,e1

NP

i2 zu versuchen

NP

en hund zu rrnaen O

PmerrITWIM.prow
 205

If the structure which underlies (4.177) has been

embedded as a relative clause on the subject. NP of the structure

-1-J-- // 1'70\
UnUeL3.y.i.u6 vt J../Uvit
1

(4.178) Der Hund ist ein Bernardiner.

'The dog is a St. Bernard.'

the rule of Relative Clause Formation must produce all three of the

clauses in the sentences of (4.179).

(4.179) a. Der Hund, den ich zu finden zu versuchen

angefangen habe, ist ein Bernardiner.21

b. Der Hund, den zu finden ich zu versuchen

angefangen habe, ist ein Bernardiner.

c. Der Hund, den zu finden zu versuchen ich

angefangen habe, ist ein Bernardiner.

'The dog which I have begun to try to

find is a St. Bernard.'

In (4.179a), only the specified node, NP3 in (4.177), has been

preposed, while in (4.179b), the phrase dominated by NP2, which

contains NP , has been preposed, and in (4.177c), the largest NP,

3
NP.. had been preposed. Note that these three NP nodes are separated
1-

by two VP nodes in (4.177), but that (4.166) still is operative. This

then indicates that it is only the node S, as was claimed above, to

which reference must be made in revising (4.166)."

In (4.180), I have modified the convention given in

(4.166) in such a way as to overcome the two inadequacies I have just

 206

discussed.

(4.180) The Pied Piping Conv2ntion23

1.
Any Lransfotma4ioa which is stated in

a way as to effect the reordering of some

specified node NP, where this node is

preceded and followed by variables in the

structural index of the rule, may apply to

this NP or to any non-coordinate NP

which dominates it, as long as there are

no occurrences of any coordinate node, nor

of the node S, on the branch connecting'

the higher node and the specified node.

A
A 4.3.2.0. The convention stated in (4.180) stipulates that any NP

above some specified one may be reordered, instead of the specified

one, but there are environments where the lower NP may not be moved,

and only some higher one can, consonant with the conditions imposed in
24
(4.180). In other words, pied piping is obligatory in some contexts.

In § 4.3.2.1, I will describe two environments in which pied piping

is obligatory, whether the specified NP is to be moved to the right

or to the left, and in § 4.3.2.2, I will cite several environments

in which pied piping cannot apply. In § 4.3.2.3, I will discus the

one environment I know of in which pied piping is obligatory if an NP

 207

Is moved in one direction, but not if it is moved in the other. In

§ 4.3.2.4, I will show how the constraints on pied piping developed

in thc.cp ,---pprinng interact with the rule of Coniunction Reduction,

and in § 4.3.2.5, I will explore the question of the theoretical

status of the various conditions on (4.180) which are discussed

in §§ 4.3.2.1 -

4.3.2.1. For English, and for many other languages, the following

constraint, which has the effect_ .of _making pied piping obligatory

in the stated environment, obtains:

. (4.181) The Left Branch Condition,

No NP which is the leftmost constituent of

44

a larger NP can be reordered out of this NP

by a transformational rule.

In other words, (4.181) prohibits the NP shown in (4.182) from

moving along the paths of either of the arrows.

(4.182) (NP X]Np

This constraint accounts for the following facts: if

the structure shown in (4.183) is embedded as arelative clause modifier

of a NP whose head noun is boy, only one output is possible --

(4.184a)
 208

(4.183)

NP VP

we NP

elected NP president
2

employer

the N duardian rs

(4.184) a. The boy whose guardian's elppl(,.17-we

elected president ratted on us.

b. * The boy whose guardian's we elected

employer president ratted on us.

c. * The boy whose we elected guardian's

employer president ratted on us.

Sentence (4.184c) is excluded by (4.181), because the

rule of Relative Clause Formation has moved the lowest NP,

from the left branch of NP1. In (4.184b), it is NP2 that has

been moved filma this branch. Since the Left Branch Condition
 209

prohibits both of these operations, only the largest NP which

(4.180) allows to be moved, NP1, can be moved to the front of

the whan this hnppArlq, (4' 184x) is the result.

Parallel facts can be adduced for non-restrictive

relative clauses, which differ from restrictives in being preceded

and followed by heavy intonation breaks. They derive 'from coordinate

sentences in deep structure, and they are formed by a different

rule than (4.135). If commas are inserted into the sentences of

(4.184), after box and j.ivesLzatcl, thus forcing a non-restrictive

interpretation of the clauses, their grammaticality is unchanged.

Another rule which is affected by this condition is the

rule of Topicalization, (4.185), which converts (4.186a) to (4.1.86b).

(4.185) Topicalization

X - NP Y
OPT
1 2 3 ------->

21/(1 0 3]

(4.186) a. I'm going to ask Bill to make the old

geezer take up these points later.

b. These points I'm going to ask Bill to make

the old geezer take up later.

If rule (4.185) is applied to (4.183), once again it will

be seen that only NP1 can be topicalized, as in (4.187a). If either

NP or NP is topicalized, as in (4.187b) and (4.187c), respectively,

2 3
 210

ungrammatical sentences result.

(4.187) a. The boy's guardikin'E employer we elected

vvr,:::104 'tart r

b. * The boy's guardian's we elected employer

president.

c. *.The boy's we elected guardian's employer

president.

A rule that was stated in (3.26), Complex NP Shift,

which performs almost the same operation as. (4.185), except that it

moves the NP in the opposite direction, is also subject to the

Left branch Condition. This rule may apply to (4.183) to move

NP1 over president (cf. (4.188a)),25 but neither NP2 nor NP3

can be so moved, as the ungrammatica:.. of (4.188b) and (4.188c)

demonstrates.

(4.188) a. We elected president the boy's guardian's

employer. N
b. * We elected employer president the boy's

guardian's. 1
c. * We elected guardian's employer president

the boy.

Finally, the Question Rule is subject to the condition:

if NP je .A.183) is questioned, it cannot be moved to the front

3

of the sentence alone -- pied piping must apply to carry NI', with

it, as (4.189) shows.

 211

(4.189) a. Which boy's guardian's employer did we

elect president?.

.11 I -14*
Do waxn Doy
il
_
gudEu.t.du we

employer president?

c. * Which boy's did we elect guardian's

-employer president?

One of the facts which supports the analysis of predicate

adjectives which is implicit in diagram (3.25) above is the fact that

when adverbs of degree which occur in pre-adjectival or pre-adverbial

position are questioned, the questioned constituent, how, cannot be

moved to the front of the sentence alone, as in (4.190a) and (4.191a) ,

but only if the adjective or adverb is moved with it, as in (4.190b)

and (4.191h).

(4.190) a. * How is Peter sane?26

b. *How sane is Peter?

(4.191) a. * How have you picked up TNT carelessly?

b. How carelessly have you picked up TNT?

These facts can be explained by (4.181) , if how is analyzed as deriving

from an underlying NP, and the adjective sane and the adverb

carelessly. are dominated by NP at the stage of derivations at

which questions are formed. Note also that if the degree adverb

that in (4.192) is questioned, pied piping must apply to move not

only tall, but also a man to the front of the sentence.

 212

(4.192) Sheila married that tall a man.

(4.193) a. How tall a man .did Sheila marry?

A4,1 maryv a man?

c. * How did Sheila marry tall a man?

These facts are accounted for if the structure of (4.193a) at the

point when the aleakejla applies is that shown in (4.194),

(4.194)

Sheila

married NP NP
2

R
a man
'

WHi-some extent tall

for (4.181) will not permit either NP3 or NP2 to be moved out

of NP .
1
One other set of facts deserves mention in connection with

this analysis of adjectives. In German, it is possible to topicalize

 213

adverbs -- thus the manner adverb genUsslich 'with pleasure' in

(4.195a) can occur at the front of the sentence, as in (4.195b).

II. Inc\
ve..4.7a, a. aueu DULL/L=LL 6eLLu:).L.J. liesh.Qu.a.u/L6=LL.

we have the beans with pleasure gobbled up.

'We gobbled up the beans with pleasure.'

GenUsslich haben wir die Bohnen verschlungen.

If an analysis in 'which adverbs are treated as being derived from NP

can be maintained, not only will it be unnecessary to complicate rule

(4.185) to' derive (4.195b) from the structure which underlies (4.195a),

but it will be possible to explain the following facts in addition.

In German, the adverb fast 'almost' normally precedes the adjective

it modifies, but it can follow it (cf. ;4.196)). The adverb sehr,

'very', however, only occurs pre-adjectivally (cf. (4.197)).

(4.196) a. Walburga ist fast hUbsch.

'Walburga is almost pretty.'

b. Walburga ist hUbsch, fast.

(4.197) a. Liselotte ist sehr hubsch.

'Liselotte is very pretty.'

b. * Liselotte ist hUbsch, sehr.

These facts suggest that whatever rule it is that moves

fast around hUbsch in (4.196) be made obligatory for degree adverbs

like sehr. If this reordering rule adjoins the adverbs which are

moved around the adjectives to the adjectives, and if this re :dering

rule precedes the rule of T( calization, the fact that fast can be
 214

that
topicalized with or without hUbsch (cf. (4.198)), bute,sehr cannot be

topicalized by itself (cf. (4.199)) is accounted for by the Left

Branch Condition.

(4.198) a. Fast hUbsch ist Walburga.

b. Fast ,ist Walburga hUbsch.

(4.199) a. .Sehr hUbsch ist Liselotte.

b. * Sehr ist Liselotte hUbsch.

Of course, it is possible to account for these facts

concerning adjectives and adverbs in other ways than by assuming

that both types of constituents are dominated by NP up to some

point in derivations, but the analysis sketched here has the virtue

of allowi g a simpler statement of the rules of apicalization and

Question and of constraints like (4.181) than can otherwise be achieved,

as far as I can see. However, since I have not made a detailed study

of adverbs, it may be the case that this analysis will have to be

excluded because it engenders complications in other parts of the

grammar.

In passing, it should be noted that Case D and Case E

of § 2.2, which provide evidence for the A-over-A, principle, are

special cases of the Left Branch Condition, which will block the

derivation of the ungrammatical (2.11) and (2.15).

Another environment in which pied piping is obligatory in

German, French, Italian, Russian, Finnish, and in many other languages,

is that stated in (4.200).

 (4.200) No NP may be moved out of the environment

EP ]NP'

In these languages, only sentences like (4.201) are

possible -- sentences corresponding to those in (4.202), where a

NP has been moved away from its preposition, are ungrammatical.

(4.201) a. On which bed does Tom sleep?

b. The bed on which Tom slept was hard.

(4.202) a. Which bed did Tom sleep on?

b. The bed which Tom slep on was hard.

Kuroda has pointed out similar facts for English with

respect to a certain class of nouns (cf. Kuroda (1964)). Kuroda

pointed out that it is just with the class of nouns that cannot be

pronominalized, i.e., nouns like time, al, manner, place, etc.,

that sentences like (4.202)are impossible. That is, the sentences

in (4.203) cannot be converted into the corresponding ones in (4.204)

by normal rules of pronominalization.

(4.203) a. My sister arrived at a time when no busses

were running, and my brother arrived at

a time when no busses were running too.

b. Jack disappeared in a mysterious manner and

Marian disappeared in a mysterious manner too.

c. I live at the place where Route 150 crosses

Scrak River and my dad lives at the place

where Route 150 crosses Scrak River too.

 216

(4.204) a. * My sister arrived at a time when no

busses were running and my brother

arrived at oue too.

b. * Jack disappeared in a mysterious manner

and Marion disappeared in one too.

c. * I live at the place where Route 150 crosses

Scrak River and my dad lives at it too.

Furthermore, prepositions cannot be left behind in such constructions

either (cf. (4.205)).

(4.205) a. * What time did you arrive at?

b. * The manner which Jack disappeared in was

creepy.

c. * The place which I live at is the place

27
where Route 150 crosses Scrak River.

The facts indicate that though the constraint in (4.200)

does not obtain for English, the modified version shown in (4.206)

does:

(4.206) No NP whose head noun is not pronominalizable

may be moved out of the environment (P ]NP

The three constraints discussed in this section - (4.181),

(4.200), and (4.206) - are all cases where the optionality which is

built into (4.180) is abrogated in favor of higher NP nodes. That

is, if NP dominates NPj, (4.180) in general allows either NP to

 217

reorder, but the above three constraints limit this freedom: they

state environments in which only the higher NP can reorder. In

the next section, I will discuss two constraints which have the

opposite effect.

4.3.2.2. After most verb-particle combinations whose object is a

prepositional phrase, such as do away with, make up to, sit in on,

,get away with, etc., while the NP in the prepositional phrase is

movable, the preposition may not be lamed with it. Thus though

the sentences in (4.207) are possible, corresponding ones in (4.208)

are not.

(4.207) a. The only relatives who I'd like to do

away with are my aunts.

b. Who is she trying to make up to now

c. That meeting I'd like to sit in on.

(4.208) a. * The only relatives with whom I'd like to

do away are my aunts.

b. * To whom is she trying to make up now?

c. * On that meeting I'd like to sit in.

For some reason which I do not understand, there are

other verbs which seem to be of exactly the same syntactic type for

which such constructions as (4.208) are permissible. Thus the sentences

in (4.209) are markedly better, for me, than those in (4.208).

 ...,A

218

(4.209) a. ? The abuse with which she puts up is phenomenal.

b. For whose rights do you expect me to speak up?

F^r pr4n^4ploa I havel npAppbr hpaltntAd

to speak out.

Similar facts obtain for such syntactic idioms as get wind

of, make light of, get hold of, etc. Normally, in my speech at

least, the preposition must be left behind for most of these idioms --

compare (4.210) and (4.211).

(4.210) a. One plan which I got wind of was calculated

to keep us in suspense.

b. Did you notice which difficulties she made.

light of?

c. Who are you trying to get hold of?

(4.211) a. * One plan of which I got wind was calculated

to keep us in suspense.

b. ?*Did you notice of which difficulties .she

made light?

c. * Of whom are you trying to get hold?

However, there are certain of these syntactic idioms for

which the preposition seems to be movable, just ad was the case with

the verb-particle combinations shown in (4.209).

R
 219

(4.212) a. The only offer of which I plan to take

advantage will give me an eleven -month paid

varnr4on.

b. 7 In the countries of which I've been keeping

track, the existing political systems are

fantastically corrupt.

c. The scenes to which the censors took

objection had to do with the mixed marriage

of a woman and a giant panda.

I believe that sentences like those in (4.209) and

(4.212) are the exception, rather than the rule, so presumably some

constraint like (4.213) must be stated for English.

(4.213) No NP with the analysis [P liP412 may

be moved if it follows an idiomatic V -. A

sequence, where A is some single constituent.

The constituent A may 'be a particle (cf. (4.207) and

(4.208)), an adjective (as in make light of, make sure of, etc.), a

verb (as in make do with, let fly at, let go of, lathold of, yet rid of

(if rid should be analyzed as a verb here)), lay claim to, hold sway over,

siege to,
Pay heed to, etc.), a noun (as in get wind of, set fire to,. lay

make use of, lose track of, tekeshatot, take umbrae:7e at, etc.), or

possibly a noun phrase (e.g., ,get the droacal, make no bones about,

set one's sights on).

 a A.V..4.17. . .......

220

There is a possibility, as Paul Kiparsky has pointed out

to me, that the difference between (4.211) and (4.212) may correlate

with whether the idiom in question has a single or a double passive.

That is, in many cases, verbs like those in (4.212), where the

preposition may be moved, allow either the first element after the

verb or the object of the preposition to become the subject of the

passive.

(4.214) a. Advantage will be taken of his offer.

b. His offer will be taken advantage of.

(4.215) a. ? In this experiment, track must be kept of

fourteen variables simultaneously.

b. In this experiment, fourteen variables

must be kept track of simultaneously.

(4.216) a. Objection was taken to the length of our

skirts.

b. ? The length of our skikts was taken objection

to.

The sentences of (4.214) - (4.216) attest to the fact

that the syntactic idioms of (4.212),. whose preposir$,ms are not

subject to (4.213), have double passives. But the idioms in (" .%0),

whose prepositions are shown to be subject to (4.213) by the v .grammati-

cality of (4.211), have only one passive, as can be seen f2-011 the

ungrammaticality of the a-versions of sentences '(4.217) - (4.219).

a
 221

(4.717) a. * Wind was gotten of a plot to.negotiate

an honorable end to the war in Vietnam.

b. A plan to negotiate an honorable end to

the war in Vietnam was gotten wind of.

(4.218) a. * Light was made of her indiscretions.

b., indiscretions were made light of.

(4.219) a. * Hold has been gotten of some rare old

manuscripts.

b. Some rare old manuscripts have been

gotten hold of.

The correspondence between the class of syntactic idioms

which allow passives like those in (4.214a), (4.215a), and (4.216a),

and the class of idioms whose prepositions' are not subject to (4.213)

is too close to be merely coincidental, but for me, at least, it is

not exact. If it were, the differences in acceptability between the

a and b-sentences below would not exist.

(4.220) a. Use was made of Sikolsky's pigeon-holing

lemma.

b. The lemma of which I will make use is due

to Sikolsky.

(4.221) a. Tabs were kept on all persons entering

the station.

b. ??The persons on whom we kept tabs all

proved to be innocent.
 222

. ."-

(4.222) a. * Faith was had in all kinds of people.

b. ? The only representative in whom I have

faith is still in the Bahamas.

But I have not made a close study of all cases which

run counter to Kipirsky's suggestions to see if they can be explained

away. I believe that it will eventually become possible to incorporate

this suggestion into a revised version of (4.213), even though I am

unable to do so now. But it is clear that some other explanation

mug be devised for the sentences of (4.209), which also constitute

counter-evidence to (4.213). The whole problem of what syntactic

properties various types of idioms have has been neglected grievously, --

I suspect that intensive research into this .problem would yield rich

rewards for many areas of syntax besides this one.

In Danish, there are many environments in which pied

0
piping is blocked. Thus, while the preposition EL 'in' can be left

behind or moved to the front of the sentence, when a manner adverb

is questioned (cf. (4.223)),

(4.223) a. Hvilken made gjorde han det pa?

which way did he it in

'How did he do it?'

b. pa hvilken made gjorde han det?

In which way did he it.

prepositions in a prepositional phrase which is immediately dominated

by VP can never be moved to the front of the sentence: (4.224c) is

I
,
4

...-T.,-

i
 223

ungrammatical.

(4.224) a: Han fandt pa den historie.

ha 4nvantorl t4Int- Qrnry

b. Hvilken historie fandt han pa?

which story invented he

'Which story did he invent?'

c. * pa hvilken historie fandt han?

This'means that in the grammar of Danish, the following condition

must be stated:

(4.225) No NP with the analysis (P NP]

may be moved if it is immediately

dominated by VP.

The full set of facts in Danish is quite a bit more complex -- a

more detailed presentation is given in Blass (1965). I will not

attempt a recapitulation of all the facts of Danish, for my purpose

here is not to suggest a complete analysis of all constructions

involving prepositions in Danish or in English, but merely to

demonstrate that just as there are environments where pied piping is

obligatory (cf. § 4.3.2.1. above), so there are environments where

it must be blocked.

4.3.2.3. The first condition on pied piping, (4.181), prevents

the reordering of an NP on a left branch of the larger NP, no

matter in which direction the NP being reordered is to move. Thus

 224

neither the rule of Topicalization, which moves noun phrases to

the left, nor the rule of Comnlex NP Shift, which moves them to

the right, can apply to NP3 or NP2 in tree (4.183), as the

ungrammatical sentences of (4.187) and (4.188) demonstrate. And

the same is true of the other conditions stated in § 4.3.2.1 --

(4.200) and (4.206). The first of these asserted that it is

impossible to "strand" a preposition in German, and various other

languages, by moving its object NP away from it. Thus, in

German, when the NP diesen Fasten 'this box' in (4.226a) is

questioned, it cannot be moved to the front of the sentence O

alone, as would be possible in English, (cf. the ungrammaticality

of (4.226b)). When the Question Rule applies, (4.200) requires

that the larger NP, in welchen Kasten, 'into which box' be .

preposed, as it is in (4.226c)

(4.226) a. Vladimir wollte das Buch [in (diesen

Vladimir wanted the book into this

Kasten] ] schmeissen.
NP Np

box throw.

'Vladimir wanted to throw the book into

this box.'

b. * Welchen XAsten wollte Vladimir das Buch

Which box wanted Vladimir the book

in schmeissen?

into throw?

I
 225

c. In welchen Kasten wollte Vladimir das

into which box. wanted Vladimir the

%in}, cni-min4ccestl? 11

book throw

'Into which box did Vladimir want to throw

the book?'

Just as it is impossible to strand a preposition in

German by moving its object NP away from it t) the left, so it is

impossible to do so by moving the NP to the right. An example

of a rule which moves NP to the right in German is the rule which

converts sentences like (4.227a) into ones like (4.227b), which,

though marginal, must be generated.

(4.227) a. Er wollte denen ein wunderbares Bilderbuch geben.

he wanted to them a wonderful picture book give.

'He wanted to give them a wonderful picture book.

b. Er wollte denen geben ein wunderbares Bilderbuch.

This rule corresponds roughly to the English rule of Luplex NP Shift,

although the English rule is not so restricted as the German one. Since

I have not studied the conditions under which such sentences as (4.227b)

can be produced, I will not attempt a precise statement of the rule

here; the formulation of Complex NP Shift which was given in (3.26)

is adequate for my present purpose.

Note that Conlex NP 'Shift, if applied to (4.226a), can

only move the larger NP, in diesen Kasten (cf. (4.228)). If the
 226

object of the preposition is moved, the impossible (4.228b) results.

(4.228) a. Vladimir wollte'das Buch'schmeissen

in diescn Kasten,

b. * Vladimir wollte das Buch in schmeissen

diesen Kasten.

This shows that (4.200), just like (4.181), constrains transformations

which move NP' to the right, as well as those which move NP to

the left.

In English, however, we find a different situation.

While prepositions may be stranded if their object NP is moved

to the left, they may not be if it is moved to the right. The rule

of Tobicalization may strand the preposition to of (4.229a), as in

(4.229b) or it may take it along, as in (4.229c).

(4.229) a. Mike talked to my friends about politics

yesterday.

b. My friends Mike talked to about politics

yesterday.

c. To my friends Mike talked about politics

yester:ay.

But Complex cannot apply to the NP a friends in (4.229a):

it can only apply to the larger NP to my

(4.230) a. Mike talked about politics yesterday to

my friends.

b. * Mike talked to about politics yesterday

my friends.
 227

Thus it can be seen that the theory of grammar must be

strengthened so that conditions making pied piping obligatory or

impossible can make reference to the direction in which the specified

NP is to be reordered. It will be necessary to add to English

condition (4.231), which is a weaker form of (4.200).

(4.231) No NP may be moved to the right out

of the environment (P 11,11).

It might appear that (4.213) would have to be modified

along these lines, in the light of such sentences as those in (4.232),

(4.232) a. ? They got wind, eventually, of the counter-

plot to fluoridate the bagels.

b. ? Carrie did away, systematically, with her

nephews from Chattanooga.

c. ??She made light, not too surprisingly, of

the difficulties-we might have at the

border.

d. ? I got hold, fortunately, of Lady Chatterley's ex.

for superficially at least, the prepositional phrases which follow V - A

syntactic idioms of the type discussed in connection with (4.213) seem

to have been moved, possibly by the rule of Complex NP Shift. I

suspect, however, that (4.213) does not have to be modified and that

some other rule than Complex NP' Shift is being used in the generation

of the sentences in (4.232). The rule in question is probably -related

to the Scrambling Rule, (3.48); it allows sentence adverbs to be

 228

28
positioned between any major constituents of a clause. Note

that the sentences in (4.232) are almost totally unacceptable if

the commas are removed, but that no commas are necessary in such

clear cases of Complex NP Shift as (4.233).

(4.233) I gave to the officer in charge the

blackjack. which I had found in the

cookie jar.

The sentences in (4.232) thus seem to be accountable for

by other means than assuming the existence of a second condition on

pied piping like (4.231), a condition in which the direction of

reordering would make a difference. So although I know of no

other facts which motivate the postulation of any other direction

dependent conditions, the .facts discussed in connection, with (4.231)

seem to require, at least for the present, a theory of language

in which such conditions can be stated.

4.3.2.4. In this section,.I will point out one puzzling fact

about the interaction between the rule of anlynction Reduction and

two of the conditions on pied piping which were discussed above --

the:Left Branch Condition and (4.231).

In 5 4.2.4.1, I gave a brief, informal description of

the rule which converts (4.118) into (4.119). Since the adjective

appears on a right branch of both conjoined sentences in

(4.118), it can be raised and Chomskyadjoined to the coordinate node

 229

by the rule of Reduction. The same is true of the two

occurrences of the NPa successful outinlatitketrack in (4.234),

of 2_19

(4.234)

a nc1S

NP VP.

am confident I NP

I-1
NP ,depends 4
i.

"of

'a successful outin at the'track

/MO
a'successful outin: at the trac

(4.235) i am confident of y and my boss depends on,

a successful outing at the track.

Since (4.235) 1.4 grammatical, some condition must be built

into (4,231) which weakens it so that it does not affect the operation

of the rule of Con unction Reduction. As (4.231) is ncwstated, it

would prevent the circled NP nodes in (4.234) from being raised,

for they are contained in the boxed NP nodes, which start with

_
 230

prepositions. I do not understand why (4.231) should not constrain

Conjunction Reduction, for it is not in general true that conditions

nn .mgerl yv4wfilft fin

r- r-r--b
not nywOar
"rr"-J
to rein4tinntinn PaAtirtinn" 7 fnllnwino
--co

example will show.

Up to this point, I have only discussed examples of the

operation of Conjunction Reduction where_the identical constituent

was on a right branch, but the rule will also work on constituents

which appear on left branches. Thus in (4.236), the circled noun

phrases can be Chomsky-adjoined to the coordinate node -- the result

is sentence (4.237).

(4.236)

an

VP

NP are intelligent are committed to

freedom

he University s. students

the University

(4.237) The University's students are intelligent

, 29
and (are) committed to freedom.
 231

'But note that if the input structure 'is; that shown

in (4.238), Conjunction Reduction must be blocked.

(4.238)

an

are intelligent is committe to

freedom

the University's students

the University's faculty_

The only identical nodes in (4.238) are the two occurrences

of the boxed NP the University's. If Conjunction Reduction is allowed

to apply to these nodes, the ungrammatical (4.239) results:

(4.239) * The University's students are intelligent

and faculty is committed to freedom.

It is not necessary to add any condition to the rule of

Conjunction Reduction to avoid generating (4.239): the Left Branch

Condition, (4.181), will prevent the boxed NP's in (4.238) from being

raised, because each is on the left branch of a larger 1114. These facts

are indicativaclearly that it is not in general the case _hat conditions

 232

on pied pipirig are aot in effect for the rule of Conjunction Reduction,

50 it will be necessary to add a clause to condition (4.231), stating

that this particular condition does not apply to the rule of

Conjunction Reduction.

For some reason, there is one environment in which (4.181)

also behaves idiosyncratically with respect to Con unction Reduction --

even though the constituents to be raised are on the left branches of

larger NP's, these constituents can be raised, if the larger NP's

are conjuncts of a coordinate NP. For example, the two circled'

NP nodes in (4.240) can be raised and adjoined to the boxed NP

node, yielding (4.241).

(4.240)

vp

P kisstlE

the boys uncle the boz's aunt

(4.241) The boy's uncle and aunt were kissing.

 - -e

233

It is not necessary that the NP being raised and

adjoined be immediately dominated by a conjunct: the NP shown

in (4.242a) can be reduced to the one shown in (4.242b), by raising

the two occurrences of the NP the

the boy's aunt's

b.

grandmother

aunt's
 234

I can think of no explanation for this strange fact --

It will simply have to appear as an ad hoc rider on (4.181). However,

this rider can be used to explain the otherwise extremely puzzling

-1 /A ni.,3-N .1-,
A4GXe-,..-
u4 &cu.. ucuwell. A.rie traTiMat.A.Ccu. v.1.4.1J4, and cam. ungrammarLed.i.

(4.243b).
..

0.243) a. The boy wnose uncle and aunt's grandmother

were kissing was furious.

b. * The boy whose uncle and Tom's aunt's

grandmother were kissing uas furious.

The relative clause in (4.243a) comes from a sentence

whose subject is the NP shown in (4.242a). If Conjunction Reduction

applies before Relative Clause Formation, thus converting (4.242a)

into (4.242b), then the circled NP the boy's in (4.242b) will be

relativizable, because it will then no longer be contained in a

coordinate structure. Since it is on the left branch of the boxed

NP, when it moves, this larger NP will pied pipe with it, as (4.181)

requires.

But the relative clause in (4.243b) would have the NP

shown in (4.244) as its subject:

 235

(4.244) O

grandmother

NP

Tom's aunt's

Since the circled NP in this tree does not occur in all conjuncts,

the rule of Conjunction Reduction cannot apply to it. Therefore, when

relativization of this NP is attempted, (4.181) will specify that

the boxed NP node in (4.244) must pied pipe, for the NP being

relativized is on its left branch. But the boxed NP is a coniunct,

and thus cannot be moved, by virtue of the Coordinate Structuxa

Constraint, (4.84). And since there f.s a clause in the Pied Piping

Convention, (4.180), which specifies that coordinate nodes cannot

pied pipe (recall the ungrammaticality of (4.172)), the top NP node

of (4.244) will not pied pipe either. Thus the circled NP node is

frozen solidly in position -- (4.181) prevents it from reordering, and

the way (4.84) and (4.181) have been stated prevent any NP node
 236

above it from pied piping -- so the rule of Relative Clause Formation,

if it applies to this circled NP, will Produce an ungrammatical

sentence. The contrast between the sentences in (4.243) is thus only

OD be explained on the basis of quite far - reaching theoretical

constructs.

4.3.2.5. What is the theoretical status of constraints like

(4.11), (4.200), (4.206), (4.213), (4.225) and (4.231)? It is

obvious that (4.200), which prohibits the stranding of prepositions,

is not universal, for prepositions may in general be stranded in

English. (4.206), which prevents the stranding of prepositions the

head of whose objects is not pronominalizable, is not universal

either, for prepositions can be stranded in this environment in

Danish, as (4.223a) shows. (4.225) is not universal, for the

prepositions of English prepositional phrases directly dominated by

VP can be stranded, as (4.245) shows.

(4.245) Who are you gawking at?

It may be that (4.231) is universal -- I know of no counterexamples

at present.

The Left Branch Condition, although it is in effect in

such languages as English, German, French, Danish, Italian and Finnish,

is not universal, for it is not in effect in Russian and Latin. In

Russian, the possessive adjective culu 'whose' can be preposed in

questions, whether or not the noun it modifies appears with it at

 237

the front of the sentence.

(4.246) a. Cuju knigu ty Clitajek?

Whose book you are reading

'Whose book are you reading?'

v
b. Cuju ty citajes knigu?

WIlose you are reading book

'lealose book are you reading?'

The same applies to the interrogative adjective skolko 'how rcany't

as can be seen in (4.247).

(4.247) a. Skolko let a nim .by11.2

how many years to him were

'How many years old was he?' (=how many

years did he have?)

b. Skolko u nim byli let?

how many to him were years

'How many years old was he?'

In Latin, too, sentences which parallel (4.246b) can be found

cf. (4.248).

(4.248) Cuius legis librum?

whose you are reading book

'Whose book are you reading?'

As far as I know, it is only in highly inflected languages, in whose

grammars the rule of Scramblins. appears, that the Left Branch Condition

is not operative, but it is not the case that it is not operative ia

 238

all such languages. In Finnish, for example, sentences like (4.248)

are not possible. At present, therefore, .1 am unable to predict

when a language will exhibit the Left Branch Condition and when not.

Thus it appears that with the possible exception of

(4.231), all of the constraints on pied piping which were discussed

in §i 4.3.2.1 - 4.3.2.4 must be stated in the grammar of each language

that exhibits them. But must each such condition be stated on each

rule which it influences? Must the Left Branch Condition be built

into the English rules of Relative Clause Formation, Appositive Clause

Formation, Topicalization,le) and auestion? To

repeat the Left Branch Condition on each of these five rules is to

make the claim that it is an accidental fact about this particular

set of five rules that they are all subject to (4.181).. I am making

the App site claim: that any reordering transformation would be

subject to (4.181) . To reflect this claim formally, the theory of

grammar must be changed. At present, the theory only permits conditions

which are stated on particular rules, like the identity condition

on Relative Clause Formation, or meta-conditions, like the Complex NP

Constraint, which are stated in the theory. But the constraints on

pied piping which are under discussion cannot be correctly accommodated

under either of these possibilities: they are not universal, and to

state them on each transformation which they affect is to miss a

generalization. What is necessary is that the grammar of every natural

language be provided with a conditions box, in which all such language-

 239

particular constra2,mts are stated once fol. the whole language. By

a univercal convention of interpretation, all conditions in the

4onditions box will be underJtood to be conditions on the operation

of every rule in the grammar.

To give some concrete examples, for English, the

conditions box will contain, among others, (4.181), (4.206) , (4.213)

'and (4.231). For French, Italian and German, it will contain (4.181) ,

(4.200) and (A.231). It should not be thought that only conditions

on pied piping will appear in this box. In FiLnish, for example, it

is the case that no element can be moved out of complement clauses

which are introduced by etta 'that'. That is, while such sentences

as (4.249a) are possible in English, no corresponding sentence is

possible in Finnish, as the Ungrammaticality of (4.249b) shows.

(4.249) a. Which hat do you believe (that) she never

wore?

b. * mita hattua z uskoit ettei ham

which hat you believed that not she

koskaan kayttanyt?

ever used.

Thus far, with one exception, which I will discuss in

footnote 15 of Chapter 5, all the constraints which I know to appear

in the conditions bc-A. of any language are constraints on reordering

transformations, but there is of course no reason to expect that no

 240

other types of constrai-its will be found to occupy condition boxes in

other languages.

4.3.3. To recapitulate the discussion of pied piping, the

existence of structures like (4.162), which allow for an in principle

unbounded number of relative clauses to be formedrclearly indicates

the need for a convention of some sort. Rather than devise some

notational convention under which an infinite family of rules like

those in (4.135), (4.164) and (4.165) could be abbreviated by some

sort of finite schema -- a notational convention which would only be

made use of to :candle these facts) I have chosen the convention stated

in (4.180, which, though still somewhat ad hoc, is weaker than a new

notational convention would be, and thus yields a more restrictive

characterization of the class of possible transformations, and hence

of the notion of natural language. In § 4.3.2 I discussed a number

of cases where pied piping is obligatory and suggested that the theory

of grammar be changed so that every particular gramma_' contairsa

conditions box in which constraints of various types)which affect all

rules of the grammar can be stated. Such constraints are intermediate

in generality between particular conditions on particular rules and

meta-constraints like the Complex /2 Constraint and the Coordinate

Structure Constraint.
 241

4.4. The Sentential Subiect Constraint

4.4.1. Compare (4.250a) with its two passives, (4450b) and

(4.250c).

(4.250) a. The reporters expected that the principal

would fire some teacher.

That the principal would fire some teacher

was expected by the reporters.

c. It was expected by the reporters that the

principal would fire some teacher.

Noun phrases in the that - clauses of (4.250a) and (4.250c)

can be relativized, but not those in the that-clause of (4.250b), as

(4.251) shows.

(4.251) a. The teacher who the reporters expected that

the principal would fire is a crusty old

battleax.

b. * The teacher who that the principal would

fire was expected by the reporters is a

crusty old battleax.

c. The teacher who it was expected by the

reporters that the principal woLad fire

is a crusty old battleax.

How can (4.251b) be blocked? A first approximation would

be a restriction that prevented subconstituents of subject noun phrases

from reordering, while allowing subconstituents of object noun phrases

... h.

'-'" irl i.r

 242

to do so. But such a restriction would be too strong, as can be seen

from the grammaticality of .(4.252).

II nen%
Vto4.04) Of which cars were Che Luods damaged by

the explosion?

The approximate structure of (4.252), at the time when the glie.asi.on

Rule applies, is that shown in (4.253).

(4.253)

were damaged by the explosion

It can_be seen that in converting (4.253) to the structure

which underlies (4.252), the boxed NP, a subconstituent of the

subject of (4.253), has been moved to the front of the sentence, so

3Z
the suggested restriction is too strong, But there is an obvious

difference between .252) and the ungrammatical ;.251b): the subject

of the latter sentence is a clause, while the subject of the former is

only a phrase. The condition stated in (4.254) rakes this difference

into account.
 21.3

(4.254) The Sentential Subject Constraint

No element dominated by an S may be

-^",4
4W.IV4U %di 4.1.44Also P
44
iL 41.0111 AIOUG

is dominated by an NP which itself is

immediately dominated by S.

This constraint, though operative in the grammars of

many languages other than English, cannot be stated as a universal,

because there are languages whose rules are not subject to it. In

Japanese, for instance, although the circled NP in (4.256), which

is the approximate structure of (4.255), falls within the scope of

(4.254), it can be relativized, as the grammaticality of (4.257)

shows.

(4.255) Mary ga sono boosi o kabutte ita koto

Mary that hat wearing was thing

ga akiraka da.

obvious is

'That Mary.was wearing that hat is obvious.'

 244

(4.256) S

NP
,

L____ \\
N SA akiraka da

NP koto

kabutte ita

Mary sono

1
boosi

(4.257) Kore wa Mary ga kabutte ita koto ga

this Mary wearing was thing

akiraka na boosi da.

obvious is hat is.

'This is the hat which it is obvious that

Mary was wearing.'

That the languages whose rules I know to be subject to

(4.254) far outnumber those whose-i-Ule-s-aie not so constrained suggests

that a search be made for other formal properties of these latter

languages which could be made use of to predict their atypical behavior

a
 245

with respect to this constraint. At present, however, whether or

not (4.254) is operative within any particular language can only be

.
treated as an idiosyncratic fact which must be stated in the

e.nnAitions box of the language in question.

4.4.2. George Lakoff has pointed out to me that on the basis

of only the facts considered fir, it would be unnecessary to state

the Sentential Subject Constraint, for it is a s2ecial case of (3.27),

the output condition which makes sentences containing internal

S unacceptable. Thus, since (4.251b) contains the internal

NP NP
clause that the principal would fire, and since this clause is dominated

exhaustively by NP, condition (3.27) would account for its unacceptability.

But the two arguments below seem to me only to be accountable for

if condition (4.254) is assumed to be operative in the grammar of

English.

Firstly, consider sentence (4.258), and its associated

constituent structure (4.259).

(4.258) That I brought this hat seemed strange to

the nurse.

1
 246

(4.259)

If .

strange to

Relativizing either of the circled NP's in (A.259) will

produce sentences which are not fully acceptable (cf. (4.260)),

(4.260) a. * The hat which that I brought seemed strange

to the nurse was a fedora.

b. ? The nurse who that I brought this hat

seemed strange to was as dumb as a post.

because both relative clauses in (4.260) will contain the boxed NP

.over S of (4.259) as an internal constituent. Conditi )n (3.27)

will be adequate to characterizing both as being unacceptable, but it

will not be able to account for the clear difference in status between

(4.2600 and (4.260b). The latter sentence is admittedly awkward,

but it can be read in such a way as to be comprehensible. The former

 X47

sentence, however, seems to me to be beyond intonational help. I

conclude that (4.260b) should ba labeled gi-emuatical but unacceptable,

but that (4.260a) must be deemed ungrammatical. To do this,

(4.254), or some more general constraint, must be assumed to be

operative in English, as well as (3.27).

The second argument for (4.254) concerns the following

two sentences:

(4.261)a. I disliked the boy's loud playing of the

piano.

b. I disliked the boy's playing the piano

loudly.
32
Lees gives a number of arguments which show these to be different.

I will assume that the derived structure of (4.261a) is that shown

in (4.262), and that of (4.261b) is that shown in (4.263).

 248

(4.262)

NP

.'/
disliked NP NP

NP

A
P

of the piano

(4.263)

V N."?

playing ar.....t.ass 1
 249

I have assumed that the word playing, in (4.262) has the

derived status of a noun, to account for the appearance of the

preposition of before the object of 32...4211.11, parallel to the of

which occurs after such substantivized verbs as construction, refusal,

fulfillment, etc. (cf. his construction of an escape hatch, our refusal,

o£ help, her fulfillment of her contract).

That the latter structure has a clausal object, while

the former does not, can be seen from the difference in relativizability

of the circled NP's in (4.262) and (4.263). This NP can be

relativized in the former structure (cf. (4.264a)), btt not in the

latter (cf. the ungramiaticality of (4.264b)).

(4.264) a. The boy whose loud playing of the piano

I disliked was a student.

b. * The boy whose playing the piano loudly.

I disliked was a student.

Although the circled NP of (4.262) is on a left branch of an NP

when the Relative Clause Formation Rule applies, pied piping can be

invoked to effect the adjunction of the boxed NP to the node $

which dominates tne clause, so a well-formed relative clause will

result.

But in (4.263), if the circled NP is moved, the boxed

NP cannot pied pipe, because there is a node S which intervenes

between the two NP nodes, and under these conditions, pied piping
 250

cannot take place, as was pointed out in 5 4.3.1 above.

Note that the object NP of.playingo the piano,

is relativizable in both (4.262) and (4.263).

(4.265) a. ? The piano which Y disliked the boy's

loud playing of was badly out of tune.

b. ' The piano which Y disliked 'the boy's

playing loudly was badly out of tune.

But if the action nominal or the factive gerund nominal appears in

subject position, as in (4.266), the NP the piano can only be

relativized out of the action nominal as (4.267) shows.

(4.266) a. The boy's loud playing of the piano drove

everyone crazy.

b. The boy's playing the piano loudly drove

everyone crazy.

?which the boy's.loud playing of

(4.267) a. That piano,
the boy's loud playing of which

drove everyone crazy, was badly out of tune.

which the boy's playing loudly

b. * That piano, the boy's playing which loudly

drove everyone crazy, was badlyout of tune.

How can (4..267b) be excluded? The bottom line of (4.267b)

can be blocked on the same grounds as (4.264b): since the subject NP

of (4.266b) dominates the node S, pied piping cannot take place. But

unless (4.454), the Sentential Subject Constraint, is added to the

grammar, the top line of (4.2673) will not be excluded. Note that
 251

even condition (3.27) cannot be invoked here, because this condition

must be reformulated as shown in (4.268).

(4268) Grammatical sentences containing an internal

NP which exhaustively dominates an S are

unacceptable, unless the main verb of that

S is a gerund.

This reformulation is necessary in any case, in order

to account for the difference in acceptability between (4.269a)

(4.269c) and (4.269d).

(4.269) a. * Did that he played the piano surprise you?

b. * Would for him to have played the piano

have surprised you?

c. * Is whether he played the piano known?

d. Did his having played the piano surprise

you?

Thus it appears that there are two reasons for insisting

that both (4.268), the revised version of (3.27), and the Sentential

Subject Const.raint be included in the grammar of English. In the

first place, condition (4.268) is not adequate to distinguish between

(4.260a) and (4.260b), aad in the second, between (4.267a) and (4.267b).

These two facts indicate the necessity of adding to the conditions

box of English something at least as strong as (4.254).

 252

4.4.3. It will be remembered, in connection with (4.249), that

in the conditions box for Finnish, there is a constraint which prevents

elements of clauses headed by ettd 'that' from being moved out of

these clauses (cf. the ungrammaticality of (4.249b)).

In her recent paper (Dean (1967)), Jane; Dean has pointed

out a condition in English that is probably related to the Finnish

condition. There is a class of verbs in English which can take that-

clauses as objects but for which the rule which normally can optionally

delete the that-complementizer cannot apply. After believe, for

example, the complementizer is optional (cf. (4.249a)), but after

verbs like quip, snort, rejoice, etc., the complementizer must be

present, as the ungrammaticality of (4.270b) shows.

(4.270) a. Mike quipped that she never wore this hat.

b. * Mike quipped she never wore this hat.

Dean discovered that no element of the complement clauses of these

verbs can be moved out of them (cf. the ungrammaticality of (4.271)).

(4.271) a. * Which hat did Mike quip that she never wore?

b. * Which girl did Mike quip never wore this

hat?

It is not clear at present how these facts should be

handled. It may be possible to assume that the English conditions

box, like the Finnish one, contains the constraint that no element

may be moved out of that-clauses, and that the object clauses of

verbs like believe do not come to be headed by*that until after all

reordering transformations have applied, while the object clauses Df

 253

verbs like quip are prefixed by that at a very early stage in derivations.

This then raises the possibility that the condition that no element be

moved out of a that-clause need not be stated in the conditions boxes

Inne QII&CFPnrad
of Finnish and English, but is instead universal.

(op. cit.) that this condition is only a subcase of a far more general

condition, (4.272).

(4.272) No element of a subordinate clause may be

moved out of that clause.

There are several difficulties with this condition which

at present prevent me from accepting it. The first is that it is not

strong enough to explain the differences among the sentences in

(4.251), and would therefore seem to have to be supplemented by the

Sentential Subject Constraint. The second is that (4.272) would

incorrectly exclude all the sentences of (2.23), which differ among

themselves in acceptability, but some of which seem perfectly

normal to me. And the third objection is that elements of clauses with

be'
Poss - Ina or for - to complementizers can be relativized, as can

seen from the grammaticality of (4.265b) and (4.273).

0.273) The only hat which it bothers me for her

to wear is that old fedora.

That such phrases must be considered to be dominated by S follows

from the fact that Reflexivization cannot "go down into" them (cf.

the ungrammaticality of (4.274)),

 254

(4.274) a. * I dislike it for him.to tickle myself.

b. * I dislike his tickling myself.

from *he fact that elements of these clauses can undergo "backwards"
33
pronominalization (cf. (4.275)) ,

(4.275) a. For Anna to tickle him drives Frank crazy.

b.. Anna's tickling him drove Frank crazy.

and from my proposed explanation of the difference in acceptability

between the sentences of (4.264). This last objection cannot be

gotten around by modifying (4.272) by attaching a condition that the

main verb of the subordinate clause be finite, for no elements of

the infinitival and gerund clauses in sentences like (4.276) can

be moved, as the ungrammaticality of (4.277) shows.

(4.276) a. We donated wire for the convicts to build

cages with.

b. They are investigating all people owning

parakeets.

(4.277) a. * The cages which we donated wire for the

convicts to build with are strong.

b. * What kind cf parakeets are they investigating

all people owning?

These three arguments against Dean's proposed constraint

strike me at present as being strong enough to reject it for the time

being. It is, however, a bold and important hypothesis, for if it can

be established, it will make my Complex NP Constraint and Sentential

I
 255

Subject Constraint superfluous, thus substantially simplifying both

the theory of language and those grammars in which the latter constraint

is operative. For this reason, a lot of future research should be

directed at the three objections to (4.272) which I have discussed,

to see if they can satisfactorily be explained away.

4.5. To summarize briefly, in this chapter I have proposed two

universal constraints, the Complex NP Constraint and the Coordinate

Structure Constraint; also, a universal convention of pied piping; and

a variety of language particular constraints, which are to be stated

in particular grammars in a conditions box, which the theory of

language must be revised to provide. I make no claim to. exhaustiveness,

and I am sure that the few conditions I have discussed are not only

wrong in detail, but in many major ways. Not only must further work,

be done to find other conditions, but to find broader generalities,

such as the condition proposed by Dean, so that the structure of

whatever interlocking system of conditions eventually proves to be

right can be used with maximum effectiveness as a tool for discovering

the structure of the brain, where these conditions must somehow be

represented.

1
 widpc.4.1.,L ti

FOOTNOTES 256

1. Subscripts indicate identity of reference.

2. This term is defined in Ross (1967a). There I argue that

pronouns may only precede the NP they refer to if they

are dominated by a subordinate clause which does not dominate

that, NP. Cf. also § 5.3 below.

3. Evidence that this rule must be placed late in the rule

ordering is given in Lakoff and Ross (op.. cit.). (. 0-44 6e itN4

4. The Japanese words"wa, etc. have been called

"particles". They correspond very roughly to case endings

and prepositions. Ga and wa are adjoined by transformations

to the rightof.istj7t,noun phrases, o to the right of

direct objecls;va-gent phrases ) etc. The syntax of these

postpositional particles and other problems in Japanese

syntax have been investigated intensively by Kuroda (cf. Kuroda

(1965)), and I will not discuss 4.t further here. In the

word-for-word glosses of Japanese examples, I will leave

the particles untranslated.

5. The structure shown in (4.25) is vastly oversimplified and

the analysis of tabete iru 'is eating' is stnply wrong: actually

iru should be the main verb of a higher sentence into which

Oa,
 257

fish eat (stem)' would

the base.string kodomo sakara tabe 'child

be embedded. Also, the determiner sono'that' would probably not

of (4c24), but rather

appear as a constituent of the deep structure

as a feature on the noun sakana 'fish' in the matrix sentence.

But such niceties are not at issue here (4.25) will serve for

the purpose at hand.

6. postal made this proposal in a talk given at the LaJolla

Conference on English Syntax on February 25, 1967.

7. Professor Barbara Hall Partee has informed me (personal

communication) that in a survey of relative clause constructions

she found
in a wide variety of languages that she conducted,

that in languages which exhibit relative pronouns which have

been moved from their original position, these pronouns

invariably appear at the end of the relative clause closest

to the head noun. Relative pronouns thus move leftwards in

English, German, French, etc., and although I at present can

cite no examples of rightward movement, Professor Partee has

assured me that they exist. It therefore seems necessary to

assume that if movement occurs in the formation of Japanese

relative clauses9 it must be movement to the right, not to

the left.
 258

These facts point to a needed change in the theory of grammar.

In order to account for the facts discovered by Professor Partee,

it is necessary to add to linguistic theory a convention for

automatically reordering the formal statement of transfonlational

rules. If such a convention is made available, the statement

in universal grammar of a relative clause skeleton rule will be

possible, for the rule of Relative Clause Formation in Japanese

is simply the mirror image of the rule shown in (4.2). In which

direction the rule will reorder constituents depends entirely

upon whether relative clauses are generated by the rule

NP S or by the rule NP S

I will present further evidence which supports this convention

for automatic reordering in a paper now in preparation, "Gapping

and the order of constituents"

8. Some speakers appear to find (4.40a) and sentences like it

grammatical, which indicates that for their dialecti, the Complex

NP Constraint must be modified somehow. I have no idea hoW

to effect a modification of this principle, which otherwise seems

to be universally valid, so I can only indicate the existence

of this problem now.

9. For an account of such segmentalization rules, see Postal (1966a).

 1 4 ..... ,,

259

10. If it should turn out to be possible to treat disjunction as

the negation of conjunction, (4.8:) will admit of simplification.

This problem is discussed in Peters (in preparation).

11. Sentence (4.92b)' is perfectly grammatical, and it means 'But

she wants to dance,'(so) I want to go home.' I have only

starred it because it is not related to (4.91).

12. There is evidence, first noted by Chomsky, that a type of

adjunction operation is required which produces one of the

two structures below, if B is adjoined to A,

or
B 'A A

depending on whether it is adjoined to the left or right of

A. The motivation for the creation of the new node A is as

follows: in such a sentence as the boy is erasing the' blackboard,

it seems clear that the result of adjoining the present

participle ending,:-.1.22, to a verb should be a node of some sort.

But the stress rules will only work properly if the formative

erase is dominated exhaustively by the node V (for a discussion

of the stress rules of English, cf. Halle and Chomsky (to appear)).

This would indicate that the correct derived structure is

 260

To distinguish this kind of adjunction from what has been called

"sister adjunction" (cf. Fraser (1963)), I refer to it as Cho mar

adjunction. It is at present an open question as to whether both

types of adjunction need be 'countenanced within the theory of

derived constituent structure. Some consequencesoif using

Chomsky-adjunction in the complement system are explored in

Lakoff and Ross (op. cit.), where the proposed analysis of

sentence coordination is based in an essential way upon this

kind of adjunction.

13. As (4.84) is presently formulated, such a rule would be impossible:

no conjunct can be moved. But in § 6.3 below I will show that

Lakoff-Peters rule oficoniuct Movement is formally different in

one crucial respect from the rules of Relative Clause Formation

and Question, and that it is this difference which makes the

former possible and the latter two impossible.

14. (4.116a) is acceptable only if strong pauses follow bought, and

him, i.e., if the second clause of (4.115) has become a parenthetical

insert into the first,clause and is therefore no longer coordinate

with it.

15. This term is Rosenbaum's. Cf. Rosenbaum (1965).

"S.2."%iit"
 261

16. Actually, it should be replaced, in (4.130) as well as in (4.126)

and (4.128), by a more abstract representation, but this fact has

no consequences for my argument.

17. It would probably be possible to order the rules which copy

the conjunction and later delete the first of the conjunctions

in such a way that at the time at which Relative Clause Formation

applied, the NP the boy in (4.133) would still be preceded by

and, so the variable would not be null and (4.84) could be

invoked to explain the ungrammaticality of (4.134). But such

a solution, even if it should prove to be possible for English,

which has not been-demonstrated, would break down in any

language whose relative clauses followed their head noun, as

in English, and whose conjunctions followed their conjuncts,

as is the case in Japanese. It does not seem unlikely that

such a language might exist, so the solution I have proposed

in the text is powerful enough to work even for such a language.

18. Of course, (4.136b) is not the correct derived structure for

the NP the boy who I saw, because many details of the correct

rule of relative clause formation have been omitted in the

formulation given in (4.135).

 262

19. I am not sure of the grammaticality of sentences conjoined

with aLd whose coniuncts contain both yes-no questions and

WH-questions, e.g.,

? Did you have a good time and what did you

bring me?

? What's for supper and is the cat back. yet?

I am sure I say such sentences often, but most of them seem some-

how disconnected. At any rate, whatever the pcact restrictions

on them may be, they are not my main concern here. .

20. I believe it is possible to restrict convention (4.166) to cases

where one noun phrase is contained within another, i.e., that

it is not necessary to generalize it so that it applies to

all category types. So until additional facts turn. up which

would force this more general version, I will propose the weaker,

one of (4.166).

21. The verb habe shave' has been moved to the eud of the 'relative

clauses in (4.179) by a rule which moves verbs to the end of all

dependent clauses:

22. Actually, there is some question as to whether the occurrences

of the node S which NP and NP dominate in deep structure

2 1
 263.

will have been pruned by the time the rule of Relative Clause

Formation applies. At present, I am, not sure that pruning must

have already applied. If it has not, the problems under

discussion multiply enormously, for then it would presumably be

necessary to distinguish between sentences with finite main

verbs and those with non- finite main verbs in the revised

version of (4.166).

23. I am grateful to Robin Lakoff for suggesting this descriptive

and picturesque terminology,' just as the children of Hamlin

followed the'Pied Piper out of town, so the constituents of

larger noun phrases follow the specified noun phrase when it

is reordered. This :hoice of terminology from the realm of

fairy tales should not, however, be construed by an overly

literal reader as a disclaimer on my part of the psych6logical

reality of (4.180).

24. There are certain nomenclative Feinschmeckers who have taken

issue with the formulation of this sentence, pointing out that

following the original Pied Piper was obligatory for all the

children of the town except one, who was lame, so that the

phrase "obligatory pied piping" is a case of terminological

coals to Newcastle. These critics sugge=t ^^"'"7".4"

(4.180) Zescribes optional accompaniment, such accompaniment

 should best be dubbed "fellow traveling," or the like, with the

term "pied piping" being reserved for cases of mandatory accompaniment,

..7,66Y&a
4.1q^e.ca
ftsavory fleten1.41.%15A
YVVYf ... 1%GaleM1
YYYYY.

While the point they make is valid, I have chosen to disregard its

eschewing an exact parallel to the fairy tale in question in the

interests of a less elaborate set of terms.

25. The fact that NP does not dominate So and that (4,188a) is
1

still grammatical, simply indicates that (3.26) is formulated

incorrectly, and that Condition i on Cilia: rule must be

Of this rule.
revised. It is abtindoned entirely in (5.57), the final statement

26. I have starred (4.190a) because it is unrelated to (4.190b) --

the how in (4.190a) does not replace to what extent, but rather

something like in what resrect or in hat war. Note also that

the echo-questions for these two sentences differ: (4.190a) is

related to Peter is sane HOW? but (4.190b) to Peter is HOW sane'

Similarly, although (4.191a) is grammatical, it is not related

to (4.191b).

27. Note that place is ambiguous: it can mean 'residence, dwelling',

and in this sense, the preposition can be left behind*(Whose

1_ .....14.7,m'ne9N
place
 265

28. This problem is discussed at some length in Keyser (1967).

0"o(1
efaWo
it ,, Aisu/
a-1...o. /A 1'27% 40 no nr,rnmat-tnnl unlAag ConiunetiOn

Reduction applies again to reduce the parenthesized are, but I.

will disregard this problem here.

30. Later rules will convert (4.242b) into the boy's uncle and

aunt's grandmother.

31. There is, however, an additional restriction which pertains to

structures like (4.253): while it is possible to move the

boxed NP, it is not possible to move the circled one --

the string *Which cars were the hoods of dasmaibylheemLosion?

is ungraatical. It is not in general the case that the

preposition of in the NP the hoods of the, cars cannot be

stranded (witness the grammaticality of Which cars did the

212Iehoodsofclosiondm?) so another clause must be added

to condition (4.206), making pied piping in the environment

[P also obligatory where the prepositional phrase is

--SNP
dominated by an NP which is immediately dominated by S; In

passing, it should be noted that the statement of this condition

will require quantifiers or some equivalent notation, such as

node subscripts. This means that the formal apparatus which

 266

is available for stating conditions in a conditions box must

be stronger than that available for stating conditions on

:2. Cf. Lees (1960), pp. 65.67. I will follow his terminology in

calling the nominalization in (4.261a) the action'nominal and

will refer to the nominalization in (4.261b) as the factive

'gerund nominal.

33. For a fuller discussion of the conditions under whicebackward",

or right-to-left, pronominalization is possible, as well as some

remarks about the notion 02 subordinate 'clause, cf. Ross (1967a).

and k3 beim.
 267

Chapter 5

BOUNDING CONNAND AND PRONOMINALIZATION

5.0. In the summer of 1966, Ronald Langacker and I, working

independently on the same general problem, arrived at highly similar.

solutions. The problem was'that of restricting variables which

appeared in the structural descriptions of various rules in such a

way that the notion of sentence under consideration could be captured.

To this end, I proposed a formal device I called 122.2.1...141a (cf. Ross

(1966b)), which will be explained in § 5.1 below. Langacker's

notion of command, which he introduces and discusses at length in

his important paper, "Pronominalization and the chaill of command"

(Langacker (1966)), seemed to me until recently to be as nearly

adequate to this end as bounding -- while there were some facts

which could be handled with command but not with bounding, there

were also facts for which the opposite was the case. Recently,

however, I have come to the realization that the latter type of

facts, which I took to be an indication if the necessity of

including the notion of bounding in linguistic theory, can in fact

be handled with command, by extending its definition in a natural

way. Langacker's notion is thus clearly preferable, and it, not

the notion of bounding, should be a part of the theory of language.

In § 5.1, I will explain the notion of bounding and

.....,...7....11.....w ,
 268

discuss the kinds of facts which it is meant to account for. In § 5.2

I'will show how all these facts can be accounted for with command,

and give several facts that cannot be handled with bounding. In

addition, I will point,out one way in which bounding is too strong.

In § 5.3 I will discuss pronominalization briefly in this context,

and show that the major condition on the rule of Pronominalization,

that it only go backward into subordinate clauses, should really be

construed as a condition on all deletion transformations of a

specified formal type.,

5.1. 222.12.12,aa

I
J.A.04.41,

5.1.1.1. Let us reconsider the rule of Extraposition, (4.126).

How is this rule to be ordered? If the cyclic theory of rule

application proposed by Chomsky (cf. Chomsky (1965)) is correctl, then

the rule of els-Replacement must be a cyclic rule, as Lakoff has

demonstrated (cf. Lakoff (1966)). This rule converts (5.1) into

(5.2), and (5.3) into (5.4) by substituting the subject of the

embedded ,sentence for the pronoun it and daughter-adjoining the

remainder of the embedded sentence to the VP of the matrix

sentence.
 269
........... .......... .M.

(5.1)

13ke girls

(5.2)

1
.42 2 ears
,
for

VP

like girls
 270

NP

we

expect N

it

H a La to VP

like gi is

NP VP

expect . for

VP

.....
....,..

like girls
 271

I will attempt to show that if It Replacement is in the cycle,.

Extraposition cannot be, for it would produce an intuitively

incorrect derived constituent structure for sentences like (5.5).

(5.5) It appears to be true that Harry likes girls.

To me, it seems clear that there is a large constituent break after

true. A plausible derived"structure for this sentence is the one

shown in (5.6)

(5.6)

NP VP

VP 12141IEIL2111:Laal

it appears to be true

If Extraposition is a cyclic rule, it will first apply

to (5.7), which underlies (5.5), on the S cycle, yielding (5.8) as

2

an output.
 - .

272

(5.7)

VP

.1.
it appears,

true

it that Harr t likes 11.412.1

be true t1121.-Ilarry/...il pc.es

 273

Now, on the S1 cycle, after the complementizers for

and to have been introduced, application of the rule of It- Replacement

will yield (5,9) as an intermediate structure underlying (5.5). The

complementizer for is deleted by a later rule.

<5.9)

NP

it appears, for

to VP that Harry'likes girls.

44, .0WWW

'be true

But (5.9) seems highly inadequate as a representation

of the intuitive structure of (5.5), for it not only makes the claim

that the strings to be true thataallylicasigi1.i's and apars to be

Ima112i.assyjaas111a are constituents, but it also makes the

claim that appears to be true is not a constituent. All of these

claims strike me as being the exact opposites of the truth about the

constituent structure of (5.5), which is captured correctly in (5.6).

v....Y.
 274

The structure shown in (5.6) can be derived from deep

2
structure (5.7) if Extraposition is a last-cyclic rule. In this

case, no rules of importance here would apply until S1. On this

cycle, after complemenLizer placement, the circled NP in (5.7)

would bedome the derived subject of, .S by LE-RsaLace.cmnt, yielding

1
the intermediate structure (5.10):

(5.10)

NP

it thaw t Harry likes girls appears for VP

be true

When Extraposition is applied to (5.10), the correct (5.6) results.

The above facts can be accounted for if Extraposition is

made a last-cyclic rule, but this is not the only means of arriving

at the correct derived structure for sentences like (5.5). Noam Chomsky

has suggested to me in conversation that it seems necessary to add

 275

certain phonologically motivated rules of adjustment to the grammar

of English, to account fc.r the intonation of such right-branching

sentences as (5.11),

(5.11) This is the dog that chased the cat that caught

the rat that ate the cheese.

to which normal transformations would assign some structure like

that schematically represented in (5.12).

(5.12)

NP S

",the 'clog. that chased

ii

NP S

the cat .that,..auElalt N

the'rat that ate'the cheese

 276

On the hypothesis that intonation rules should correlate length

3 (5.11) would not be
of pause with size of constituent break ,

assigned its observed intonation pattern, where pauses of roughly

equal size precede each occurrence of that, unless some rule were

to operate on the nested syntactic output structure of (5.12) to

turn it into the roughly coordinate phonological input structure

which the normal pause pattern of (5.11) would indicate. Such

rules Chomsky proposes to call "surface structure adjustment rules",

and he suggests thatthe same rule which raises the nested sentences

of (5.12) to make them coordinate with the highest sentence there

might be formulable in such a way that it would also raise 53

to the level of S in (5.9), thus producing (5.6), the correct

I
derived structure of (5.5), from (5.7), even if the rule of

Extraposition is made a cyclic rule.

Until some detailed work has been done on the.problem

of such adjustment rules, it is not possible to accept or reject

this proposal conclusively. However, even if Chomsky's proposal

should prove to be correct, there is another argument, independent

of this one, which indicates that Extraposition cannot be a cyclic

rule.

Consider such intercalated structures as (5.13).

(5.13) Ivan figured it out that the bridge would hold.

This sentence derives from the structure shown in (5.14).

 277

(5.14)

NP VP

Ivan

'figure out

that the bridge would hold

To this structure, the two rules of Particle Movement,

(3.9), and Extraposition apply. From the arguments given above, in

4.2.4.2, it follows that Particle Movement must apply first,

moving the particle out to the right of the circled NP of (5.14);

for Extraposition cannot apply "vacuously" to attach the circled node

S somewhere higher up the tree, if sentences like the ungrammatical

(4.132b) are to ba avoided.

However, if we assume Extraposition to be cyclic, since

Particle Movement precedes it, it must also be cyclic. But if

Particle Movement is cyclic, then the problem arises as to how

sentences containing ungrammatical action nominalizations like the

one in (5.15a) are to be excluded.

(5.15) a. * Her efficient looking of the answer up

pleased the boss.

I
 278

b. Her efficient looking up of the answer

pleased the boss.

Sentence (5.15b) demonstrates that the ungrammaticality of (5.15a)

does not reside in an incompatibility between verbparticle

constructions and action nominalizations in general, and that it can

only be attributed to the fact that Particle Movement has applied

when the sentence in the underlying subject of (5.15a) was processed,

but not. when the one in the subject of (5.15b) was. I believe the

claim to be warranted that action nominalizations are derived from

embedded sentences -- that is, that there are two passes made

through the transformational cycle in processing (5.15b) -- and

not, as Chomsky suggested in course lectures in the spring of 1966,

by means of lexical derivation rules; but I cannot go into this

problem here. I mention the matter merely because (5.15a) could

rather easily be excluded if the subject NP of (5.15b) had

been produced in the lexicon: if the word looking. in (5.15) is

best considered to be a derived noun, which seems to me to be an

open question, then Particle Movement could not apply to it, and

even if looms must be considered to be a verb, (3.9) could be

made to block because of the presence of an intervening of. But if

action nominalizations-are desentential, as I believe to be the case,

no such easy explanation is available. It would of course be impossible

to impose the condition upon (3.9) that it not operate in any sentence

which was embedded in whatever the correct underlying structure for

!..V.,
 279

action nominalizations *urns out to be, for by the principle of

operation of the transformational cycle (cf. Chomsky (1965), p. 134-

135), contexts from higher sententuts than the ore being processed cannot

be referred to in cyclic rules. This would mean, then, that Particle

Mbvealent would have to be allowed 0 apply freely, and that some ad hoc

condition would have to bá imposed upon Action Nominalization so that

it would block incase Particle Movement had applied on the previous

cycle. This is not impossible-, merely laboured, inelegant and

undesirable.

The obvious way out of this latter difficulty is to

make Particle Movement a last-cyclic rule, and to order it after the

rule which forces action nominalizations. If this rule has applied,

Particle Movement will be blocked by a constraint which is necessary in

any case: particles cannot be moved over an object NP which starts

with a preposition.. _Thus the particle ....razart may not be moved over

the NP with her father in (5.16a).

(5.16) a. She did away with her father.

b. * She did with her father away.

It is necessary to claim that idioms like do away with, sit in on, etc.,

which were mentioned in § 4.3.2.2 above, consist of a verb-particle

combination followed by a prepositional phrase, and not simply of a

verb followed by two prepositions and a noun phrase, for it is the case

that only that subclass of prepositions which can function as particles

(e.g. along, hit, on, in, off, 211., etc., as opposed to at, among, for, etc.)

can occur as the first member of such a two-preposition chain.

 280

5.1.1.2. Thus if Particle Movement is last-cyclic, (5.15a) will

be excluded without any additional complication of the rules of

MU,. *weal^ Ac
W414:iio QWWW6 UWG.
Action Nominalizarion or Particle Movement.

Extra osition? Since it follows Particle Movement, it is last-cyclic:

what then will prevent it applying to (5.17) to produce the

ungrammatical (5.18)?

(5.17)

'is not "true

was obvious

it that Bob was

(5.18) * That it was obvious is not true that Bob was

lying.

For sine: ±,.....!re are variables in the structural index of Extraposition,

when it applies on the last cycle, it can either operate to move S2

out of NP1, in which case, the grammatical (5.19) will result,

 281

(5.19) It is not true that that Bob was lying was

obvious.

tie it can operate to move e out of Ni' arioldino the unrammatical

-2'

(5.18).

This problem is highly reminiscent of the one discussed

in Case C of § 2.2, which'was given as supporting evidence for

the A-over-A principle. But since the facts given in § 2.1 show

the principle to be too strong, I have tried to find alternative

explanations for all the cases given in support of it in § 2.2.

Cases A and B have been accounted for by the Complex NP Constraint,

Cases D and E by the Left Branch Condition on pied piping (4.181),

and Case F has been shown to be a special case of the Coordinate

Structure Constraint. Only Case C remains.

The problem discussed in Case C was how the rule of

Extra osition from NP should be constrained so that it will apply to

(2.7) to produce (2.8), but not (2.9), all of which I reproduce here

for convenience©
 282

(2.7)

VP

was 'given,

VP

Ni) had been made

the *claim 'that VP

John had lied

(2.8) A proof was given that the claim that John

had lied had been made.

(2.9) * A proof that the claim had been made was

given that John had lied.

Just as was the case with (5.18), (2.9) results from S

3
being extraposed "too.far". It happens that (2.9) can be blocked with

machinery that is already available, but this is not true of (5.18).

For notice that NP in (2.7) is complex, and that the Complex NP

1

Conctraint will therefore not allow S3 to be moved out of S2. But

chat will stop S3 fr.a being extraposed out of S2 in (5.17)? It

 283

is not the case that constituents of clauses dominated by noun phrases

whose head noun is the pronoun it cannot be'moved out of these clauses,

as the grammaticality of (4.13a) s'..ows. And even supposing that it

were possible to formulate some revised version of the aew-overA

principle which was strong enough to exclude (5.18), but weak enough

to avoid the counterevidence in S 2.1, the problem would remain. For

.consider structure (5.20)x;

(5.20)

N1
i
is possible O

1///

it NP

that Sam didn't pick

those packages which are to mailed tomorrow

 O

284 .

Since the rule of Particle Movement must be last-cyclic,

for the reasons discussed above in connection with (5.15), it is

obvious that Fxtraposition from NP must also be; for it, like

Extraposition, must follow Particle Movement. But now the question

is, how will Extraposition from NP apply to (5.20)? As this rule

is presently formulated, the variables in it will allow the extraposition

of S to the end of S1, with (5.21) as the ungrammatical result.

3

(5.21) * That Sam didn't pick those packages up is

possible which are to be mailed tomorrow.

How can this sentence be blocked? Even if it were assumed that the

two rules of extraposition were the same, and could be collapsed into,

one (I will show why such an assumption would be incorrect immediately

below), the A-over-A principle could not be invoked to. block (5.21).

For this principle dictates that transformational rules must apply to

a tree uniquely, and always in the highest possible environment. Since

both NP and NP would meet the structural index for a collapsed

1 2

rule of Extraposition, the A-over-A principle would prediCt that this

Extraposition could only affect the higher NP, NP1, moving S to

2
5
the right of is possible . But in fact, either clause can be extraposed

to the end of "the first sentence up", independently of whether the

other has been. Thus if neither has been, (5.22a) results; if only S2

has been, (5.22b) results; if only S3 has been, (5.22c) results; and

if both have been, (5.22d) results.

 285

(5.22) a.?* That Sam didn't pick those packages which

6
are to be mailed tomorrow up is possible.

b. * It is possible that Sam didn't pick those

packages which are to be mailed tomorrow up.

c. That Sam didn't pick those packages up which

are to be mailed tomorrow is possible.

d. It is possible that Sam didn't pick those

packages up which are to be mailed tomorrow.

Thus, since S3 must be allowed to extrapose, so that

(5,22c) and (5.22d) can be generated, it seems to me inconceivable that

any version of anything resembling the A-over-A principle can be devised

which could exclude (5.21).

5.1.1.3. A final nail in the coffin of any such proposal is provided

by the following argument, which shows the two rules of extraposition

to be necessarily distinct, because another rule, Question, must

intervene between them. That is, the rules must be ordered as in (5.23).

(5.23) 1. Particle Movement (3.9)

2. Extraposition (4.126)

3. (question (4.1)

4. 1trai
postionfrojLE)NP(1.10)

The necessity for this ordering can be seen in connection

with (5.24), which derives from the intermediate structure (5.25), a

structure only minimally different from (5.20).

 286

(5.24) Which packages is it possible that Sam didn'E

pick up which are to be mailed tomorrow?

which packages which are to be mailed tomorrow

It should be obvious that Extraposition must precede

kapstion, for if S2 has not been moved out of NP1 to the end of

S the questioned element, NP will be contained in a sentential

l' 3

subject, NP1, and will be prohibited from moving out of it by the

constraint stated in (4.254). But it is not so obviow that Extraposition

from NP must fallow glaslisa.

 287

For if it is assumed that (5.21) can somehow be avoided,

it might be argued that a collapsed rule of extraposition could

operate to move both embedded sentences to.the ends of the appropriate

higher sentences, yielding a structure like (5.26)7

(5.26)

VP

N is possible that NP

it Sam didn't pick

which packages

which are to be mailed

tomorrow

But notice that if the questioned NP, NP2, is now

moved to the front of S by the rule of Question, and the subject

1

and copula are inverted, the resulting structure is (5.27), not the

intuitively correct (5.28).8

I
 288

(5.27)

eishPA,shr Is it V that NP O

ages

possible Sam didn't pick up

which are to be mailed

tomorrow

(5.28)

VP

which aackages is

it possible that Sam didn't pick up. which are to be mailed

tomorrow
 289

The structure shown in (5.27) makes the incorrect claim

that the string didn't pick up which are to be mailed tomorrow is a

constituent, while (5.28) correctly reflects the fact that there is

a large constituent break after the particle 12. It might appear that

the same method of avoiding this undesirable result that Chomsky

has proposed for avoiding the similar intuitive inadequacy of (5.9),

namely having some surface structure adjustment rule obligatorily

convert (5.27) to (5.28), just as (5.9) would be converted to (5.1),

could be made use of in this case.

To see that this is impossible, consider (5.29) and an

intermediate structure underlying it, (5.30):

(5.29) Sam didn't pick those packages up which are to

be mailed tomorrow until it had stopped raining.

(5.30)

NP until it had stopped raining.

didn't pick NP

those packages which are to be mailed tomorrow

 290

How does the rule of Extraposition from NP apply to

(5.30)? If some constraint can be stated on this rule which has the

4rt 4,11
enA of
G.11 . W1 wa.s.y 0.11.1VW.1.446 4.4AG Gebt.A.401,WOIA griiaMOG myvv

the first sentence up, then the rule could apply to (5.30) to produce

the derived structure (5.31).

(5.31)

NP until it had stopped raining

Sam didn't. pick NP

2

those packages which are to be mailed tomorrow

Since some such constraint will be necessary in any case,

so that (5.21) can be avoided, the grammaticality of (5.29), where

the extraposed relative clause immediately follows the particle u2.,

provides some support for the structure shown in (5.30) , 1 ,Uch S4

is not a constituent of S2. The facts of do so pronomink._Ization

(cf. Lakoff and Ross (1966)) indicate that S could LA be dominated

4

by VP9, for do so stands for a whole VP, and until-clauses are

outside the VP, as is shown by the grammaticality of (5.32).

 293.

(5.32) Sam picked those packages up .which are to be

mailed tomorrow rest might, but he didn't want

to do so until it had stopped raining.

If S4 were directly dominated by S2 in (5.30) , then wa would expect

that the most normal version of this sentence would be (5.33), not

(5.29).

(5.33) ?* Sam didn't pick those packages up until it

had stopped raining which are to be mailed

tomorrow.

In my speech, (5.33) is impossible unless heavy intonation breaks

surround the until-clause, in which case it fairly acceptable. But

such a sentence should clearly be analyzed-as a stylistic variant

derived from (5.29) by the optional rule which positions adverbs in

9
various positions between major constituents of a sentence, not as

the most normal form for this sentence.

But now notice what happens if a structure like that

shown in (5,,30), except that which replaces those, is embedded in

place of S in (5.25). Two variants of the resulting structure,

2

(5.34), are possible: (5.35a), in which the relative clause S3

has not been extraposed away from its head NP, which packages,

and (5.35b), in which it has.

(5.34)

is possible

it that

NP

Sam didn't pick NP

which packages which are to be mailed tomorrow

(5.35) a. Which packages which are to be mailed

tomorrow is it possible that Sam didn't

...

pick up until it had stopped raining?

b.?? Which packages is it possible that Sam

didn't pick up until it had stopped raining

which are to be mailed tomorrow?

While it is clear that (5.35a) is the more comfortable

version of the two, by far, I think (5.35b) should be treated as being

 293

grammatical but of low acceptability. For notice that the acceptability

of (5.35b) can be improved by lengthening the extraposed relative

clause, as in (5.36).

(5.36) Which packages is it possible that Sam didn't

, pick up until it had stopped raining which he

had arranged with his agents in Calcutta to

send to him here in Poplar Bluff because of

his fear that someone in Saint Louis might

recognize him?

Note that in (5.35b) the extraposed clause follows the until-clause,

which the ungrammaticality of (5.33) shows not to be possible when the

structure underlying (5.29) is-not-embedded. But more important is

the fact that the preferred order in the non-embedded case, i.e.,

with the relative clause preceding the until-clause, as in (5.29), is

absolutely impossible in the embedded case, as the ungrammaticality

of (5.37) shows.

(5.37) * Which packages is it possible that Sam didn't

pick up which are to be mailed tomorrow until

it had stopped raining?

In fact, if a relative clause has been extraposed away

from its head NP, that NP cannot be questioned. So compare

(5.29), which contains such a head NP, with the ungrammatical

(5.38), in which this NP has been questioned:

 294

(5.38) * Which packages didn't Sam pick up which are

to be mailed tomorrow until'it had stopped

raining?

Elsewhere (cf. Ross (1966a)), I have pointed out that

no elements of an extraposed relative clause may be relativized or

questioned. For an-example of this restriction, consider (5.39) and

its derivedctructure (5.40).

(5.39) A girl came in who had worn this coat,

(5.40)

a girl came in NP

who

this coat

That the circled NP in (5.40) cannot be relativized is

apparent from the ungrammaticality of (5.41).

(5.41) * The coat whidh a girl came in who had worn was

torn.
 295

The ungrammaticality of sentences like (5.37), (5.38) and

(5.41) seems to call for the adoption of a new'confcraint, such as

the one stated in(5742):

(5.42) The Frozen Structure Cor_;traint

If a clause has been extraposed from a noun

phrase whose_head no'iu is lexical,. -his noun

phrase may not be moied, nor may any element of

the clause be moved out of that clause.

The formulation of this cc'straint is reminiscent of the

formulation given in (4.20) -- the Complex NP Constraint. A moment's

reflection on the content of the former constraint suffices to reve-1

why this should be so: what (5.42) says, in effect, is that elements

of complex noun phrases, which are prohibited from being moved before

the rule of Extraposition from NP has applied are also prohibited after

this rule has applied. In other words, (5.42) must duplicate the

constraints which are stated in (4.20) and (4.181), if Extraposition

from NP is allowed to precede transformations like Question and

Relative Clause Formation. The solution is obvious: the Frozen

Structure Constraint can be dispensed with if the rule of Extraposition

from NP is made a last cyclic rule (recall that there is independent

evidence that this rule is not cyclic, since it must follow Particle.

Movement) , and if it follows all movement rules, in particular ,Question

and Relative Clause Formation.

 296

5.1.1.4. Since the structure of the argument I have just presented

is highly complex, a review of the main points may prove helpful.

1. Extraposition is last-cyclic.

There are two arguMents for this: (a) if

it were cyclic, sentences like (5.5) would be

.assigned the wrong d.c.s., unless some independently

motivated surface structure adjustment rule can

be formulated in such a way as to automatically

convert (5.9) into (5.6), and (b) it must

follow Particle Movement, which ,the facts of

sentence (5.15) show to be last-cyclic.

2. If Extraposition is last-cyclic, unless it

is constrained in some new way, deep structures

like (5.17) will be converted into ungrammatical

strings like (5.13) .

3. The A-over-A principle, though it might be

used to block (5.18), cannot be used to block

sentences like (5.21), which involve both

Extrapostiton and Exraposition from NP, unless

it can be argued that these two rules should be

collapsed into one rule.

 297

4. Extraposition must precede ausgm,

because while no .elements of subject clauses

may be moved out, of these lauses, by virtue of

the Sentential Subject Constraint, (4.254), if

these clauses have been extraposed, elements in

:them become movable (compare (4.251b) and (4.251c)).

5. All movement rules, in particular guessioE.,

must precede Extraposition from NP, or else the

Frozen Structure Constraint, an.otherwise

unnecessary condition, which in essence repeats

provisions of the Complex NP Constraint

and the Left Branch Condition, must be added to

the theory of grammar.

6. Since one precedes and the other foliows

Question, Extraposition and Extraposition from NP

cannot be collapsed into one rule. In the

derivation of sentences like (5.35b), the four

rules of Particle Movement, Extraposition,

Question, and Extraposition from NP must all

apply, in the order 1:;.sted.

 'c '

298

7. Therefore, ungrammatical sentences like

(5.21) cannot be excluded by any version of

the, A.-ver-A pr4n^4p1e,.

Conclusion: some new type of restriction on rules

must be devised and added to the

theory of grammar.

5.1.2.

5.1.2.1. Sentences like (5.21), which the argument above shows

not to be excludable by any presently available theoretical mechanism,

can be blocked if rules can make reference to the boundaries of the

first sentence above the elements being operated on. I will refer to

a rule as being upward bounded if elements moved by that rule cannot be

moved over this boundary. To give a concrete example, the rule of

Extraposition must be marked as being upward bounded. This means

that when the structure shown in (5.43) is inspected to determine

whether the structural description shown in (4.126) is satisfied, and

if so, how the operation of the rule_is.to be carried out, by universal

convention, the variable Y in term 4 of (4.126) will be interpreted

as ranging over all nodes of the tree which are below the first

double line above the nodes Of (5.43) which could be affected by the

rule -- S2, S3, and S And the instruction in the structural

4.

change of (4.126), that the S .of term 3. is to be adjoined to the

 299

variable in term 4, will be interpreted to mean that the S is to ^be

adjoined to the largest part of the tree consistent with this convention.

That is, the S will move to the right,up to the first double line.

Thus depending on whether Extraposition moves S2, S3, or S4, or

any combination of these, (5.43) will become one of the eight sentences
10
of (5.44).

(5.43)

is not true

wou con use t_e_mards

for Herschel to throw a fit

 300

(5.44) a. That that for Herschel to throw a fit

would confuse the guards was obvious is

not true.

b. It is not true that that for Herschel to

throw a fit would confuse the guards

was obvious.

c.. That it was obvious that for Herschel to

throw a lit would confuse the guards is

not true.

d. It is not true that it was obvious that for

Herschel to throw a fit would confuse the

guards.

e. That that it would confuse the guards for

Herschel to throw a fit was obvious is not

true.

f. It is not true that that it would confuse

the guards for Herschel to throw a fit was

obvious.

g. That it was obvious that it would confuse

the guards for Herschel to throw a fit is

not true.

h. It is not true that it was obvious that it

would confuse the guards for Herschel to

throw a fit.
 301.

The ungrammaticality of (5.21) shows that the rule of atraposition

from NP must also be designated as an upward bounded rule.

5.1.2.2. It seems that it is necessary to postulate yet a third

extraposition-like rule, to account for related pairs of sentences

like those in (5.45)'.

(5.45) a. A review of this article came out yesterday.

b. A review came out yesterday of this article.

It seems possible that the maximally general formulatiaa of this rule,

which is given in (5.46) may prove correct.

(5.46) Extraposition of PP

X - NP]Np - Y
OPT
1 2 3 -

1 0 3+2

Arguments similar to those given in § 5.1.1 show this

rule to be necessarily last-cyclic. Firstly, if it were in the cycle,

it would convert (5.48), which underlies (5.47) into (5.49) , instead

of converting it into (5.50).

(5.47) A review seems to have come out yesterday

of this article.
(5.48)

NI' *vP

,,'///(
it for NP

NP to have come out yesterday.

a review NP

of this article

(5.49)

a review

seems NP

77'
to have come out yesterday
/
P

of this article
(5.50)

NP NP

a review

seems to have comajltatai.15Ierla of this article

Like (5.9) and (5.27), (5.49) makes incorrect claims about

intuitions of constituency -- it claims that the string to have come

otitiesterlaoftlzA.sarticle is a constituent -- but unlike these.two

previous structures, it seems unlikely that the rule which converts (5.12)

into a coordinate structure can be extended to effect the conversion of

(5.49) into(5.50). Thus if Extraposition of PP is made a cyclic

rule, some new surface structure adjustment rule will be necessary.

Secondly, in order to produce intercalated structures

like those of sentences (5.51),

(5.51) Why don't you pick some review up of this article?

it will be necessary to order Extraposition of PP, after the last-cyclic

rule of Particle Movement. Thus it too must be last-cyclic.

, And finally, unless it is last-cyclic, it will be necessary

to add the constraint stated in (5.52) to the theory of grammar,

(5.52) If a prepositional phrase has been extraposed

out of a noun phrase, neither that noun phrase

nor any element of the extraposed prepositional.

phrase can be moved.

 .11...0 4 -

304

for if (5.53a) is converted by (5.46) into (5.53b) neither of the

underlined NP's in (5.53b) can be questioned, as the impossibility

of (5.53c) and (5.53d) shows.

(5.53) a. Ann is going to send a picture of Chairman

Mao to her teacher, as soon as she gets

home.

b.' Ann is going to send a picture to her

teacher of Chairman Mao, as soon as she

gets home.

c. * Which picture is Ann going. to send to her

teacher of Chairman Mao as soon as she

gets home?

d. * Who is Ann going to send a picture to her

teacher of, as soon as she gets home?

But just as condition (5.42) can be dispensed with by

making Extraposition from NP last-cyclic, so (5.52) can be if

Extraposition of PP is last-cyclic.

But if the above three arguments are correct, then the

fact that (5.54) can be converted into the structure underlying

(5.55a), but not that underlying (5.55b),Ahows that it too must be

designated as being upward bounded.

(5.54)

vP

11 is catastrop ic

it that NP VP

NP came out yes

a review

of this article

(5.55) a. That a review came out yesterday of this

article is catastrophic.

b. * That a review came out yesterday is catastrophic

of this article.

It seems to me to be passible to collapse ....2Copleas.NP Shift,

(3.26), and Extraposition of PP, removing condition 1 on (3.26), which

specifies that only NP dominating S can undergo the rule, and

stipulating that condition 2 applies only if the NP to be shifted

does not begin with a preposition. The removal of the first condition

will mean that (5.56b), which results from the application of the rule
 306

to (5.56a), will not be considered to be ungrammatical, but rather

unacceptable, and to be so designated by Output Condition (3.41).

(5.56) a. I'll give some to my good friend from Akron.

b. * I'll give to my good friend from Akron some.

I will henceforth refer to this rule, which is stated in (5.57), as

NP Shift.

(5.57) NP Shift

X - NP - Y
OPT
1 2 3

1 0 3+2

Condition 1: This rule is last-cyclic.

Condition 2: BLOCKS if 3 = X
1 + [ +Adj,i + X2 2

where there exists no NP which

+V
dominates f

g# [P NP)NP.

5.1.2.3. Whether or not I am correct in assuming that Complex NP

Shift and Extra position of PP are the same rule is not of great

importance at present. The generalization stated in (5.58) remains

true no matter how many rules (5.57) must be broken down into
 (5.58) Any rule whose structural index is of the form

... A Y, and whose structural change specifies

that A is to be adjoined to the right of . Y, is

upward bounded.

I know of no exceptions to this generalization.

It is probably impossible to maintain that all rules

which adjoin terms to the left of a variable are upward bounded,

unless the following facts can be explained in some other way than

the one I will propoie below.

Observe first that sentence (5.59) is ambiguous.

(5.59) I promised that he would be there around

midnight.

The adverb around rid12ight can either modify be as in (5.60),

which is the d.c.s. of one of the readings of (5.59), or it can

modify promised, in (5.61), which is the d.c.s. of the other

reading.

(5.60)

promised

that NP

he would be there aroun nigt.

*a.
 308

(5.61)

NP

NP around midnight

pis d 1

tha

he would be there

If the adverb is preposed to the front of (5.59), with

normal intoLation, the resulting sentence, (5.62), is unambiguous:11

(5.62) Around midnight I promised that he would be

there.

(5.62) can only be derived from (5.61). This can be demonstrated

by a consideration of_(5.63).

(5.63) I promised that .he would be there tomorrow.

This sentence, unlike (5.59), is unambiguous, and can only be assigned

a structure similar to (5.60), for tomorrow cannot modify the past

tense verb 22.2119122.d. Now note that the rule of Adverb Preposing, which

converts (5.59) into (5.62), cannot convert (5.63) into (5.64), for

(5.64) is ungrammatical unless. tomorrow bears heavy stress.

 309

(5.64) Tomorrow I promised chat he would be there.

The adverb tomorrow can be proposed, but only to the front of the

eirml,c,AAc.A #4,^
...so. .c.44
/C 401
vreva/flo

(5.65) I promised that tomorrow he would be there.

Similarly, on the reading of S5.59) where the adverb modifies the

embedded verb, as in (5.60), it can be preposed to yield (5066).

(5.66) I promised that around midnight he would be

there.

Thus it seems that we must propose the following rule:

(5.67) Adverb Preposing12

X- f+Adverbi - 'Y

OPT>
1 2 3

2 + 1 0 3

Condition 1: This rule is last-cyclic.

Condition 2: This rule is upward bounded.

It should be obvious why this rule must be last - cyclic:

if it were cyclic, it would cause the structural descriptions of such

cyclic rules as Equi NP Deletion, Complementizer Placement, Passive

and It Replacement to be complicated. However, if it is a last-cyclic

rule, the only way to prevent the adverb around midnight from

incorrectly being proposed to the front of 01 in (5:60) , instead of

to the front of S2,. J.s to mark it as being upward bounded.

 310

But now let us reconsider sentences (5.62) and (5.64),

when tae preposed adverbs have heavy stress. Sentence (5.62) becomes

ambiguous, and sentence (5.64), ungrammatical without such a stress,

becomes grammatical Such stress and intonation also appears in such

sentences as those in (5.68):

(5.68) a. Beans I don't like.

b. Proud of hip. live never been.

Such sentences are generated by (4.185), the rule of

Topicalization. Topicalization is-not a_bounded rule, as such examples

as (5.69) show.

(5.69) Beans I don't think you'll be able to

convince me. Harry has ever tasted in his

life.

In light of these remarks about Topicalization, it seems

reasonable to suppose that the versions of (5.62) and (5.64) in

which the preposed adverbs have heavy stress shoulA be analyzed as

resulting from the applicatica, cf the rule of 22picalization, not

Acl?.erkPre252s1.2.n. Thus these facts seem to indicate that there is a

syntactic minimal pair here: while all rules which adjoin elements

to the right of variables are upward bounded, rules which adjoin

elements to the left of variables must be marked idiosyncratically,

for some are upward bounded, and some are not.

There is, however, one possibility of avoiding such a

conclusion. It is possible that topicalized sentences such as (5.64),

Opiwile

33.1

(5.68), and (5.69) should not be derived directly by the rule of

Topicalization which was stated in (4.185), but rather from such

"cleft sentences" as those in (5.70), by means of a rule which deletes

the it, the copula and the relative pronoun in these sentences

(sometimes obligatorily), thus converting them into the corresponding

topicalized sentences.

(5.70) a. It was tomorrow that I promised that he

would be there.

.b. It is beans that I don't like.

c.?* It is proud of him that I have never been.

d. It is beans that I don't think you'll be

able to convince me Harry has ever tasted

in his life.

But while such a derivation is possible, I know of no

compelling arguments which indicate that it is necessary. And until

such arguments can be found, the generalization stated in (5.58) cannot

be extended. Nevertheless, the fact that (5.58) holds in all cases

I know of in which terms are permuted rightwards around variables

means that it is not necessary to complicate the formulations of the

three rules of Losition, Extraposition from NP, and NP Shift

which would have to be given in the grammar of English or of any other

particular language. In other words, while neither the principle of

the transformational cycle, nor the AoverA principle, nor any of the

constraints discussed in Chapter 4)is poWerful enough to block the

 312

derivation of such sentences as (5.21) or (5.55b), this can be accomplished

by defining a notion of bounding and adding the empirical generalization

the Tr, rho fnlinwin3 sections I will

coLah , aeu la (5.58) to
_ _ .1
theory.

show that the notion of bounding is necessary.to account for other

kinds of facts as well.

A 6. it, 3

5.1.3.1, In this section, I will show that the notion of bounding

is useful in restricting the power of rules which introduce features,

E 5.1.2. One well-

as well as movement rules like those discussed in

known rule of this type is the rule of Indefinite Incorporation, (5.71)

which Klima proposed in his important article "Negation in English"

(Klima (1964)).

(5.71) Indefinite Incorporation

a. X - (+ Affective] Indeterminate] --

1- 2 - 3 - 4 - 5

1- 2 - 3 - 4 -1 - 5
Indef site

b. X - Indeterminate] Y (+ Affective]

1-
1 2 4 - 5
Indeterminate 3

In this rule, negatives, questions, the word only in

refers to as
certain contexts, and certain lexical items which Klima

"adversatives" (op. cit. p. 314) trigger the change from indeterminate

 313

quantifiers like some, to indefinite ones like ay: Klima uses the

feature Affective] to mark those elements which can trigger this

change. Some examples of the effects of (5.71) can be seen by comparing

the sentences of (5.72) with their corresponding members in (5.73).

.(5.72) a. * I won't have some money.

b.. I fwill; ask you to believe that he tried to

tdon t

force and to give her some money.

c. Do you think that he sometimes went there

alone?

d. at he sometimes went there alone is

certain
odd
e. Do you be eve (the claim) that somebody

was looking for something?

f. I never met that man who somebody tried

to kill.

(5.73) a. I won't have any money.

b. I ask you to believe that he

won t
trie'd to force me to give her any money.
13
c. Do you think that he ever went thre alone?

d. That he ever went there alone is

odd
Do you believe (*the claim) that anybody

was looking for anything?

f. * I never met that man who anybody tried to

kill.
 314

The ungrammaticality of (5.72a) shows that there are

cases where the rule is obligatory. The ungrammaticality of (5.73b),

44 16.&41.4.V. 1J
^
&4V 64GgcsiVG .1.44 1.4iG
Jw J.wAlwwdb.J...w we IMUw
ALJA4.1.0211.iVG Vd. 11141G a.w.:41.

that some's can be converted into any's indefinitely far away from

the triggering (+Affective] element. (5.73c) shows that the change

can take place in questions, and (5.73d) shows why rule (5.71) must

be formulated in such a way that the change can operate backwards as

well as forwards, and also that the adjectives certain and odd must

differ in their marking for the frature [Affective]: the first must

be marked (-Affective], the second (+Affective].

With respect to such sentences as (5.73b), which show

the infinite scope of (5.71), Klima remarks.thatthe change can take

place in the same clause as the one in which, the (+Affective] element

appears, or in any clause subordinate to it. The definition of

"subordinate" which he proposes makes use of the notion in construction

with, which I will discuss in' §-5.2.2 below, but this notion is not

powerful enough to block (5.73f) or the version of (5.73e) in which

the head noun the claim appears. The fact that (5.71) will neither go

down into clauses in apposifion to sentential nouns nor into relative

clauses makes it similar to .:eordering transformations like Question

and Relative Clause Formation in a way which I will argue in § 6.4

is anything but coincidental.

NOt1 3/4:e that there are other environments in which some

is not converted to am. The sentences in (5.74) must not be operated

upon by rule (5.71) to produce the ungrammatical strings of (5.75).

' - - r rii:'

315

(5.74) a. Tom told somebody that he wasn't sick.

b. That Sam smetines didq't sleep must have

pleased somebody.

c. Buffy couldn't do 100 pushups and somebody

laughed.

(5.75) * Tom told anybody that he wasn't sick.

b. * That Sam sometimes didn't sleep must have

pleased anybody.

c. * Buffy couldn't do 100 pushups and anybody

laughed.

The sentences in (5.74) have the structures shown in

(5.76).

(J.76) a.

NP

Tom NP

told ,somebody that NP

he wasn't sick

Amiiip.IMIN11100.0.111111
 316

(5.76) b.

must have pleased somebody.

VP

sometimes didn't sleep

(5.76) c.

Billy couldn't do 100 pushup

If one thinks of rule (5.71) in slightly metaphorical terms,

imagining the (+Affective] element as being a source which "broadcasts"

the feature (+Indefinite] through the tree, the ungrammatical sentences

in (5.75) can be blocked, provided that this broadcasting is upward

bounded, and is not permitted to cross the first double line up from

the (+Affective] source. In other words, while rule (5.71) can effect
changes indefinitely far down the tree from the element that causes the

change, no elements of sentences higher, up the tree than this element

will be affected.

Restricting the rule of Indefinite Incorporation by

making it upward bounded, in .the sense I have just discussed, is adequate

to the task of excluding the sentences in (5.75), but it is not strong

The
enough to block (5.73e) and the ungrammatical version of (5.730.

problems posed by these sentences will be taken up again in §

below. What concerns us at present is not a more precise statement

of rule (5.71), but rather the following generalization about all

rules of the same form as this rule:

(5.77) All feature-changing rules except Pronominalization

rules are upward bounded,

By "feature-changing rule" I mean any rule whose

structural index is of the form (5.78a), and whose structural change

if of the form o.f. (5.78b) or (5.78c).

(5.78) a. ... Al 0.. A2 ...

2

b. Al [ A21
J

CO 0,0 [ Al] Odle A ...

2
4.F

That it is necessary to specifically exclude rules of

pronominalization from the generalization in (5.77) can be seen from

(5.79a) and (5.79b), which are of exactly the same syntactiz type as
 318

(5.74b) and (5.74c). The latter two become ungrammatical if rules

like (5.71) are allowed to apply to them, .while the former two cause

no problems under pronominalization operations, as the grammaticality

of the sentences in (5.80) shows.

(5.79)a.ThatSam.sometimes didn't sleep must have

pleased Sami.

b. Billyi couldn't do 100 pushups, and Billyi

broke dow' and cried.

0.80a.That8am.sometimes didn't sleep must have

pleased himi.

b. Billyi couldn't do 100 pushups and hei broke

down and cried.

It is at present an unexplained mystery why it is that

rules of pronominalization do not conform to (5.77). It will be seen

in § 6.4 below that these rules violate another extremely general

constraint on feature-changing rules, again, for no presently explicable

reason. But the larzbe number of feature-changing rules which are

upward bounded, of which the rules in the next section constitute

a small sample, suggest to me that (5.77) is essentially correct, and

that other factors must be involved in pronominalization.

5.1.3.2.
5.1.3.2.1. As a second example of an upward bounded feature-changing

rule, let us consider facts from Finnish which are closely related to

the facts of Indefinite Into oration L. English.

 319

The Finnish verb tuomaan 'to bring' normally takes an

accusative direct object, as in (5.81).

(5.81) (ginU)1.16in kirjan.

brought the book (ace.):

Although it is possible to construct sentences such

as (5.82), where the object NE's is in the partitive case, such

sentences are unusual and would only be used to convey some such

meaning as "I spent my whole life bringing the book."

(5.82) To in kirjaa.

brought the book (part.).'

But if sentence (5.81) is negated, as in (5.83), the object NP must

be converted, to the partitive case.

(5.83) En tuonut kirjaa.

Not I brought the book (part.) .

'I didn't bring the book.'

The presence of a negative in a higher sentence can

cause accusatives to change to partitives in sentences indefinitely

far down the tree from the negative morpheme. (5.84) shows a simple

case where an element of an originally embedded S changes its case.

(5.84) En pyytanyt hUnta tuomaan kirjaa.

not I asked him to bring a. book (part.).

'I didn't ask him to bring a book.'

Inspection of various other facts, which I will not take

up in detail here, reveals that the Finnish rule, unlike the English rule,
 I 1 In
320
allt

cannot go backwards, so the rule can be formulated, in first

approximation, as in (5.85).

(5.85) Finnish Partitive Iatroduction

Q,
X - [-i-Neg] Y - [-i-acc] - Z
OBLIG

- 2 -3- 4 -5
I. - 2 -3- [ 4 I -- 5
+parti

Since this rule has the form of (5.78), (5.77) will

make it upward bounded. That this is necessary can be seen from the

following sentences. If (5.84) is changed so that the negative morpheme

en is reuoved, and the subject mina 'I' is replaced by a NP containing

a relative clause in which a negative appears, as in (5.86),

kirjan)
(5.86) Poika joka ei mennyt pyytUnyt hentU tuomaan
kirja4

Boy who not went asked him to bring book

The boy who didn't go asked him to bring a book.'

then it is no longer possible to have the object NP of the verb

tuomaan'to bring' in the partitive case, except with the unuzual sense

of (5.82). The structure of (5.86) is that shown in (5.87)

- -
 321

(5.87)

_NP

poika NP

.1.21ca rei V
L+Neg.I 0.

jensit

and since the negative morpheme ei is to the left of and below the

double line emanating-from S2 in (5.87), if (5.85) is upward bounded,

the NP kirjan (acc.) 'book' will correctly be prevented from being

converted to kiriaa (part,) 'book'. Another case showing the same

restriction is that of (5.88a), which rule (5.85) must change to

(5.88b), but not (5.88c).

(5.88) a. En tuonut kirjan, mutta toin lehden.

Not I-brought book (add.), but I brought paper (acc.

b. En tuonut kirjaa, mutta toin lehden.

Not I brought book (part.), but I brought paper (ace.

'I didn't bring the book, but I brought the paper.'

 322

c. * En tuonut kirjaa, mutta toin lehted,

Not I brought book (part.), but I brought paper (part.

Since the structure of (5.88b) is that shown in (5.89),

it is clear that upward bounding will once again suffice to prevent

the undesired change from taking place.

(5.89)

en tuonut kir aa- mutta'toin lehden

5.1.3.2.2. In Russian, too, there_is_a rule which changes case in

the presence of negatives. So while the direct object to 'this' in

(5.90a) is accusative, if the negative inorpheaie ne is introduced, eto

(acc.) is changed to etovo (gen.).

(5.90) a. ja eto sdelal

this (acc.) did

'I did this.'

b. ja etovo ne sdelal

this (gen.) not did

'1 didn't do this.'

 323

A negative in a higher clause can cause cases to change

in infinitival complements, under various complicated conditions which

I will not deal with here. (5.91) .s one example of such a change.

eto sdelat1
(5.91) ja ne xocu
etovo

I not want this (aL. or gen.) to do

'I don't want to do this.'

It is not clear to me that examples (5.91), where

the genitive case depends on a higher negative, can be extended to

any desired length, as is the case in English and Finnish (cf.,

(5.73b)), for the restrictions on this Russian rule have to do

with the verbs of the sentences separating the negative element from

the accusative noun phrase wh:,,ch the rule is to operate on. For

example, the verb xotet !want' allows the negative to affect noun

phrases in Its complement, while the verb nacat. 'begin' does not.

The class of verbs like xo,_ tet, appears to be small, and it may not

be possible to construct sentences of any desired length in which

there are unbroken sequences of adjacent sentences who main verbs

not
are of this class. If this impossible, it may be possible to

reformulate the rule I give below in (5.92) in such a way that no

variable is necessary between terms 2 and 4. In this case, the facts

of (5.93) and (5.94) would not constitute proof that (5.92) must

be upward bounded, so these facts from Russian could not be used

in support of (5.77).
 324

(5.92) Russian Genitive Introduction15

X- [.+Negi Y [+acc] - Z

1- 2 - 3 - 4 - 5

1- 2 3 I 4 -5
L--genj

If it is necessary to state this rule with a variable

as term 3, then facts which parallel those of (5.86) and (5.88) can

be adduced to show that (5.92) must be upward bounded. While the

rule can change vodku (acc.) 'vodka' to vodki16 , thereby converting

(5.93a) to (5.93b), it must be prevented from converting eto to

etovo to yield the ungrammatical (5.93c).

(5.93) a. 61ovek kotoryj ne pil vodku sdelal eto.

man who not drank vodka (acc.) did this (acc.

'The man who didn't drink vodka did this.'

b. celovek kotoryj ne pil vodki sdelal eto."

man who not drank vodka (gen.) did this

'The man who didn't drink vodka did this.'

c. * Celovek kotoryj ne pil vodki sdelal etovo.

As was the case in Finn...sh, since the negative morpheme is

in a relative clause, it can effect no changes in higher levels of the

tree -- (5.92) must be upward bounded. And for the same reasons that

(5.88a) could be converted to (5.88b), but not to (5.88c), (5,94a)

must be converted to (5.94b), but cannot be converted to (5,94c).

 (5.94) a. ja eto ne sdelal, no eto ja sdelal

I this (ace.) not did but this (acc.) I did

'I didn't do this, but I did do this.'

b. ja etovo ne sdUal, no eto ja sdelal.

I this (gen.) not did, but this (acc,) I did

'I didn't do this, but I did do this.'

c. * ja etovo ne sdelal, no etovc ja sdelal.

The structure of (5.94b) is that shown in (5.95):'

(5.95)

ja etovo ne,delal no eto is sdelal

Since the negative morpheme ne is upward bounded, the

eto (acc.) in the second clause will be prevented from being converted

to etovo (gen.), and the ungrammatical (5.94c) will not be generated.

5.1.3.2.3. As was noted in footnote 15, the Russian rule of

Reflexivization can affect noun phrases which were in different

clauses in deep structure. An example of the operation of this rule

is provided in (5.96), where (5.96a) is obligatorily converted to (5.96b).

 326

(5.96) a. * on uvaYajet jevo

i i

he respects him (ace.)

b. oni uvazajetsebjai

'He respects himself.'

III
An example showing the conversion of an NP which is

the object of an infinitive into a reflexive pronoun is the optional

change of (5.97a) into (5.97b).

17
ljevoi
(5.97) a. on. sostavil menja uvazat,

i::::jai
b. 'He forced me to respect
himself

The rule which effects these changes is approximately

that stated in (5.98).

(5.98) Reflexivization

X - NP - Y - NP - =

1 2 3 4 5

1 2 3 4 1 5
L+Refl-.1

Condition: 2 = 4

By the generalization in (5.77), this rule will be marked

as being upward bounded. That this is necessary can be seen from the

fact that (5.99a) cannot be converted into (5.99b) by rule (5.98).

 327

(5.99) a. vze,raina kotoruju on ljubil sostavila

i

woman who he loved forced'

menja uvazat, jevoi.

me to respect him

'The woman who he loved forced me to respect

him.'
v
b. * zenscvv ina kotoruju on ljubil sostavila
i
menja uvaIat, sebjai.

The string of words in (5.99b) is a grammatical sentence,.

and can mean either 'The woman who he loved forced me to respect her',

or'The woman who he loved forced me to respect myself.' But it

cannot be synonymous with (5.99a), which is the reading which is

of Literest here. Since (5.99a) has the structure shown in (5.100),

the fact that (5.98) is upward bounded will prevent this undesired

conversion from taking place.

(5.100)

NP////7

V V71
Y
ZengLna S sostavila NP

kotoraia NP
f1/4 VP uvazvat
.

i2Y21

.oni ljubq
 328

Similarly, (5.101a) must not be converted into (5.101b).

(5.101). a. oni: ljubit YenMinu, i ja uvadju jevoi.

'Hei loves the woman, and I respect himi.'

v vv,
b. * oni ljubit z enscInu, i ja uvazV skju sebjai.

Once again, (5.101b) has a meaning, but not the same meaning that

(5.101a) has. It means 'He loves the woman, and I respect myself.'

Since (5.101a)has the structure shown in (5.102),

(5.102)

on liubit'zenscinu i j a uvazaju levo

this conversion will be prevented by the fact that rule (5.98) is

upward bounded.

At present, (5.98) is still too strong, for it will

allow (5.103a) to be converted into (5.103b).

(5.103) a. on znaet sto ona ljubit jevo.2.

'Hei knows that she loves himi.'

b. * oni znaet sto ona ljubit sebjai.

While (5.103b) can mean 'He knows that she loves herself',

it cannot be synonymous with (5.103a). Therefore, reflexives must

 329

somehow be prevented from being introduced into subordinate clauses.

I will defer discussion of this pro-.?em until § 6.4 below.

5.1.3.2.4. In Japanese, the reflexive pronoun zibun, which, like

sebia, is used for all persons, can be introduced into clauses, as

the conversion of (5:104a) into (5.104b), whose structure is shown.

in (5.105), demonstrates:

(5,104) a. Mary wa kare ga byooki de aru to

i i

Mary she , sick is that say

koto o sinzite iru.-

fact believing is.

'Meryi believes that shei is sick.'

b. Mary wa zibun ga byooki de aru to iu

i
koto o sinzite iru.

(5.105)

loyooki de aru

tt
 330

As a first approximation, it appears that the Japanese

rule of Reflexivization can be stated the same way the Russian rule

was. And, just as the Russian rule is, the Japanese rule must be

upward bounded. This can be seen from the fact that (5.106a), whose

structure is shownfin (5.107), cannot be converted to (5.106b).

(5.106) 'a. Maryi ga byooki de atta to iu koto wa

Mary sick was that say fact

karei ni akiraka de atta.

she to obvious was.

'That Maryi was sick was obvious to here'

b. * Maryi ga byooki de atta to koto wa

18
zi buni ni akiraka de atta.
.

(5.107)

5) to iu koto wa

karei.ni a iraka de atta

Mary g byooki de atta

.-

*-

f
 331.

Since the circled antecedent NP in (5.107) is to

the left of and below a double line, as seen from the boxed NP,

upward bounding will prevent rule (5.98) from converting this structure
---

into (5.106b) .

5.1.3.2.5. For a sixth example of a feature-chanoing rule which is

upward bounded, let us return to Finnish. The rule stated in (5.108)

(5.108) Finnish Nominative Introduction

[+acci y.
V x -Pro_i
1 2 3 4

1 2 3 1 4

accounts for the fact that in sentences whose subjects have been

deleted, e.g., in impersonal sentences,or ia imperatives like (5.110b),

all non-pronominal noun phrases in the accusative case are converted

to the nominative case. Thus in (5.109), which at this stage of the

derivation still has a subject mint, 'I', the direct objects of the

verbs zyyte 'to ask' and tuomaan 'to bring' appear in the accusative

case.

(5.109) Mina koetin pyytaa- pojan tuomaan

'I am trying to ask the boy (acc.) to bring

kirjan.

the book (acc.).'

But in the structure underlying an imperative sentence, after the

subject NP 'sins 'you' has been deleted, as in (5.110), the direct

objects must be converted to the nominative case. Thus (5.110a) must

become (5.110b) .
 332

(5.110) a. Koeta pyytad pojan tuomaan kirjan.

try to ask the boy (ace.) to bring the book (acc.)

b. Koeta pyytad poika tuomaan kirja.

'Try to ask the boy (nom.) to bring the book (nom.).

That (5.108) must be upward bounded can be seen from the fact that
19
(5.11a), whose structure is that shown in (5.112), must be converted

into (5.111b) , and not into (5...111c).

(5.111) a. Tuo kirjan, ja mita tuon

you bring the book (ace.) and I will bring

lehden.

the paper (acc.)

b. Tuo kirja, ja mina, tuon lehden.

'Bring the book (nom.), and I'll bring the paper (acc.

c. * Tuo kirja (nom.), ja mina tuon lehti (nom.).

(5.112)

VP NP

V NP mina

t U0 'tuon lehden
 So".4.011Metasra.

333

5.1.3.2.6. The last feature - changing rule which I will discuss in

support of (5.77) is the rule which changes tense, ill some contexts

obligatorily, so that it agrees with the tense of some other verb in

the sentence. Thus while both is and was are possible in (5.113),
20
only was is in (5.114) .

(5.113) I. believe that the sun s out.

as

I believed that the sun *is out.

(5.114) ws
a

Although much more research must be done on this,

traditional phenomenon of sequence of tenses, it seems reasonable

to me to assume that the rule which effects the change of tense must

be formulated roughly as shown in (5.115).

(5.115) 'Sequence of Tenses

a. X -144/ Y HAT] Z
L aTensei

1 2 3 4 5

1 2
3 [aT:nse]

4-11
b. X - PV] Y aTensej

2 3 4

1 [ 2 j 3 6 5
aTens

It is necessary to formulate this rule so that it can apply

in both diretions, so that sentences like (5.116) will be excluded.

(5.116) * That the sun is out was obvious.

.,&
11:

That this, rule is far too strong can be seen from the

fact that it would only allow the version of (5.113) ia which is appears

to be generated. This indicates that the tense agreement which rule

(5.115) effects is much too simple-minded a change, and that the correct

rule will have to provide fox a much more complex mapping.

It is equally obvious, upon a moment's introspection,

that (5.115) must be upward bounded, so that it will allow the generation

of both versions of (5.117).

is
(5.117) Taat I believed that the sun was out obvious.
was

If rule (5.115) were not upward bounded, it would make

all the tenses in (5.117) agree with believed (or with one of the other

verbs in (5.117)), thus making the incorrect claim chat sentences cannot

"mix tenses", and that the version of (5.117) containing is is ungrammat-

ical.

The six examples in this section of upward bounded feature-

changing rules provide compelling evidence that the generalization

expressed in (5.77) is a correct one. Furthe: consequences of this

generalization will be taken up in § 6.4 below.

5.1.4.. In § 5.1.2 and § 5.1.3, I have presented evidence

which indicates that it must be possible to limit the upward range of

application of both reordering transformations and feature-changing

rules. In this section I will discuss three cases which suggest that

, .
 335

it is also necessary to be able to limit the downward range.

For a first example, let us ,*c.,14,-°ct our attention to

the English rule of Refle.x4v4-":4^,.. rn & 41;6 above; I mentioned

that in Lees and Klima (1963), the term "simplex sentence" is us ad to

restrict the scope of application of this rule. The question which

should now be raised is the following one: should both this notion

and the notion of upward bounding be defined in the theory of grammar?

Or should the former notion be defined as a conjunction of upward

bounding and a new kind of bounding -- downward bounding? A rule is

upward bounded if it cannot permute constituents into,, or change

features in, a higher clause, and, correspondingly, a rule would be

downward bounded if it could not effect such changes in lower clauses.

It seems to me to be desirable to "decompose" the notion

of simplex sentence into the two notions of upward and downward

bounding, for the following reasons. Firstly, the arguments in the

previous sections indicate that regardless what decision is made with

respect to the English rule of Reflexivization, the notion of upward

bounding must appear as an element of the theory of grammar. To

characterize the difference between the English and the Japanese rules

of Reflexivization, some auxiliary primitive term must be added to the

theory -- either simplex sentence or downward boundtaa. If the former

term is chosen, then the fact that the restrictions on the English

rule are in part universal cannot be captured. For the fact that

elements of higher clauses cannot be reflexivized in English is a

 336

consequence of (5.77), since Lailexivization; is a feature-changing

rule, The only way to express the fact that the English rule is

partly universal, within a theory which only contains the primitives

lipward boundiag and simplex is to complicate (5.77) in an

ad hoc way, as has been done in (5.118).

(5.118) 'All feature-changing rules are either upward

bounded or restricted to apply within a simplex

sentence.

Since the notion of simplex sentence would be unanalyzed within such

a theory, it would be impossible to capture the intuition that the

a

English rule is identical to the Japanese rule (or to the Russian

rule -- all three can be stated as in (5.98)), except for containing

21
an additional restriction which is not stated on the latter two rules.

So for the purposes of the present discussion, I will assume that the

theory contains as primitives the notions of upward and downward

bounding. This assumption will be modified in § 5.2 below.

The second example of a rule which requires the use of

the notion of downward bounding is the Scramblin&lule, (3.48), which

was discussed in § 3.1.2 above. As noted in the condition on (3.48),

major elements in a Latin sentence can scramble, provided that they

are in the same clause. This restriction on (3.48), the statement

of which required quantifiers (cf. Ch. 3 footnote 7), can now be

achieved by marking (3.48) as a rule which is upward and downward

bounded.

<<
 337

The third case where downward bounding seems to be

necessary, although not sufficient, is in connection with the rule of

Serbo-Croation Clitic Placement, (3.63), which was discussed in 5 3.1.4

above. There I pointed out that clitics must be moved so that they

follow the first constituent of the first sentence up -- thus the

rule must be upward bounded. However, it is also necessary to stipulate

that (3.63) be downward bounded, so that the clitics cannot be inserted

after the first element of a sentential subject clause. In other

words, the circled clitic in (5.119) must not be allowed to follow

V"'

the path of the dotted arrow, but only that of the solid arrow.

(5.119)

NP VP

`1
///\\
NP VP

EP,

Such an incorrect positioning of a clitic can be avoided if (3.63) is

22
marked as being downward bounded, in addition to being upward bounded.

The three cases I have just discussed indicate that an

adequate theory of bounding 2.st countenance both upward and downward

bounding. At present, however, there is a puzzling redundancy, which

--^
 338

cries out for explanation: all downward bounded rules are upward

bounded, but the converse is not true. That is, while there are

rules whose scope extends indefinitely far down the tree from the

triggering element or context, but does not extend upward,there are

no rules whose scope extends indefinitely far up the tree, but not

downward. I will present the first steps toward. an explanation of

this asymmetry in § 6.4 below.

5.2. Command

5.2.1.

5.2.1.0. In § 5.1, I discussed several problems which necessitated

the addition to linguistic theory of some new mechanism, and to this

end I proposed the particular device of bounding. In this section,

I will show that Langacker's notion of command can account for all the

facts adduced in support of bounding, and in addition, facts which

cannot be accounted for with bounding. Furthermore, I will show that

Klima's notion in construction with is too weak to account for all

facts which can be handled with command.

5.2.1.1. Langacker defines command in a definition which is

equivalent to that staffed in (5.120) (cf. Langacker (1966) , p. 11):

(5.120). Node A of a phrase marker commands node B

if neither node dominates the other, and if

node B is dominated by the first node 'S

above A.
 339

To give an example, in phrase-marker (5.121),

A commands and is commanded by B, C, and D; and C and D

command each other. E, F, and G command S2, A, B, C and D.

S2, F, and G command each other, as do S2 and E. 14 and N

command each other, and are commanded by only Sl. Nodes A, B,

C, D, E, F, G and S2 neither command nor are commanded by 11

and N.

A moment's reflection will convince one that command can

be used in place of upward bounding in all feature-changing rules. For

instance, to say that Indefinite Incorporation, (5.71), is upward

bounded is to say that the feature [4, Affective] cannot "broadcast"

the feature Indefinite] upwards across double lines in a phrase-

marker. Rephrased in terms of command, the restriction would be that

 340

the Affective] element must command any (+ Indeterminate] element

to which it adds the feature Indefinite]. It is simple to replace

ai a a- a - , - I. J:._
A.CD43.1A06.Vi4 VA UpW4A04 UVUALU..0
r_ -
A.Vi.
- a al.. a -a _2. & a a
&AUG V6liCir DAA AGAUUMG-41a46..a6
al a

rules discussed in § 5.1.3.2: the rule of Finnish Nominative

Introduction, (5.108) must have the restriction imposed on it that

term 1 command term 3, rule_ (5.115b.) must be restricted so that

term 4 commands term 2, and the condition, which must be imposed on

the other five rules is that term 2 command term 4.

Furthermore, just as it could be predicted that all

feature-changing rules are upward bounded, the conditions stated

in the last paragraph can be derived automatically from (5.122),

which is the analog to (5.77).

(5.122) Except for rules of pronominalization, in all

feature-changing rules, elements to which

features are added must be commanded by any

non- variable terms appearing in the structural

indices of the rule in question.

5.2.1.2. Langacker cites the rule of Indefinite Incorporation as

an example of the usefulness of command, and on pp. 27-32, he

discusses two examples of rules which move constituents and their

relationship to his important notion of control. He does not consider

rules such as Exaapsition, which the discussion in § 5.1 showed

to be necessarily upward bounded. But once again, it is easy to

.- -r 1"-^1%.-, zvt,
 343.

dispense with upward bounding as a device for preventing extraposed

constituents from going too far. If the, definition in (5.120) is

i &4446Ultioi .hat the relati-- ^1 ^^"""'" hAlA'

not only between one node and another, but may hold between one node

and a sequence of nodes, if and only if the first node commands

each of the nodes in sequence, then instead of designating a rule

A
such as Extr4ositior., (4.126), as being upward bounded, we can

impose the condition on it that the clause to be extraposed command

the variable in the fourth term of its structural index. Moreover,

the generalization expressed in (5.58), that all rules which adjoin

a term to the right of a variable which occurs on the right end of at

structural description are upward bounded, can be equally well expressed

in ,terms of command, as in (5.123):

(5.123) In all rules whose structural index is of the

form ... A Y, and whose structural change

specifies that A is to be adjoined to the

right of Y, A must command Y.

Having stated this generalization in the theory of grammar, it is not

neccssary to attach any conditions to the rules of Extrapositinn from NP,

(1.10), Extraposition, (4.126), and NP Shift, (5.57): (5.123) has

the effect of constraining the structural changes of these rules the

same way the conditions would. And it is evident that the operation

of the upward bounded rule of AdverbtraogaR, (5.67), can be correctly

discinguished froLkthat of the unbounded rule of:Topicalization, (4.185),

 342

if a condition that term 2 command term 1 is imposed upon the

former rule, but not upon. the latter. .

Finally, note that all the cases presented in s 5.1.4

in support of downward bounding, which Y originally believed not to

be accountable for within a theory of grammar in which only command

was available, can in be .accounted for by stating two conditions

in terms of command. That is,, instead of ensuring chat only elements

of the same clause can be scrambled by designating the rule of

Scrambling, (3.48), as being upward and downward bounded, this effect

can be achieved by requiring that terms 2 and 3 of rule (3.48)

command each other. This condition makes it impossible for the elements

being permuted in (3.48) to be in different clauses: if A were a

member of a clause which did not contain B, then A would not

command B, and conversely. To specify that two nodes command each

other is to specify that each is dominated by the first node S above

23
the other, and because of the formal properties of trees, these

S nodes must be the same. That is, two nodes which command each

other are in the same simplex sentence.

Although Langacker remarks in passing that it is possible

to restrict the scope of a transformation by the use of double command

conditions, he gives no examples where this device is necessary. It

should be clear that the other two examples cited in § 5.1.4,

the English. rule of Reflexivization and the rule of Serbo-Croatian

Clitic Placement, can also be formulated in terms of double command.

 343

Thus a theory in which command is an available primitive is at

least as powerful as a theory which provides upward and downward

bounding. Before showing that the former theory is stronger than the

latter in a crucial way, I will digress to show that Klima's notion

in construction with is not qtrong enough.

5.2.2.. Klima's notion is defined as in (5.124) (cf. Klima (1964),

p. 297):

(5.124) Node A of a phrase - marker is in construction

with node B if B is dominated by the node

which immediately dominates A.

That this relation is stronger than command can be seen from (5.121),

where E, F, and G command S2, A, B, C and D, but where

only E is in construction with these latter five nodes. Klima

proposes to constrain the operation of rule (5.71) by imposing on

it the condition that the [-I- Affective] element be in construction

with the j.f Indeterminate] element which is to be changed. That

this condition is too strong can be seen from (5.125a), which (5.71)

must be able to convert to (5.125b).

(5.125) a. That Jack sometimes slept is impossible.

b. That Jack ever slept is impossible.

c. * That Jack ever slept is possible.

The ungrammaticality of (5.125c) shows that it is the negative prefix

int- that contains the feature Affective] and triggers the change.
 344

But the structure which Klima would assign to (5.125) (cf., e.g.,

op. cit. p. 298, fig. 4) is that shown in (5.126),

UMW

(5.126)

NP Pre. cate

Aux

that Jack sometimes slept_ tns Pre cate

Prs

fective] possible

and in this structure, the circled node geg, which carries the feature

[4. Affective], is not in construction with the occurrence of sometimes

in the subject clause, although the latter word is commanded by the

circled node. Thus with respect to rule (5.71) there is at least

one structure for which Klima's notion produces the wrong results, and

Langacker's notion the correct ones. Langacker's notion must therefore

be chosen even if only the facts connected with rule (5.71) are taken

into consideration.

But there are even more important respects in which the

notion of command is superior to the urtion in construction with. As

0
 345

showed in 6 5.1.3.2, all feature-changing rules except rules of

pronominalization are upward bounded. This extremely powerful gener-

alization, to which I know of no counterexamples, can be restated

- in terms of the notion of command, as was done in (5.122). But this

generalization cannot be reformulated in terms of the notion in

construction with. (5.127), in which I have stated such a reformulation,

is too strong.

(5.127) In all feature-changing rules, non-variable terms

are in construction with the terms to which the

features are added.

To see that (5.127) is too strong, consider (5.128), the

structure of (5.129a).

(5.128)

I NP

talked

Wins tons about Win tan

i
 346

Winston
(5.129) a. * I talked to Winstoni about i
him

b. I talked to lelinstoni about himself

Since the English rule of Reflexivization is a feature-changing

24
rule, and since the circled NP node in (5.128) is not in

construction with the boxed NP node, generalization (5.127) would

incorrectly prevent Reflexivization from converting (5.129a) into

(5.129b). But Reflexivization is obligatory in such structures as

(5.120, so (5.127) must be wrong.

Another rule which provides counterevidence to (5.127)

is the rule for Sequence of Tenses, (5.115). String (5.130a) must

be converted into (5.130b) by this rule,

(5.130) a. * That the sun is out was obvious.

b. That the sun was out was obvious.

but since the structure of (5.130a) is that shown in (5.131),

(5.131)

NP

41 obvious
that the sun is'out 4-Pasti

was

.
 1
347

where the tensed yell) was is not in construction with the verb is
25
in the sentential subject, the generalization in (5.127) would not

.11ou the change to take place.

The third argument for choosing command over in

construction with is that while the important notion of simplex

sentence can be captured by the use of two conditions making use

of command, tLis cannot be done with the notion in construction with.

To say that two nodes command each other is to say that they are elements

of the same simplex sentence, but to say that they are in construction

with each other is to say that they are siste'*s.

The above arguments indicate that the notion of command

cantlt be replaced by the notion in construction with, but of course

they do not show that the latter notion cannot supplement the former

in linguistic theory. To account for the facts in § 5.1 and § 5.2.3,

the notion of command, or its equivalent, must be defined in linguistic

theory. While the notion in construction with is not the equivalent

of the notion of command, it is possible that phenomena will come to

light whose analysis will necessitate the inclusion within linguistic

theory of the former notion. At present, no such facts are known.

5.2.3.

5.2.3.1. In this section I will discuss two problems which can be

solved within a theory in which command is defined, but not within

one in which only bounding is available.

a
 AV/

348

Consider first the following facts about identity:

so did Mall'
(5.132) John scratched his arm and
Mary did (so) too

The second clauses of the sentences in (5432) are ambiguous - they

could be derived from the structure underlying (5.133a) or the one

underlying (5.133b).

(5.133) a. Mary scratched her arm (too).

b. Mary scratched John's arm (too).

Thus it appears that linguistic identity must be defined

in such away that the difference between his arm in the first clause

of (5.132) and her arm in (5.133a) is "disregarded." However, it is

not the case that all differences between pronouns can be disregarded:

(5.134a) cannot be transformed into (5.134b).

(5.134) a. John scratched his arm and the boy whc. knew

Mary scratched her arm.

b. John scratched his arm and the boy who Nary

knew did so too.

These facts can be accounted for if the following definition

26
of identity is adopted in the theory of grammar:

(5.135) Constituents are identical if they have the same

constituent structure and are identical morpheme -

for- morpheme, or if they differ only as to

pronouns, where the pronouns in each of the identical

constituents are commanded by antecedents in the

non-identical portions of the phrase-marker.

.
 349

Thus in (5.136), which underlies one reading of (5.132),

the circled NP's John and Mary command the circled pronouns his and

her, so deletion is possible under the definition given in (5.135).

his arm

On the other hand, in (5.137), which underlies (5.134), J_ ohn commands

his, but the boxed NP Mary does not command its pronoun her, so

(5.135) will not let the deletion go through.

 350

(5.137)

scratched

The same facts obtain for right-to-left pronominalization:

(5.138a) can be derived from (5.138b) ox (5.138c) , because the circled

27
noun phrases command the pronouns which refer to.them.

(5.138) a. That the fuzz wanted him worried Inhni but

it didn't worry Mary.

b. That the fuzz wanted him worried,

that the fuzz' wanted didn't worry

c. That the fuzz wanted him worried ohm but

Ch-t 43. fuz- Tri

^h Ynryt -
 351

Note, however, that just as (5.137) cannot be converted

into (5.134b), (5.139) cannot be converted into (5.140) a for while

the circled NP 'John (5.139) commands its circled pronoun, him,

the boxed NP Lam. does not command its boxed pronoun, her.

(5.139)

but N' VP

didn't worry NP
N_

that the fuzz wanted

that the fuzz wanted NP

thelia who

(5.140) That the police wanted him worried Johns but it

didn't worry the boy who Mary knew.

know of no reason to assume that the relation of identity must be

defined in language-particular terms, so some revised version of (5.135)

will appear in the theory of grammar. And since (5.135) makes crucial

use of the notion of command, this definition provides strong support

for the hypothesis that command is a primitive term of the theory of

 352

grammar, and not the notion of bounding. For notice that bout_ing,.

was devised to restrict the scope of a process -- it has to do with

the structural changes of rules which move constituents or features- -

and that here some static relation is necessary, in order for the

conditions under which a process can take place to be established. It

is because of this difference in function that bounding is intrinsically

unsuited to the task of defining linguistic identity.

5.2.3.2. It is for the same reason that command, but. not bounding,

can handle the following facts. There is a well-known restriction

28
that excludes negatives in than-clauses. Somehow, all the sentences

in (5.141) must be excluded, while the ones in (5.142) must be allowed.

(5.141) a. * John is prouder of having gone than nobody

expected me to believe he would be.

b. * ....than John didn't expect me to believe ...

c. * ....than John expected nobody to believe ...

d. * ....than John expected me not to believe ...

e. * than John expected me to believe not all

my friends were.

f. * ....than John expected me to believe that

he wasn't.

(5.142) a. John is prouder of having gone than people

who don't know him would expect me to believe

he would be.
 b. ....than Sally expected Joan tobelieve

that the man who didn't shave would be.

c. .,.than I expected you to believe he would

be of not having fallen asleep.

In other words, to exclude all negatives from than-

.

clauses would be to incorrectly exclude the sentences in (5.142). The

difference between (5.141) and (5.142) can be expressed naturally if

conditions on rules can be stated which make use of command. To

exclude the sentences in (5.141) it is sufficient to say "The feature

u29
(+negative] may not command the compared element in the than-clause.

Since the negative dements in (5.142a) and (5.142b) are in relative,

clauses, they will command only the other elements of these clauses.

And the not of (5.142c) is one clause lower than the compared adjective,

proud, so all the sentences of (5.142) will be generated. Bit in each

of the sentences in (5.141), proud is commanded by a negative element,

so all will be blocked by the condition stated above.

Once again, since what is required here is the statement

of a static precondition for the operation of a rule, these facts cannot

be accounted for with bounding. Therefore, in conjunction with the

facts about identity discussed above, and the rules which Langacker

discusses on pp. 27-33 (op. cit.), which require Langacker's principle

of control for their correct application (this principle is also not

susceptible of reformulation in terms of bounding), these :acts abo'it

comparatives seem to me , make the choice between bounding and command

 354

obvious: command, as defined in (5.120), is a part of the theory of

gra=ar, while bounding is not.

5.3. Pronominalization

5.3.0. Thus far, in this work, I have discussed constraints on

variables in reordering transformations (in Chapter 4 and in 5S 5.1.1 -

5.1.2) and constraints on variables in feature-changing rules (in

5.1.3 and § 5.2). There is another kind of process whose scope

is unbounded, the statements of rules for which also make crucial use
30
of variables -- pronominalization. In § 5.3.1, I will discuss

several kinds of pronominalization and show that not all transformations

which delete under identity make crucial use of variables. In § 5.3.2,

I will argue against Langacker's contention (cf. Langacker (op. cit.))

that constraints on variables in rules of pronominalization can be

stated in terms of command. In § 5.3.3, I will discuss four 'rules

of pronominalization, which appear, at least at the present state of

knowledge, to have to be stated as distinct processes, showing that

they obey the same constraint which the rule that introduces the

definite pronouns is subject to. Finally, in § 5.3.4, I will show'

that they obey no other constraint thus far discussed, and discuss

the possibility that_the constraint stated in § 5.3.2 is universal.

5.3.1. The most natural definition of pronominalization is

deletion under identity. This definition covers a number of operations,

 355

which, though unbounded in scope, do not made crucial use of variables

and will not be dealt with here. For instance, the rules which convert

the sentences in (5.143) into the corresponding ones in (5.144) must be

formulated as schemata, and I will not discuss such rules here.

(5.143) a. Tom knows it and Dick knows it and Harry

knows it.

b. Tom washed the car, and Dick waxed the car,

and Harry polished the car.

c.. Tom ate, and Dick drank, and Harry sang.

d. Tom ordered bacon, and Dick ordered lettuce,

and Harry ordered tomatoes.

(5.144) a. Tom, Dick, and Harry know it.

b. Tom washed, and Dick waxed, and Harry

polished the car.

c. Tom, Dick, and Harry ate, drank, and sang,

respectively.

d. Tom ordered bacon, and Dick lettuce, and

Harry tomatoes.

Although rules like Gapping, the rule which converts

(5.143d) into (5.144d)1 can apply to delete the verb of an indefinitely

large number of consecutive conjoined sentences)it cannot be formulated

with a variab16, for otherwise it would convert (5.145a) into the

ungrammatical (5.145b).

A
 356

The superficial similarity of (5.148) to a feature-

changing rule should not be deceptive. For the feature [ +Pro] is

not a feature like the Lt Indefinite] of (5.71) or the Nom] of

(5.108) -- it is an instruction to delete all or part of the

constituents of the node to which it is attached.' So if some

rule of the form of (5.148) converts (5.149a) into (5.149b), by

adding the feature [+ Pro] to the circled NP,

(5.149) a.
v.

N?

who we watched

the prize
/
NP

which

the man who we watched

 ti

.(5.145) a. Tom ordered bacon, and Dick ordered lettuce,

and I think that Harry ordered tomatoes.

b. * Tom ordered bacon, and Dick lettuce, and

think that Harry tomatoes.

.There are also a nuzber of rules which reduce identical

elements if these occur in designated constructions. For instance,

(5.146a), may be converted into (5.146b) by the operation of one

I
such rule.

(5.146) a. Joe is taller than Mary is.

b. Joe is taller than Mary.

However, this rule must not be stated in a way that ',lakes crucial tse

of variables, or else (5.147a) would be converted into the ungrammatical

(5.147b).

(5.147) a. Joe is taller than I think .Nary is.

b. * Joe is taller than I think Nary

- I will therefore restrict my attenticin to those rules of

pronominalization whose structural index is like that shown in (5.148a),

and 'hose structural change like one of the versions of (5.148b) or

(5.148c)

(5.148)32a. ...A ..X...A .... Condition: Al = A

1. 2 2

b. ...A ...X.. [-A2 c] ...

1
, ..
+Pr
0
.

c. ..*. f[A1 j. X......A

2" .
 357

b.

V T%

the man who we watched

which [ NP
-1.Prol_

tc'..- wanted

. the mans who we watched

---

some later rule or convention must reduce all of the NP so marked to

33
the single word he. In other cases, the deletion is complete, as

in the conversion of (5.150a) to (5.150b).

(5.150) a. Mike will sing if you will sing.

b. Mike will sing if you will.

Furthermore, rules of pronominalization are not upward bounded, as

was shown with reference to the sentences in (5.80), and they will

be shown, in § 5.3.3, not to be subject to the constraints of Chapter 4,

 358

which appear to constrain all other feature-changing rules '(cf. 6.4

below).

5.3.2. Most rules of pronominalization produce paradigms like

the one in (5.151).

(5.151) am. will go if he feels good.

b. * Hei will go if Jim. feels good.

c. If Jimi feels good, hei will go.

cLutle.feelsgood,nra.%.7ill go.

I have argued elsewhere (cf. Ross (1967a)), that the

constraint which is operative here is the one stated in (5.152):

(5.152) Condition on backward ;pronominalization

If one element precedes another, the second caa

only pronominalize the first if the first is

dominated by a subordinate clause which does

34
not dominate the second.

There are two instances of right-to-left, or "backward"

pronominalization in (5.151) - (5.151b) and D.151d). Since the

if-clause is a subordinate clause, the latter is grammatical, while

the former is not.

Langacker propoEes a dif.erent condition on backward

pronominalization (cf. op. cit. pp. 11-22), the gist of which is

stated in (5.153).

(5.153) One noun phrase may pronominalize another unless

the first both precedes the second and.is commanded

by it.
 359 .

These conditions are almost identical, but not quite. To see this,

consider the two sentences of (5.154) (these are the sentences numbered

(72) And (71_ rpgpectivalv_ In LangnnkAr. (nn_ cit,)),

(5.154) a. I gave the book to Harveyi because hei

asked me to.

* I gave the book to himi because Harveyi

asked me to.

Langacker derives (5.154a) from the intermediate structure shown is

(5.155) :

(5.155)

P;DP

NP PD.P
%mallows.*

Harveyi Harvey

Since the circled NP in this structure upth precedas

and commands the boxed NP, the condition on pronominalization stated

in (5.153) will suffice to prevent (5.155) from being converted to (5.154b).

 But this explanation of the ungrammaticality of (5.154b)

is only as good as the constituent structure on which it depends,

so let us iaquire as to the adequacy of the represeatatioa ia (5.155) .

In all traditional accounts, what 'would be said about

(5.154a) is that it contains two clauses, the main clause being I gave

the book to Harvey, and the subordinate clause being because he asked

me to. Such a parsing would yield some structure like that shown in

(5.156).

(5.156)

ave the book to

This structure is surely in far better accord with

intuitions about the constituency of (5.154a) than is (5.155): the

latter makes the counterintuitive claims that the major break in

(5.154a) occurs after the pronoun I, and that I gave the book to Harvey

is not a constituent. But Langacker's condition on pronominalization,

(5.153), is not strong enough to block (5.154b), if the structure

underlying it is like (5.156), rather than like (5.155). For while

the circled NP in (5.156) precedes the boxed NP, it is not commanded

by it and (5.153) .blocks pronominalization only if both of these

conditions obtain.
 361

There is another reason.io believe (5.156) to be correct,

and (5.155) incorrect. In Langacker (op. cit. footnote 13),

tangnnkwr Aict,tieciac vhe 1-11,-mD oanveinnmc o f is 1c71

. (5.157) a. That I might want to leave never occurred to

Harvey because he is insensitive to other

people's desires..

b. It never. occurred to Harvey that I might

want to leave because he is insensitive

to other people's desires.

c. * It never occurred to Harvey because he is

insensitive to other people's desires that,

104.
I might want 414.U6.1,,w40

Langacker correctly concludes that the structure underlying

ig.7n1 ga mnrA nearly basic than the one underlying (5.157b), but

he proposes to derive both from (5.158).

(5.158)

ever occurred to Harvey

it t at I might want to e e.
because he is insensitive to other
people's desires

- a
 362

Having assumed such a structure, he is forced to conclude that the

rule of Extraposition must be formulated to permute S2 around VP,

and not around a variable, to the end of S1. However, if Extraposition

is stated in this restrictive manner, it will be necessary to state in addition

another rule, so that sentences like those in (5.159) can be derived,

(5.159) a.. I figured it out that she was lying.

b. I explained it to Bill that she was lying.

c. I took it for granted that she was lying.

d. I regret it exceedingly that she was lying.

for here, the extraposed clause does not move over a VP.

Since it is clearly wrong to treat (5.157b) and the

sentences of (5.159) as being produced by different processes, another

solution to the problem of excluding (5.157c) must be sought. The

most satisfactory analysis, in my view, is to derive (5.15:h) from:

(5.160).

(5.160)

never occurred to Harve

it that, I might want to leave because he is insensitive to other

people's desires
 363

The clause to be extraposed, S2, must command any string over which

it is permuted (by the generalization stated in (5.123)), and aincc

S2 commands IT_ in (5.160); and does not command S-. (5.157b)

1
S
can be generated when EjssrakoA.tion appliel to (5.160), but not (5.157c).

Therefore, since (5.160) produces none of the d.c.s. inadequacies

noted in connection with (5.155), and since it requires no unpalatable

proliferation of. rtles of extraposition, I conclude that it, and not

(5.158), represents the correct structure of (5.157a) , and that

similarly (5.156) and not (5.155), the correct structure of (5.154a).

If (5.156) and (5.160) are correct structures, then

backward pronominalization cannot be blocked by Langadker's condition,

(5.153), although it can be blocked by (5.152). It is for this reason

that I have rejected condition (5.153) in favor of (5.152), but it

should be noted that there are a number of interesting facts having

t A^ with varyine degrees of naturalness in pronominalization (cf.

Langacker (op. cit.) pp. 16-18), which can be accounted for with the
0

former condition on pronominalization but not with the latter. I

therefore regard the matter as anything but closed, and my assumption

1
below that (5.152) is correct should be treated as. being only provisional.

5.3.3.

5.3.3.0. Below, I will discuss briefly four kinds of pronominalization

which produce paradigms like the one in (5.146). It may turn out that

they only appear dissimilar and can really be shown to be subcases of

 364

the same rule, but I will not attempt such a proof here. I will merely

show that they are similar to the rule which produces l.finite

pronouns in that all ate subject t the condition stated in (5.152),

and that none are subject to the constraints of Chapter 4 or § 5.1.3.

5.3.3.1. While the rule which produces the definite pronouns of

(5.151) requires identity of reference, the rile which inserts the

pronoun one does not. That this rule is subs ct to (5.152) can be

seen from the sentences of (5.161):

(5.161) a. He'll bring me a hotdog if he sees one.

b. * He'll bring me one if he sees a hotdog.

c. If he sees a hotdog, he'll bring me one.

d. If he sees one; he'll bring me a hotdog.

Sentences like those in (5.162) are obligatorily converted

into the corresponding sentences in (5.163), under conditions which

need not concern us here.

(5.162) a. * Seven more soldiers came in after ten

ones had left.

b. * Seven more ones came iu after. ten soldiers

had left.

c. * After ten soldiers had left, seven more ones

came in.

d. * After ten ones had left, seven more soldiers

came in.
 365

(5.163) a. Seven more soldiers came in after ten had

left.

b. * Seven more came in after ten soldiers had

left.
1

c. After ten soldiers had left,.seven.nore

came in.

d. After ten had left, seven more soldiers came

in.

5.3.3.2. The rule of S Deletion, which deletes a sentence which

is a sister of the abstract pronoun it, if this sentence is identical*

to some other sentence in the phrase-narkei, is also subjc.ct to

36
condition (5.152), as (5.164) shows.

(5.164) a. Harry believes that Sally is innocent,

although noone else believes it.

b. * Harry believes it, although noone else

believes that Sally is innocent.

c. Although noone else believes that Sally is

innocent, Harry believes it.

d. Although noone else believes it, Harry .

believes that Sally is innocent.

If sentence (5.165) is derived from a structure like that

shown in (5.166), as I will argue is correct, in Lakoff and Ross (in

preparation a),

-..
 366

.(5.165) Webster touched a sword.

(5.166)

NP

Webster

1
did

it

Webster V NP

touch a sword

then the sentences of (5.167) can be derived as a special ease of

S Deletion.

(5.167) a. Webster touched a sword after Henry had

done it.

b. * Webster did it after Henry had touched

a sword.

c. After Henry had touched a sword, Webster did

it.

a. 3.3;2, o3,34. F,1 4

 367

d. After Henry had done it, Webster touched

a sword.

If the analysis implicit in (5.166) cannot be maintainadi

then some additional rule of pronominalization, which replaces verb

phrases having the feature ( Stative] with do it., will have to be

formulated to account for.these cases. Which analysis is correct .

is not my concern here.

5.3.3.3. There is another rule which pronominalizes sentences

under identity, replacing them with the morpheme so. It may eventually .

prove to be possible to collapse this rule with the rule of S Deletion,

although sentences like those in (5.168) make this seem unlikely.

so}
(5.168) a. Did the Mets win? If , I've lost
*it
$500,000.

b. The doctors say that she's co ng along

well, but it didn't seem .to one.

*iot
/
Whether So Insertion is the same rule as S Deletion or not, it is

subject to (5.152), as the sentences In (5.169) show.

(5.169) a. Harry thinks that Sally is innocent, although

noone else thinks so.

b. * Harry thinks so, although noone else thinks

that Sally is innocent..

 -2 224 4111IIINA.~11/*

368

c. Although noone else thinks that Sally is

innocent, Harry.thinks so.

d. Although noone else thinks so, Harry thinks

that Sally is innocent.

Once again., if the analysis implicit in (5.166) is correct,

37
the pro-VP do so can be generated as a special case of So Insertion.

If not, a special rule inserting these forms must be added to the

grammar. This rule will also be subject to (5.152), as (5a70) shows.

(5.170) a. Webster touched a sword after Henry had

done so.

b. * Webster did so after Henry had touched

a sword.

c. After Henry had touched a sword, Webster

did so.

d. After Henry had done so, Webster touched

a sword.

5.3.3.4. The fourth type of pronominalization is the rule which

converts sentences like those in (5.171) to the corresponding

sentences of (5,172)

can work on it
you work on it
(5.171) I'll work on it if
noone else has worked on it
am will be working on it
(I can
Iy do
(5.172) I'll work on it ie nou
oone else had
Sam will be too

a
 369

In past generative treatments, this rule would have been formulated

in such a way that it deleted a verb phrase under identity. In

Lakoff and Ross (in preparation a) (cf. also Ross (1967b)) , I will

propose a reanalysis of the auxiliary system under which this rule

will become a special case of So Insertion, with an additional rule

deleting the pro-sentence so when it follows an auxiliary verb. But

whichever of theanalyses is correct, the rule is subject to (5.152),

as the sentences of (5.173) show.

(5.173) a. I'll work on it if I can.

b. * I will if I can work on it.

c. If I can work on it, I will.

d. If I can, I will work on it.

5.3.4. Rules of pronominalization of the form shown in (5.148) are

not upward bounded, as will be evident from the sentences of (5.174).

(5.174) a. The boy who Maryi loves hates heri.

b. The man who ordered a hotdog got one.

c. Tom says that it's going to rain but I

don't believe it.

d. He said he would leave and now he's done it.

e. I think that Mort's a swell guy, and Lenny

thinks so too.

f. Why can't the man who usually cuts the grass

do so today?

g. Mickey and Roger have signed, and Whitey

will tomorrow.
 MOMONaJMN.11. 'UMW*. Mir

370

The sentences in (5.175) show that the rule which

introduces definite pronouns can go down into complex noun phrase,

0464.4.sc.4.01, itAL6CK noun pwkabcb, cum

sentential subject clauses.

(5.175) a. These shoes won't fit into the trunk

i

.theyi're next to.

b. Ronald scoffs at the belief that he

i i

would run if nominated.

c. Romeo conceded that he and Juliet were

i

going steady.

6...lockoicarefullybrushedoffhis.tongue.

e.Onedentist.felt that for hi rai to swim

without a bathing suit would be too daring.

The major constraints proposed in Chapter 4 thus do not

constrain the variable-in this rule. That they also do not constrain

the variables in the rules discussed in § 5.3.3 is indicated by

the grammaticality of the examples in (5.176).

(5.176) a. I lost a jwpcsi,oc olide-rule, and the fact

that Peter now has one.I regard with suspicion.

b. The earth is flat, but will all those who

don't believe it please raise their hands?

c. Pilots who can fly barrel rolls say that for

me to 4y to do it in a glider would be

hazardous.

a
 371

d. The passengers who had known that the train

was not on fire said that those who had

thestighr en linA INnrritatioti t-liginIcelltrete 4n

the bathrooms.

e. Playing with matches is; lots of fun, but

doing, so...and emptying gasoline from one

can to another at the same time is a

sport best reserved for pyromaniacs.

f. Swimming is fun, and I believe that people

who carOt should be taught to.

In these examples, I have not shown for each type of

construction that it is not subject .o each of the four constraints,

but the examples given here should provide a sound enough basis.for

this generalization.

Although there are other constraints on particular rules

of the form shown in (5.148), the condition stated in (5.152) seems

to be the basic one governing all pronominalization rules which make

38
crucial use of variables. Condition (5.152) appears to be operative

in French and German, as well as in English, but there are apparently

languages in which only forward pronominalization is possible. In

Finnish, and in Ijo and a, two languages of West Africa, this seems

to be the case. I know of no language, however, in which backward

pronominalization is as free as forward pronominalization, and it seems

possible, at least at the present state of syntactic knowledge,

to claim that if a language exhibits

 111 11.11.K.

372

backward pronominalization at all, then such pronominalization is

subject to condition (5.152).

5.4. To summarize briefly, in this chapter I have argued that

there are reordering transformations which make crucial use of variables,

but which cannot be restricted correctly by either the piinciple of

the transformational cycle or by the constraints developed in Chapter 4.

I have provided additional evidence in support of Langacker's notion

of command, showing that in addition to being necessary to restrict

the operation of all feature - changing rules except pronominalizations,

it can be extended in a natural way so that it correctly restricts

the scope of the problematic reordering transformations. Finally, I

have argued that Langacker's proposal to restrict with the notion of

command the rule which introduces definite pronouns is inadequate, and

that this rule, as well as all rules of pronominalization which make

crucial use of variables, is subject to a different condition, which

I stated in (5.152). Thus far, in my survey of restrictions on

syntactic variables, for all constraints except those developed in

Chapter 4, I have specified the formal properties of the rules which

were subject to the constraints in question. Thus all pronominalizations

which have the form of (5.148) are constrained by (5.152); all rules in

which elements are permuted rightwards around, or adjoined to the right

of, a variable term at the right end of a structural index, and all

feature-changing rules, which have the form given in (5.78), are upward

I
 373

bounded. In the next chapter, 1 will attempt such a formal specification

of the class of all rules which are subject to the constraints of

Chapter 4.
 374

Chapter 5

FOOTNOTES

1: At present, there is no known principle of rule ordering, or

combination of such principles, which can correctly account for

all relevant facts of ordering. The difficulties which arise, by

,

and large, have to do with various kinds of pronominalization.

For an extended discussion of this area of study, cf. Lakoff and

RQSS (in preparation b).

2. Evidence that certain rules must be constrained not to apply

until the 7.ast pass through the transfor-:tional cycle, where they

may precede rules which apply on each pass through the cycle, is

given in Lakoff (1966).

3. A detailed investigation of German intonation along these lines

can be found in Bierwisch (1966).

4. For expository purposes, I have shown in (5.20) not an underlying

structure, but an intermediate structure, to which the rules of

Relative Clause Formation and Particle Movement, among others,

have already applied.

5. Actually, it is not clear to me whether Chomsky's formulation of

the principle, which I quoted in i 2-.0, was meant to be strong

.
 375

enough to have this effect, or whether a slightly stronger

version would be necessary. For the present discussion, it is

immaterial which is the case.

The fact that sentences (5.22a) and (5.22b) are of low accepta-

bility, if not completely impossible, is accounted for by the

Output Condition on Post-verbal Constituents (3.41) and is of no

relevance to the present discussion. For the reasons I discussed

in § 3.1.1.3.2, both of these sentences must be considered to be

fully grammatical, though unacceptable.

7. The question of whether the extraposed S3 should be dominated

oee4
directly by S2 or by the VP of S2 CM not concern us here.

8. In (5.27) and (5.28), I have assumed that the rule of Question

has been reformulated along the lines of (4.135) Relative Clause

Formation, so that the questioned constituent is Chomsky-adjoined

to the sentence headed by Q. It is this operation of Chomsky-

adjunction which is the source of the new node S0 in (5.27) and

(5.28).

9. For some discussion of the many exceedingly difficult problems

concerning this rule, cf. Keyser (1967).

 376

10. The fact that various sentences in (5.44) are rendered less than

fully acceptable by the output condition stated in (3.27) need

not concern us here - all should be considered to be grammatical.

11. This problem was brought to my attention by Michael L. Geis.

12. For the purpose of stating this rule, I will make the dubious

assumption that there is a feature (+Adverb] which is assigned to

all adverbs. Though trees (5.60) and (5.61) do not indicate the

presence of this feature, it should be assumed to appear in them.

13. Klima analyzes ever in such sentences as (5.73c) as an obligatory,

morphophonemic variant of anytimes.

14. In Finnish, as in many other inflected languages, non-contrastively

stressed subject pronouns are normally deleted.

15. David Perlmutter has called to my attention the fact that this

rule is obligatory for accusatives in the same clause as the

negative element (but cf. fn. 16)) and optional for elements of
/civet-
what were clauses in deep structure. He points out that this
A
restriction is shared by the Russian rule for reflexivization,

which must have the same restriction imposed on it. Th.i.s is the

only case I know of where a restriction which seems to have to be

in a conditions box is not,a restriction on a reordering transfor-

 377

16. I have drastically oversimprfied the. facts in my presentation

of this example. For example, while both (5.93a) and (5.93b) are

possible, they have different meanings. If vodku (acc.) appears,

the rOica
nl means 'whn never drank vodka'; with vodki (gen.): it

means 'who didn't drank any of the vodka.'

17. Since the reflexive pronoun sebia ii used for all persons, the

sentence on sostavil men a uvazat, sebja can also mean 'He forced

me to respect myself.' For the present discussion, this reading

can be disregarded.

18. The string in (5.106b) is a grammatical sentence, but it means

'That Mary was sick was obvious to me.' The fact that here zibun

can only refer to the first person suggests that in the deep

structure of (5.106b) must contain an earlier occurrence of the

pronoun watakusi 'I'. Precisely this position is argued for in

my forthcoming paper "On declarative sentences" (Ross (1967c)),

where I present arguments that all declarative sentences must, in

deep structure, be clauses embedded as the object cf.a verb of

communication, like alyr or declare, with a firstperson subject.

19. The reasons for not Pruning S1 in (5112) will be gone into in.

Lakoff and Ross (in preparation b),

20. I am grateful to Paul Kiparsky for calling to my attention cases

like (5.114), in which the tense' -charging rule is obligatory.

a-
 378

21. Further research may reveal that it is normal for reflexivization

rules to b2 both downward anu upward bounded. ILL this case, the

theory would have to mark the English.rule as being normal, and

the Japanese and Russian rules as being idiosyncratic in having

an imusually wide range of application.

.22. There are many other complex conditions which have, to do with

clitic placement, and these have extremely important- consequences

for the theory of grammar. This problem will receive intensive

discussion in a forthcoming paper by E.'Weyles Browne, III, and

David M. Perlmutter.

23. For a formal definition of the notion tree, cf. Zwicky and

Isard (1963) :

24. The problem of why rules of reflexivization should behave not like

rules of pronominalization, to which they are formally similar

(cf. § 5.3 :below), but like other feature-changing rules, with

respect to the generalization in (5.122), will be taken up in § 6.5

below.

25. Note that even if it le argued that the analysis implicit in

(5.131) is incorrect, and that the category S must be expanded by

the rule S 4 NP Aux:VP, and the category Aux by the rule

Aux + Tns (X) (Perf) (Prog), the notion in construction with will
I
- .

379

not allow the required change to take place if (5.127) is

included in the grammar, under the assumption that the node on

which the feature [Tense] is marked is the node Tns.

26. This definition, is inadequate in that it does not come to grips

with the problems.brough up in footnote 19 of Chapter 3.

. 27. Anthony Naro has pointed out to me the extremely interesting

fact that the sentence That the fuzz wanted to question John

worried him, but it didn't worry Mary is ambiguous in the same

way that (5.138a) is. This means that the definition of linguistic

identity given in (5.135) must be revised in such a way that not

only commanded pronouns can be disregarded, but also that noun

phrases which have entered into an anaphoric relationship with

some other noun phrase and pronamiralized it can be disregarded

under certain circumstances. I will not attempt such a revision

here, for a full treatment of the many complex issues connected

with the definition of identity is far beyOnd the Scope of this

work.

28. All the following remarks apply equally well to the as-clause

of the comparison of equality.

29. At present, I know of no way of defining the term "the compared

element." This thorny problem I will bequeath to future

 38o

researchers on the grammar of comparatives.

. 30. By the phrase "crucial use of variables", I mean all rules whose

ctrtte.ttirn1 inclAy e.entaina a clihtring of thA form eA_ X A

-2-

or whose structural change specifies that some term is permLted

around, or adjoined to, some term which contains a variable. Thus

the rules of Indefinite Incorporation, (5.71), and Question, (4.1),

make crucial use of variables, while the rule of It Deletion,

(4.128), does not. This distinction between rule types has

important consequences. For instance, it can be shown that no

rules which make crucial use of variables are governed - that is,

they can have no lexical exceptions.

31. For some discussion of this rule, cf. Ross (1967d).

32. In this rule, the letter A is a variable over node types, not

strings.

33. In Postal (1966a), some concrete proposals of rules to effect

these changes are made.

34. It is at present unknawr as to Thether a universal definition of

the notion subordinate clause can be given, or whether it will be

necessary to give a language-particular definition for each

language in which this condition appears.

 " ...

381

35. Indeed, the assumption that pronominalization should be effected

by a syntactic rule, rather than by a semantic one, is also

provisional. For arguments pro and con, see Lakoff (1967) and

Jackendoff (1966a,.b). There are so many mysteries connected

with various kinds of pronominalization that almost nothing about

it seems free of serious doubt.

36. For a discussion of some of the consequences of assuming that

this is a syntactic rule, cf. Lakoff (1967).

37. For discussion of this construction, cf. Lakoff and Ross (1966),

and. Anderson (1967).

38. One interesting, if poorly understood, exception is the rule which

produces anaphoric noun phrases like that idiot in such sentences

as Wilfred, raised his hand and th,n that idiot, even tried to

answer the question. This rule appears not to work backwards at

all (witness rho: ungrammaticality of *After that idiots had shut

3122._eyel:zatiledp..) and to work forward only under

certain circumstances (cfs, *Wilfred, said that that idiot. was

ao.ilaso_2&I..back at us.). The special nature of this rule was

first pointed out to me by George Lakoff.

 382

Chapter 6

ON THE NOTION "REORDERING TRANSFORMATION"

6.0. In Chapter 4, I presented evidence which showed that the

rules of Relative Clause Formation and Question are subject to a variety

of constraints. Since the facts cited in § 5.3.4 above show that

these constraints do not affect rules of pronominalization, the question

arises as to whether there are other rules than just the two studied in

Chapter 4 which are subject to the constraints, and if so, whether it

is possible to predict from the formal statement of a rule whether that

rule will obey the constraints or not. This question has already been

begged: the constraints in Chapter 4 were stated not in terms of the

specific rules of Relative Clause Formation or _Question, which were

used to exemplify the effect of the constraints, but rather in terms

of "reordering transformations". In this chapter, I will give a

precise characterization of this presystematic term.

In § 6.1, I will describe briefly a large number of

rules, some apparently related, some not, showing that each is subject

to the constraints. In § 6.2, I will show that transformations which

reorder a constituent, but leave behind a pro-form, to indicate

the

the
place the copied constituent occupied before the operation of

rule, are not affected by the constraints, and that it is rather

transformations which "chop" a constituent and move it from its

subject to the
original position without leaving any trace, which are

constraints. In § 6.3, I will show that even chopping transformations

 383

are not subject to the constraints unless the chopped constituent it

moved over a variable. In § 6.4, I will show that the feature-

nUell.%ev4v.r.
Iwt.c44.5J.4a6
A4c.nynn^A 4." 5 5.1.3 also obey the constraints. Th4s

fact leads to a theory of islands, the maximal domains of chopping

and feature-changing rules. In § 6.5, a brief summary of the

characterization arrived at is given.

6.1. Some Rules Obeying the Constraints

6.1.0. At the outset of my research on variables, I noticed that

the German rule which preposes various types of constituents to the

front of a sentence, thereby triggering a rule which inverts subject

and verb (thus (6.1a) becomes (6.1b) , (6.1c) , or (6.1d)),

(6.1) a. Ich sprach gestern mit Orje {fiber Liebe.

I spoke yesterday with Orje about love.'

b. Gestern sprach ich mit Orje Uber Liebe.

c. Mit Orje sprach ich gestern {fiber Liebe.

d. 'fiber Liebe sprach ich gestern mit Orje.

obeyed the same constraints as the rules of Relative Clause Formation

and Question, and the rules involved in cleft sentences, like (6.2),

and pseudo-cleft sentences, li,;e (6.3).

(6.2) Es war gestern, dass ich mit Orje {fiber Liebe sprach.

It was yesterday that I with Orje about love spoke.

'It was yesterday that I spoke with Orje about love.'

 384

(6.3) Wortiber ich gestern nit Orje sprach war Liebe.

Where about I yesterday with Orje spoke was 1oveL

frY1..-.. V
W& all. 1 birvaN.= Wll.11 VLJG caVV46 ycQ,6ctuay INGQ,

At that time, I concluded that the way to explain the similarity of

the constraints on these rules was to assume that one rule was basic,

and was a component of the operations of the other three rules. But

roam Chomsky pointed out to mcs. an Alternative possibility: this

similarity of constraints ight be derivable from .iome formal property

shared by the four rules, rather than from some assumed common

function or component. My further research proved Chomsky correct:

there are a large number of transformations which obey the same con-

straints as the four rules that I had originally noticed, rules whose

vperations are far too dissimilar for it to be possible that there

is one rule which is basic to each of these.

In my brief discussion of each of these rules, I will

first give an example which is sufficiently complex to suggest that

the scope of the rule is unboundedly large, and then give examples

to show that the rule is subject to the Complex NP Constraint

(CNPC), the Coordinate Structure Constraint (CSC), the Sentential

Subject Constraint (SSC), and, where possible, the Left Branch

Condition on pied piping (LBC). I have partitioned the rules into

three arbitrary groups: the rules in § 6.1.1 produce clauses which

resemble questions or relative clauses, some of which may derive

from rules which can be collapsed with the rules of Question and

Relative Clause FormatiOni. The rules in § 6.1.2 share only the

property of producing structures which in no way resemble relative

clauses. The rules in § 6.1.3 constitute the only counter-evidence

I know of (but cf. § 6.4) to the claim that only "reordering trans-

formations" are subject to the constraints of Chapter 4.

6.1.1,

6.1.1.1. One rule which results in question-like structures is

the rule which produces exclamatory sentences, like those in (6.4);

(6.4) a. How brave he is!

b. How surprisingly well he dances!

c. The bravery of our boys in Vietnam, Thailand,

Cambodia, Korea, Malaya, Iceland, Nepal.,

Egypt, Turkey, Kazakhistan, Morocco, Haiti,

Peru, Chile, Quebec, the Honduras,

Baffinland, Monaco, and all the other

places in the world :.;here freedom needs

protection!

I imagine that sentences like (6.4c), which consist of a

single abstract NP, spoken reverentially, will derive from sentences

like (6.4a), where he is replaced by our boys in Vietnam, etc., but

I do not know how the rules that effect this conversion should be

...11".".Ff,r1R.Carmavy,
formulated.

Although the sentences in (6.4) resemble questions, they

are much more limited, for there are many question words that cannot

head an exclamatory sentence, as (6,5) shows.

(6.5) a. *Whether he left!

2
b. *Why he knows tne ..Inswer!

c. *Which boy is tall!

It seems likely to me that the restriction which is

operative here is that it is only sentences with degree adverbs which

can function in exclamatory sentences. This is indicated by the fact

that if the word bravery, which is derived from a lexical item

allowing degree modifiers (very brave), is replaced in (6.4c) by an

abstract noun like arrival, whose underlying lexical item does not

admit of degree modification ( *vim arrive, *arrive very), the sen-

tence becomes ungrammatical. But there are several classes of counter

examples to this generalization (cf. e.g., the sentences in (6.6)),

and although these seem intuitively to be different from the sentences

in (6.4), I have no convincing arguments which show this to

necessarily be the case.

(6.6) a. When my daughter came home last night13

b. What my husband eats!

:c, Where my son and that girl he married are living!

But no matter what the source for such sentences as

 386

those in (6.4) is, it is clear that the rule which forms them must

be able to move the wh-ed ,constituents tothe front of the sentence

from indefinitely deeply embedded structures (cf. (6.7)).

(6.7) How brave everybody must think you expect me

to believe he is!

That this rule is subject to the CNPC, the CSC, and the

SSC, can be seen from (6.8), (6.9), and (6.10), respectively.

(6.8) a. *How brave I know a boy who is!

b. How brave they must believe (*the claim) that

4
you are!

(6.9) a. *How brave he is tall and!

b. *How brave Mike is cowardly and Sam is!

(6.10) a. *How brave .that Tom is must be believed!

5
b. How brave it must be believed (?that) Tom is!

That it is also subject to the LBC can be seen from the

fact that it is (6.4a) that is grammatical, and not (6.11).

(6.11) 'CHow he is brave!6

The reason that (6.11) is ungrammatical is the same as the one F'ven

for the ungrammaticality of (4.190), in § 4.3.2.1 aboite.

6.1.1.2. The first constructions which exhibit relative-clause-like

structures are clauses introduced by where, when, after, before, since,

until, and while. Michael L. Geis has proposed7 that all of these
 387

clauses be treated as deriving from relative clauses on such head

nouns as place or time. Thus at the time at which becomes at the

time when, which may, by deletion of-ehe _NP at the time, result in

a clause introduced by the single word when. That the source in

the constituent sentence for the phrase at that time, from which this

word derives, can be indefinitely far down ."the tree can be seen from

(6.12),

(6.12) Bill left when everyone will believe that the

police have forced me to confess that I shot

Sandra.

where the word when refers to the time of the shooting of Sandra. That

the rule which forms such adverbial clauses, if it is different from

the rule of Relative Clause Formation, which I doubt, is subject to

the CNPC, the CSC and the SSC can be seen from (6.13), (6.14), and

(6.15), respectively,

(6.13) a. *Bill left when I am looking at a girl who vomited.

b. Bill left when I believe (*the claim) ( ?that)

the bomb had just exploded.

(6.14) When I am awake (*at that time) and Susan

is asleep, Bill will leave.

(6.15), a. *Bill left when that noone else was awake

8
is certain.

b. Bill left when it is certain that noone else

was awake.
 388 01.

Sentences similar to these, which show the other adverbial

clauses mentioned to be subject to the three major constraints, can

also be constructed, but I will not undertake this here.

6.1.1.3. The second type of relative-clause-like construction is

exemplified in (6.16):

(6.16) Here's a knife for you to cut up the onions with.

For to phrases can modify noun phrases in the same way as relative

clauses. The subjects of these clauses can be deleted under inter-

esting conditions (cf. (6.17)).

myself
(6.17) a. I brought a razor to shave with.
*himself

b. I brought_a razor with which to shave ,myself

himself
*myself
c. I brought John a razor to shave with.
himself
*myself
d. I brought John a razor with which to shave himselft
7
The presence of the relative pronoun which in (6.16b)
.7
and (6,16d) suggests that whatever rule forms these clauses always

preposes this pronoun to the front of the clause, deleting it

obligatorily just in case the embedded subject has not been deleted.

Thus (6.16) would be derived from the structure which underlies (6.18).

(6.18) *Here's a knife which for you to cut up the

onions with.

Somehow the rule which forms these clauses must prevent a preposition
 389

which precedes the NP to be relativized from pied piping, unless the

subject of the clause has been (or will be?) deleted -- nothing can

save a structure like (6.19), where the preposition with has pied

piped, except possibly some ad hoc rule to reinsert the preposition

where it came from, arule unstateable under present conventions,

in any account.

(6.19) *Here's a knife with which for you to cut up

the onions.

Constituents can be moved by this rule from indefinitely far down the

tree, as (6.20) shows.

(6.20) Here's a plate for you 'to make Bob try to

begin to force his sister to leave the

0
cookies an.

I am not sure whether this rule can relativize elements from within

that-clauses at al.3..,_but if so, it is only elements dominated by VP

in such clauses, not subjects, that can be relativized. (6.21a) may

be grammatical, but (6.22b) is almost certainly not.

(6.21) a. ?Here's a knife for you to say that you cut

up the onions with.

b. *Here's a knife for you to say was on the table.

Thus we see that this rule, even if it should someday prove

to be collapsible with the rule of Relative Clause Formation, will have

to have a number of special restrictions imposed on it. And yet the

 390

sentences in (6.22), (6.23), and (6.24) show it to be subject to the

CNPC, the CSC, and the SSC, respectively.

(6.22) a. *Here's a pole for you to kiss the girl who

tied the string around.

VO ?tJF% .Ale,
.4D GI.
,ifrA.VVAL
eves 10. ". VW
YVUO +-
GOLOOVtAA.Al /4e4.4..^
N.VOLG

possibility) that you will shave with.

(6.23) *Here's a razor for you to chop up these

nuts with this cleaver and.

(6.24) a. *Here's a razor for that you will be shaved

with to be announced.

b. ??Here's a razor for it to be announced

that you will be slaved with.

Whether or not the LBC can be shown to be operative for

this rule will depend upon it being possible to set up a contrast

between such sentences as those in (6.25).

(6.25) a. ?I loaned Maggie a Swiss Army knife with

whose corkscrew to open the padlock.

b. *I loaned Maggie a Swiss Army knife

whose to open the padlock with corkscrew.

While it is clear that (6.25b) is word salad, I am not

sure that (6.25a) is fully grammatical. If not, this rule cannot be

shown to be subject to the LBC.

7YL1,....
 391

6.1.1.4. It is well-known that appositive clauses obey the same

restrictions restrictive relative clauses dc, but it may not have

been observed before that sentential clauses, like those in (6.26),

also do.

(6.26) a. Fluffy is sick, which few people realize.

b. Fluffy is sick, which I'm not sure you

know Sarah expects me to believe Joan realizes.

Sentence (6.26b) suggests that this rule must be able to prepose the

relative pronoun which, which stands for the sentence Fluffy is sick,

from indefinitely deeply embedded positions, and sentences (6.27),

(6.23), and (6.29) show that it too is subject to the CNPC, the CSC,

and the SSC.

(b.27) a. *Fluffy is sick, which I slapped a boy who

wouldn't acknowledge.
.

b. Fluffy is'sick, which I believe (*the claim)

that few people realize,

(6.28) *Fluffy is sick, which I fell asleep and

Tom suddenly realized.

(6.29) a. *Fluffy is sick, which that noone here

realizes is certain.

b. Fluffy is sick, which it is certain that

noone here realizes.

The same -estrictions apply to sentential as-clauses: the

word as can be substituted for which in sentences (6.26) - (6.29) with

 gr. e4
474

no change in grammaticality, although this is not in general true.

The sentences in (6.30) show that the rule which forms as-clauses must

be sensitie both to the presence of certain types of negation and to

the syntactic environment from which the constituent which as replaces

9
cames.
which
(6,30) a. Fluffy is sick, nobody knows.
*as
which
b. Fluffy is sick, not everybody knows.
as
hick
c. Fluffy is sick surprises me.
* as

These restrictions on as-clauses are unlike any known to

obtain on relative clauses, restrictive or appositive, so I an highly

doubtful that the rule which forms as can be collapsed with

other rules which form relative clauses.

6.1.1.5. The rules that form cleft sentences, pseudo-cleft sen-

tences, and topicalized sentences are also subject to the constraints.

The sentences in (6.32) show them all to be subject to the CNPC, and

those iu (6,33), (6.34), and (6.35) show them to be subject to the

CSC, the SSC, and the LBC, respectively, while the sentences in

(6.31) show their scope to be unbounded.

(6.31) a. It was this hat that Tom said Al thought you

wanted me to make Jack put on.

b. What Tom said Al thought you wanted me to

make Jack put on was this hate

lortgograk01711TrivriTrivtmiwIlmsft "ftmeMPIIMITATtrir011
 s n.)
474

c. This hat Tom said Al thought you wanted me

to make Jack put on.

(6.32) a. *It is this hat that I know the boy who is wearing.

b. It is this hat that I believe (*the claim)

that he was wearing.

c. *What I know the boy who was wearing is this hat.

d. What I believe (*the claim) that he was

wearing is this hat.

e. *This hat I know the boy who was wearing.

f. This hat I believe (*the claim) that he

was wearing.

(6.33) a. *It is this hat that the gloves and ware on

the table.

b. *What the gloves and were on the table was this hat.

c. *This hat the gloves and were on the table.

(6.34) a. *It is this hat that that he was wearing is certain.

b. It is this hat that it is certain that he

was wearing.

c. *What that he was wearing is certain is this hat.

d. What it is certain that he was wearing is this hat.

e. *This hat that he was wearing is certain.

f. This hat it is certain that he was wearing.

(6.35) a. *It was John's that I stole bike.

b. *The one whose I stole bike was John's.

c. *John's I.stole bike.

. ,
.43,114r14110"/PWRISIMWRIMIMMANNicV
S
 394

Because of the many additional similarities shared by

these constructions, I am inclined to think they all derive from the'

same deep structure source, although I can propose none that is

convincing. But all that is at issue here is the fact that the set

or sets of rules that produce these constructions are all subject to

the constraints of Chapter 4.

6.1.1.6. The next relative-clause-like construction I will consider

is that exemplified in (6.36).

(6.36) Maxwell isn't (half) the doctor that his

father was.

The fact that the element half can precede the modified NP in

(6.36) shows that this sentence cannot be considered to be an

instance of a predicate nominal modified by a relative clause, as

in (6.37),

(6.37) Maxwell is the man who won the Nobel Prize

for horoscopy.

for if half is present in (6.36) , the relat-Lve clause" must be

10
present, as the ungrammaticality of (6.38) indicates.

(6.38) *Maxwell isn't half the doctor.

It seems probable that (6.36) can be related to such

sentences as those in (6.39),

(6.39) a. Maxwell is quite doctor.

t

b. Maxwell isn't much of a doctor.

 c. Maxwell is more of a doctor than his son is.

but no analysis of these constructions has been deep enough for this

to be established positively. One final pant of inte,est about these

constructions is that the "relativized" element seems 1 have to

4

follow the copula be in both the matrix and constituent sentences.

When this strange constraint 'is violated, ungrammatical sentences such

as those in (6.40) result.

(6.40) a. *Maxwell isn't (half) the doctor that was here.

b. *Maxwell isn't (half) the doctor that

polished off the vodka.

c, *(Half) the doctor that Maxwell's father

was sat down.

As (6,41) suggests, the that-clause of (6.36) is not

bounded in length:

(6.31) Maxwell isn't (half) the doctor that

feared Marge would realize Tom had confessed

that he knew Bill expected hii to be.

Whatever rule it is that forms such clauses, it is subject to the

CNPC, the CSC, and the SSC, as sentences (6.42), (6.34), and (6.44),

respectively, show-..

(6.42) a. *Maxwell isn't half the doctor that I know

an African chief who is.

b. Maxwell isn't half the doctor that people

around here believe (*the claim) that his

father was.
 396

(6.43) *11.;Ixwel1 isn't half the doctor that his sister

is a psychologist and his father was.

(6.44) a. *Maxwell isn't half the doctor that that he

would be if he studied is certain.

b. Maxwell isn't half the doctor that 1:- is

certain that he would be If he studied.

6.1.1.7. The last two cases of relative-clause-like constructions

that I will discuss are those exemplified in (6.45).

(6.45) a. He's the happiest that I've ever seen him.

b. The hardest that it ever snowed was last

January 12th.

I have grouped these two constructions together only on the basis of

the fact that they both contain superlatives. What their deep

structures are in fact, and whether the same rules are used in

forming each, is anyone's guess. The grammar of superlatives, if it

is not the most poorly understood of all problems yet investigated

within the framework of generative grammar, is certainly not far off

11
the pace.

That both of the that-clauses in (6.45) can be extended

without bound is suggested by the random degree of complexity attained

in (6.46).

(6.46) a. He's the,. happiest that any .of,myz.friends

could estimate anybody would expect you to

believe that I've ever seen him.

 397

b. The hardest that I think I remember him

ever telling me that he had heard of it

snowing around here was last January 12th.

The rules that produce such constructions are subject

to the three constraints of Chapter 4, as sentences (6.47)(6.49) show.

(6.47) a. the happiest that we ever talked to

the boy who had seen him.

b. He's the happiest that I believe (*the

claim) that he's ever been.

c. *The harda7t that I ever knew a man who said

that it had snowed was last January 12th.

d. The hardest that I believe (*the claim)

'hat it ever snowed was last January 12th.

(6.48) a. *He's the happiest that I've ever seen him

drunk and.

b. *The hardest that all the power lines were down

and it snowed was last January 12th.

(6.49) a. *He is the happiest that that he has ever

been is believed.

b. He is the happiest that it is believed

that he has ever been.

c. *The hardest that that it has snowed here is

believed was last January 12th.

d. The hardest that it is believed tFat-. it has

snowed here was last January 12th.

 39 8

6.1.2.

6.1.2.0, While no arguments are available (and I doubt that any

are forthcoming) that all the above structures are offshoots of

either the rule of Relative Clause Formation or the rule of gasaka,

since all the constructions discussed exhibit some clause headed by

a wh-word or the word that, it is at least logically possible that an

analysis will someday be discovered which makes use of one of these

two rules to derive all of the above constructions. But in the case

of those constructions that I will discuss in this section, such an

analysis would be inconceivable, for the structures produced contain

relative-clause-like structures only incidentally, if at all.

6.1.2.1. The rule of Extraposition from NP, (1.10), because of its

formal structure, is upward bounded, so it is impossible to show with

such sentences as (4.18) that it is subject to the CNPC; the same

obtains for the SSC. It is, however, possible to show that it

must be subject to the CSC. For consider structure (6.50):

(6.50)

VP

met in Vienna

a friend of mine
who was woslas in Europe, azi.r1

who was from his home town

 399

If the rule of Extraposition from NP applies to this structure to

move S2 out of NP1, or S3 out of NP4, one of the ungrammatical

sentences in (6.51) would be generated.

(6.51) a. *A friend of mine and a girl who was from

his home town met in Vienna who was working

ins Europe.

b. *A friend of mine who was working in Europe

and a girl met in Vienna who was from his

home town.

A similar example can be constructed to show that

Extraposition, (4.126), must also be subject to the CSC.

(6.52)

NP
NP NP
was tragically evident
and
it

that she loved him

that he loved another

If Extraposition does not apply to this structure, the rule of

(klas),
It Deletion which was stated in itV.M',;,Z.Z.,,,`Irtr-2.=}73=2 will delete

both occurrences of it in (6.52), and the grammatical (6.53) will

result.
 400

(6.53) That she loved him and that he loved another

was painfully evident.

However, if Extraposition were allowed to apply to either S2 or

S in this structure, one of the ungrammatical structures in. (6.54)

3

would be produced.

(6.54) a. *It' and that he loved another was painfully

evident that she loved him.

b. *That she loved him and it was painfully

evident that he loved another.

The CSC must be invoked to block the generation of the sentences in

(6.51), and it can also block the generation of those in (6.54).

However, since it is not known what the relative ordering of the rules

of Extraposition and Conj unction Reduction is, it might be that the

rules could be ordered in such a way as to prevent (6.54) without the

CSC being necessary. But such a rule-ordering explanation is not

available in the case of (6.51)s-for-if the analysis presented in

Lakoff and Peters (1966) is correct, the conjoined NP subject of

such verbs as meet, similar, etc. is derived from a conjoined NP

in deep structure. It therefore seems inescapable that the CSC must

constrain the operation of at least one rule, Extraposition from NP,

which cannot be argued to be a subcase of the rules of Relative Clause

Formation or allstion.

6.1.2.2. Although the rule of NP Shift, (5.57), cannot be shown

to be subject to the CNPC or the SSC, because it, like the two
 401

extraposition rules, is subject to the stronger restriction of being

upward bounded, it can be shown to obey the CSC, for the a-sentences

below must not be converted into the b-sentences.

(6.55) a. Mary and [an ol4 friend who comes from

Miami]NP kissed.

b.' *Mary and kissed an old friend who comes

from Miami.

(6.56) a. I gave a picture of a covered bridge and

[a hundred hikers from Hoboken] to my sister.

NP

b. *1 gave a picture of a covered bridge and to

my sister a hundred hikers from Hoboken.

(6.57) a. Joan plays [a wonderful old guitar from

Spain]NP and sings folksongs.

b. *Joan plays and sings folksongs a wonderful

old guitar from Spain.12

That the rule of NP Shift is also subject to the LBC was

argued in §4:3.2.1 above, in connection with the ungraugaticality

of (4.188b) and (4.188c).

6.1.2.3. The rule of Conjunction Reduction, whose operation was

described informally in § 4.2.4.1. above, is stated roughly as in

(6.58).

(6.58) Conjunction Reduction

a. - [X -

.. 61. 1,....

1 2 3
[1 2 0]B#3
 .....tomormomPrea 1 . _. - . ..-........2=...

.4,2

b. [and - - x1: IB

OPT
1 2 3

21/fl 0 31B

Condition: all occurrences of A are identical.

This notation should be interpreted to mean that in any

coordinate node of the category B, which dominates any number of

conjuncts which are also of the category B, and each of which either

ends or begins with a constituent of category A, where all occurrences

of A are identical, all of these occurrences of A are superimposed,

and adjoined to the conjoined node B. Thus (4.118) could be converted

into (4.119) by the operation of this rule.

This rule must be formulated in such a way as to reorder

each instance of the category A, adjoining it to the coordinate node,

for otherwise the following facts cannot be explained. If my intui-

tions are correct, (6.59a) cannot be converted into (6.59b), and

(6.60a) can be converted into (6.60b) only if the parenthesized NP,

the claim, is not present.

(6.59) a. Sally might be pregnant, and I know a girl

who definitely is pregnant.

b.?* Sally might be, and I know a girl who

definitely-is, -pregnant.

(6.60) a. Sally might, be pregnant, and I believe (the

claim) that Sheila definitely is pregnant')

 403

b. ?Sally might be, and I believe (?*the claim)

that Sheila definitely is, pregnant.

Some speakers claim to find no difference between the

version of (6.60b) in which the claim is present and the one in which

it is not, or between (6.59b) and either of these, If all are held to

be ungrammatical, then rule (6.58) must simply be restricted in such

a way that the nodes A cannot be dominated by a that-clause. However,

if all are held to be grammatical, then there is a serious inadequacy

in my analysis, for I would hold that if a rule is subject to one

of the constraints of Chapter 4, it must be subject to all. And it

seems clear that at least the CSC must constrain the operation of

rule (6.58), for I know of noone who finds the result of the conver-

sion of (6.61a) into (C 61b) grammatical.

(6.61) a. The younger woman might.have been tall

and blonde, and the older one definitely

was blonde.

b. **The younger woman might have been tall and,

and the older one definitely was, blonde.

But the picture is complicated by the existence of such sentences as

those in (6.62) and (6.63).

(6.62) a. Sally is tall, and maAoe blonde, and Sheila

is short;--and definitely is blonde.

b. .?*Sally is tall, and maybe, and Sheila is

short, and definitely is,.blonde.

 404

(6.63) a. Hank plays the guitar and finds arrangements

for all the old folksongs which are still

sung in these hills, and Ernie writes down

all the old folksongs which are still sung

in these hills.

b..??Hank plays the guitar and finds arrangements

for, and Ernie writes down, all the old folk-

.

songs which are still sung in these hills.

In my speech, (6.62b) and (6.63b) are clearly far better

than (6.61b), but I am not confident enough of this judgment to assert

that they should be considered fully grammatical. However, if all

three are to be considered ungrammatical, as well as (6.59b) and

the version of (6.60b) in which the NP the claim appears, at least

the rule which converts (4.118) into (4.119) must be formulated as a

reordering rule, and be subject to the CNPC and the CSC. That this

rule must also be subject to the LBC was pointed out in § 4.3,2.4

above, in connection with the ungrammaticality of .(4.239) (but cf.

also the discussion of sentence (4.241)').

6.1.2.4. The next rule I will discuss in connection with the con-

straints'of Chapter 4 is the rule which converts (6.64a) to (6.64b)

by preposing a VP which immediately follows an emphatically

stressed auxiliary verb, under various conditions which need not

concern us here.
 405

would pay up, and he wfilipay up

(6.64) a. They said that Tom had gone home, and he lad gone home
was working, and he is working

'' would pay up, and pay up he e

w1 11i
b. They said that Tom ?had gone home, and gone home he hi a .

L was working, and working he az j

The statement of this rule must make crucial use.of a variable, as

(6.65) suggests.

(6.65) They said Tom would pay up, and pay up I'm

sure everybody will tell you that his lawyers

expect me to believe he did.

The rule is subject to the CNPC, the CSC, and the SSC, as can be seen

from (6.66), (6.67), and (6.68), respectively.

(6.66) a. They said nobody would pay up, but I know

a boy who did pay up.

b. *They said nobody would pay up, but pay up

I know a boy who did.

c. They said that Tom would pay up, and pay up

I believe (*the claim) that he did.

(6.67) a. They said that Tom wouldn't pay up, but he

did go to the bank, and he did pay up.

b.. *They said that Tom wouldn't pay up, but pay

up he did go to the bank and he did.

(6.68) a. *They said that Tom would pay up, and pay up

that he did is well-known.

b. They said that Tom would pay up, and pay up

it is well-known that he did.

 406

6.1.2.5. The statement of the rule which converts (6.69a) into

(6.69b) also must make crucial use of variables, as the complexity of

(6.70) suggests.

(6.69) a. Although Dick is handsome, I'm still going

marry Herman.

b. Handsome though Dick is, I'm still going

to marry Herman.

(6.70) Handsome though everyone expects me to try

to force Bill to make Mom agree that Dick is,

I'm still going to marry Herman.

That this rule is subject to the CNPC, the CSC, and the SSC can be

seen from sentences (6.71), (6.72), and (6.73), respectively.

(6.71) a. *Handsome though I know several boys who are,

I'm still going to marry Herman.

b. Handsome though I believe (*the claim) that

Dick is, I'm still going to marry Harman.

(6.72) *Handsome though Dick is fair, Nordic,

strong and, I'm still going to marry Herman.

(6.73) a. *Handsome though that Dick will be is likely,

I'm still going to marry Berman.

b. Handsome though it is likely that Dick will

be,_ I'm still going to marry Herman.

6.1.2.6 Whatever rule it is that derives sentences like (6.74)

must make
from some equally unknown deep structure, its statement
.'
 407

crucial use of a variable, as such sentences as (6.75), if they are

grammatical, would suggest.

111 (6.74) The more contented' we pretended to be, the

more we grew angry at the doctors.

(6.75) ,
?The more contented the nurses began to try

td persuade us to pretend to be, the more

angry we grew at the doctors.

That this rule is subject to the CNPC, the CSC, and the

SSC can be seen from sentences (6.76), (6.77), and (6.78), respectively.

(6.76) a. *The more contented I laughed at the nurse

who thought that we were becoming, the more

angry we grew at the doctors.

b. ??The more contented the nurses began to

believe (*the claim) that we were going

to pretend to be, the more angry we grew

at the doctors .13

(6.77) *The more contented we pretended to be

better fed and, the more angry we grew

at -the--doctors.

(6.78) a. *The more contented for us to pretend to be

became possible, the more angry we grew

at the doctors.

b. ?The more contented it became possible for

us to pretend to be, the more angry we

grew at the doctors.

-""*Y._, ""'
 408

6.1.2.7. The next rule I will consider in this section is the

rule which converts such sentences as (6.79a) into (6.79b), provided

that the object of the preposition de has been pronominalized.

(6.79) a. J'ai une photo de cette maison.

I have a picture of this house.

b.\ J'bn ai une photo.

I of it have a picture.

'I have a picture of it.'

This rule seems to be able to operate over a potentially indefinitely

large portion of a tree, as (6.80b), which results from (6.80a) if

14
the NP la table the table' has been pronominalized,. shows.

.80) a. %Se vois le bout du toit de l'aile

I -see the end of the roof of the wing

gauche de la maison.

left of the house.

'I see the end of 'the roof of the left wing of the

b. J'en vois le bout du toit de l'aile gaud

'I of it see the end of the roof of the wing left

'I see the end of the roof of its left wing.'

This rule is subject to a stronger constraint than the combination

15
of the CNPC and the SSC -- it is upward bounded. It can be shown to

be subject to the CSC by the fact that (6.81a) cannot become (6.81b)

if the NP la maison 'the house' has been pronowinalized.16

 '

409

(6.81) a. -.re vois la porte du garage et le toit

I see the door of the garage and the roof

de la maison.

of the house.
17
b. *J'en vois-la porte du garage et le toit.

6.1.2.8. The last rule I will-deal-with in this subsection, the

rule which produces structures like (6.82),

(6.82) I have some papers to grade.

also seems not to be able to move NP's-out of tensed clauses

(cf. (6.83)),

(6.83) ?*I have some papers to announce that I've

got to grade.

although this rule appears to be able to range indefinitely far down

into a tree, as (6.84) suggests.

(6.84) I have some papers to try to finish grading.

It is not clear to me whether sentences (6.82) and (6.84)

(6.85b),
can be argued to be synonymous with any reading of (6.85a) and

respectively.

(6.85) a. I have to grade some papers.

b. I have to try to finish grading some papers.

If their meaning is correct, they are the most obvious source for

(6.82) and (6.84). But if they cannot bethe source for these

sentences, I am at a loss to suggest what might be. It seems unlikely

that a structure like that shown in (6.86) can serve as a source;

 410

(6.86)

I V ,N1

have some papers

CT

grade some Tapers

for there are sentences like (6.87),

(6.87) I have getting into college to consider.

where the NP that directly follows have in surface structure is

abstract, and I know of no other verb which takes an NP S object

(e.g., verbs like compel, motivate, challenge, etc.) ,where the NP

can be inanimate.

However, no matter what the source of such sentences is,

the fact that the rule that produces them obeys the CSC and the LBC

can be seen from the ungrammaticality of the sentences in (6.88)

and (6.89).

(6.88) a. *I have some papers to grade these exams and.

*I have some voice exercises to play the

guitar and sing.

if
(6.89) *I have John's to grade paper'.

6.1.3.

6.1.3.0. In § 4.1.4 above, I argued from the fact that the rule

which forms relative clauses in Japanese is subject to the crossover

 411.

condition, (4.30), and to the CNPC (it is also subject to the CSC, .but

not to the SSC, as Z showed in § 4.4.1) to the conclusion that the

rule must be formulated as a "reordering transformation" (in a sense

which will be made more precise in § 6.2 and §6.3 below). This is

only one of the possible conclusions: the other is that is not the

case that the crossover condition and the constraints of Chapter 4

only affect "reordering transformations"; rather, there are some

transformations whose only effect is to delete constituents under

se
identity, but which are nonetheless still subject to the constraints.

The question then arises as to how such deletions are to be distin-

guished from other rules of pronominalization, which I showed, in

§ 5.3.4, not to be subject to the constraints of Chapter 4. This

questiou will be taken up in § 6.5 bela4."

6.1.3.1. The first two pronominalization-like rules I will

consider are those which produce those comparative constructions

which exhibit the morphemes -er...than and as...as. Since these two

constructions behave alike in all respects of interest here, I will

give examples of only .the former construction.

As (6.90) suggests; than-clauses of any desired length

can be constructed.

(6.90) Wilt is taller than I imagine anybody would

ever, guess that people had begun expecting

Red to announce that he was.

 412

One of the operations that takes place in the formation of than-clauses

is that the compared' element in the'than-clause is obligatorily deleted

if it is identical to the element of the main clause with which it is

compared. Thus in (6.914, because the two compared adjectives arc

dissimilar, the one in, the than-clause is. retained. In (6.91b),

however, since the compared adjectives are identical, the parenthe-

sized occurrence in the than-clause is obligatorily deleted.

(6.91). a. The sofa was longer than the room was wide.

b. The sofa was longer than the desk was (long).

This deletion operation is subject to the GINIPC, the CSC, and the SSC,

as the sentences in (6.92), (6.93), and (6.94) show.

(6.92) a. *Wilt is taller than I know a boy who is.

b. Wilt is tallerthan I believe (*the claim)

that Bill.is.

(6.93) a. *Wilt is taller than Bill is strong and.

b. *Dean drank more booze than Frank ate

Wheaties and Sammy drank.

(6.94) a. *Wilt is taller than that Bill is is

generally believed.

b. Wilt is taller than it is generally

believed that Bill is.

subject to the
There is another deletion rule which is

best treated as being a special case

constraints and which is probably
In sentences containing
of the rule which forms comparatives.
 -er...than or inherently comparative verbs like increase, diminish,

outrun, overthrow, etc., it is possible to have 122:phrases, like those

5
in (6.9k), which make precise the amount by which the compared ele-
18
ments differ.

(6.95) a. Wilt is taller than Bill by 7 millimeter.

b. The raise which Scrooge generously gave

Tom's father increased his yearly salary by

five cents.

c. The hare outran the tortoise by so much

that he forgot the latter was even in the

race any more.

d. Who knew Mickey would overthrow home plate

by that much?

If two sentences contain such by:phrases, as is the case

with the sentences of (6.96),

(6.96) a. Wilt is taller than Bill by that much.

b. Big 0 is taller than the Cooz by that much.

then it is possible for' one sentence to appear as a subconstituent

of the other, superficially, at least, as a degree modifier of much.

Thus (6.96b) can become a modifier of the occurrence of much in (6.96a),

as in (6.97).

(6.97) Wilt is taller than Bill by as much as Big 0

is taller than the Cooz.

The objects of the preposition can also be compared, as is the case

in (6.98).

r4r,'"'"-- "5.'4171' 73777711`wr.

. ,
 414

(6.98) Wilt is taller than Bill by more than Big 0

is taller than the Cooz.

Exactly what the rule is which is at wrk.here is not my concern: for

my present purposes it is sufficient to point out that this apparent

rule of deletion has an unbounded scope (this is suggested by (6.99)),

(6.99) Wilt is taller than Bill by as much as

everybody seems to expect me to admit to

having publicly proclaimed that I believed

Big 0 to be taller than the Cooz.

and that it is subject to the CNPC, the CSC, and the SSC (cf. (6.100),

(6.101), and (6.102), respectively).

(6.100) a. *Wilt is taller than Bill by as much as I

know a boy who thinks that Big 0 is taller

than the Cooz.

b. Wilt is taller than Bill by as much a

Peter believes (*the claim) that Big 0 is

taller than the Cooz.

(6.101) *Wilt is taller than Bill by as much as I

watch all the games and I know Big 0 is

taller than the Cooz.

(6.102) a. *Wilt is taller than Bill by as much as that

Big 0 is taller than the Cooz is believed.

b. Wilt is taller than Bill by as much as it

is believed that Big 0 is taller than the

Cooz.
 415

6.1.3.2. The second deletion yule which obeys the constraints is

optionally,
the rule which converts (6.103a) into (6.103b), sometimes

sometimes obligatorily.

(6.103) a. ?The rock was too heavy for me to pick it

up.

b. The rock was too heavy for me to pick up.

that this rule must be

I am not entirely sure of this, but I believe
far down in a tree
allowed to delete elements which are indefinitely.

(cf. (6.104)).

(6.104) a. This rock is too heavy for me to begin to

decide about helping Bob to try to pick it

up.

b. ??This rock is too heavy for me to begin to

decide about helping Bob to tryto pick up.

long examples of this

Even if it is possible to find indefinitely
elements
construction, a restriction must apparently be stated so that

be deleted: no grammatical
of clauses containing finite verbs will not

sentences like (6.105) appear to exist.

(6.105) *This rock is too heavy for us to try to

claim that we picked up.

it must be made subject to

If this rule is formulated with variables,
(6.106), (6.107) (if grammatic4 sen-
the CSC, the SSC, and the LBC, as

tences like (6.107b) exist), and (6.108) show.

(6.106) a. Sodium is a little too peppy for me to want

to try mixing it and water in a teacup.

pe 40 tia 4.o J4 7 '4:Awl

b, 5bdtvpii ,r peppy -or

Go1/4 et:fie;-
 416

(6.107) a. *That piece of ice is too big for for him to be

able to pick up with a teaspoon to be likely.

W 441LACIA. r^^ 1.%4N 4^r gr re, 1..0

likely for him to be able to pick up with

a teaspoon.

(6.n8) a. Bsab is too thin for me to be able to squeeze

into his jacket.

b. *Bob is too thin for me to be able to 'squeeze

into jacket.

The rule which is at work here can probably be collapsed

with the rule which converts (6.109a) into (6.109b),

(6.109) a. This rock is light enough for. Marcia to

pick it up.

b. This rock is light enough for Marcia to

pick up.

for the grammaticality of sentences (6.103)-(6.108) is not affected

by the substitution of Adj+enough for too+Adj.

6.1.3.3. A rule possibly related to this last rule is the one

which converts (6.110a) into (6.110b):

(6.110) a. The socks are ready for you to put them on.

b. The socksfare ready for you to pug. ln.

Once again, although it is difficult to construct long examples, it

may be the case that this deletion rule can operate over indefinitely

long stretches of phrase markers (Cf.'(6.111)).

 417

(6.111) a. The socks are ready for you to go about

beginning to put them on.

b. ?The socks are ready for you to go about

beginning to put on.

As was the case with the previous rule, this rule seems not to be

able to delete elements of clauses containing finite verbs (cf. (6.112)).

(6.112) a. Tht socks are ready for you to announce

that you will put them on.

b. *The socks are ready for you to announce

that you will put on.

If this rule must be stated with variables, then it must also be

subject to the CSC and the LBC, as (6.113) and (6.114) show. Sentence

(6.115a) shows that it is not possible to delete elements of senten-

tial subject clauses, but I have not been able to 'find sentences like

(6.115b), where the deletion has become possible after the extra-

position of the clause, so it may be that this ruleis subject to'a

stronger constraint than the previously discussed rules in this

section.

(6.113) a. The socks are ready for you to try them

and the garters on.

b. *The socks are ready for you to try and the

garters on,

(6.114) a. Pfc. Golliwog is ready for you to inspect

his bunk.

b. *Pfc. Golliwog is ready for you to inspect

bunk.
 ft p . Siaa4AMPIPPAPIPPlPP1P....-

418

(6.115) a. * The socks are ready for for you to put

on to be planned.

* f^r 4r 1-nha pinnnaA

for you to put on.

The facts that I have brought out here in,connection

with ready hold true for a small class of similar adjectives, luch

as suitable, fit,- convenient, etc., none of which can be provided

with a plausible deep structure source at present.

They also hold true for adjectives like easy, difficult,

hard, etc., which occur in constructions like (6.116).

easy
(6.116) It is difficult to play sonatas on this
hard
violin.

T.t has been assumed in previous transformational studies

(Cf., e.g., Rosenbaum (1965)) that sentences like those in (6.117)

are to be derived from the structure underlying (6.116) by a reordering

transformation which substitutes some NP in the extraposed clause

of (6.116) for the subject of (6.116), the pronoun it.

easy
(6.117) Sonatas are difficult to play on this
hard
violin. .

Recently, however, several new facts have come to light which cast

doubt on the correctness of this analysis. Klima has pointed out to

me that bor.L :':,.117) and (6.118), which are not synonymous, would be

derivable from the structure underlying (6.116).

 419

easy
(6.118) This violin is difficult, to play sonatas
hard
On.

Similarly, Perlmutter has observed (cf. Perlmutter (op. 'cit.)) that

the sentences of (6.119), which would have the came deep structure,

are also not synonymous.

(6.119) a.. Imade John easy to get along with.

b. I made it easy to get along with John.

A more serious problem is posed by such sentences as

(6.120).

(6.120) John tries to be easy to get along with.

Perlmutter (op. cit.) argues that it is incorrect to analyze Imes

being lexically marked in such a way that the rule of Emil. NP Deletion

must apply to delete the superficial subject of the next sentence

down, as was proposed in Lakoff (1965). He presents a number of

convincing arguments, all of which suggest that in the correct analysis

ror

of la, the fact that such sentences as (6.121) are ungrammatical

(6.121) * John tried (for) Bill to play whist.

will be attributed to a deep structure restriction that the verb la

requires its deep subject tG be the same as the A222. subject of the

complement sentence.

If Perlmu.tter's hypothesis that the constraints on La

are to be stated in terms of deep structure, rather than in terms of
is correct
derivations then the fact that (6.120) is grammatical forces the
'A

conclusion that the deep subjects of'asy. in (6.117) and (6.118) are
 420

sonatas and violin, respectively. And the underlying structure of

the constituent sentence in (6.120) would be roughly that shown in

(6.122):

(6.122)

NP

John is st

one gets along with John

Thus the rule that forms such sentences as (6.117) and

(6.118) is a deletion rule, like the other rules discussed in § 6.1.3,

and not a reordering rule, like those discussed in §§ 6.1.1 - 6.1.2,

unless the above arguments can be gotten around.. This. rule appears

not to be able to delete elements of clauses containing finite verbs

(cf. (6.123)),

(6.123) 7* These flowers would be easy for you to

say that you had found.

and to be subjectto the CSC (cf. (6.124)).

(6.124) * My mother is easy to please my father and.

 421

As is the case with adjectives like ready, a stronger constraint

than the SSC seems to be operative here, for neither (6.125a) nor

(6.125b) is grammatical.

(6.125) a. * Bill would be easy for for you to chat

with in Moscow to become expensive.

b. *.Bill would be easy for it to become

expensive for you to chat with in Moscow.

6.2. Shapina21L11
6.2.0. In §§ 6.1.1 - 6.1.2, I gave a large list ofnreordering

transformations" -- rules whose structural change specifies that

some term of the structural index is to be moved around some other

term of it -- and showed that each was subject to the constraints

of Chapter 4. In this section, I will demonstrate that there

are rules which perform such an operation, but yet are not subject

to the constraints. It is possible, however, to find an important

formal difference between reordering rules which are subject to

the constraints, and reordering rules which are not: in rules of

the first type, if a term of the structural index is adjoined to, or

permuted around another term, the original term is deleted or

substituted for. But in rules of the second type, the original

term is not deleted, but remains behind in pronominal form, as a

kind of place-marker.

.
 422

6.2.1. A clear example of the contrast between these two

types of rules can be seen from a comparison of the rule of

Topicalization, (4.185), which Y have repeated for ease of

reference, and the rule of Left Dislocation, (6.126).

(4.185) Topicalization

X NP Y
OPT
1 2 3

(6.126) Left Dislocation19

X NP Y
OPT
1 2

24 [1 [ 2 ] 3
+Pro

This latter rule converts the structure underlying

(6.127) into any of the structures underlying (6.128)

(6.127) The man my father works with in Boston

is going to tell the police that that

traffic expert has set that traffic

light on the corner of Murk Street

---- far too slow.

(6.128) a. The man my father works with in Boston;

he's going to tell the police that ...

 423

b. eMy father, the man he works with in

Boston is going.to tell the police that ...

therei
c. (in) Boston, the man my father works with *its it

is going to tell the police that ...

a. The police, the man my father works with

in Boston is going to tell them that

e. That traffic expert, the man my father

works with in Boston is going to tell the

police that he has set that traffic light

on the corner of Murk Street far too slow.

f. That traffic light on the corner of Murk Street,

the man my father works with in Boston is

going to tell the police that that traffic

expert has set it far too slow.

g. (?On) the corner of Murk Street, the man my

father works with in Boston is going to tell

the police that that traffic expert has set

that traffic light there far too slow.

h. Murk Street, the man :my father works with .in

Boston is going to tell the police that that

traffic expert has set the traffic light

ion the corner there

on that corner f &r too slow.
.

* on it J
 424

The fact that the versions of (6.128c) and (6.128h)

which contain the definite pronoun it is obvi.ously the same as the

ML r MS..w
1..G
^ r a.111.
IA 1114N ^v.^ nnA 1*1^f4

would be excluded by some restriction along the lines of that

proposed in Kuroda (1964). Another restriction on this rule is

that it only places constituents at the head of main clauses:

while (6.129) is grammatical,

(6.129) My father, he's Armenian, and my mother,

she's Greek.

to my ear, the sentences in (6.130) sound unacceptable.

(6.130) a. * That my father, he's lived here all

his life is well known to those cops.

b. * If my father, he' comes home late, my

mother always grills him.

c. * It started to rain after Jackie and me,.

we bad finally gotten to our seats.

This restriction is somewhat too strong, for sentences in which

this rule has applied in certain object clauses seem to be acceptable

(compare (6.131a) with (6.131b)), and mysteriously, sentences like

(6.130b) seem to be improved if the rule has applied in both clauses

(cf. (6.132)).

(6.131) a.?* I acknowledged that my father, he was

tight as a hoot-owl.

b. I said that my father, he was tight as

a hoot-awl.
 425

(6.132) If my father, I- cumeu, " ' M"

_

mother, she always grills him.

Note in passing that the same restriction about

subordinate clauses also obtains for Topicalization. Thus such

sentences as those in (6.133) are ungrammatical.

(6.133) a. * That beans he likes is now obvious.

b. * I'm going to write to the Game Warden

if more than one deer my neighbor brings

back.

c. * I don't know the boy

who the flowers Mary gave tot
the flowers who Mary gave toi

Again, topicalization is sometimes possible in clauses

and object position, though not in clauses and subject position.

(6.134) a. ? The Revenooers claim that informers they

never use.

b. * That informers they never use is claimed

by the Revenooers.

As my purpose is not to present a maximally correct

formulPtion of each of these rules, I shall disregard these improvements

and pass on to the main business at hand: a comparison of the

constraints to which (4.185) and (6.126) are subject.

Notice that noun phrases can be dislocated out of

complex NP (cf. the b, c, a, and h-versions of (6.128)), out of

coordinate structures (cf. (6.135)), out of sentential subject clauses

 426

Ccf, (6.136)). and out of left branches of larger NP (cf. (6.137)).

And the distance that the dislocated NP has. traveled in (6.128h)

suggests that the statement of the rule must make crucial use of

a variable.

(6.135) a. My father, I hardly ever see him and

my mother when they're not glaring at

each other,

b. This guitar, I've sung folksongs and

accompanied myself on it all my life.

c. Poor Jonesy, it had started to rain and

he had no umbrella,

(6136) My father, that he's lived here all

his life is well-known to the cops.

(6.137) My wife, somebody stole her handbag

last night.

Thus Left Dislocation is not subject to the CNPC, the

CSC, the SSC, or the LBC. But I showed in S 6.1.1.5 and in § 4.3.2.1

that Topicalization is subject to all these constraints. Since both

rules reorder term 2 of their structural index, some formal

distinction between them must be found, if the generalization that

all reordering transformations obey the constraints is to be retained.

A distinction which appears to be adequate is that between

copying transformations and .staaim transformations (cf. (6.138)).

 427

(6.138) If the structural index of a transformation

has n terms, al, a , a , it is a

2 n
reordering transformation if its structural

change has any ai as its kth term, or if

a is adjoined to its kth 'term! where

i .

i 0 k.

If a transformation reorders ai, and its

structural change substitutes the identity

element or some ak, i -14 ko for the ith

term of the structural index, the transforma-

tion is a chozijignation. Other

reordering transformations are called

copying transformations.

For example, if the structural index of a transformation

were that shown in (6.139), it would be a chopping transformation (or

rule) if any of the lines in (6.140) were its structural change, but

it would be a copying rule if any of the lines in (6.141) were.

(6.139)____ a - a - a - a4
1 2 3

1 2 3 4

(6.140) a. 1 3 2 4

b. 1 2130-4

c. 1 0 3 4 +2

d.441 0 312 0]

etc.
 428

(6.141) a. 2+1 2 3 4

b. 1+2 2 3 4

c. 1 2 3 4+2

etc.

The generalization for which this distinction is crucial

is that stated in (6.142).

(6.142) Chopping rules are subject to the constraints

of Chapter 4; copying rules are not.

Siace Topicalization is a charring rule, it is subject to the constraints.

Since Left Dislocation is not, it is not subject to them.

The generalization in (6.142) is really a shorthand way

of regarding all the constraints of Chapter 4. Thus the CSC, (4.84),

instead of stating "... no conjunct may be moved....", should state

... no conjunct may be chopped ...", and similarly for the other

constraints of Chapter 4. Such a restatement will be postponed

until § 6.5 below.

6.2.2. For another clear contrast between copyingsand chopping

rules, consider the rule of Right Dislocation:

(6.143) Right Dislocation

X -
[Pro] Y
1 2 3

El 2
[4.Po]

This rule converts the structure underlying (6.144) into

 429

any one of the structures underlying (6.145).

(6.144) The cops spoke to the janitor about

that robbery yesterday.

(6.145) a. They spoke to the janitor about that

robbery yesterday, the cops.

b. The cops spoke to him about that robbery

yesterday, the janitor.

c. The cops spoke to the janitor about it

yesterday, that robbery.

This rule is, as (5.12.3) would predict, upward bounded.

This can be seen from the contrast in grammaticality between (6.146)

and (6.147):

(6.146) a. That they spoke to the janitor about that

robbery yesterday, the cops, is terrible.

b. That the cops spoke to the janitor about it

yesterday, that robbery, is terrible.

(6.147) a.?* That they spoke to the janitor about that

robbery yesterday is terrible, the cops.

b.?* That the cops spoke to the janitor about

it yesterday is terrible, that robbery.

Sentences like those in (6.146) show that this rule is unlike the

rule of Left'Dislceation in that it can copy a constituent at the

end of a subordinate clause, while Left Dislocation must be restricted

to main clauses.
 430

The specification in term 2 of (6.143) that the NP

to be right-dislocated not be a pronoun is necessary to exclude such'

sentences as those in (6,148).

rh 'I
e
(6.148) a. * They let him go yesterday,
him
1

b. I like beer, *I
?*me

we
c. * We'll go together,
us

they
d. * They can't stand each other,
them

The restriction is stated somewhat too strongly, at

present, for it would not allow the generation of such sentences

as those in (6.149), unless a coordinate NP, all of whose conjuncts

have the feature [+ Pro] can still be argued to have the feature

11Prol - which seems unlikely to me.

'

(6.149) I
a. We'll do it together, you and
ne

he and she
b. They can't stand each other,
him and her

Note that the,rule of Left Dislocation does not require

the NP to be dislocated not to be a pronoun -- the sentences in

(6.150), which correspond to those in (6.148), are gr.ammatlical. 20

(6.150) ati
*He they let him go yesterday.
Him

b. *1 I like beer.
die
 431

c. we'll go together.
Us
r
. J )
_ *They! ,
d. C } , they cant stand each other.
I Theml
l )
Once again, hawevei, I am not concerned with fine points

in the formulation of Right Dislocation -- my main purpose here is

to show how the constraints'on this copying rule differ from those

. on the rule of N.? Shift, (5.57); for except for the various minor

conditions stated on each. rule, their only difference is that the

former is a copying rule, while the latter is a chopping rule.

Since both rules are upward bounded, they will of

course both be subject to the CNPC and the SSC, The sentences

in (6.151) are a syntactic minimal pair: the ungraMmaticality of

(6.151a) and grammaticality of (6.151b) shows that the CSC restricts

the operation of only the rule of NP Shift. And the sentences in

(6.152) show the same to be true of the LBC.

(6.151) a. * I saw Mary and downtown yesterday your

friend from Keokuk.

I saw Mary and him downtown yesterday,

your friend from Keokuk.

(6.152) a. * I noticed car in the driveway last night

your_friend from Keokuk.

b. I noticed his car in the driveway last

night, your friend from Keokuk.

 432

In 5 4.3.2.3. above, I presented evidence showing that

a constraint is necessary, to the effect that no NP can move

rightwards out of a prepositional phrase, thereby stranding the

preposition (cf. (4.231)). In connection with my remark that the

generalization in (6.142) is a.shorthand way of rewording the

constraints of Chapter 4, condition (4.231) should be reinterpreted

but only on
as a constraint not on all reordering transformations;

chopping transformations. The sentences in (6.153) constitute another

minimal pair which shows the need for this distinction: that (6.153a)

is ungrammatical, but not (6.153b), shows that only NP Shift, and not

Right Dislocation, is subject to (4.231).

(6.153) a. * I spoke to about the war yesterday

that guy who's always following us.

b. I spoke to him about the war yesterday,

that guy who's always following us.

6.2.3. Distinguishing between copying and chopping rules will

also provide an explariation of the following fact, which is otherwise

puzzling. There is a dialect of English in which all the sentences

in (6.154) are perfectly grammatical.

(6.154) a. I just_saw that girl who Long John's claim

:.that E3w:.,.s a Venusian made all the

headlines.

;,'nor
c,

.4
 433

b. All the students who the papers which Fi

submitted were lousy I'm not goikg to allow

c.
to register next term.
1
Didn't that guy who the Gam.; Warden and him
ormssmosIbe

had ,seen a flying saucer ;rack up?

d. Palmer is a guy who for ,nim to stay in

;101110

school would be stupid.

e. The only kind of car which I can never

seem-to-get-T.1;1 carburetor adjusted right

4

is them Stanley Steamers.

f. King Kong is a movie which you'll laugh

yourself sick if you see it .

The rule that forms this type of relative clauses would

appear to differ from (4.135), the more usual rule, only in that the

structural change of (4.135) specifies that term 4, the relativized

element, is to be deleted, whereas this rule MN only pronominalize5

illillii
term 4. Thus this rule 411 a copying rule, while (4.135) is a chopping

rule. And, as (6.142) predicts, this rule is subject to none of the

constraints: in (6.154a) and (6.154b), elements of complex NP's have

been relativized; in .(6.154c), a conjunct has been, and in (6.154d), a

constituent of a sentential subject clause. In (6.154e), an NP on the

left branch of a larger NP has been relativized, and in (6,1540, an

element of a subordinate clause has been. If any of the boxed pronouns

1119.01M,....1.44 Vr"

I
 434

1.11 (6.154), which this rule leaves behind, are deleted, as would

be the case if (4.135) had applied, none of the resulting sentences

is grammatical.

Such sentences as those in (6.154), while common in

almost everyone's speech, are regarded as sulotandard by normative

grammarians. But there are languages whose relative clauses are

normally formed by a copying rule like the one responsible for the

sentences of (6.154), and in these languages, such sentences are

regarded as fully grammatical. Michael Brame has informed me21

that this is the case in several dialects of Arabic.

6.2.4.1. If the correct analysis-of appositive clauses is

that implied in § 4.2.3., above, where I stated that the second

conjoined S of (4.115) could be inserted into the first, in

apposition to the NP Pietro, then the rule, which forms these

r.its
clauses is a chopping rule, and it violates the CSC. =OM rule
e 1,e 44' q1.10 a. re

gigigrojounk
/5
chopping rulelI know of whichillft not subject to

all the constraints of Chapter 4. It therefore merits very careful

scrutiny.

There are two arguments for deriving appositive clauses

from coordinate structures. The first is that there are cases where

such clauses can begin with :and, as in (6.155).

.1
 434

in (6.154) , which this rule leaves behind, are deleted, as would

be the case if (4.135) had applied, none of the resulting sentences

ic grnmplatic=11.

Such sentences as those in (6.154), while common in

almost everyone's speech, are regarded as substandard by normative

grammarians. But there are languages whose relative clauses are normally

formed by a copying rule like the one responsible for the sentences of

(6.154), and in these languages, such sentences are regarded as fully

21
Michael Brame has informed Me. that this is the case in
grammatical.

severAl dialects of Arabic.

6.2.4.

6.2.4.1. If the correct analysis of appositive clauses is that

S
implied in § 4.2.3. above, where I stated that the second conjoined

44 the NP
of (4.115) could be inserted into the first, in apposition
rule,
Pietro then the rule which forms these clauses is a chopping

and it violates the CSC. This rule would be one of the two chopping
are,
of
rules I know of which al= not subject to all the constraints

Chapter 4. It therefore merits very careful scrutiny.

There are two arguments for deriving appositive clauses

from coordinate structures. The first is that there are cases where

such clauses can begin with and, as in (6.155).

 435

who.
"(6.155) Enrico, s the smartest of us all,
and h
. nn& f-to wnal,ca SeVen Qof'Anik,

The second argument is that after NP's whose determiners

are any, no, every, etc., appositive clauses cannot appear (cf. (6.156)),

A
Any }
who
(6.156) student, wears socks, is
and he
i°
Every 1

a swinger.

and that in these cases are the corresponding conjoined sentences also

impossible:

Any
(6.157) * No student is a swinger and he wears
Every "1.

socks.

These arguments are valid, and the facts they are based

on must be explained somehow.

But there is a problem here: how are sentences like

(6.158) to be generated?

(6.158) Is even Clarence, who is wearing mauve

socks, a swinger?

This sentence cannot be derived from the structure shown in (6.159),

 436

(6.159)

and

ILEvan Clarence is a swinper .arence s wear npauve socks

for the arguments in § 4.2.4.3 showed ",hat such deep structures must

be rejected on the basis of some constraint stated in terms of deep

structure, not in terms of transformational operations.

The gravity of the two problems connected with deriving

sentences like (6.158) from structures like (6.159) -- namely the fact

that if it is a chopping rule that is involved in the conversion)it

is not subject to the constraints, and the fact that such senten0,:is

as those in (4.149) seem only to be excludable if structures like (6.159)

are also excluded as deep structures -- suggests that this derivation

must be wrong, and that another source must be found for appositive

clauses.

At present, the only solution that comes to my mind is

a very radical one. Since it appears that there must be rules of some

kind which convert one sentence into two (how else can the second

sentence in (4.90a) be derived than from a conjunct?), it may be

that there are also some rules which reverse the process. That is,

it may be that the source for (6.158) is the sequence of structures

underlying the sentences in (6.160).

 437

. (6.160) Is even Clarence a swinger? Clarence

is wearing mauve socks.

If this analysis is adopted, it will still be possible

to account for the fact that the sentences of (4.156) are ungrammatical,

for the corresponding sentences sequences are also.

fAny
(6.161) *)No is a swinger. He wears
Istudent
very

socks.

However, the first argument that appositive clauses come from conjoined

structures (i.e., the fact that appositives can be introduced by and)

cannot be gotten around in this reanalysis, at least, not in any way

I can see at present. I am, therefore, very diffident in proposing

this reanalysis. It looks like the best analysis of appositives that

is presently available, but one which is none too good.

6.2.4.2. There is only one other chopping rule that I know of

which in any way provides counterevidence to (6.142). This is the

rule of There Replacement. It seems reasonable to assume that after

the rule of There Insertion has converted (6.162a) into (6.162b),

some rule should operate on the structure underlying this latter

sentence to convert it into the structure which underlies (6.162c),

by substituting some NP for the derived subject, ee.

 438

(6.162) a. Seven pine trees are behind that barn.

.b. There are seven pine trees behind that barn.

22
c. That barn has seven pine trees behind it.

There are two arguments which support this analysis. The

first is that just as the rule of There Insertion requires an indefinite

subject NP to apply (cf. the strangeness of (6.162b) if the is

inserted before seven, and the ungrammaticality of (6.163b)),

(6.163) a. There will be a hole in Jack's pocket.

b.* There will be the hole in Jack's pocket.

sq sentences like (6.162c) require the object of have to be indefinite.

Thus if the precedes seven, (6.162c) is as odd as (6.162b), and the

sentences in (6.164) parallel exactly those in (6163), from which

they are derived.

(6.164) a. Jack will have a hole in his pocket.

b.* Jack will have the hole in his pocket.

The second argument has to do with the fact that such

sentences as (6,162c), while they cannot contain reflexives (cf.

(6.165a)), must contain a pro-form of the subject NP as the

object of the preposition (cf. the ungrammaticality of (6.165b)

and (6.165c)).

(6.165) a.* That barn has seven pine trees behind

itself.

b.* That barn has seven pine trees behind the

COW
 439

23
c.* Jack will have a hole in my pocket.

That the rule of There Replacement must have a variable

in its structural index was pointed out to me by Nary Bremer: not

only can the structure underlying (6.163a) be converted into that

underlying (6.164a), but also into the one underlying (6.166).

(6.166) Jack's pocket will have a hole in it.

And the structure underlying (6.167) can eventually become any one

of the sentences of (6.168), all of which I believe to be fully

grammatical, but some of which are rendered unacceptable by an

output condition.

(6.167) ?? There is a hole in John's quilt's upper

right-hand corner.

(6.168) a.?? John's quilt's upper right-hand corner

has a hole in it.

b. John's quilt has a hole in its upper

right-hand corner.

c.?? John has a hole in his quilt's upper right-

band corner.

d. John has a hole in the upper right-hand

corner of his quilt.

Notice that since the rule of hereTlent substitutes.

some NP for the derived subject there, it is a chopping rule, by

definition (6.138). We would therefore expect it to obey the CNPC, the

CSC, and the LBC (I have as yet not been able to construct examples
 440

to show it to be subject to the SSC). The fact that (6.169a) cannot

be converted into (6.169b) or (6.169c) shows it to be subject to the

CSC,

(6.169) a. There are seven holes in the door and

window.

b. * The door has seven holes in it and the

window.

c. * The window has seven holes in the door

and it.

but the fact that (6.163a) can be converted into (6.164a) , and that

(6'.167) can be converted into (6.168c) and (6.168d) shows this rule

not to obey the LBC. To complicate things, ho;4ever, if the possessive

NP i3 an inalienable possessor: the rule apparently is subject to

the LBC: (6.170a) cannot be transformed into (6.170b), though it

may be transformed into (6.170c).

(6.170) a. There is a blemish on the end of Jerry's

sister's nose.

b. * Jerry has a blemish (At the end of his

sister's nose.

c. Jerry's sister has a blemish on the end

of her nose.

It seems to be the case that only animate NP can be

copied out of complex NP's. 'Thus while the sentences in (6.171)

can be transformed into those in (6.172), those in (6.173) cannot be

 441

transformed into those in (6.174).

(6.171) a. There is a hole in the rug which Toby

bought in Butte.

b. There was an error in the proof Prof. Hiatus

presented.

c. There was a snake behind the car Fred was

sitting in.
he
(6.172) a. ? Toby has a hole in the rug which
j
bought in Butte.
he
b. Prof. Hiatus had an error in the proof
*Sarah

presented.
jhe 1.
c. Fred had a snake behind the car was
Joe

sitting in.

(6.173) a. There was a yellow collar on the dog which

the car injured.

b. There's a hole in the tarpaulin which that

stone is holding down.

c. There was a snake behind the car the time

bomb was sitting in.

(6.174) a. * The car had a yellow collar on the dog

which it injured.

b. * That stone has a hole in the tarpaulin

WO,

which it is holding down.

c. * The time bomb had a snake behind the car

25
which it was sitting in.
 442

liot only does this rule unexpectedly fail to obey the

CNPCfand the LBC under certain conditions, it also appears to obey

stronger constraints. Thus while the boxed NP in (6.175a) can be

relativized (cf. (6.175c)), it cannot be substituted for there, as

(6.175c) shows.

(6.175) a. There were several hundred people yelling

for me to put down gently.

b. The hot potato which there were several

rr hundred people yelling for me to put down

gently turned out to have been filled with

TNT.

c. * The hot potato had several hundred people

yelling for me to put it down gently.

6.2.5. Except for the two rules discussed in § 6.2.4 I know

of no chopping rule that does not obey all the constraints of Chapter 4.

And I know of no copying rule which does obey them. Thus the distinction

made in (6.138) appears to have a basis in linguistic fact, as long as

there are so many unresolved problems in the analysis of the two

constructions discussed in § 6.2.4. I will provisionally assume,

therefore, that the generalization stated in (6.142) is correct.

6.3. Reordering over Variables

6.3.1 In § 4.2.3 above, I discussed the rule proposed in

 443

Lakoff and Peters (1966) which I will refer to as Conjunct Movement.

It is stated approximately as in (6.176).

(6.176) Conjunct Novement26

[NP - [and NP ]NP 7 VP

1NP

1 2 3

1 . 0 31/2

This rule must apply to (6.177), which underlies (6.178a),

to move the circled NP along the path shown by the arraw,"eventually

producing (6.178b).

(6.177)

11-- VP

and 1NP and RI danced

1
Bartlett Toli

(6.178) a. Bartlett and Toni danced.

b. Bartlett danced with Toni.

But as I pointed out in footnote 13 of Chapter 4, as

the CSC is presently stated, such. an operation is impossible, for

 444

Conjunct Movement is a chopping rule, dnd the subject NP of

(6.178a) is a enordinate node.

It is not .possible to claim that somehow this

particular subject NP is not affected by the CSC, for it is

impossible to move either boxed NP to the end of (6.177) by

the rule of NP Shift, (5.57), as is shown by the ungrammaticality

6.179).
05
(6.179) a. * Bartlett and danced Toni.
/

b. * (And (and)) Toni danced Bartlett.

Since it is not this particular construction that is

exempt from the CSC, it must be some feature of the rule. The

operation of the two rules of Conjunct Movement and pi) Shift

is virtually the same -- in each, some NP gets moved to the

end of a sentence. But there is a significant difference in the statement

of the rules; while the latter rule permutes to the end of the first

sentence up any NP (because term 2 of (5.57) is surrounded by variables),

the former rule specifies that the second conjunct of the conjoined

subject NP may be moved to the end of its VP.

In other words, the first rule makes crucial use of

variables, while the second does not. At present, I believe it

to be the case that the constraints of Chapter 4 never affect any rule

unless that rule reorders one of its terms around a variable. This

generalization is stated in (6.180).

 445

-(6.180) Only rules 141 which terms are reordered

around variables are subject to the constraints

of Chapter 4.

In the case just discussed, it is possible to imagine

an alternative solution involving rule ordering. Thus it could be

argued that if either the first and of (6.177) has been deleted, or

if the second has been converted into a preposition, the subject node

of (6.178a) would no longer be coordinate, so the CSC would not be in

effect any longer. But if this is the correct explanation, it must

be possible to order the rule of NP Shift early, so that it precedes

all these changes, and I do not know whether such an ordering can

be maintained.

However, even if such an analysis can be carried through

for English, there are languages, like Japanese, where the conjunction

is not rewritten as a preposition by the rule which corresponds to

(6.176), so such an explanation will not be possible in general. And

there are two additional cases, from English, which seem to require

the generalization stated in (6.180). These will be presented immediately

below.

6.3.2. In sentence (6.181), the N1 her cannot be relativized, as

(6.182) shows.

(6.181) It bothers me for her to wear that old fedora,

 446

(6.182) a. * The only girl for whom it bothers me to

wear that nld fedora 3s Annnhellp,

b. * The only girl who it bothers me (for) to

wear that old fedora is Annabclle.

It is not the case that no element of an extraposed for -

to phrase can be chopped, as (4.273) shows. It therefore seems to be

necessary to add (6.183) to the conditions box for English.

(6.183) No element in the environment [for -- VP]

%(1 .- :
Vy chopped

But now consider the rule of It: Replacement, which was

discussed in S 5.1.1.1. The formal statement of this rule, which

raises interesting theoretical problems which I will not take up here

(they are discussed brifly in Lakoff (1966)), contains as a subpart

the rule shown in (6.184).

for
(6.184) X - [Al NP
oss VP 1S ]
NP

1 2 3 4 5

1 4 3 0 5

This ride will convert the structure underlying (6.185a)

into the one underlying (6.185b).

 447

IF inrw I would prefer it for Ithere) to be no

W.J.W.,

tarkiAg.

b. 1 would prefer there to be no talking.

Notice Chat the boxd NP of (6.185a), even though it

is in the environment which is specified in (6.183), has been chopped

by rule (6.184). Once again, however, there is a cortrast in the

formal statement of the rules in questit5a. The rule of Relative Clause

Fomation, which is subject to (6.183), as the ungrammaticality of

(6.182) shows, permutes the relativized NP around a variable, while

in (6.184), the chopped term merely moves over the constants in term 3.

Thus the fact that (6.185b) is grammatical, and (6.182) ungrammatical,

provides further evidence for the correctness of (6.180).

A "I 1 In § 3.1.1.3.1, above, t pointed out that it was necessary

to constrain the rule of NP Shift somehow, so that sentences like

(3A0b), (3.35b), and (3.36b) would not be generated. But the condition

I stated there, (3.34), can be generalized, for while the underlined

NP in (6.186a) can be questioned (cf. (6.187a)), if the indirect object

precedes the direct object, as in (6.186b), the indirect object cannot

be questioned (cf. (6.187b)).

(6.186) a. He gave my binoculars to that girl.

b. He gave that girl my binoculars.

 448

-(6.187) a. Which girl did he give my binoculars to?

ti
27
b. * Which girl did he give my binocu/ars?

Since it is not universally the case that indirect

objects cannot be chopped (for instance, in German the sentence

Welchem Nadchen b er meinen Feldstecher ?, which translates (6.187b),

is grammatical), it would appear that some condition like that stated

in (6.188) must appear in the conditicms.box for English.

(6.188) No element may be chopped out of the

environment [NP V NP ]S, unless the

following NP begins with a preposition.

However, if this condition is correct, how can both

versions of (6.186) be passivized, as the grammaticality of the

sentences in (6.189) indicates is necessary?

(6.189) a. Ny binoculars were given to that girl-by

him.

b. That girl was given my binoculars by him.

The answer is obvious: since all reordering rules which

are subject to (6.188) make crucial use of variables, while the .

Passi.:2 Rule, however it is to be stated, need not do so, If the

generalization expressO In (6.180) is added to the theory of grammar,

the contrast between (6.187) and (6.189) z9n be naturally accounted

for. Therefore, on the basis of these facts, and the evidence

presented in §§ 6.3.1 - 6.3.2, I tentatively propose the addition of

(6.180) to the theory of grammar.

 449

6.4. Islands
6.4.0. The fundaMental insight of this section is due to
Paul Kiparsky. In connection with some extremely important, but
still unpublished, research on complement constructions which he
is conducting, he pointed out that the that-clause in (6.190r1) has
a factive meaning, while this is not the case in (6.19010.
(6.190) a. Bill confirmed that Roger has eaten
[the cake I
b. f had eaten
Bill alleged that Roger ?has
;the cake)
One who utters (6.190a) is not only reporting an action of Bill's,
he is himself asserting that the content of the that-clause is true.
This is not the case with (6.190b) -- there the speaker merely
comments on Bill's action, without himself taking any stand on the
truth of the embedded sentence. One of the many ways that Kiparsky
has discovered this semantic difference to be paralleled by
syntactic differences is in the behavior of elements of the two
.kinds of that-clauses under chopping rules. Thus while the boxed
NP in (6.190b) can be questioned (cf. (6.191b)), the boxed NP
of (6.190a) can only be questioned with difficulty, if at all,
(cf. (6.19.1a)).
(6.191) a.?? What did Bill confirm that Roger had
eaten?
 450

b. 'That did Eill allege that Roger had

ontrIn?

For the purposes of the present discussion, Kiparsky's most important

observation was that the restrictions on a feature- changing rule like

z,cor7:9aii)r
r
Indelf4 kj 4.- (S 71) exactly parallel those on the rule of

Question, a chopping rule.

(6.192) a.?* Bill didn't confirm that Roger had eaten

anything.

b. Bill didn't allege that Roger had eaten

anything.

These facts can be generalized trivially, to yield the

hypothesis in (6.193).

(6.193) All feature-changing rules obey the same

constraints as chopping rules.

The rest of 6.4 is devoted to exploring the oonsequem:es

of this hypothesis. In § 6.4.1, I will discuss a few of the many

pieces of confirming evidence that I know of, and in § 6.4.2, I

will discuss all the-disconfirming evidence that has come to light

thus far. Finally, in § 6.4.3, I will examine the converse of (6.193)

and define the concept island.

6.4.1.

This section is divided into four parts. In the first

6.4.1.0.

three, 1 will show how various feature-changing rules are subject to

 451

the CNPC, the CSC, and the SSC, respectively, and in the fourth,

.0 ruicse whiph annear

I will show [low various restrictions on ^!"""-14Pa cs _

in the conditions boxes of a number of languages also affect the

operations of feature-changing rules.

6.4.1.1. If the rule of IndefilliajasaamIum, (5.71), is

subject to the CNPC, the contrast between the sentences of (6.194)

is accounted for (cf. also (5.73e)) .

(6.194) Waldo didn't report (* the possibility)

that anyone had left.

The CNPC also correctly predicts that sentences like (5.731), where

rule (5.71) has gone down into a relative clause, are ungrammatical.

There are, however, relative clauses which can contain

words like an, ever, and at all, which typically occur in environments

where rule (5.71) operates. The sentences in (6.195) are a

representative sample of such clauses.

(6.195) a. Nobody who hates to eat anything should

we -L in a delicatessen.

b. Anybody who ever swears at me better

watch his step.

c. Everybody around here who ever buys

anything on credit talks in his sleep.

d. I want all the students who have ever

trled to pat Nacavity to show me their scars.

 ,

452

The only
e, travelers who anybody has
*Only the

.ever robbed don't ci:rry machetes,

What seems to be going on here is that indeterminates

can become indefinites in a relative clause which modifies an NP

whose determiner belongs to the set no, an v, a, sma, all, the first

(but not the second, third, etc.) the last, the Ad &. 4. est (cf. the

best steak I ever ate) (but not only the), etc., whether or

not the sentence containing the clause is negated. That this rule

cannot be the same as (5.71) is indicated by the following facts.

The word an cannot appear in the relative clause of

(6.196), because the determiner some of the NP this claust, modifies

is not one of the set mentioned above.

(6.196) * I can't remember the name of Rbody who

had (*any) misgivings.

But if the boxed [I- Affective] element of (6.196) has triggered the

change of the boxed some to any., then the environment for the rule

which allows indefinites to appear in relative clauses will be met,

and this rule can go down into the relative clause, as has happened

in (6.197).

(6.197) I can't remember the name of Anybody

who had any misgivings.

It is therefore evident, since the rule in question

must follow (5.71), that the two rules cannot be collapsed into one.
 453

Incidentally, sentence (6.198) shows that this rule must be able to

apply to its own output,. in a rather interesting way.

(6.198) lEver+ody who has ever, worked in any

1

office which contained any typewriter

2

which had ever3 been used to type any3

letters which had to be signed by any4

adniniStrator who ever worked in any

5 5

department like mine Will know what I

mean.

The element which allows the presence of all the ayls

and ever's in this sentence is the boxed determiner every. The first

time the rule in question applies to the structure underlying

(6.198), it will produce evert end . But now, the result of this

first application, the determinerAng , provides a new environment for

the rule to reapply in (recall that this rule could not have gone down

into a relative clause on an EP whose determiner was some (cf. (6.196))).

The rule must then be ablc to produce any2 on its second application,

and this any will provide yet a third environment for 'he rule to

reapply in, and so on down the tree. This is the only rule I know of

which applies in this''anti-cyclic" way, eating its way from higher

sentences into lower ones, in sequence, so to speak, instead of the

normal type of rules, which process embedded sentences first, and

then the sentences that contain them. This rule is therefore

eminently worthy of very detailed inve.-..tigation, which would be

 454

beyond the scope of this section, so that it can be determined

whether this apparently necessary anti-cyclic ordering is in fact

necessary.

The second fact which demonstrates the impossibility

of collapsing this rule and (5.71.) can be seen from a comparison

of the sentences in (6.199).

(6.199) IT° student who [everl goes to Europe

{*Every

has enough money.

As sentences (6.195a) and (6.195c) demonstrate, both

no and every, belong to the set of determiners which can cause

indeterminates in relative clauses to be converted into indefinites

(cf. the boxed ever). However, the fact that only the negative

determiner no can cause the indeterminate sometimes in the main

clause to change to the circled ever shows once again that the rule

which produces the sentences in (6.195) must be a different rule

from (5.71).

But, it might be asked, even granting that the two

rules are different, why are not both subject to the CNPC, since

both are feature-changing rules? The answer to this question is

that both are: the CNPC is stated in (4.20) in such a way that

it prevents a constituent from being chopped out of a sentence

dominated by a complex NP and from then being moved out of the NP.
 455

For it is possible, as Ceorge Lakoff has pointed out to me, for

elements to be moved out of the complex NP's sentence, as long as

they stay within the NP itself (cf., e.g., rule (4.135)). To

say that a feature-changing rule obeys the CNPC is to say that no

elomnt not -1,,,,14natP,1 by a complex NP can ncf-..:t changes In the

sentence dominated by that NP. Thus the determiners under discussion,

since they are dominated by the NP, can cause the introduction

of the feature [-h Indefinite into a relative clause, as is the case

in (6.195), while [4-Affective] elements which are outside the NP

cannot.

There are two other sets of facts which can be accounted

for readily if the hypothesis stated in (6.193) is correct. In § 3.1.3

above, I pointed out that the Case Markin; Rule must be restricted so

that no elements of relative clauses are assigned the case of the

head NP, and I stated an ad hoc condition (lit which subscripts had

to be used) to this effect on rule (3.58). However, once it has been

stated in (6.193) that all feature-changing rules like (3.58) are

subject to the CNPC, no restriction need be stated on rule (3.'8).

Similarly, in § 4.1.6, I claimed that it was universally

true that reflexives do not go down into relative clauses. I know

of only one language, Japanese, which contradicts this generalization

(the Japanese rule of Reflexivization will be investigated briefly

in § 6.4.2 below), so though the generalization must be reformulated

in a weaker way, it appears to contain an important truth, a truth

 456

which can be explained if Reflexivization is subject to the CNPC.

I hope that it will turn out to be the case that if there are

other languages whose rules of reflexivization can go down into

complex NP, it will be possible to point to some formal property

shared by all such languages, on which this unusual behavior can

be made to depend. At present, however; this is no more than a

hope, so the Japanese facts constitute clear counterevidence for

(6.193).

6.4.1.2. To see that rule (5.71) is subject to the CSC, it

is sufficient to observe that the boxed some of (6.200) cannot be

converted into my if (6.200) is negated: while (6.201a) is

possible, (6,201b) is not.

(6.200) I ate the ice cream and [some cake.

(6.201) a. ? I didn't eat the ice cream and some

cake.

b. * I didn't eat the ice cream and any

cake.

Similar facts obtain for sentence (6.202): if negated,

as in (6.203a), thr! boxed some of the second conjunct cannot be

converted into all,

(6.202) I realized that it had rained and

crops had been destroyed.

 457
i

(6.203) a. I didn't realize that it had rained and

some crops had been destroyed.

b. * I didn't realize that it had rained and

any crops had been desLroyed.

Interestingly, there appears to be a phenomenon he

which is reminiscent of the "across-the-board" rules that were

discussed in § 4.2.4.1 above. Thus indefinites can appear in

conjuncts if they are conjoined with or, instead of and, as in

(6.204). .

(6.204) I didn't eat any ice cream *or any

d

cake.

It seems to me that such sentences as those in (6.205), where

indefinites appear only in one conjunct, are all ungrammatical in

varying degrees, but I am not sure of this intuition.

* any ice cream or IlarY's cake

the
(6.205) I didn't eat ?* the cake or any ice cream

? Mary's cake or any ice cream

J
Even if it should prove to be correct that some kind of

across-the-board constraint is operative here, I can see no way of

accounting for the differences between the sentences of (6.205), or

for the fact that only or can appear in such sentences as (6.204) and

(6.205). Clearly a great deal of further research is needed here.

I
 458

The CSC appears to restrict feature-changing rules

not only in that the feature [I- Indefinite] cannot go down into a

conjunct, but also in that the [I. Affective] element which broad-

casts the L+ Indefinite] features cannot be in a conjunct. In

Lakoff and Peters (op. cit.), (6.206a) and (6.206b) are derived

from the same underlying structure, the only difference being that

in the derivation of (6.206b), two rules have applied which do not

apply in the derivation of the more basic (6.206a) -- the rule of

Conjunct Ilbvenent (6.176), and a rule which deletes the preposition

with which was originally in front of the superficial object

Maxine.

(6.206) a. Cottlob and Naxime met in Vienna.

b. Cottlob met Maxine in Vienna.

Now note that if the determiner few appears in a

conjunct of such a conjoined NP subject, rule (5.71) cannot

introduce the feature [+Indefinitej into the second conjunct (cf.

the ungrammaticality of (6.207a)), but that if the rule of Conjunct

Movement has applied, to break up the coordinate structure, the

moved conjunct can be converted into an indefinite (cf. (6.207b)).

(6.207) a. * Few writers and any playwrights meet

in Vienna.

b. Few writers meet any playwrights in

Vienna.
 459

The situation seems to be a great deal more complicated

than the above facts would indicate, however. So note that (6.207a)

is not improved by replacing any with some, as might be expected.

And while (6.208a) is ungrammatical, (6,208b) is grammatical.

(6.208) a. * My brother and few Americans meet in

Vienna.

b. My brother meets few Americans in

Vienna.

Also, while (6.209a) is grammatical, (6.209b) is not.

and no
(6.209) a. No writer, playwright, speaks
nor any
clearly.

b. * No writer, and no 1playwright, meets

nor anyj
in Vienna.

These sentences raise so many problems that I can only call attention

to them here -- I have no idea what processes are at work.

That the Reflexivization Rule is subject to the CSC

is immediately apparent from the sentences in (6.210).

* Mary and himself

(6.210) a. Bill understands
* himself and Mary

b. * Bill and Mary washed himself.

c. * Andy pinched Sarah and tickled herself.

d. * The gun and a description of itself

lay on the bureau.

 ,
460

Ii A particularly clear example is provided by (6.211),

whose underlying structure is that shown in (6.212).

1,,
(K 414)111, Bill believes that Anna and he are

similar.

.re

A
1
believes

it

A /
VP

and NP NP be similar

I I
Anna Billi

If the rule of Itieslasenent does not apply, this

structure will undergo various rules, and will finally emerge as

the grammatical (6.211). If It Replacement does apply, however,

and the circled NP has been substituted for it in (6.212), it would

 461.

be expected that the leftmost occurrence of Bill would be able

to r flexivize the rightiimost occurrence, for each commands the

other. That this does not happen (cf. the ungrammatical version

of (6.213)) is explained if the CSC also constrains feature-changing

rules.
him to be
(6.213) Bill believes Anna and
* himself

similar.

I believe it to be the case that feature-changing

rules are also subject to the SSC, but the pieces of evidence I

have been able to find to support this claim are based on very

delicate intuitions, and these may not be shared. For instance,

I believe it to be true that while Indefinite Incorporation

can go down into that-clauses, it cannot go down into them if

they are in subject position. Thus (6.214a) is ungrammatical,

and (6.214b), where the embedded subject clause has been extraposed,

is grammatical.

(6.214) a. * I deny that that McIntyre has

mat is certain.

b. I deny that it is certain that McIntyre

has any money.

The problem is this: since the underlined phrase in

(6.214a) is a sentence which is dominated exhaustively by NP,

 462

acceptability of (6.214a).
outpu,t condition (3.27) will lower the

Does, therefore, the fact that rule (5.71) has applied to produce the

The
boxed mx in this sentence contribute to its unacceptability?

answer to this question will lie in a comparison of (6.214a) and

(6.215), which is identical to the former sentence except for the

fact that any has been replaced by some.

(6.215) ?? I deny that that McIntyre has some

money is certain.

I myself find a clear, if small, difference between (6.214a) and

(6.215): while both are unacceptable, I would judge the former to

be ungrammatical in addition. If these are the correct facts, it

is to the SSC that the difference between (6.214a) and (6.215)

must be attributed.

The second set of facts that seem to indicate that

a feature-changing rule is subject to the SSC has to do with

Klima's rule" of NelaumlamEmEELisa (cf. Klima (op. cit.)),

which can optionally convert the structure underlying (6.216a) into

the one which underlies (6.216b),

(6.216) a. Tom will not force you to marry any

student.

b. Tom will force you to marry no student.

and which obligatorily converts the structure underlying (6.217a)

into the one underlying (6.217b).

 463

(6.217) a. * The writers of any of the reports

didn't know the answer.

b. The writers of none of the reports

ten
s.aacz.by
414
"e"""".
C.4.44.7%11&11

Klima slipports his claim that (6.216b) and (6,217b) are instances

of sentence negation by showing that both may be followed by

neither-tags, as in (6.218),

(6.218) a. Tom will force you to marry no student,

and neither will I.

b. The writers of none of the reports

knew the answer, and neither did the

writers of any of the chronicles.

a property which he demonstrates elsewhere in the article to be

restricted to sentences whose main verb is negated.

Since both (6.216b) and (6.217b) are grammatical, the

rule of \e alive must be able to operate forward and

backward. And since it can operate forward into an extraposed

clause, changing (6.219a) into (6.219b),

(6.219) a. It is not certain-that you'll marry any

(particular) student.

b. It is certain that you'll marry no

student.

the fact that it cannot, if my intuitions are correct, operate

backwards into a subject clause ((6.220a) cannot become (6.220b)),

mo.4^1.
 464

requires explanatinn,

(6.220) a. That you will marry any (particular)

student is not certain.

.b. * That you will marry no student is

28
certain.

The fact that the SSC can block (6.220b), if the rule of 1.1egatart
29
Incorporation is formulated as a feature-changing rule, thus

provides further support for the hypothesis that all feature-

clanging rules obey the same constraints as chopping rules.

6.4.1.4. In § 5.1.3.2.3, in connection with the sentences

in (5.103), I pointed out that the Russian rule of Reflexivization,

(5.98), could not go down into clauses headed by the word Xto 'that'.

But it is necessary in any case to state in the Russian conditions

box that no elements of sto-clauses can be chopped out of these

clauses. For instance, the NP Ilengginu 'woman' in (6.221) cannot

be relativized, as the ungrammaticality of (6.222) shows.

v vv.
(6.221) ja znal sto on ijubil aenscinu.

'I knew that he loved the woman.'

v vv.
(6.222) * vot z enscina kotoruju ja anal

here is the woman who I knew

V
sto on ijubil.

that he loved.
 465a

Since some condition must be stated in' the grammar of Russian in

any case, so that (6,122) will not be generated, if the hypothesis

in (6.193) is adopted as a principle of the theory of language,

thp tingrnrilmtirality of (5,103h) can hp explained_ The fact that

the rule of Russian Genitive Introduction, (5.92), also does not go

down into tto-clauses (cf. the sentences in (6.223)),

V
(6.223) a. ja ne znal s to on eto sdelal.

I not know that he this (acc.) did

'I didn't know that he did this.'

b. * ja ne znal vsto on etovo sdelal.

I not know that he this (gen.) did

is of course to be explained on exactly the same basis. Similarly,

it can be shown that the two Finnish rules which were discussed in

§ 5.1.3.2. -- the rules of Finnish Partitive Introduction, (5.85),

and Finnish Nominative Introduction, (5.108), also do not go down

into clauses headed by etta 'that', a fact that can be explained

on the basis of hypothesis (6.193) and the restriction in the

Finnish conditions box that no elements can be chopped out of

etta-clauses (cf. the ungrammaticality of (4.249b)).

Finally, if (6.193) is in the theory of grammar, the

fact, noted in § 4.1.6 above, that there is a parallelism between

the relativizability of elements after picture nouns and their

 465b

ref exivizability (cf., e.g., the parallelism between (6.224) and

(6.225)),

a
(6.224) The man who I gave John ??this picture
*Ed's
of was bald.
1

a
(6.225) I gava Jack ?this picture of myself.
*Ed's
1

can be explainLtd, and the correct prediction can be made that other

featlre-changing rules will be subject to the same curious constraints

involving the determiners of picture nouns (cf. (6.226)).

a
(6.226) I didn't give Jack *this picture of
*Ed's
30
anybody.

6.4:2. While the facts presented in § 6.4.1 provide very

strong evidence that (6.193) is correct, there are still some

puzzling countercases. Thus while (6.193) would predict that no

features of NP's which are on the left branch.of larger NP's

could be changed, this in fact can happen, as (6.227) indicates.

(6.227) I hope I'm not treading on anyone's

toes.
 466

Secondly, while sentences like (6.210a) show that the

normal rule of Reflexivization cannot go down into conjuncts, there

is an interesting rule which produces emphatic reflexives, in free

31
variation with non-reflexive pronouns, which can do so. (cf. (6.220).

(6.228) Abernathy admitted that thR poison pen

letter had been written by my sister'

t

and him
thimself 11.

Thirdly, while the facts presented in § 6.4.1.3 show

that there are environments in which features cannot be changed in

subject clauses, as the SSC and (6.193) would predict, it is obvious

that there are circumstances in which features can be changed. Thus

the rule of Sequence of Tenses, (5.115), must operate backwards in

(6.229) to change the ungrammatical is of the subject clause to was.

(6.229) That the sun is out was obvious.

was

A particularly puzzling fact, in light of the contrast

between (6.214a) and (6.215), is the fact that indefinite Incorporation

can go backwards into the subject Clauses of negated verbs and

adjectives, or EE Affective] verbs and adjectives, as (6.230) shows.

in not known
is not certain
(6.230) That.anybody ever left at all
is impossible
'surprises me
is odd
 467

.
In Japanese, it appears to be possible to violate

at least the CNPC, with *respect to the rule of Reflexivization.

Thus the boxed NP of tree (6.231), which underlies (6.232), can

be reflexivized, yielding (6.233).

(6.231)

VP

NP

/\ S

VP sakana
tabeta

Biru. NP
0----1

sakana katta

(6.232) Biru
i
wa kare ga katta sakana o tabeta.
i
Bill he bought fish ate

'Bill ate the fish he bought.'

0.233) Biru wa zibun ga katta sakana o tabeta.

i i
Bill self bought fish ate.
 468

The same situation appears to obtain with respect to

sentences in apposition to Sentential nouns like syutyno 'claim'.

Thus in (6.234), either the reflexive pronoun zibun 'self' or

the third person non-reflexive pronoun kare 'he, she, it' can

be used to refer back to the subject of the sentence, Biru 'Bill'.

fkare4 1
(6.234) Biro_ wa ga kono sakana o
1
zibun
l i

he 11

Bill this fish

self
..;

katta to iu syutyoo o sinzita.

bought that say claim believed.

'Bill believed the claim that he had

bought the fish.'

1-do not know what the facts are in Japanese with

respect to whether Reflexivization can violate the CSC; but if it

can, the obvious conclusion is that (6.193) cannot be universal, and

that particular grammars must designate in their conditions boxes

whether (6.193) is operative In the language or not. That is, (6.193)

would be a language-particular "option".

Whatever the outcome of the investigation of the question

as to whether (6.193) is a universal condition (which now seems unlikely),

or an option, it seems reasonably clear that it is operative in English.

 469

In the next section, I will investigatc.the consequences of assuming

the converse of (6.193) also to be operative in English.

6.4.3.

6.4.3.1. The converse of (6.193) is stated in (6.235):

(6.235) All chopping rules obey the same

constraints as feature-changing rules.

The only constraint that I know to hold for all feature-

changing rules is the one which was stated in (5.77), and then

restated in (5.122) in terms of command: if an element A in a

phrase marker is to have the feature f+ F] added to it, the element(s)

which triggers this change must command A.

Graphically, then, (5.122) says that if A, at the

bottom of the schematic phrase marker shown in (6.236), is to be

changed, then the triggering element must lie within the shaded

"strip" of (6.236), for it is only elemmts of this strip that

command A.
 .
470

6)

s s

s
AA
LAA

A
A
There is an independently motivated principle of derived

constituent structure, which restricts reordering transformations

in a way highly reminiscent of (5.122): this principle is stated in

(6.237).
 471

(6.237) If the structural change of a transformation

specifies that one term of the structural

index is to be adjoined to a variable,

pick the highest proper analysis which the

variable allows, and adjoin the term to

32
this string.

Instead of attempting a formal definition of the term

"highest proper analysis", which would be straightforward, if

difficult, I will illustrate the effect this principle has with an

example.

Supposing that (6.238a) is converted to (6.238b) by the

rule of Adverb Prepos-Incv, (5.67) .

(6.238) a. What Bob cooked yesterday still tastes

good tonight.

b. Tonight, what Bob cooked yesterday still

tastes good.

If (6.238a) is assumed to have the structure shown in (6.239) (whether

(6.239) is correct in all details -- in particular whether the adverb

tonight should be dominated by VP, S1, or by some other node, is

not important), then which of the possible derived constituent structures

shown in (6.240) should be assigned to (6.238b)?

 472

(6.23
fNNS

NP VP NP

r I
still tastes rood VAdverb

tonight

Bob

cooked

(6.240)

./4NP
//
/
/,
I

/
still tastes good
//
,141)
1 2
NP

what
 473

Intuitively, of course, it is clear that the preposed

tonight can only be the daughter of S; if it were dominated by

NP or S2, the couatc.rintuitive claim would be made that the

1
string tonight what Bob cooked yesterday is a constituent, and if

it were dominated by NP2, that tonight what is a constituent.

Syntactic evidence is available to show that tonight

cannot be immediately dominated by NP1, S2, or NP2. Since

Adverb Preposing must precede all rules of pronominalization (cf.,

e.g., the paradigm in (5.151), where the subject of will go can only

be pronominalized if the adverbial if-clause has been preposed by

(5.67)), (6.241b) will only be derivable from (6.241a) if the string

whatBol2ccLokecczrclay is a constituent, for it is clear that

this string is what the it of (6.241b) refers to, and pronominalization

is restricted to delete constituents under identity.

(6.241) a. Tonight, what Bob cooked yesterday

still tastes good, so tonight, what Bob

cooked yesterday will be eaten up.

b. Tonight, what Bob cooked yesterday still

tastes good, so tonight it will be

eaten up.

If tonioht were dominated by S2 or NP2, the string what Bob

:ecizescootesclay would not be a constituent, and if NP dominated

1

tonight, while this string would be a constituent, it would not be

 474.

an NP. Since it seems most reasonable to analyze the it of

(6.241b) as being a pro-NP, the only place the adverb tonight

can be attached is as a sister to NP1, connected by the highest

dotted line in (6.240) to S1. Since principle (6.237) would

ensure that this d.c.s., and none of the other counterintuitive

possibilities indicated by the other dotted lines of (6.240)

would result, there is good reason to believe that (6.237), or

its/equivalent, must appear in any adequate theory of grammar.

I But now note that (6.237) will also ensure that if

element A of phrase marker (6.236) is permuted around a variable,

it will not move out of its strip. It is of course theoretically

possible to state a reordering rule which makes crucial use of

variables and which can move an element out of its strip; one

such rule is stated in (6.242).

16.242) VP ]s - X - NP Y

1 2 3 4 5

4 4- 1 2 3 0 5

This rule could apply to a structure like (6.243a) and convert it

to (6.243b), moving the circled NP off its shaded strip in (6.243a).

 475

(6,243) a,

if.Y2P try

x_t
] EilLa.k.usc to
th..

King Ron

Ring Kong you will laugh yourself

sick

see
 476

The question is, will the grammar of any natural

:Imagine ever have to contain such a rule? My present answer

to such a question, an answer based on all the rules I know of

is an unequivocal "no". Not only must the "highest proper analysis"

principle of (6.237) be stated in the theory of grammar, but some

formal constraint must be stated so that rules like (6.242)can

never be stated in any grammar. So little is known at present,

however, that it is pointless to propose a formal constraint to

this effect at the present juncture.

To point up the close conceptual parallels between

(5.122) and (6.237), a paraphrase which makes use of command may

prove helpful. (5.122) asserts that if the feature [4- F] is

added to an element A, the cause of the change commands A

(is in the strip above A). (6.237) asserts that if A moves,

it will move to a position which commands (is in the strip above)

its original. position.

Actually, this last paraphrase of (6.237) is inaccurate,

for if it is only required that a preposed adverb command its place

of departure, the adverb tonight could be attached as the daughter of

S or NP in tree (6.240): only if it were to become a daughter

1

of S2 or VP in (6.240) would it no longer command its point of

2 00

departure. Thus (6.237) is a stronger condition, for reordering

transformations, than (5.122) is for feature-changing transformations.

 477

If we accept both (6.193) and (6.235) as 'working hypotheses, then,

since (6.237) is necessary in any event, as the discussion of

(6.238) and (6.241) showed, it should be possible to logically

deduce (5.122) from the stronger (6.237). In other words, if

the conditions on feature-changing rules are all and only the

conditions on reordering rules (but cf. the discussion on

Japanese in § 6.4.2), then the asymmetry mentioned at the end

of § 5.1.4 above, that while there are upward bounded rules

which are downward unbounded, there are no downward bounded,

upward unbounded rules,should follow from the "highest analysis"

principle of (6.237). Intuitively. (5.122) "feels" the same as

(6.237), although I have as yet been unable to construct a

rigorous proof, along the lines sketched above, that the former

is a consequence of the latter.

6.4.3.2. As I showed in phrase marker (6.,236), the converse of

the relation command selects for each element A of phrase marker

P na1strj2.ofAthernaxiiint. Element A cannot be moved

off its maximal strip, nor can any element of P which is not on

this strip cause any feature to be added to A. In other words,

the maximal strip of A is the maximal ___domain of application for

all chopping or feature-changing rules.

 478

But how do the constraints of Chapter 4 affect the

maximal strips of a. phrase marker? The answer is easy to see: if

the main branch of the maximal strip of A (that is, the branch

consisting of all and only those nodes of P that dominate A)

contains one of the types of nodes which is specified in the

statement of the CNPC, the CSC, the LBC or the SSC as not permitting

the chopping of one of its subconstituents, then the maximal strip

is cut into a smaller strip at that node. That is, if the main

branch contains a complex NP with a lexical head, a coordinate

node, an NP on the left branch of a larger NP. or a sentence in

subject position, the main branch (and the strip it is a part of)

is cut at ell'a node. The resulting substrips I call islands, and it

is these islands that the feature-changing and chopping rules are

constrained to operate within.

6.5. Summary,

The rules of pronominalization which were discussed in

§ 5.3 above, and copying rules, like Left Dislocation, (6.126), or

the rule which forms relative clauses with a "returning pronoun",

like those in (6.154), are the rules which can cross island

boundaries. But what of the deletion rules of § 6.1.3, which were

shown not to be able to cross island boundaries? Under the

ela.remely broad definition of pronominalization that was given in

(5.148) of § 5.3.1, the rules of § 6.1.3 would be characterized

 479

chopping
as pronominalizations, and would not obey the constraints on

. and feature-changing rules which were developed in Chapters 4 and 5,

but inst-aA ;m1,1 be QflhjPnt to the less restrictive condition which

,

is stated in (5.152).

There is, however, one formal difference between the

rules of § 5.3 and the rules in § 6.1.3: while the former rules

can delete under identity in either direction, the latter rules

are stated to delete only in one direction. The English rules

mentioned in § 6.1.3 all deleted from the left to right (that

is the element on the right was deleted), while the Japanese

1

rule of Relative Clause Formation deleted only from right to left.

And the rule of Reflexivization, (5.98), can, in every language I

know of, be formulated unidirectionally ; so the puzzling fact noted

in footnote 24 of Chapter 5, that Reflexivization obeys the constraints

on feature-changing rules rather than the normal constraint on

pronominalization, can also be accounted for. It is at present a

total mystery as to why unidirectional pronominalizations should

obey the constraints of Chapters 4 and 5, but it does seem to be

the case in the few languages I have studied.

Summing up, then, the results of the investigation of

formal properties exhibited by rules which are subject to the constraints

of Chapters 4 and 5 can be expressed as in (6.244) below, where I

have used the term "cross" in an undefined, but I think intuitively

clear, sense:
 480

Variables in chopping rules, feature-

changing rules, and unidirectional rules

of deletion cannot cross island boundaries;

variables in other rules can.

1
 44., .4 4,,,r 4
, -, ,....

I:

.
481

Chapter 6

FOOTNOTES

1. It has been assumed since the inception of transformational

grammar (cf., e.g., Harris (1957), section 11.2) that these two

rules are the same, an assumption that I find extremely dubious.

The arguments that have been used are that the relative pronouns

(except for that) are a subset of the wh-words used in questions,

and that both rules are subject to the same constraints. But
1
l

/
if the main argument of this chapter is correct, that all chopping

transformations which move constituents over variables are

subject to the same constraints, then the second argument for

assuming the existence of a "WH-Rule", such as Chomsky's rule

(6), which I quoted in § 2.4.0 above, can be disregarded. And

the first argument for such a rule, which is essentially a

is weak.
morphological onea Although there are many parallels between

the uses of wh-words in questions and in relative clauses, there

are also puzzling differences. So while it is desirable to

relate the fact that who replaces human nouns in questions, and

the fact that it also does so in rcomr-hrpq, the fact that whose

can be used for both human nouns (the boy whose body was lithe

snored on) and non-human nouns (trie car whose body was dented

still runs) in relatives, but only for human nouns in questions

(Whose body was lithe? ici,lhose body was dented?) causes problems
 482

for those who assume that the two rules are the same. A more

importarit argument against identifying these rules can be

derived from the following considerations.

In sentences introduced by the expletive there, the

subject NP cannot be relativized (*The two men who there were

Euarding the door wore shoulder holsters). It cannot be argued

that sentences beginning with there are frozen to relativization,

for such strings as This is a problem which there are a lot of

people working on are grammatical. Nor can it be argued that

there is a restriction in the English conditions box which pro-

* hibits any reordering transformation from moving the subject of

a there-sentence, for such subjects can be questioned (How many

men were there guarding the door?). To me, it seems most likely

that the reason that such subjects behave differently under

Relative Clause Formation and Question will be connected with

the fact that subjects of there-sentences are always indefinite,

and a restriction on the former rule that the identical NP

in the constituent sentence always be definite. But whether or

not this analysis proves to he correct, unless the facts just

presented can be explained even on the assumption that the

rules of Question and Relative Clause Formation are the same,

it seems to me that the only arguments I know of which argue

for this are far too weak to be regarded as having established

such an identity.

e
 483

2. This sentence is of course perfectly grammatical as an expression

of surprise, but on such a reading; the wh-word why does riot

replace a purpose adverb, as it usually does in questions

(witness the grammaticality of Why, he left for that reason

after all!), and can be followed by a pause, unlike the word

how in (6.4a) and (6.4b). These facts are indicative of the

clear intuitive difference between this reading of (6.5b) and

the exclamatory sentences of (6.4).

3. The six-pointed star which I have prefixed to these examples,

one of McCawlPy's many bahnbrechenden ErfindunRen (cf. NcCawley

(1964), fn. 2), indicates that these sentences are only gramma-

tical if Yiddish. A particularly clear example of such a

sentence, for which I am indebted to David. M. Perlmutter, is

4 Egg', creams you want, bananas you'll get.

4. In sentence (4.18) above, 1 showed that while elements of

clauses which follow believe can be relativized, elements of

clauses which follow believe the claim cannot. Since such

sentences provide such a clear case of the operation of the

CNPC, I will use them as a paradigm example of this con-

straint throughout § 6.1.

 484

5. For some reason I cannot explain, elements cannot be extracted

by the rule which makes exclamatory sentences from most extra-

posed clauses, although elements can be relativized here.

Compare, e.g., *How brave it is certain that Tom is! with Here is

a house which it is certain that Tom lived in.

6. This sentence is acceptable with the meaning "I don't see how

he is so brave", if prefixed by the six-pointed star discussed

in fn. 3. It cannot, however, have the intended meaning of (6.4a)

7. Personal communication.

8. Of course, since (6.15a) contains an internal sentence which is

exhaustively dominated by NP, the output condition stated in

(3.27) will lower its acceptability. But it should not be

considered to be merely unacceptable, for the following sentence,

where when modifies had been established, while awkward, is

still far better than (6.15a): Bill left when that noone else

was awake had been established.

9. These facts were first pointed out by Katharine Gilbert, in

Gilbert (1967).
 485

10. This fact was pointed out to me by Morris Halle.

II. A rough estimate of the perils that await the unwary grammarian

who stumbles into this quagmire can be obtained from a quick

perusal of the myriad confusions and inconsistencies in Ross (1964).

12. This sentence cannot be blocked by any ordering of the rules of

NP Shift and Conjunction Reduction if the analysis presented in

Peters (in preparation) is correct. Peters argues that on the

reading of (6.57a) where the meaning is that the playing of

the guitar and the singing areS simultaneous, the conjoined VP

node should derive from a conjoined node in deep structure.

13. If both versions of (6.76b) are felt to be ungrammatical, this

rule must have the general constraint imposed upon it that no

element'of a clause containing a finite verb can be preposed.

14. These facts were brought to my attention by Maurice Gross.

15. That is, the morpheme en'of it' must command the verb to which

it is to be prefixed as a clitic. For a detailed treatment of

the grammar of clitics in several Romance languages, cf.

.Perlmutter (in preparation):

 486

16. In fact, if la maison is pronominalized fully, not merely to

some form such as celle -la that one there', nothing can save

(6.81a) from ungraMmaticality. The CSC will not allow the

clitic to be moved, but the rule which moves clitics to preverbal

position will not let it stay where it is. In such an impasse,

no matter which rule wins out, an ungrammatical sentence will

result.

17. As a rough indicator of the superficiality with which I heve

discussed this construction (indeed, all th.e constructions in

§ 6.1), consider the following facts, which were pointed out

to me by Sylvain Bromberger. in the sentence below,

Je vois les fenAtres de la maison et la porte 211.LgAgaf.

'I see the windows of the house and the door of the garage.'

while it is not possible to pronominalize and convert into en

either of the underlined phrases in isolation, if both are pro-

nominalized, a grammatical sentence results:

J'en vois les fenetres et la porte.

I of it see the windows and the door.

I of it
'I see the windows and the door
thereof

What particularly interesting is that the en here seems to

refer neither to de la maison 'of the house', nor to du garage

of the garage', but rather to the set, or gestalt, or individual

(to use Nelson Goodman's terms consisting of them both, a

concept only roughly translatable into English by such locutions

 487

as the house-gaIme. Notice that the reason that the CSC can

be "violated" here. is, in a strange new way, the same reason that

across-the-board rules (cf. § 4.2.4.2) can "violate" it. I

cannot deal further with this extremely interesting problem here.

18. The grammar of comparatives in general, and of these lzphrases

in particular, has been intensively examined by Austin Hale.

Cf. Hale (1965), Hale (to appear).

19. This term is due to Maurice Gross.

20. The ungrammatical versions of the sentences of (6.150), where

the pronouns are in the nominative case, can be blocked by

imposing the condition on Left Dislocation that the dislocated

NP be marked with the feature [4. objective ] . This feature

will only produce a phonetic difference if the NP to which

it is attached is one of the pronouns I, he, she, we, Iha.

21. Personal communication. Classical Arabic grammarians refer

to pronouns like the boxed ones in (6.154) as "returning

pronouns."
 488

22. That the rule which converts (6.162b) to (6.162c) changes be

to have.should occasion no surprise. There are a number of

deep ways in which these two verbs behave the same under

transformational rules, but a discussion of these facts would

be out of place here. One interesting rule of Italian, which

changes have to be in certain circumstances, will be discussed

in Perlmutter (op. cit.)

23. That this sentence may be acceptable to some, with the meaning

"Jack will cause a hole to appear in my pocket", need not

concern us here.

24. This sentence is grammatical if Joe appears in the relative

clause, but T am not surf it is an instance of the saue

construction.

25. I am not sure that the contrast in acceptability between

(6.172c) and (6.174c) is great.

26. I have greatly oversimplified the statement of this rule.

Lakoff and Peters (op. cit.) argue, e.g., that the and in

term 2 of (6.176) should have been converted into some preposi

...... , .....

tion (cf. He left. With her, She is similar to him, I am

different from her) before this rule applies. Also it is an open

...
 489

question as to whether term 2 should be Chomsky-adjoined or

daughter-adjoined to term 3.

27. There are some speakers who appear to find no difference in

acceptability between the sentences in (6.181), but I know

of no one for whom sentences like (3.20b), (3.35b), and

(3036b) are grammatical. I cannot explain this asymmetry.

28. Of course, (6.220b) is not ungrammatical on all readings. It

can mean 'That your spouse won't be a student is certain', but

this meaning is not related to the structure underlying (6.220a).

29. Klima postulates a negative constituent, net, so his rule of

agative Incorporation is not a feature-changing rule but rather

a chopping rule which inserts the chopped nea into some other

part of a phrase marker. But I know of no valid argument for

treating negation as being anything but a feature; Klima's

main argument that negation is a constituent has to do with his

notion in construction with, which I have already argued

(cf. § 5.2.2 above) is not adequate to the task of accounting

for the facts of Indefinite Incorporation, to say nothing of

restrictions on the other members of the class of feature-

changing rules. Even " Klima's analysis is right, however, so

that Negative Incorporation has be considered to be a rule

 490

which chops and inserts, it would still be possible to account

for the difference between (6.219b) and (6.220b) by broadening

i

the hypothesis stated in (6.193) so that it covered all kinds

of chopping rules.

Note also that the contrast between (6.220b) and (6.217b)

provides an additional argument for pruning. Thus if the NP

the writers of some of the reports is sententially derived,

/
which I believe is inescapable, then by the time the rule of
1
d iiesotivi applies, the sentence must have been
pruned, for otherwise the SSC will not allow (6.217a) to be

converted into (6.217b).

30. I have no explanation at present for the differential behavior

of the sentences in (6.224), (6.225) and (6.226) , if the

determiner of picture is this.

31. In Ross (1967c), I show how this rule provides evidence that all

declarative sentences are embedded as the direct object of a verb

like sax, whose subject is I, in deep structure. Note, by the way,

that this rule is unlike the normal rule of Reflexivization in that

it can go down into clauses.

32. For a definition of the term 'proper analysis', cf. Chomsky (1955),

Fraser (1963).
 491

Chapter 7

CONCLUSION

This thesis has been an attempt to provide the theory of grammar

with a more adequate notion of syntactic variable, a notion which I

showed in Chapter 1 and elsewhere to be absolutely essential if the

central fact of syntax -- that there are unbounded syntactic processes --

is to be captured. In Chapter 2, I argued that the earliest attempt

at limiting the power of variables, Chomsky's A-over-A principle, is

both too strong and too weak. A far more serious inadequacy in this

principle than those I discussed in Chapter 2 is the fact that it

cannot be extended in any natural way, as far as I can see, to account

for the phenomena which led me to construct a theory of syntactic

Islands. In Chapter 3, I gave a preliminary sketch of a theory of

node deletion, or pruning -- a theory which interacts closely with the

constraints developed in later chapters. In this chapter, I also

gave some evidence that a rather substantial revision in the syntactic

component was necessary -- that many conditions previously thought to

te best stated as restrictions on particular rules should instead be

regarded as static output conditions, with the rules in question being

freed of all restrictions. These output conditions effect no changes

on final derived constituent structures -- rather they lower the

acceptability of sentences output by the transformational component, if

these sentences exhibit certain formal properties which are specified in

 492

the conditions. Thus the relationship between grammaticality and

acceptability must become.much more abstract than has been assumed.

In Chapter 4, I formulated two putatively univc.rsal constraints

and one putatively universal convention, as well as a number of

language-particular constraints, which I showed to be intermediate

in generality between conditions on particular rules and universal

constraints, and thus to necessitate a further addition to the

syntactic component -- the conditions

In Chapter 5, I showed that various facts made necessary the

adoption of a new mechanism into the theory of grammar, so that rules

whose variables would otherwise be too strong could be correctly

stated. Langacker's notion of command, with suitable extensions,

was demonstrated to be adequate to this task, and a number of

interesting restrictions on types of rules were shown to be stateable

in terms of this notion. Various rules of pronominalization were

discussed, and it was shown that while these rules did not obey the

constraints of Chapter 4, they also did not obey restrictions which

could be stated in terms of command.

And in Chapter 6, I discussed a large number of rules, showing them

all to be subject to the constraints developed in Chapters 4 and 5. A

close examination of all rules subject to these constraints reveals

that not only are feature-changing rules and unidirectional deletion

rules subject to the same constraints as the chopping rules for which

the constraints were first developed, but that it is only rules which
 493

make crucial use of variables which are subject to them. Thus, in

a sense, it is wrong to speak of constraints on rules -- the

constraints in Chapters 4 and 5 are rather to be construed as

limiting the power of variables that can appear in a certain type

of rules. In conjunction with the notion command, the constraints

divide up phrase markers into islands, the maximal domains of rules

of the type in question.

JAll the proposals I have made should be regarded as being

extremely tentative, for our present knowledge of syntax is

ridiculously small. This thesis has raised far more questions than

it has attempted to answer. Among thaam are: Why should rules

which adjoin terms to the right side of a variable be upward bounded,

and not those which adjoin terms to the left of a variable? Why

should it be that chopping rules, feature-changing rules and uni-

__directional deletion rules share the property of being subject to

the constraints, to the exclusion of other rules? Why should there

be a difference between unidirectional and bidirectional pronominaliza-

tion? Why should it be that the constraints are all "downward-

oriented" that is, why should it be that there are phrase marker

configurations that prevent elements indefinitely far below them

from undergoing various syntactic operations, whereas there are

no configurations which affect elements indefinitely far above them?

Why should complex NP's) coordinate nodes, sentential subject clauses,

 494

and IIP's on the left branches of larger NP's all function the
i

same in defining islands? Can islands be shown to behave like

psycholinguistic entities?

While none of these questions can now he answered, the fact

that they can now be asked is a major result of the thesis. For

as e.e. cummings has said, "always the more beautiful answer who
/
asks the more beautiful question."
i

if
 495

BIBLIOGRAPHY

Anderson, Stephen R. (1957),. "How do so does so." Unpublished paper,

M.I.T.

Bach, Emmon (1967). "ProWominalization." Unpublished mimeograph,

University of Texas, Austin, Texas.

Bierwisch, Manfred (1966). "Regeln fur die Intonation deutscher

S'gtze." Studia Crammatica, VII, pp. 99-201. Akademie Verlag,
East Berlin.

Blass, Birgit (1965). A Comparison of the Adverbs and Prepositions

of Danish and English. Master's thesis, Columbia University.

Bolinger, Dwight (1967). "The Imperative in English." To appear in

Festschrift for Roman Jakobson. The Hague: Mouton & Co.

Browne, E. Wayles, III.(1966). "On the problem of enclitic placement

in Serbo-Croatian." Unpublished paper, M.I.T.

Chomsky, Noam (1955). lhelogicalstrt.tgziuisticTheor.

Mimeographed, M.I.T. Library, Cambridge, Masse

Chomsky, Noam (1957). Syntactic Structures. The Hague: Mouton & Co.

Chomsky, Noam (1961). "On the notion 'rule of grammar'." In

R. Jakobson (ed.), Structure of Language and Its Mathematical
Aspects, Proceedings of the Twelfth SyTposiumipApplied
----Mathematics, Vol. 12, pp. 6-24. Providence, R.I.: American
Mathematical Society.

Chomsky, Noam (1962). "A transformational approach to syntax." In

A.A. Hill (ed.), Proceedings of the 1958 Conference on
Problems of Linguistic Analysis in English, pp. 124-148.
Austin, Texas. Reprinted in Fodor and Katz (1964).

Chomsky, Noam (1964a). "The logical basis of linguistic theory."

In H. Lunt (ed.), Proceedings of the Ninth International
Congress of Linguists, Cambridge, Mass., 1962Q The Hague:
Mouton & Co.

The
Chomsky, Noam (1964b). Current Issues in Linguistic Theory.
Hague: Mouton & Co.
 496

Chomsky, Noam (1965). Aspects of the Theory of Syntax. M.I.T.

Press, Cambridge, Nass.

English and
Dean, Janet (1967). "Noun phrase complementation in
German." Unpublished paper, M.I.T.

Fillmore, Charles J. (1965a). "On the notion of 'equivalent sentence

Analysis, Report No. 11.
structure'." Project on Linguistic
Ohio State University, Columbus, Ohio.

in
Fillmore, Charles J. (1965b). Indirect Object Constructions
English and the Orderina of Transformations. The Hague:
Mouton & Co.

Fodor, Jerry A., and Jerrold J. Katz (1964). The Structure of Language:_
Readings in the Philosophy of Language. Englewood Cliffs,
N.J.: Prentice-Hall.

Fraser, J. Bruce (1963). "The linguistic framework for a sentence

recognition and analysis routine: Transformation structure."
Working Paper W-6266. Bedford, Mass.: Mitre Corporation.

Fraser, J. Bruce (1965). An Examination of the Verb-Particle

Construction in English. Unpublished Ph.D. thesis, M.I.T.

Gilbert, Katharine F. (1967). "Which-sentential relatives and as--

clauses." Unpublished paper, M.I.T.

Hale, Austin (1965). "Quantification and the English comparative.'

Unpublished mimeograph, University of Illinois.

Hale, Austin (to appear), quantification and the English Comparative.

The Hague: Mouton & Co.

Halle, Morris, and Noam Chomsky (to appear). The Sound Pattern of
English. New York: Harper & Row.

Harris, Zellig S. (1957). "Co-occurrence and transformation in

linguistic structure." Language, 33, pp. L93-340. Reprinted
in Fodor and Katz (1964).

"Pronominalization, reflexivizatic. ,Ind the

Jackendoff, Ray (1966a).
semantic component." Unpublished paper, M.I.T.

Jackendoff, Ray (1966b). "The erased NP in relative clauses and

complements." Unpublished %paper, M.I.T.

Katz, Jerrold 3., and Paul M. Postal (1964). An Integrated Theory of

linguistic Descriptions. Cambridge, Mass.: M.I.T. Press.

Positions in
Keyser, S. Jay (1964). Review of Jacobson, Adverbial
kcal..Lat. To appear in Language.'
 497

Kiefer, Ferenc (1966). "On emphasis and word order in Hungarian."

Unpublished paper, M.I.T.

Klima, Edward S. (1964). "Negation in English." In Fodor and Katz

(1964).

Kuroda, Sige-Yuki (1964). "A note on English relativization."

Unpublished paper, M.I.T.

Kuroda, Sige-Yuki (1965). Generative Grammatical. Studies in the

Japanese Lannage. Unpublished Ph.D. thesis, M.I.T.

Lakoff, George (1965). On the Nature of Syntactic

Mathematical Linguistics and Automatic Translation, Report
No. NSF-16 to The National Science Foundation, Harvard
University, Computation Laboratory.

Lakoff, George (1966). "Deep and surface grammar." Unpublished

paper, Harvard University.

Lakoff, George (1965). "Pronominalization and the analysis of

adverbs." To appear in Rosenbaum, Teter S., and Roderick
Jacobs (ed.), Readings in English Trnsformational Grammar.

Lakoff, George, and P. Stanley Peters (1966). "Phrasal conjunction

and symmetric predicates." Mathematical Linguistics and
Automatic Translation, Report No. NSF-17 to The National
Science Foundation, Harvard University, Computation Laboratory.

Lakoff, George, and John Robert Ross (1966). "A criterion for verb
phrase constituency." Mathematical Linguistics and Automatic
--Translation, Report No. NSF-17 to The National Science Founda-
tion, Harvard University, Computation Laboratory.

Lakoff, George, and John Robert Ross (in preparation a). The Abstract-
ness of Underlying Structures.

Lakoff, George and John Robert Ross (in preparation b). The Trans-
formational Component.

Langacker, Ronald W. (1966). "On pronominalization and the chain of

command." Unpublished multilith, University of California at
San Diego, La Jolla. (Page references are to this version.)
To appear in Schane, Sanford A., and David A. Reibel (ed.),
Modern Studies in English. Englewood Cliffs, N.J.: Prentice-
Hall.
 498

Lees,.Robert B. (1960). The Grammar of English Nominalizations. The

Hague: Mouton & Co.

Lees; Robert B. (1961) . "Grammatical analysis of the English com-

parative construction." Word, 17, pp. 171-185.

Lees, Robert B., and Edward S. Klima (1963). "Rules for English
pronominalization." Language, 39, pp. 17-28.

McCawley, James D. (1964). "Quantitative and qualitative comparison

in English." Paper presented at the Winter Meeting of the
Linguistic Society of America, December 29, 1964.

Perlmutter, David M. (in preparation). Deep and Surface Structure

Constraints in Grammar. Ph.D. thesis, M.I.T.

Pester :, P. Stanley(in preparation). Coordinate Conjunction in English.

Ph.D. thesis, M.I.T.
1

Postal, Paul M. (1964). "Underlying and superficial linguistic

structure." Harvard Educational Review, 34, pp. 246-266.

Postal, Paul M. (1966a). "On so-called 'pronouns' in English." In

Dineen, F.P., S.J. (ed.), Report of the Seventeenth Annual
Roundtable Meeting on Linguistics and Language Studies.
Washington, D.C.: Georgetown University Press, pp. 177-206.

Postal, Paul M. (1966b). "A note on 'understood transitively'."

International Journal of American Linguistics, 32.1 (1966).

Qualls, Brandon (to appear). Ph.D. thesis, M.I.T.

Quine, Willard V. (1960). Word and Oblect. Cambridge, Mass.:

M.I.T. Press, and New York: Wiley.

Rosenbaum, Peter S. (1965). The Grammar of English Predicate Comple-

ment Constructions. Unpublished Ph.D. thesis, M.I.T.

Ross, John Robert (1964). A Partial Grammar of English Superlatives.

Master's thesis, University of Pennsylvania, Philadelphia.

Ross, John Robert (1966a). "Relativization in Extraposed Clauses

(A Problem which Evidence is Presented that Help is Needed to
Solve)." Mathematical Linguistics and Automatic Translation,
Report No. NSF-17 to The National Science Foundation, Harvard
University, Computation Laboratory.

Ross, John Robert (1966b). "On the nature of bounding in syntactic

rules." Unpublished paper, M.I.T.

MEM.
...M1111
 - --4,,t1.16111.000)Par

499

Ross, John Robert (1966c). "Adjectives as noun phrases." Unpublished

ditto, M.I.T.

Ross, John Rcbert (1966d). "A proposed rule of tree-pruning."

Mathematical Linguistics and Automatic Translation. Report
No. NSF-17 to The National Science Foundation, Harvard
University, Computation Laboratory.

"On the cyclic nature of English pronominali-

Ross, John Robert (1967a )..
zation." To appear in Festschrift for Roman Jakobson.
The Hague: Mouton & Co.

Ross, John Robert (1967b). "Auxiliaries as main verbs." Unpublished

utimeograph,-M.I.T.

Ross, John Robert (1967c). "On declarative sentences." To appear in

Jacobs, Roderick and Peter S. Rosenbaum (ed.), in
English Transformational Grammar.

Ross, John Robert (1967d). '!Gapping and the order of constituents."

To appear in the Proceedings of the Tenth International Congress
of Linguists.

Smith, C. (1961). "A class of complex modifiers in English." Language

37, pp, 342-365.

Vendler, Z. (1962). "The order of adjectives. ". Transformations and

Discourse Analysis Project paper no. 31, University of
Pennsylvania.

Warshawsky, F. (1965a). "Reflexivization." Unpublished paper, M.I.T.

Warshawsky, F. (1965b). "Reflexivization." Unpublished paper, M.I.T.

Winter, W.(1965). "Transforms without kernels?" Language, 41.3,

pp. 484-489.

Zwicky, A.M., Jr., and S. Isard (1963). "Some aspects of tree theory."
Working Paper W-6674. Bedford; Mass.: Mitre Corporation.
 500

BIOGRAPHY

I was born on May 7, 1938, in Boston, Massachusetts, the

son iof Dr. Douglas Allen Ross and Eleanor Campbell Mott Ross. I

lived in Montreal and then in Sudbury, Massachusetts, until I was nine,

when we moved to Poughkeepsie, New York. I had the good fortune to

go to the Poughkeepsie Day School from the third grade to the eighth

grade, graduating in 1952. My luck continuing, I was accepted at,

arid managed, despite many disciplinary problems, to stay in,

Phillips Academy, Andover Massachusetts, where I graduated in 1956.

As I entered Yale in the fall of that year, intending

to major in mathematics, I stumbled by chance into a brilliant

and fascinating introductory course in linguistics -- which I had

never heard of -- a course taught by the late Bernard After

I had failed out of mathematics, he allowed me to piece together

a special undergraduate major in linguistics, and became my adviser.

It is to his understanding, humor, and patience that I owe the fact

that I am now a linguist.

After graduating from Yale in 1960, I received a grant

from the Deutscher Akademische Austauschdienst and went for two

semestersz to the University of Bonn)and for one semester to Berlin,

to the Free University and to the Technical University, where I

studied a little linguistics and a lot of everything else.

Having returned to the United States, I received a

Woodrow Wilson Fellowship to study at the University of Pennsylvania,

 501

where Zellig Harris put me forever profoundly in his debt by

introducing me to the fascinatingly complex realm of syntax. Under

his tutelage, I wrote a Master's Thesis entitled "A P:rtial Grammar

of English Superlatives", receiving the degree in MAy 1964.

Since January 1964, I have been a student at the

Massachusetts Institute of Technology, where 1 nave had the privilege

of studying with Foam Chomsky, Morris Halle, .:oman Jakobson, Paul

Kiparsky, Edward Klima, G. Hubert Matthews, aid Paul Postal.

 4ti A

.1.1441,0(
eP111r 6-1k - 1 1:-/-:-.2 C/J4,2:). r
_I

, .
(7... a, '. Alt te / a )
a 4 . 3? a7P.PC

. .t-C
*.
j*2)

I, (7 RoC -fe. 4.2..3)1 I I 6ti-c06.4,4

tv

..s, .
I . o 1 21 7471 r`r is
k4 Io 4 4 111.
-V iT:-..e 1 I 1
.
4k I "C3 .t
(3.0-e) so Irc.(1.3.-/-.-1,)) etc
r /
Fe '311 3t: .3 31 re. C3 6)

i, .3 iii: riP. (.., eC 3. -.'_, ) 33?) .31C-1.

11."/I: I 1XA.' IV P 5 14.j

'1."'":"'..."'"*--....."..".""7.
1 / .:. /
. 1 (
1
1'
-
. r2
i
iI' Cf.--,:/,.. e i
'''
1'4
/ 1,";',,:%**./..;
....... -......II2,,........,........------------
p ......% ...,
I ..,
,- 1/ !NI)
(d;* .1-.Si: 1'73 c.t.q.n.. t ,o)
2 Alfa- 1. 3

(,!--),!,.,...a."!1-4 t..!* e .0, .:i /6o

: I
-----:"--:-----"
.
-1: ( 1,.
. ..!-
-:.!.i...,-.,e,I,IAN,
-., ..)
r,---..."1-17.".i-.-
I -:.
,

(ic.Di-) . 3 c1 4-1 /.1.7:'2=1

, ,
4.
1.7.1
4 S.5-

(.)
a. )

.........
62 t:j /
34. !,,o

of) 1gal 30e, it11

/7 )

3 eigi:c (i C. i.

i
al./
krICA

ftit./Hg..
-

-..,
ia
.0.4,..c.s,,,

i/1 1."1',J.:^"k

I ..

6 ,, . / 6- 71.
.
tr.

f:t
v

ir'irs....'..-ATt'.'t:
I I'

...
,,i.,...,,..,;.......,,,,......,......,...
I ejt
?",:,..I. C : VC 1,..
'II'

1
It '
Ir 1)

.....
-.I
. .r.-.0. 4

/"-C. tel.
(5 1(6)
f Cu) ..1.'`j
V' 1-

D.14:1
C(
33:14 -c
i .. 1.. ( .4:,
( - P1 ,..",:i
a

f.) " ft

7 ='
5

1 i 17 C 1 / 2^
iv!

, ,i)
,
b. 1...),I. 4.
...,...,1
,

-.3.)0) L-#
4.)

.31.0. 31 AN- ?..?o, 331

t 11
-
) .--)-

( ,.--;z)--,:i.p.3.6-)) 3/0)36-6) 465.4%.

-rle
-, /
,,
iCti.-
-- )

.1. ',....'4..; I.... ;A-0

.,,..,..........4..............................7.
1,, 7
#.?*:

I 1144),1 If 4 a(?) 1,P6-./Qoa, CA, 5 ft. 30 , 3

q- C§ 1.1,a), '10

r
 1... 1
4 AA14, j
t

. ;

(Lta 6) loa-Cr(^ C.21) ) ) 1(30

II I,
t
13(7--r6!f-./41

ff. 06.1,4.3)) (4.6 .60.1

tt S

57) f.y,. 3.1)63 ga 3uis(f

f4.LlfCsc C. . 2. ) 31.1',/:0L/1.(c).)) 11it-fP4(11,341))447#.0.1.3.s

"'''-e r

,tti.,(:).
11
,j
47 -12.e. (:-; 3 .

la ?, 130)
1,1)

13-) /4-2
3 -tr;..

/7:')
17T ar - aet)
zi-47cf.01.3,3)
)
.!
: (%1 . 3 -(1. .
/o 121 1 31 a Pi .:

Ism 4 j
.-=. 1.%
.;.

.4 .
I

::".......
A.
,
S.. 3)
)
h,-/ "

2 J-,/ --/IA I, 4.1.

1.3)% 473 oSffi
7 7 /
.1.
r)

) ti?; i;z3
1.1
I 2. , 1,/

; .
1:".. ,,,,

/ 3p.),,
,A) _3%4
rn 447 14.50
)

,- ...),/f/AP-1,4?)1s.,A%;-
-.am* .i. i.; , id (i :./;:)r Ricp-,--.. a& ../ 6 ) 1
4.-53 3 -!0 ( 5.1?..Q . .

0 ) 469- 1)0
335- L3 » `-) ,34.1....) 341 JP-e-1 14
L,477
1%. 3 I Li 0- 92

.,..... "1.
,.
, -I .
,f
JJ p 1.
11.1 -
v.) e..1'1.1.)
1
,

LI ,. S4,i t'
v,,. tt .4s-7)/ )
, ,
44i)si......,.-,',0. ) ....,,,,
4 .3 i ...{..-- _., (4.1 i.
I....1; a cA4i 1f,A 71
. %
) }
)
ILO! 7? I LpC7-
) .. I :.fr..---
,) 1.41) r.e ( i t; .-1
t . 41 .
1 . ..1 ,,,.,) )i1' '1'.)/ ... ,','', t :it.)i
1 4t. 1
.4". 2.,2 t 1- 2- 3/
) ) ":2 )

-
CI
i.:"
Q Tit I 2
to
1 C.1t. 8e(1( be 4,

3/P t/i'd trici

3
0 ;Is.) ".°
L... :-.;
p ad -0) 76) tivc

(t.iti3)
, .....amerta
" *0 : .

)k)
/,n
'1
61.3,a,Q) 3q0 117CA-
3 ele
ofts

1
; / ri 1 t ( sl.sa) C_._, .J .
)
» )1 33 6T3'tf-
G c
(2...
-
1 if `'. . 0 \ :7 2,2 C

1,L.a.4)) ye )44
" '''..."- ...1.
`a I .--, :.:...) !

a
, h,,17
6..3.3.3-- 6.3.3.0 )
-:". .y
ter t,
I \-1
s. -.
a
bil4,..../i
i
t f /4:1 ? °O)S g
ftrom. 6i e
.,. if ! , if lit13 c4). (f:c.).ci....-)..)
,1. 130.

it 41J / -1-1. . it. a)

''11. -..........*
' V !.. 4.

dit /; ,
,qt.ft.,24.:e))gratif at-az s
(5f6; 13)

2 Lit 6) .tr, V if4 \

') ) ?; e:e. 1'4"
114. 5%