ESSAYS TOWARD

REALISTIC SYNTAX
LINGUISTICS RESEARCH
MONOGRAPH SERIES

Volume 1
Base Generated Syntax, M. K. Brame

Volume 2
Essays Toward Realistic Syntax, M. K. Brame
 ESSAYS
TOWARD
REALISTIC SYNTAX

M. K. BRAME

NOIT AMROFER SEATTLE

Noit Amrofer Publishing Co.
P. O. Box 15176
Seattle, Washington 98115

© 1979 by Michael K. Brame

All Rights Reserved

Library of Congress Card Number 79-67347

ISBN 0-932998-01-1
For my plant
 CONTENTS

Introduction 9

Realistic Grammar 19

Chomsky/Lasnik Filters
Are Special Cases of Functional Deviance 67

The Opacity Condition,

The Nominative Island Condition,
And the Myth of Control 111

Quantifiers,
Reciprocals,
And Raising 131

The Base Hypothesis

And the Spelling Prohibition 151

Alternatives to the Tensed S

And Specified Subject Conditions 183
 INTRODUCTION

More than two decades ago, Chomsky made the following remarks at the
Third Texas Conference on Problems of Linguistic Analysis in English.

Transformational rules are, of course, an important part of traditional

grammar. E.g., O. Jespersen argues, on what we will reconstruct as
transformational grounds, that “the doctor’s arrival” is different in
structure from “the man’s house,” despite superficial similarity, be¬
cause of its relation to the sentence “the doctor arrives.” This obser¬
vation, which, I think, is entirely correct is criticized by E. A. Nida
... as a “serious distortion and complication of the formal and
functional values . . .”

A decade later, in “Remarks on Nominalization,” we find that the “super¬

ficial similarity” between the doctor’s arrival and the man’s house might
not be so superficial after all. Poor Jespersen’s “entirely correct” trans¬
formational analysis can now be reconstructed according to the latest line,
presumably. At any rate, the move to the surface had begun.
The essays collected in this volume are ripples set in motion by the
stone cast into the quiet pool of transformational grammar of the ’60s.
They are, in some measure, a reaction to work which fails to carry the
move to the surface to its logical limit and to recognize the lexicon as the
locus of syntactic distributional generalizations. Seen in broader perspec¬
tive, they are a natural outgrowth of Chomsky’s flirtation with the lexicon
in Aspects of the Theory of Syntax and “Remarks on Nominalization”.
They are my own stepping stones to a vantage point which those who
cling to transformations and conditions and constraints and filters will
continue to resist.
My first negative reaction to transformational grammar was inspired by
a reading of an introduction to syntax by Roberts in 1964. Roberts, having
learned of transformational grammar, wrote a programmed course in
which he provided an orthodox approach to infinitival complements. This
analysis struck me as somewhat contrived at the time. My reaction, then,
as it is now, was one of discomfort and reluctance to accept the alleged
motivation for the abstract analysis.
9
10 Realistic Syntax

Upon entering MIT as a graduate student in 1966, I, like so many

others, was subjected to the party line on abstract subjects and Equi.
Although I soon fell into the rhythm of MIT factory life, and even came
under the influence of generative semanticists, who taught me introduc¬
tory syntax, I continued to harbour lingering doubts about the whole
question of Equi and abstract subjects. But preoccupation with phonology
and teaching did not allow time to explore my heretical ideas, so it was not
until 1972, in preparation for my Fulbright-Hays lectures in Holland, that
I began to reassess transformational syntax with some systematicity. My
thoughts of the period are recorded as Part II of Conjectures and Refuta¬
tions in Syntax and Semantics (hereafter CRSS).
Besides the work of Chomsky, mentioned above, I can clearly discern
three related, though quite distinct, influences on my thinking. Emonds,
more than anyone else, provided the most convincing demonstration of
the importance of phrase structure at a time when transformations were
the focus of attention. Rosenbaum’s dissertation had served as an impetus
for work on complex sentences in English, and as a consequence, up until
Emonds, transformations were the rage. Although I later took issue with
Emonds’ structure-preserving hypothesis, cf. CRSS, I still feel that his
work, which culminated in his dissertation Root and Structure-Preserving
Transformations, had an enormous influence on much subsequent work;
this influence is seldom acknowledged today, nor is it common knowledge
that Emonds adopted as early as 1968 an analysis of passives as base
generated structures and suggested an analysis of infinitival complements
without abstract subjects, cf. CRSS, p. 144, fn. 7. But influences were
such that these analyses, so far ahead of their time (still unrecognized by
trace theorists), were abandoned in his dissertation proper.
A second influence derived from my own work in phonology. I had
come to the conclusion during the course of my work on Maltese in ’68
that subject pronoun prefixes must be provided in the lexicon and that
these prefixes should be combined by lexical rules with their associated
stems prior to lexical insertion. How else could it be explained that sub¬
ject pronouns were analyzed in terms of the initial cycle to account for the
interaction of stress and segmental processes? Publication of my work on
Maltese was delayed until I had the opportunity to confirm the data with
native speakers in Malta during the summer of 1970. My work on Maltese
finally appeared in Contributions to Generative Phonology in ’72. In the
published paper, evidence is provided for treating the subject pronoun
prefixes as analyzed during the course of the initial cycle; however, the
lexical rules of combination which juxtapose the prefixes and stems were
not provided. In fact, it was not until ’78 that I returned to this topic and
realized the full implications of this omission—that phonological rules,
 Introduction 11

like transformations, really should not alter words, stems, or affixes, but
should instead serve as instructions for the proper combination of the full
range of phonetic alternants, or alternatively, as filters on their proper
juxtaposition. This work was cited as “Lexical Phonology” in Yildirim’s
dissertation, A Functional Interpretive Approach to Turkish, although the
paper was completed only recently. Yildirim’s dissertation and Guerssel’s
work are much in the spirit of this conception, although both have con¬
tributed their own important insights, analyses, and proposals.
The realization that phonological rules should not alter words, men¬
tioned above, can be viewed as a generalization of the Spelling Prohibi¬
tion, which was proposed in “The Base Hypothesis and the Spelling
Prohibition”. This article appeared in Linguistic Analysis. It is reprinted
here as the second to last essay.
A further influence of my work in phonology on my thinking in syntax
concerns research on Arabic phonology. I had discovered an extremely
complex set of relations between the so-called weak stems of Arabic,
including hollow stems, doubled stems, and what I called lame stems. The
generalization was not adequately expressed in my dissertation on Arabic
phonology, but I soon discovered a means of capturing the generalization
after submission, defence, and acceptance of the original. Part of a semi¬
nar at the University of Washington during the spring of 1971 was devoted
to this topic, and I also lectured on the same material in Utrecht in ’73-
’74. Expressing the generalization entailed positing an intermediate repre¬
sentation as the optimal underlying representation, together with two sets
of rules, one mapping the optimal representations onto superficial repre¬
sentations and one mapping them onto abstract representations, a view
not incompatible with traditional phonemics of the structuralists, though
differing in many respects. This work was delivered with handouts as
“Interpretive Phonology” at C. L. Baker’s conference at the University
of Texas during the spring of’72. It was never published; related, though
somewhat different, ideas have subsequently been developed quite inde¬
pendently by Eliasson in Sweden and Leben and Robinson at Stanford.
Around 1978,1 was to see that the generalization is best expressed within
a framework of lexical phonology. If one compares the outlines of the
analysis presented as “Interpretive Phonology”, I think one will see its
relation to the syntactic analyses provided in CRSS.
In addition to the aforementioned influences, I should not fail to men¬
tion Bresnan’s sparkling paper “Sentence Stress and Syntactic Transfor¬
mations”, published in Language in 1971. Although this paper dealt for
the most part with stress and related phenomena, it was the almost casual
remarks on VP which attracted my attention and rekindled my interest in
this topic. In her paper, Bresnan argued that Object Shift (=Tough
12 Realistic Syntax

Movement) constructions, which I have subsequently called scopal predi¬

cate constructions, should be analyzed in terms of VP. It is wrong to
assume an abstract S or S with concomitant abstract subject. Upon read¬
ing this paper, memories of my dissatisfaction with Equi were awakened,
and I immediately saw that the VP-hypothesis should be extended to all
infinitival constructions. This step, in turn, had serious consequences for
transformational grammar and, therefore, it is understandable that those
bent on retaining transformations should ignore these consequences and
fail to explore the uncharted terrain. The consequences were recorded in
my “A Reanalysis of a Class of Transformations,” cited in my “On the
Abstractness of Syntactic Structure: The VP Controversy”, the latter
appearing in Linguistic Analysis, several years after having been solicited,
and then rejected, for the Benveniste festschrift. The former article I sent
to Bresnan in '14 and I believe it had some influence on her later thinking.
It was ultimately published as Chapter 5 of CRSS under a different title.
While in Holland I attempted to formulate a complete range of un¬
bounded transformations as lexical redundancy rules; however, I aban¬
doned this attempt, mistakenly it seems, for stupid reasons, adopting
instead an “inverse” treatment, as laid out in CRSS. Although the
analyses provided in CRSS are breathlessly quick and lacking in detail, I
think one can discern, nevertheless, a series of steps in the right direction,
bridged by later work. One can, for example, find an attempt to generalize
a range of unbounded constructions as a single rule of COMP-Insertion
and a discussion of the generalized filtering problem. This work antedates
Chomsky’s own attempt at generalizing the full range of unbounded trans¬
formations. It was sent to him in 1974 and he was kind enough to provide
extensive comments. I have not failed to notice, however, that Chomsky
has repeatedly cited my critique of generative semantics favorably while
consistently omitting discussion of Part II, which has implications for his
own theoretical perspective.
During my tenure at Utrecht, Chomsky’s “Conditions on Transforma¬
tions” was a central topic of discussion by the local inmates of the In-
stituut. Their enthusiasm I failed to share, for I disagreed with the basic
assumptions underlying this work. In the article, Chomsky referred to
Bresnan’s “Sentence Stress and Syntactic Transformations”, noting that
“her proposal is to take the bracketed expressions as VPs, an approach
which ... has far-reaching effects on the formulation of many rules” [p.
265]. How could such far-reaching consequences be dismissed without
serious discussion? It was clear to me that the consequences were even
more serious than intimated, since an extended VP-hypothesis gave the
lie to a full range of transformations, and, hence, a fortiori, to the condi¬
tions on transformations that Chomsky had proposed.
 Introduction 13

My disapproval of this approach was given expression in “Alternatives

to the Tensed S and Specified Subject Conditions”, which was delivered
during the course of a seminar at the University of Washington during the
fall of 1975. It was published in Linguistics and Philosophy two years later
and brings up the rear of the selection offered in this volume. Although
significant progress has been made since the appearance of this article, I
think a good deal of the criticism leveled against the constraints approach
ring true today.
Chomsky and Lasnik pursued the theory which had originated in
“Conditions on Transformations” in their article “Filters and Control”,
which appeared in Linguistic Inquiry. Here they extended the theory to
include a wide range of ad hoc filters with no appreciable formal analysis
of control. My response to this work is provided in the second essay of
this collection, “Chomsky/Lasnik Filters Are Special Cases of Functional
Deviance”. A fragment of this paper was delivered at the Sloan Work¬
shop at Stanford in January of this year (1979) with Chomsky and Lasnik
in attendance. A hint of Chomsky’s response to my presentation is pro¬
vided in the initial essay of this collection, “Realistic Grammar”, which
was delivered at the Conference on Current Approaches to Syntax in
Milwaukee, Wisconsin during March of this year. This paper I originally
submitted to Linguistic Inquiry where it was rejected.
One of the papers offered here deals with Chomsky’s latest paper “On
Binding”. It shows, I believe, that his latest revisions and refinements
of trace theory offer no real improvement over his original conditions. All
lead to significant loss of generalization and suffer from ajailure to recog¬
nize the integrity of lexical items. Here, for perhaps the first time, one
begins to perceive an affinity of his work with that of generative seman¬
tics. Abstract case is introduced without a single explicit provision for the
actual treatment of lexical items. The formal account of the actual content
of words is taken to be mere detail, a trait of generative semantics in its
heyday, cf. Part I of CRSS. As time now runs out on trace theory, one
sees ever more far-fetched devices proposed to accommodate coun¬
terexamples that genuinely follow from more realistic approaches. Just as
generative semanticists were inspired to propose global rules and other
prophylactic devices to immunize their theory against refutation, so also
trace theorists have begun to follow suit by adopting theoretical con¬
structs which are seldom made explicit. It is now impossible to evaluate
claims relating to the mystical level of logical form, for lack of explicit¬
ness, and Chomsky more recently has even proposed a generative
semantics-like analysis of examples such as John struck me as like Harry
and John regards me as like Harry, cf. “On Binding”.
The articles collected here are, with two exceptions, compiled in re¬
verse chronological order. The piece critical of “On Binding” was
14 Realistic Syntax

presented during the spring quarter of this year at a seminar at the Univer¬
sity of Washington. It seemed more appropriate following the articles
“Realistic Grammar” and “Chomsky/Lasnik Filters Are Special Cases of
Functional Deviance”, although the latter antedate it. The paper entitled
“Quantifiers, Reciprocals, and Raising” was actually composed prior to
the final two essays and is cited in the concluding paper. But since this
paper has not previously appeared in print, it has been placed out of
chronological sequence, before the two essays which are reprinted here.
My original intention was to extensively rework this article, but too much
time has elapsed and my interest has waned; there are a number of points
in this paper, nevertheless, which are still worthy of consideration; so I
include it here as the third to last essay.
Lest I convey the impression that I consider the approach offered here
as anything definitive, let me hasten to emphasize that ideas continue to
evolve; as always, there are many alternative ways of formulating bind¬
ing, anaphora, and other processes of interest in these papers. Whatever
the final outcome of the precise form which the best theory will assume in
the future, I am laying odds that it will be a theory which accords with that
presented here in at least the following respects.

i. Abstract subjects or PRO (or [e]NP, etc.) along with the

accompanying abstract S or S will play no role. Infinitival
complements will be analyzed as infinitival complements,
i.e. some version of VP.

ii. Transformations in the sense of the standard theory, ex¬

tended standard theory, or trace theory will play no role.
There will come a time when linguists agree that transfor¬
mations do not exist.

iii. Conditions, constraints, metaconditions, or

metaconstraints such as the Specified Subject Condition,
Propositional Island Constraint, Opacity Condition, Tensed
S Condition, Nominal Island Constraint, Subjacency, A/A
Principle, Relativized A/A, Recoverability of Deletion, etc.
will play absolutely no role. There will come a time when
linguists agree that the effects of such conditions follow
from more lexically based theories.

iv. Lexical items will not be altered by transformations, nor,

even by phonological rules. There will come a time when
linguists agree that words are not to be abused by case
marking rules and other ad hoc processes.
 Introduction 15

These pronouncements may seem bold at this stage of our understanding,

but these are the objectives toward which my own research has been
channeled in recent years. Of interest for serious minded syntacticians
will be a comparison of alternative theories which do not incorporate
transformations, while satisfying the desiderata of (i)-(iv). Perhaps these
essays can provide some stimulus towards this end, an interesting and
intriguing prospect for the ’80s.
■
Sensei, do these facts really follow?

John, who I wanted to see and who I did see, is my friend.

*John, who I wanted to see and who I did, is my friend.
They say John is eating beans, which he is.

John is eating beans.

Who is?
*What is he?

Yes, gakusei, read on.

Introductory Remarks

1. Linguistic Problems
1.1 Binding
1.2 The Multiple Ambiguity Problem
1.3 The Problem of Generalizing Anaphora
1.4 Interaction of Binding and Ellipsis
2. A Sketch of the Theory

3. Tentative Solutions
3.1 Binding
3.2 VP-Ellipsis
3.3 Pronominalization and Ellipsis as Anaphora
3.4 Binding and Anaphora
4. Conclusion
4.1 Apparent Arguments Against Pure Interpretive Theories
4.2 Nonexistence of Empty Nodes and Traces
4.3 The Missing Argument Phenomenon
4.4 Chomsky/Lasnik Filters
4.5 Nonexistence of Ordering
4.6 New Frontiers: Realistic Phonology
4.7 Linguistic Problems
Notes

References
 REALISTIC GRAMMAR*

Discussion of goals of syntax and linguistic theory will probably never

amount to much. I have yet to see anything profound emerge from de¬
bates over introspection vs. techniques of elicitation, competence vs.
performance, mentalism vs. antimentalism, and other high-level
theoretical oppositions. My perspective is to let such topics lie and to turn
to the really exciting domain of activity which I will characterize as lin¬
guistic problems. My view is that the ultimate fate of any theory of natural
language will be determined not so much by metatheoretical arguments
for it as by its ability to deal with the significant linguistic problems in a
revealing manner, by its success in bringing together apparently disparate
sets of facts and in providing explanations for apparent irregularities.
There are a number of linguistic problems which have crystallized over
the past decade or so of research within the framework of transforma¬
tional grammar and it is to these problems that theoretically minded lin¬
guists should be devoting their energies.
In harmony with this view, I therefore wish to focus on several of these
problems in my talk. I consider these problems to be central to the evalua¬
tion of any theory. They must ultimately be confronted regardless of
theoretical bias.
The plan of my talk is the following: First, I will lay out four problems
which I have chosen as representative of significant linguistic problems.
Next I will provide a sketch of the theory I wish to advocate. Then I will
offer solutions to these problems within the proposed framework. Finally,
I wish to conclude by contrasting the proposed theory with trace theory.

1. Linguistic problems

What I consider to be the most significant problems of linguistic theory

are outlined and developed in some detail in Brame (to appear a). In what

*This paper was presented at the Conference on Current Approaches to Syntax, Milwaukee,
Wisconsin in March 1979.

19
20 Realistic Syntax

follows, I have selected four of these problems as central to the syntax of

English.

1.1 Binding

The question of binding is one which provides a challenge to any lin-

gustic theory. Syntactic structures in English that exhibit binding include
questions, relative clauses, focus constructions, topic structures, and
comparative clauses, as illustrated below.

(1) a. What did Mary put into the garage?

b. the car which Mary put into the garage
c. It was the car that Mary put into the garage.
d. This car Mary put into the garage.
e. more cars than Mary put into the garage

A number of properties seem to suggest that these different construc¬

tion types should be treated in a general way. For example, as is well-
known, all exhibit a kind of deviance when two such constructions inter¬
act in a crucial way. Relevant examples illustrating this deviance in¬
clude the following.

(2) X-question

a. * Where did she ask which car Bill put_? question-question

b. *the girl who John asked the boy who hit_ rel-question
c. *It was in the garage that she asked which car Bill put
-foc-question
d. *That car she asked where Bill put__top-question
e. *John drank fewer beers than Mary asked who ate _
apples, com-question
cf. John drank fewer beers than Mary ate_apples.

(3) X-rel

a. *Which car did she see the boy who put _ into the
garage? question-rel
b. *the boy who Mary likes the girl who hit_ rel-rel
c. "It was this car that she saw the girl who drove_foc-rel
d. *This car she saw the girl who drove top-re I
e. "John drank fewer beers than the boy who ate apples
left, com-rel
 Realistic Grammar 21

(4) X-foc

a. *Which car was it in the garage that Mary put_? question-

foe
b. *the car which it was in the garage that Mary put_rel-foc
c. *It was the car that it was in the garage that Mary put
-foc-foc
d. *This car it was in the garage that Mary put_top-foe
e. *John drank fewer beers than it was Mary that ate _
apples, com-foe

(5) X-top

a. * Where did this car John put_ __? question-top

b. *the garage into which this car John put_rel-top
c. *It was into this garage that this car John put_foe-top
d. *Into this garage this car John put_top-top
e. ^John rode fewer bikes than into the garage Mary put_cars
_com-top

(6) X-com

a. *Who does John like Bill more than Sue likes_? question-
corn
b. *the boy who John likes Bill more than Sue likes_rel-com
c. *This boy John likes Bill more than Sue likes_top-com
d. *It was this boy that John likes Bill more than Sue likes
_foc-com
e. *1 have more apples than John saw fewer girls than_boys eat
_can-com

Let us call the property which distinguishes all of these examples, the
accessible scope property. It is clear that it is not the accessible scope
property alone that accounts for the deviance of at least some of these
examples. For example, there seem to be independent reasons for the
deviance of the X-top examples, since topic constructions are not nor¬
mally permitted in embedded contexts, cf. Emonds (1976).

(7) a. *1 realize that this car John put_into the garage.

b. *Mary always says this car she puts_into the garage.

Nevertheless, the accessible scope property will be crucial to a satisfactory

account of the majority of examples listed in (2)-(6).
 22 Realistic Syntax

A second property which is invariant across the full range of construc¬

tions adduced above is a property which I have elsewhere called the
filtering property. The filtering property is illustrated for the full range of
constructions below.

(8) a. *Who(m) did John see Mary? question

b. *the boy who(m) John saw Mary rel
c. *It was John that Harry saw Mary. foe
d. *This hat Mary saw John. top
e. *more boys than the girls com

The fact that all of the above-cited constructions exhibit both the acces¬
sible scope property and the filtering property suggests that we are dealing
here with a single phenomenon, which I will term binding. Several propo¬
sals have been advanced to express the underlying generalization and
these will be touched on below in section 3.
The problem of binding, then, is to provide a general characterization of
the accessible scope property and the filtering property while at the same
time providing an analysis which generalizes across construction types. In
addition, the analysis must be sufficiently general to cut across category
types, since categories other than NP are often involved. Thus, in addi¬
tion to NP-like heads, we encounter PP-like heads in the case of relative
clauses, questions, focus constructions, and topic structures.1

(8) a. Into which bottle did Mary put the coins? question
b. the bottle into which Mary put the coins rel
c. It was into this bottle that Mary put the coins . foe
d. Into this bottle Mary put the coins. top

We also find sentential heads, as illustrated in the following paradigm

(9) a. What did Mary claim? question

b. Mary left, which was surprising. rel
c. It was simply that John failed that Bill thought
had frightened Harry. foe
d. That Mary left I believe surprised Harry. top

Other categories as well figure in the binding framework and the complete
range of permissible argument types should be subsumed under one gen¬
eral analysis. A general treatment is once again suggested by the fact that
 Realistic Grammar 23

we encounter the accessible scope property and the filtering property in

connection with such examples. For example, in connection with the
sentential case, we find the following deviant examples.2

(10) a. *That Mary left Bill saw the boy who believes _ surprised
Harry.
b. *That Mary left Bill believes that Sue is smart surprised Harry.

In summary, the problem of binding is that of presenting an analysis

meeting the following desiderata.

i. The analysis must generalize across construction types, e.g.

relative clauses, questions, focus constructions, topic struc¬
tures, comparative clauses in English.
ii. The analysis must generalize across category types where rele¬
vant, e.g. NP, PP, etc. in English.
iii. The analysis must account for the accessible scope property in a
general way.
iv. The analysis must account for the filtering property in a general
way.

It should be noted, in passing from this topic, that not all languages treat
each of the aforementioned construction types as in instance of binding.
Thus, languages exhibiting question particles typically do not reflect the
binding property. For example, in Japanese the question is typically
marked with the particle ka, not only in the case of yes/no questions, but
also in the case of wh-questions.

(11) a. Mary-ga sakana-o tot-ta

Mary subj. fish obj. catch past
Mary caught the fish.
b. Mary-ga sakana-o tot-ta ka?
Did Mary catch the fish?
(12) a. dare-ga sakana-o tot-ta ka?
who subj. fish obj. catch past Q
Who caught the fish?

If Japanese questions are not instances of binding, we should not expect

to encounter the accessible scope property when binding constructions
interact with questions. Apparent binding constructions in Japanese in¬
clude the relative clause, topic construction and focus structure, which
are illustrated below.
 24 Realistic Syntax

(13) a. Mary-ga tot-ta sakana rel

Mary subj catch past fish
the fish that Mary caught
b. Sakana-wa Mary-ga tot-ta. top
fish top Mary subj. catch past
The fish Mary caught,
c. Mary-ga tot-ta no wa sakana da. foe
Mary subj catch past nom fish is
It was the fish that Mary caught.

There is no related accessible scope violation when questions interact

with relative clauses, focus constructions, and topic structures, as illus¬
trated below.

(14) a. Mary-ga dare-ga tot-ta ka to tazuneta sakana rel-question

Mary subj who subj catch past Q comp asked fish
*the fish that Mary asked who caught
b. Sono sakana-wa Mary-ga dare-ga tot-ta ka to tazuneta top-question
that fish top Mary subj who subj catch past Q comp asked
*That fish Mary asked who caught.
c. Mary-ga dare-ga tot-ta ka to tazuneta no wa sono sakana da foc-ques
Mary subj who subj catch past Q comp asked nom top that fish is
*It was that fish that Mary asked who caught.

Other interactions, however, do yield the accessible scope property of

deviance.

(15) a. *Mary-wa tot-ta sakana. top-rel

Mary the fish that_caught
b. *Mary-wa tot-ta no wa sakana da. top-foe
Mary, it was the fish that_caught_

It is interesting to note that there are examples illustrating the filtering

property which correspond to the apparent binding constructions, al¬
though there is no such example corresponding to the question in
Japanese.
 Realistic Grammar 25

(16) a. *Mary-ga booru-o tot-ta sakana

*the fish that Mary caught the ball
b. *Sakana-wa Mary-ga booru-o tot-ta. top
*The fish Mary caught the ball.
c. *Mary-ga booru-o tot-ta no wa sakana da. foe
*It was the fish that Mary caught the ball.

Perhaps it is a property of all binding constructions that they exhibit

apparent “holes” in their syntax. Thus, for example, when we compare
the relative clause Maiy-ga tot-ta with the sentence Mary-ga sakana-o tot-ta,
we see that the former exhibits a hole, just as in English we find a hole in
the relative clause who John saw. The same is true of the focus construc¬
tion in Japanese, e.g. (13c). It does not hold for the question in Japanese,
however, as contrasted with English.
If it is true that a structure must exhibit a hole to qualify as a binding
construction, we might then expect that constructions exhibiting the re¬
sumptive pronoun phenomenon do not involve binding. There are many
languages that include such constructions. Many apparently do not ex¬
hibit the accessible scope property, indicating that we are not dealing with
binding, but many apparently do. Clearly, binding across languages is a
topic which invites detailed research; it would be premature to simply
write off the resumptive pronoun constructions as instances of “predica¬
tion” or some other phenomenon. At the same time, it is not clear that
they are genuine instances of binding. For some discussion, cf. Brame
(1979a).

1.2 The Multiple Ambiguity Problem

Examples such as the following have been the focus of a good deal of
activity within the transformational camp, cf. Wasow (1972) and Sag
(1977).

(17) a. Mary has eaten kabsa and Bill has too.

b. John is playing the oud and Hamza is too.

There is an interesting puzzle related to such constructions. Consider first

the fact that the following examples are ambiguous.

(18) a. They are flying planes.

b. This chicken is too hot to eat.
c. She sees Harry more often than Sally.
d. Tom looked up the pipe.
 26 Realistic Syntax

The ambiguity can be elucidated in terms of the following examples.

(19) a. They are flying these planes,

b. They are those flying planes.
(20) a. This chicken is too hot for me to eat.
b. This chicken is too hot to eat anything.
(21) a. She sees Harry more often than she sees Sally,
b. She sees Harry more often than Sally sees him.
(22) a. Tom looked the pipe up in the catalogue,
b. Tom looked way up the pipe.

Now a well-known puzzle arises in connection with the following exam¬

ples.

(23) a. These are flying planes and those are too.

b. This chicken is too hot to eat and that one is too.
c. She sees Harry more often than Sally and Mary does too.
d. Tom looked up the pipe and Sue did too.

Since the examples in the left conjunct are 2-ways ambiguous, one might
have expected the resulting ellipsis cases to be 4-ways ambiguous, corre¬
sponding to the following four examples, taking (18a) as representative.

(24) a. These are [flying planes]NP and those are [flying planes]NP too.
b. These are [flying planes]Vp and those are [flying planes]Vp too.
c. These are [flying planes]NP and those are [flying planes]Vp too.
d. These are [flying planes]Vp and those are [flying planes]NP too.

However, contrary to expectations, the examples in (23) are not 4-ways

ambiguous; they are only 2-ways ambiguous. They exhibit only the paral¬
lel interpretations. Thus, again taking (18a) as representative, we see that
it can have readings analogous to (24a) and (24b), but not to (24c) and
(24d).
The multiple ambiguity problem [MAP] is that of accounting for the fact
that examples such as those listed in (23) are not 4-ways ambiguous. The
familiar transformational answer to this riddle is based on the notion of
identity. If string identity is associated with deletion rules, then we would
expect 4-way ambiguity under the assumption that the relevant examples
are derived by means of a deletion rule. Because 4-way ambiguity is not
observed, linguists have proposed stronger identity requirements and
 Realistic Grammar 27

some have even gone so far as to propose global identity, cf. Chomsky
(1968).
Any solution to MAP must be sufficiently general so as to extend to
discourse. This conclusion follows from the observation that the same
problem arises in connection with discourse, as illustrated below.

(25) a. These are flying planes,

b. Those are too.

It is clear that this discourse is 2-ways ambiguous, not 4-ways ambiguous.

Again, only the parallel interpretations are possible. A theory which is not
sufficiently general to include coarser discourse relations will not succeed
in providing a satisfying solution to MAP.

1.3 The Problem of Generalizing Anaphora [PGA]

For some time, it has been known that there is a generalization cutting
across pronouns and ellipsis. Consider first the well-known paradigms
involving pronouns.

(26) a. I saw Bill after Mary pointed him out to me.

b. John thought Mary was tall when he saw her.
(27) a. *1 saw him after Mary pointed Bill out to me.
b. *John thought she was tall when he saw Mary.
(28) a. After Mary pointed him out to me, I saw Bill.
b. When John saw her, he thought Mary was tall

Example (27a) and example (27b) show that two NP’s cannot be coin¬
dexed when the second is not a pronoun and the first commands the
second. We find a similar state of affairs with ellipsis of the type discussed
above in subsection 1.2.

(29) a. John has certainly tasted the beans if Bill has.

b. Sally can’t leave if Bill can’t.
(30) a. *John has if Bill has tasted the beans.
b. *Sally can’t if Bill can’t leave.
(31) a. If Bill has, then John has probably tasted the beans too.
b. If Bill can’t, then Sally can’t leave either.
28 Realistic Syntax

Examples (30a) and (30b) are deviant under the intended interpretation,
just as (27a) and (27b) are deviant under the intended interpretation.
(None of the examples is deviant, however, under a different interpreta¬
tion where requisite antecedents are supplied.) Similar conditions hold
between the verbal elements of (30) as hold between the pronouns of (27)
with respect to command, precedence, and fully specified anaphor. Hence
the deviance of the examples in both (27) and (30) should be expressed in a
unified way.
If there is indeed a generalization to be expressed here, we might expect
to find a similar distribution of data in discourse. This seems to be true.
For example, the following parallelism seems to hold.

(32) a. John saw Bill,

b. I saw him.
(33) a. John has eaten beans,
b. I have too.
(34) a. John saw him.
b. I saw Bill too.
(35) a. John has.
b. I have eaten beans too.

In (32) and (33) we find that anaphora is well-formed in both cases.

In (34) and (35) we find that anaphora is not possible. This suggests that a
very similar phenomenon is being observed.3
The problem of anaphora, then, is to generalize pronouns and ellipsis so
as to express the parallelism observed in terms of intrasentence and dis¬
course relations.

1.4 Interaction of Binding and Ellipsis

The two processes discussed under subsection 1.1 and 1.3 converge to
provide an extremely interesting problem. Consider the following dis-
courses.

(36) a. John is eating beans.

b. Who is?
(37) a. John is eating beans.
b. *What is he?

In (36) we see that the subject of (a) can be questioned and that ellipsis is
simultaneously permitted. By contrast, this is not so in (37) when the
 Realistic Grammar 29

object is questioned. These observations hold, of course, for other forms

of the auxiliary, including have, modals, do, and infinitival to. What un¬
derlies this distinction? The interaction problem is that of providing an
analysis which yields insight into this question.
There are many other puzzles of interaction, cf. the examples on the
main title page. Several of these will be discussed below in section 3.4.

2. A SKETCH OF THE THEORY

The theory which I have in mind is one which is currently in the process
of being developed. The spirit of the approach is shared I believe, by a
number of colleagues, although I hasten to add that their emphasis and
formalization may well differ from my own in important respects.4
The basic idea behind some of the previous work is that structures
which were previously derived by means of transformations should in fact
be base generated. This point of view was adopted in Brame (1976),
although a set of inverse transformations was posited to take over the
work of the classical transformations. In Bresnan (1978) the base
hypothesis is adopted together with a more interpretively based account
of several alternations which were previously treated by transformations,
in particular, the passive. Bresnan introduces functional structures as
components of lexical specifications in conjunction with operations on
functional structures. She assumes that the passive construction is to be
treated by a universal operation on functional structure. In Brame (1978a,
1978b), the functional framework is elaborated and refined and pursued in
terms of binding structures. In Yildirim (1978) an investigation of causa¬
tive constructions is undertaken.
In what follows, I would like to take what I consider to be a natural next
step. Let us eliminate the phrase structure rules in favor of a still more
surface approach. We will then provide a set of rules of composition in
accord with the suggestion in Brame (1978a, p. 122, fn. 17). We will then
be working directly off the surface strings of lexical items, or more accu¬
rately, of phonological words. Within such a theory, it is no longer possi¬
ble to simply gloss over lexical specifications of the words that play a role
in the analysis. Rather, lexical specifications must be provided to ensure
the proper functioning of the grammar as a whole.
In this spirit, partial lexical specifications corresponding to John, man,
you, the, met, and left are provided below.

(38) John; An[x]

30 Realistic Syntax

(39) man; Anom[sl

(40) the; Fdet, __(Anom[xl)
(41) met; Fv,_(An[0])
(42) left; Fv, _ or __(An[0])

Instead of phrase structure rules, we define the composition function f .

A tentative and fragmentary formulation is provided below.

(43) a. For each phonological word a, compose a as some grammatical

function Ax or Fy.
b. Compose Fv or Fv Ax as Fvp
c. Compose Fd Ax as An(x].
d. Compose Ax Fvp as As.

In (43a), the x and y range over appropriate values for arguments and
functions. For example, arguments include nominal singular arguments
such as man in (39), np arguments such as John in (38), etc. Functions
include verbal functions such as met (which is really complex in this case)
in (41), preposition functions such as into which is Fptdir], etc. It is quite
conceivable that some of the rules in (43) can be generalized. For exam¬
ple, (43b) and (43c) are similar in that they compose head functions with
their arguments.
We now ask whether the following example is a well-formed sentence
of English. Since it obviously is, we must illustrate the means by which a
syntactic interpretation is built up.

(44) The man met you.

Applying rule (43a) to the string of phonological words, one possible

result is the following.

(45) pdet Anom[s] An[x]

I I II
"the man met you

Applying rules (43b) and (43c) to (45), we obtain the following.

 Realistic Grammar 31

(46) An[x] Fvp

the man met you

Finally, applying rule (43d) to (46), we obtain the following.

the man met you

Further rules will specify the x of the two instances of An[xl as 5 or o,

depending on whether the function is subject or object.
We now wish to check the result for well-formedness. This involves an
indexing procedure in conjunction with a convention for establishing
equivalences.

(48) a. Let T be a term dominating a lexical category a. If (a; T, ...) e

Lexicon, then assign T a distinct index i, which we may write as
T: i.
b. Let T be a term directly dominating terms Tx ... Tn and let Tt be
indexed by i. Then set T equivalent to the functional structure of
the lexical item indexed by i and substitute the indices of Tl5 T2,
..., Tn for corresponding terms.

Applying (48a) to our example, we obtain the following result.

(49)

the man met you

32 Realistic Syntax

We now apply (48b) to (49). The lexical item the corresponds to the lexical
entry (the; Fdet, _(Anom[x1))- The complete functional structure is the
function Fdet taken with its argument Anom[x], which is Fdet (Anom[x]).
Flence we set Anlsl equivalent to Fdet (Anom[sl) and substitute correspond¬
ing indices to get/(?'), which can be indicated as k, the index of An[sl. A
similar procedure applies to the Fvp function and we obtain the following
partial interpretation.

(50) as

An[s]5 Fdet(Anom[s]);f(i) = k pvp - ) :g(j ) = h

' An[°]: j
I
the met you

By a similar procedure, we set As equivalent to Fvp (An[s)) and substitute

corresponding indices to yield the complete interpretation/?^), which can
be represented as the propositional index p.
With this fragmentary grammar, we succeed in providing a functional
interpretation for the simple example provided in (44). By contrast, an
example such as (51) cannot be provided a complete interpretation.

(51) *The man met.

Incompleteness follows from the fact that met rquires an object as indi¬
cated in its lexical specification in (51).5 When we attempt to interpret
(51), we obtain the following partial interpretation.

(52)
Af_=J^(An[s]) S (FV(An[°]))(An[s])=|;g(An[o])(k)|
An[s] - Fdet(Anom[s]):f(i) „ k
I™ = Fv(An[o]):g(An[o])
Fdet:f An°m[s ] ^ Fv:g
i
the met

The absence of the required object is evident when we look at the final
interpretation structure (g(An[0])) (k), which has been boxed in (52).
To account for the deviance of such examples, the initial part of the
following principle was proposed in Brame (1978a).6
 Realistic Grammar 33

(53) Incompleteness
A string at ... an is incompletely interpreted with respect to an
interpretation <p provided there is a term T in corresponding to no
index or provided there is no index in ip corresponding to a{ for
some i, where 1 i n.

In the earlier work, the filtering device was termed the Principle of Func¬
tional Deviance. The second part of (53) will account for the deviance of
cases such as John slept the mouse and a full range of examples, as we
shall see below in section 4.

3. Tentative solutions

Partial solutions to the problems mentioned in section 1 can be provided

within the proposed framework provided we extend the theory in natural
ways. Let us turn first to the question of binding.

3.1 Binding

Consider once again the deviant example given in (51). The deviance
can be ameliorated if we append additional words to get the following.

(54) the girl who(m) the man met

Let us consider the embedded sentential structure who the man met.
Although there is no phonological word following met which could func¬
tion as the object required by its lexical specification, intuitively we wish
to interpret who as the object in this case. In fact, this can be done without
making the traditional assumption of transformationalists, i.e. without
assuming that (54) derives from a more abstract source with post-verbal
object. The relative clause of (54) can be interpreted directly provided we
posit an appropriate lexical specification for the relative pronoun who in
conjunction with a general operation on functional structure. A partial
lexical specification is provided below in (55).

(55) who; An[x], (Frel (_)) (As)

34 Realistic Syntax

Although who is taken to be an unspecified np argument, it has the ulti¬

mate character of an operator, for it is composed to yield the relative
operator Frel (An[xl) taken with its argument As. After the interpretive
procedures apply, we obtain the following partial interpretation.

(56)
(Frel(An[x]))(AS):(Frel(x))((g(An[°]))(j))

An[x]:x AS = FVP(Ants]):g(An[°])(j)

n[s]
:f(i) = F = FV(An[°]):g(An[o]
I
,nom[s] .
A :i FV:g
I
who the met

Here we obtain the partial interpretation indicated in the box to the right
of the colon in the top line of (56). It is to this functional structure that the
general rule of binding proposed in Brame (1978a) applies. This rule is
repeated below as (57).

(57) Operator Binding [OB]:

Let F{a) be an operator with index a and let (... T ...) be its
scope, where T is a term in the accessible scope of F(a) and of
the same type as that indexed by a. Then identify T with a and
index F.

The result of applying OB to the partial interpretation of (56) is illustrated

below in (58).

(58) (Frel (x)) ((g(An[°i)) 0)

(h(x)) ((g(x)) (j))

Since (56) is now completely interpreted, the principle of incompleteness

will be inapplicable. The relative clause itself can be taken as an argument
As[rel] and provided a single index/?. It itself must then be composed with
its head the girl in (54), by a rule of anaphora.
Operator Binding is a general rule. It is not formulated so as to apply
only to relative clauses. Other operators will be provided an interpretation
utilizing OB and such operators include the question operator Fq(a), the
topic operator Fm(a), the focus operator Ffoc(a), and the comparative
operators Fc[x](x). There are perhaps others as well, such as the
exclamatory operator Fex(o) and the contrastive operator Fcon(a). If
 Realistic Grammar 35

heads of syntactic constructions can be interpreted as any of these

operators under appropriate conditions, then OB will apply to bind the
appropriate arguments. This is done without any necessity for the usual
assumption (of transformational grammar) that there is an NP category
following the verb in relative clauses such as (54). The object argument is
interpreted in terms of the lexical specification of the verb met in this
case.
Not only does OB generalize across operator types, it also generalizes
across category types. Thus, if the operator suffix happens to be a prepo¬
sitional phrase argument or a sentential argument, OB will apply nonethe¬
less. This analysis succeeds to some extent, then, in satisfying criteria (i)
and (ii) of the binding problem elaborated in subsection 1.1. Points (iii)
and (iv) are also covered by this analysis. To see this, let us return to the
deviant examples provided in (2)—(6) above.
These examples violate the accessible scope condition associated with
OB. Accessible scope is defined as follows:

Def. Let 0(a) be an operator with index a and let ( ... T ...) be its scope
(i.e. its argument structure), where T is a term. If P(b) is an
operator falling in the scope of 0(a), we say that the accessible
scope of 0(a) includes all material falling in the scope of 0(a) which
does not fall in the scope of P(b) for all P(b).

The effect of incorporating accessible scope into OB is that of prohibiting

certain functional terms from being identified with indices of operator
suffixes. This in turn gives rise to incomplete interpretation by the princi¬
ple of incompleteness noted above.
Note also that the filtering property displayed in terms of (8) is also
accounted for. In the case of such examples, there is no term which could
be identified with the operator suffix. Consequently the binding rule will
not apply, from which it follows that the operator prefix in each case will
not be indexed. Hence, again, the principle of incompleteness will apply
and such examples will be declared deviant.
Above it was noted that OB generalizes across categories. However, it
is necessary that the associated lexical representations of the operator
words reflect the operator structure before an interpretation can be built
up which will give rise to binding. Some wh-words, for example, function
as determiners and require the presence of their argument structure. Such
examples include the following questions.

(59) a. Whose books did Mary read?

b. Which clepsydra did Bill use?
c. How many eggs can you eat?
 36 Realistic Syntax

A wh-word such as whose will have to be given a partial lexical specifica¬

tion such as (60)

(60) whose; Fpos, Fq(_(Anom(x])) (As[q])

Now the presence of Anom[x] in the functional structure of wh-determiners

such as whose, which, and how many will ensure that the following exam¬
ples are not completely interpreted.

(61) a. *Whose did Mary read books?

b. *Which did Bill use clepsydra?
c. *How many can you eat eggs?

To be sure, whose, which, and how many also function as np arguments,

thus providing for examples such as the following.

(62) a. Whose did Mary read?

b. Which did Bill use?
c. How many can you eat?

But this simply entails an elaboration of the lexical specifications of these

words to include the additional argument structure. For example, whose
must include in addition to the structure in (60), a representation as An[x).
In neither case, however, will the examples in (61) receive a complete
interpretation. In the latter case where whose is interpreted as An[x] we
will be left with an example characterized by the filtering property remi¬
niscent of (8). In this case, we will be left with an uninterpreted operator
prefix. In the former case, where whose is interpreted as Fpos, we will be
left with an uninterpreted argument Anom[x]. Hence both types of exam¬
ples are incompletely interpreted. Consequently examples such as those
provided in (61) are deviant.
This result follows without ad hoc theoretical elaboration. There is no
need for special constraints such as the Left Branch Condition proposed
in Ross (1967) or the Relativized A/A proposed in Bresnan (1976). Rather,
the lexicon in conjunction with general interpretive procedures plays a
central role. There is some hope that the vast array of constraints and
conditions of transformational grammar can be dispensed with.
A key to the realization that long-distance transformations should be
eliminated in favor of more general nontransformational analyses is that
questions, relative clauses, focus constructions, and others do not really
 Realistic Grammar 37

involve structural distortions which cannot be treated on the surface. A

first attempt at generalizing the full range of constructions can be found in
Brame (1976), where a general rule of COMP-Insertion was proposed
together with base generated questions, relative clauses, and comparative
clauses, with syntactic heads in situ. The generalized filtering problem
was accounted for by incorporating an inverse cyclic checking procedure.
Thus, although an example such as the boy who(m) Mary saw Bill could
be base generated within the proposed framework, after COMP-Insertion,
a violation of subcategorization resulted, viz. the boy [Mary saw who(m)
Bill] or the boy [Mary saw Bill who(m)], etc. Although this analysis is
deficient in various ways, I think it can be seen as a step in the right
direction. In particular, there is an obvious similarity here between the
subcategorization violation and incompleteness within the framework ad¬
vocated above.
A second proposal for treating the examples cited in subsection 1.1 can
be found in Chomsky (1977). Although this analysis includes much in¬
teresting discussion, in my opinion it is in some respects a step backwards
from the earlier proposal incorporating base generated structures. The
analysis clings to tradition and provides for underlying relative clauses,
questions, etc. as “complete sentences”. It incorporates an inverse var¬
iant of my earlier rule of COMP-Insertion, now called “Move wh-
phrase”. This analysis clearly misses an important generalization since a
second “interpretive” rule is required to introduce variables in the posi¬
tion of traces left behind by the movement rule, cf. Chomsky (1977, rule
(38)). The analysis also involves a number of theoretical elaborations,
including traces, special constraints such as SSC, PIC, RAOAC, special
housekeeping rules, and many many other assumptions, cf. Brame
(1978a, 1978b). Later developments include a proliferation of filters, cf.
below and Brame (1979).
A third analysis is that detailed in Brame (1978a, 1978b) in which base
generated structures are again provided together with the general treat¬
ment outlined above. The analysis provided here differs only insofar as
base rules are eliminated in favor of rules of composition.
To my knowledge, there are no convincing arguments against the “di¬
rect” approach to long-distance structures as opposed to the traditional
approach involving movement rules. In Chomsky (1977), there is a com¬
ment in footnote 16 to the effect that a direct treatment is not “a meaning¬
ful alternative to the transformational analysis as a movement rule, for
reasons discussed in Chomsky (1975c).” When we consult this reference,
however, we find no argument against the direct treatment. By contrast,
Wasow (1977) attempts to provide arguments against a direct approach to
questions and relative clauses by appealing to well-known facts about
38 Realistic Syntax

contraction. A detailed discussion of these arguments is provided in my

forthcoming book Linguistic Problems in which an analysis of contracted
elements is provided. Within the framework of this analysis contractions
are contractions and Wasow’s facts are handled in a straightforward way.
It turns out that examples such as *Where does John think the party’s,
*John has eaten more beans than Bill’s, etc. cannot be provided a com¬
plete interpretation.
Although I think there are no real arguments against a direct approach
to questions, relative clauses, and other binding constructions, it does
appear that there are arguments against the traditional indirect approach,
including its modern variant proposed in Chomsky (1977).
The most obvious argument against the indirect approach within the
framework of Chomsky (1977) has already been mentioned, namely, it
misses an important generalization. Both a movement rule is proposed to
move wh-phrases into COMP position and an interpretive rule is needed
to identify the deposited trace with the index of the COMP phrase. Cer¬
tainly this duplication is to be avoided. The obvious move is to eliminate
the movement into COMP altogether and to state matters once and for
all, as is done above in terms of Operator Binding.
A second argument against the indirect transformational approach is
noted, interestingly, in Chomsky and Lasnik (1977), where examples such
as the following are drawn from Bresnan (1977).

(63) a man who Mary called_an idiot and June called_a cretin

Chomsky and Lasnik remark: “A question arises, then, as to how the

wh -phrase can be moved simultaneously from the two positions marked
by — [in this example] to the single italicized position ...; no existing
theory permits anything of the sort” (pp. 490-1). Indeed, this is so, indi¬
cating that theories incorporating movement from the positions marked_
to the italicized position approach the question in the wrong way. It
should be noted that the analysis advanced above in terms of Operator
Binding can account for (63) without further ad hoc elaboration. The rule
of OB does allow binding into conjoined sentences when the arguments
are of the same type! By contrast, Chomsky and Lasnik must make an
additional assumption: ”... the wh-word appears [to derive] from the first
clause while some sort of deletion applies in the second” (p. 491). This
new assumption involving “some sort of deletion” is a consequence of
assuming a movement analysis as opposed to a direct interpretive ap¬
proach.
Now if the w/z-phrases really derive from a more abstract sentence
 Realistic Grammar 39

internal position, as in transformational grammar, we might expect them

to exhibit a distribution sentence-internally identical to that of the wh-
phrases in head-position. The distribution does not quite coincide, how¬
ever, as noted in Brame (1978a).

(64) a. What the hell did you see?

b. *You saw what the hell?
c. *You saw something the hell.
(65) a. Who the hell is he talking about?
b. *He is talking about who the hell.
c. *He is talking about someone the hell.

Before turning to the next topic, let us consider for a moment examples
of the following type, involving scopal predicates.

(66) a. Sonatas are easy to play on this violin.

b. This violin is easy to play sonatas on.
c. It is easy to play sonatas on this violin.

It is well-known that a traditional assumption is that such examples are

related transformationally. By contrast, in Brame (1976), examples such
as (66a) and (66b) are not derived from (66c). Rather, (66a) and (66b) were
base generated. This analysis is a more radical departure from tradition
than the analysis mentioned in Ross (1967) and subsequently adopted in
Lasnik and Fiengo (1974) which incorporates structures such as this violin
is easy to play sonatas on it corresponding to the underlying representa¬
tion for (66b), for example.
Now it is interesting that examples such as (66a) and (66b) illustrate the
filtering property. Thus, the following examples must be ruled out.

(67) a. *Sonatas are easy to play tunes on this violin.

b. *This violin is easy to play sonatas on this cello.

An account of this property is provided in Brame (1976). However, the

fact that such examples exhibit the filtering property should perhaps clue
us to the possibility that such examples involve binding. If so, then we
should expect to find the accessible scope property as well. This we do
find, as illustrated below.
40 Realistic Syntax

(68) question-scoped predicate

a. *Which sonatas is this violin easy to play_on_?
rel-scopal predicate
b. *the sonatas which this violin is easy to play_on_
top-scopal predicate
c. *These sonatas this violin is easy to play_on_
foc-scopal predicate
d. *It is this sonata that this violin is easy to play_on_

Such examples illustrate the prohibition against operating into scopal predi¬
cation contexts by questioning, relativizing, topicalizing, and focussing.
It is also not permitted to scopally predicate into question contexts, rela¬
tive clause contexts, etc.

(69) scopal predicate-question

a. ’’This violin would be easy to ask Mary which sonatas she plays
_on_
cf. It would be easy to ask Mary which sonatas she plays on this
violin.
scopal predicate-rel
b. *These sonatas are easy to tune the violin that Mary plays _ on

cf. It is easy to tune the violin that Mary plays these sonatas on.

Similarly, it is not possible to scopally predicate into a scopal predication

context.

(70) scopal predicate-scopal predicate

*This violin would be hard for these sonatas to be easy for Mary to
play_on_
cf. It would be hard for these sonatas to be easy for Mary to play
on this violin.

There are, then, excellent grounds for concluding that we are dealing
with another instance of binding. One of the virtues of the analysis pro¬
posed by Chomsky in his wh-movement paper is that there is an attempt at
generalizing the scopal predication cases with the others. However, as
reiterated here, these processes do not involve movement. Nor do the
scopal predication cases involve w/j-phrases.
 Realistic Grammar 41

Within the theory offered here, it would not be implausible to treat the
formative to as a kind of operator, perhaps as an operator prefix of a
scopal predicate operator Fsp (An[xl), as indicated below in terms of a
partial lexical specification for to.

(71) to; Fsp, (_(An[x])) (Avp)

(Of course to will be provided with additional functional structure.) We

would then interpret an example such as John is easy to please as some¬
thing approximating (72).

(72) f(g((Fsp (An[x])) (h(An[0]))) (i)

where John : i, is :f,g: easy, and h : please

The operator prefix Fsp is indexed and a rule is required to identify the
subject index i with the operator suffix An[x]. Finally, OB will then be
applicable and will bind An[o1 as i.
Such an analysis would involve no special w/i-phrase nor a special
deletion rule as proposed in Chomsky (1977).

3.2 VP-Ellipsis

Let us now turn to the problem enunciated in subsection 1.2, termed the
Multiple Ambiguity Problem. In particular, let us first consider the lack of
4-way ambiguity of the following example.

(73) These are flying planes and those are too.

To treat such examples it is necessary to introduce a partial lexical spec¬

ification for conjunctions such as and.

(74) and; Fcon[&1, _(Xl5 X2, ..., Xn)

It is intended that a lexical item such as and is interpreted as a kind

of function which selects any number of arguments of type X.
It is also necessary to provide an analysis of the auxiliary system of
English. Within the framework advocated here, there are no transforma¬
tions. In particular, there is no affix-hopping transformation as in the
42 Realistic Syntax

standard theory, cf. Chomsky (1957). Rather, we must provide for the
relevant auxiliary phrases directly, making more use of the lexicon, as in
Brame (1978c).
To see how this might be accomplished, let us take the phrase flying
planes in the verbal sense. A representation for the verb fly might be
something like (75).

(75) fly; F\_(An[0])

Of course it is necessary to provide for additional functional structure to

account for the intransitive use of fly, etc.
Now we would like to say that flying, like fly, can select a direct object
argument. Suppose we enter the suffix ing in the lexicon as a kind of
function, the progressive function.

(76) ing; Fprog, _(FV)

Now the fact that flying selects the identical argument structure as/?y can
be expressed by the following redundancy rule which operates on lexical
specifications to yield new lexical entries.

(77) Aux Redundancy:

If (a; Fx, —(Fv)) and (b; Fv, ... ) e Lexicon, form a new lexical
entry L = (ba; Fx (Fv), ... ).

Applying the redundancy rule (77) to (75) and (76), we obtain the new
lexical entry flying, indicated below.

(78) flying; Fprog (Fv), _(An[0])

We might even provide the composition function FproE (Fv) with a coarser
specification such as Fprog part and call it the progressive participle func¬
tion. More importantly, we may consider/?ymg planes as a special kind of
argument, a progressive participle argument, represented as Av[prog]. We
could then provide the auxiliary verb be with the following lexical rep¬
resentation.

(79) be; Fv, _(Av[prog])

 Realistic Grammar 43

In addition to the lexical specifications provided, it is necessary to

extend the rules of composition to include the following.

(80) a. Compose Fcont&] Xn as Fcon[&] (Xx, X2, Xn.1}

Xn) ^
b. Compose Fpros (F'’)'ax as (Fprog(Fv))(Ax)

With these lexical specifications, composition rules, and indexing proce¬

dures, we may now derive an interpretation for the left conjunct of exam¬
ple (73). Rather than provide the complete tree, the terminal interpreta¬
tion structure will be provided.

(81) (Fv(Av[prog]))'(Ants]) : (f(i)) 0)

where/ — are, i = flying planes, j = these

Notice that / indexes flying planes in the progressive participial sense, not
the np sense of a subject complement. To derive the other interpretation,
we must extend the lexical specification of the verb be to permit subject
complement arguments.

(82) be; Fv, _(Av[prog]) or _(An[scl) or _(Aa[sc])

Noun phrases interpreted as An[xl will qualify as subject complements and

adjectives which are interpreted as Aa[sc] in their predicative function (vs.
their attributive function in which they are Fa) also qualify as arguments
of be. The second reading for the left conjunct of example (73) is then the
following.

(83) (Fv(An[sc]))(An[s]) : (f(i))(j)

where/ = are, i = flying planes,j — these

Under this interpretation, flying planes is a subject complement, not a

progressive participial phrase.
Consider now the right conjunct of (73), repeated below as (84).

(84) Those are too.

We may derive two partial interpretations for (84).

44 Realistic Syntax

(85) a. (Fv(Av[progl))(An[sl) : (f(Av[prog]))(j)

where / = are, j = those
b. (Fv(An[sc]))(An[s]) : (f(An[sc]))(j)
where / = are, j = those

If there is no context, (84) will be declared deviant under either of the two
interpretations of (85). Clearly, we require an operation on functional
structure to identify the unindexed term Av[prog] in the partial interpreta¬
tion of the right conjunct with the index i of (81) and to identify the
unindexed term An[sc] of (85b) with the index / in (83). To see this more
clearly, let us collect together the two conjuncts as they would be inter¬
preted. There are in fact four possibilities.

(86) a. Fcon[&](F)' (Ay[progl))(AIJ[s]), (F£(Av2[prog]))(A/s])):

h((f(i))(j), (g(A^progJ))(k))
where h = and,f = are, i = flying planes, etc.
b. Fcon[&1((F)(A/sc]))(A i[sl), (F|(A§[sc]))(A/s1)):
h((f(i))(j), (g(A3Csc]))(k))
where h = and,f = are, i = flying planes, etc.
c. Fcon[&]((Fy(Av[prog]))(Ai[s]), (F^(An[scl))(A§[s])):
h((f(i))(j), (g(Antsc]))(k))
d. Fcon[&]((Fj(An[sc]))(Ap[s]), (F^(Av[prog)))(Ap[s]))
h((f(i))(j), (g(Av[progl))(k))

Now in order to ensure the necessary identification between the unspec¬

ified arguments in (86a) and (86b), the following operation on functional
structure can be adopted.

(87) Ellipsis:
Let i and Ax be included in a partial interpretation ip such that i
indexes an argument of the same type as Ax for some x. Then
identify Ax with /.

By employing this rule of Ellipsis, we succeed in deriving a complete

interpretation for (86a) and (86b), but we fail to derive a complete in¬
terpretation in the case of (86c) and (86d) since the index i in (86c) does
index an argument of the same type as An[scl and the index i does not index
 Realistic Grammar 45

an argument of type Av[progl in (86d). It follows, then, that the related

example (83) is 2-ways ambiguous, not 4-ways ambiguous. This consti¬
tutes a solution to MAP discussed in subsection 1.2 and can be extended
to examples (23a-d).
The theory developed here seems to be suitable for handling the dis¬
course examples as well, cf. (25). Coarser discourse relations are certainly
compatible provided we append suitable rules of discourse composition.
Some initial work with discourse relations was attempted in Brame
(1978b).
This analysis should also extend to examples such as the following.

(88) a. John has eaten more beans than Mary has.

b. Sally will learn less than Mary will.
c. His mind is not as quick as yours is.
d. Sue is running faster than Ed is.

Some details have not been discussed. For example, it is necessary to

provide a lexical representation for too and an appropriate rule of compo¬
sition to ensure that a complete interpretation is provided for example
(83). We might take too and its negative counterpart either to be functions
and provide lexical specifications to reflect this characterization. A
number of questions arise in the course of attempting such an analysis.
For example, there are apparently a number of restrictions on the dis¬
tribution of too and either as the following examples illuminate.

(89) a. John has eaten or Bill has too.

b. John eats less than Bill does too.
c. John has eaten or Bill hasn’t either.
d. John eats less than Bill doesn’t either.

The functions too and either seem to involve a kind of parallelism which
when absent leads to deviance. In (89), it appears that or and than imply
nonparallelism of the requisite type. Whether or not we should make
parallelism a necessary condition for indexing of too and either is a ques¬
tion which must ultimately be confronted.

3.3 Pronominalization and Ellipsis as Anaphora

As noted in subsection 1.3, the pronominalization facts and the ellipsis

facts point to a deeper generalization and pose somewhat of a dilemma for
46 Realistic Syntax

the standard theory of transformational grammar. Both Lakoff (1976) and

Bresnan (1978) have recognized this fact.
What is unexplained in the transformational theory is why verb
phrase ellipsis is subject to the same conditions as pronominali-
zation ... [Bresnan (1978, p. 47)]
Bresnan points to what appears to be a promising solution. She proposes
that Lasnik’s Noncoreferentiality Rule be formulated in terms of index¬
ing, presumably generalized to accommodate the ellipsis facts in addition
to the pronoun cases. (Cf. Lasnik (1976), Reinhart (1976) and DeCarrico
(1978) on noncoreferentiality.) Lasnik’s rule is repeated below.

(90) Noncoreferentiality Rule:

If NP, precedes and commands NP2 and NP2 is not a pronoun, then
NP, and NP2 are disjoint in reference.

The analysis developed in the preceding subsection is in the spirit of

Bresnan’s suggestion. This suggests that we attempt to extend the ellipsis
rule (87) to include pronouns and the relevant precede and command
relations.

(91) Let i and Aytx] be included in a partial interpretation <p. Then index
Ay[xl as i provided the following conditions hold:
(i) i indexes Aytz] for some z
(ii) Ay[x] does not precede and command Aytzl in the tree as¬
sociated with (p

This rule allows us to index an uninterpreted pronoun or verb phrase

argument with i provided the argument indexed by i and the pronoun or
verb phrase argument are of the same general type An[xl or Av!x]. It fur¬
thermore prohibits indexing when the uninterpreted argument precedes
and commands the argument indexed by i. Rule (91), which we may
call Anaphora, should also generalize to nominal arguments of functions
such as some, any, many, etc.
Now if it is correct to generalize the latter phenomena with ellipsis, we
might expect to find an analogue of the multiple ambiguity problem in
these domains. In fact we do encounter MAP here as well. This is illus¬
trated in the following examples.

(92) a. I saw some boy[z] flying planes and Bill saw some too.
b. Bill has many pipes which you must look up and you have many
too.
 Realistic Grammar 47

The ambiguity of such examples can be illuminated in terms of the follow¬

ing.

(93) a. I saw some boys flying planes,

b. I saw some boys’ flying planes.
(94) a. Bill has many pipes up which you must look.
b. Bill has many pipes which you must look up in the catalogue.

Again, as with the earlier examples of 1.2, we see that there is no 4-way
ambiguity in (92), only 2-way ambiguity. It appears, then, that we are
indeed dealing with a more far-reaching process of anaphora which cuts
across argument structure.

3.4 Binding and Anaphora

Turning back to the problem of the interaction of binding and ellipsis,

let us consider the following discourse.

(95) a. John is eating,

b. Who is?

There is no difficulty in interpreting this discourse. The relative pronoun

who stands for the subject and does not therefore interact with the
anaphor of (95b). On the other hand, the following example does exhibit
such interaction.

(96) a. John is eating,

b. *What is he?

The partial interpretation associated with (96b) is provided below.

(97) Fq(An[x])((Fv(Av!prog])(An[s])) : Fq(x)((f(Av[prog])(i))

where x — what,/ = is, i = he — John

In (97) it is assumed that a prior application of Anaphora has identified he

with John. This identification is not of concern. Now, if we apply
Anaphora to (97), we can in fact identify Av[progl with the index of eating
beans in the interpretation associated with (96a). This identification, how-
48 Realistic Syntax

ever, has the consequence that Operator Binding will be inapplicable to

(97) . Hence the operator prefix Fq will not be indexed, with the net result
that (96b) will not be fully interpreted. Hence by the principle of incom¬
pleteness, this example is ruled out. On the other hand, we can also
operate on a finer representation of the partial interpretation associated
with (96b). For example, the following is equivalent in the intended sense
to (97).

(98) Fq(An[x])((Fv(Fprog'part(An[0]))(An[s1)) : Fq(x)((f(Fpr0g-part(An[0]))(i))

where x = what,f = is, i = he

To this partial interpretation OB is now applicable, yielding the following

partial interpretation.

(99) g(x)((f(Fprog'part(x))(i))

But now Anaphora is inapplicable since the coarser argument Avtprog] is no

longer a completely uninterpreted argument. Hence there is again no way
of providing a complete interpretation for (96b). In consequence, this
example is deviant. Or to put it differently, it simply cannot be inter¬
preted.
Now if binding is correctly expressed as a general process like OB, then
we expect to encounter similar interaction results and we do.

(100) Root Q
a. *What did he?
cf. What did he eat?

Embedded Q
b. *1 asked what John had eaten but not what Bill had.
cf. I asked what John had eaten but not what Bill had eaten.
I knew that John had eaten but not that Bill had.
Relative Clause
c. *1 saw the boy who Harry is arguing with and who Sam is.
cf. I saw the boy who Harry is arguing with and who Sam is
arguing with.

Topic alization
d. *This boy John might nominate and that boy Bill might.
 Realistic Grammar 49

Pseudo-Cleft
e. ¥What John is eating and what Harry is may not be fit for you.
cf. What John is eating and what Harry is eating may not be fit
for you.
Focus
f. *It was salt that Harry objected to and pepper that Mary did.
cf. It was salt that Harry objected to and pepper that Mary ob¬
jected to.
Root Exclamation
g. *What big teeth you have and what big eyes you do!
cf. What big teeth you have and what big eyes you have!
Embedded Exclamation
h. *I’m surprised at what big teeth you have and at what big eyes
John does!
cf. I’m surprised at what big teeth you have and at what big
eyes John has.
Free Relative
i. *F11 buy what you are selling and what Bill is.
cf. I’ll buy what you are selling and what Bill is selling.
Appositive Relative
j. *John, who Mary likes but who Bill doesn’t, is my friend.
cf. John, who Mary likes but who Bill doesn’t like, is my
friend.

In (100) I have attempted to give clear-cut examples of the deviance that

results when interaction is attempted. Some of these examples can per¬
haps be made to appear better if the mind is misdirected with additional
adverbial and scopal-like elements. My feeling is that the examples should
be judged deviant. If so, then, again, the parallelism suggests that it is
correct to treat binding in a general way as suggested in 3.1.
Several examples of interaction have been noted in Sag (1976) and
Williams (1977), where explanations are attempted. Williams’ analysis has
more in common with that proposed here than Sag’s.
Now the theory advanced here makes an interesting prediction. To see
this consider the following examples.

(101) a. I wanted to see John and I did see John,

b. I wanted to see John and I did.
(102) a. I wanted to see John and did see John,
b. I wanted to see John and did.
50 Realistic Syntax

The examples listed in (102) involve conjoined verb phrases and are not
derived from more abstract underlying structures.
The following contrast is predicted in accord with the explanation pro¬
vided above.7

(103) a. John, who I wanted to see and who I did see, is my friend,
b. *John, who I wanted to see and who I did, is my friend.

But a further prediction follows. The appositive relative related to (102b),

as opposed to (101), should be fully grammatical.

(104) John, who I wanted to see and did, is my friend.

The operator who is the head of the sentence / wanted to see and did.
Since this sentence involves a conjoined verb phrase, not a conjoined
sentence, there is no operator associated with the right conjunct per se.
Hence, after the index of who has been associated with the object argu¬
ment Ant0] of see, Anaphora can proceed to supply an appropriate index
for the argument of did, corresponding to see who. The problem that
arises in connection with (103b) simply does not arise here in connection
with (104). This difference is predicted by the theory provided here.
There is still a further prediction made by the theory advocated here.
Recall, above in subsection 1.1, it was noted that operators tend to
generalize across argument types. We now have in hand a new set of
argument types, namely, the Av[x] arguments, of which we have had occa¬
sion to consider Av[progl. Others include Av[perf], Av[pas], etc., correspond¬
ing to eaten beans, eaten by John, etc. We might therefore expect to find
some wh-type operator which subsumes this type of argument as well.
And in fact we do, as shown in the following examples.

(105) a. They say John is eating beans, which he is.

b. They say John has eaten beans, which he has.
c. They say John will eat beans, which he will.
d. They say John eats beans, which he does.

The striking point about these examples is that they are grammatical, even
though they appear to exhibit the interaction property. But, in fact, the
grammaticality of such examples is predicted by the theory. We let
operator words such as which be interpreted as Av[x] arguments in addi¬
tion to An[x] and let Anaphora identify the pronominal-like argument head
 Realistic Grammar 51

with the preceding argument index. It is this index which is then identified
with the argument structure of the auxiliaries in (105). In other words,
Anaphora does not function to fill in the missing auxiliary arguments in
(105); Operator Binding does. This accounts for why the interaction prop¬
erty is not observed in these cases.
In conclusion, it seems that the realistic approach offered here does
yield some insight into the interaction problem. It also leads to additional
predictions which appear to be borne out by the data.

4. Conclusion
4.1 Apparent Arguments Against Pure Surface Theories

In 3.1 several putative arguments against a direct surface approach to

binding were surveyed. Are there any arguments that can be leveled
against the more lexically based theory advocated here? For the most
part, trace theorists have simply ignored initial work in purer approaches
to syntax which eliminate classical transformations. However, more re¬
cently Chomsky is reported to have advanced two arguments against
surface grammar during the course of his Kant lectures “Rules and Rep¬
resentations” at Stanford (January 1979). I was present for the Sloan
Foundation Workshop following these lectures where I learned of these
arguments. Apparently Chomsky argued that (i) the “asymmetry” of
idioms and (ii) the cross-over facts discussed by Wasow (1972) argue
against pure interpretive theories. In the discussion following my own
presentation at which Chomsky was present, this assessment was borne
out and my work was given the familiar notational variant treatment.
Since there are many differences between realistic grammar and trace
theory, it will be pointed out in this closing section how realistic grammar
surpasses trace theory in providing a much more adequate representation
of natural language.
Discussion of the two apparent counterarguments to realistic grammar
must be postponed to such a time when the arguments are laid out in an
explicit and comprehensible fashion. As I understand the asymmetry ar¬
gument, Chomsky has claimed that idioms exhibit an asymmetry with re¬
spect to deep structure vs. surface structure: there are idioms which
reflect deep structure canonical properties only, but there are no idioms
that reflect surface structure canonical properties only. Within the theory
developed here, there is no distinction between surface and deep struc¬
tures. Hence Chomsky’s argument must somehow translate to the struc¬
ture of the lexicon if it is to hold. What are the facts concerning this
52 Realistic Syntax

asymmetry? Is there truly some asymmetry? (Tom Wasow informs me

that my examples of binding in (64)-(65) may constitute an outright
counter-example to Chomsky’s argument.) In the absence of published
discussion of examples, it is not possible to assess Chomsky’s claim.
It is interesting to turn back to an earlier argument against a specific
surface analysis provided in Lakoff (1976). This argument concerns the
unity of pronominalization and ellipsis, discussed above in 3.3.
One might be tempted to conclude that since one anaphoric de¬
vice, definite pronouns, cannot be derived by a deletion trans¬
formation, therefore no anaphoric devices can be derived in that
way. That is, one might be tempted to assume that anaphoric
devices must be derived in some uniform manner.
Of course this is what is claimed above in 3.3 and is suggested in Bresnan
(1978). Lakoff goes on to say
However, it is not at all clear that this is true, since it can be
shown that at least one type of anaphoric device must be trans¬
formationally derived ... one realizes that definite pronouns
and omitted VPs have deep structures of an entirely different
nature. The Bach-Peters case seems to show that definite pro¬
nouns cannot be derived from repetitions of the full NPs that
they refer to. However, omitted VP’s must [emphasis by Lakoff]
be derived from the full VPs to which they refer. The omitted VP
cannot just be represented by a blank or a reference index in
deep structure. Although definite pronouns cannot be derived by
a transformational rule, missing VPs must be ... The reason
for this is fairly obvious. Missing VPs can refer to VPs that are
derived by transformation, and are not present in deep structure.
For example, consider the following:
[14] If John is shot by Max, Harry will be.
[15] If John is expected by Sam to be shot by Max, Harry will
be.
[16] If Shakespeare translates easily into Japanese, Marlowe
will.
[17] If John is likely to leave, Bill will be.
All of the underlined VPs in the preceding examples are trans¬
formationally derived; they do not occur as such in deep struc¬
ture. Since they are not even constituents in deep structure,
there is no way to refer to them on that level of analysis. Thus,
there is no way of indicating in deep structure what the missing
VP will refer to. So, a solution for missing VP’s parallel to the
no-full-NP solution for definite pronouns ... is not possible.
[Lakoff (1976, pp. 332-3)]
 Realistic Grammar 53

The fallacy of Lakoff’s argument is not unclear. The alternative to the

“fairly obvious” conclusion which Lakoff draws is that the examples
listed in his [ 14]—[ 17] are not transformationally derived. In fact, such facts
argue rather strongly for the nontransformational theory advocated
above. (Such an argument has been advanced in Brame (1978c) and in
Schachter (1978).) We see that Lakoff’s observation turns out to be an
argument against transformational grammar.
This argument against transformations is similar to that provided in
Brame (1976) where it was noted that once Equi is abandoned, the full
range of transformation falls with it. Once VP-Deletion as a transforma¬
tional rule is abandoned, so also a full range of transformations must be
eliminated in favor of surface analysis, where lexical items assume a
major role.

4.2 Nonexistence of Empty Nodes and Traces

One of the basic drawbacks of trace theory is the fact that it is burdened
with an excess of theoretically elaborate devices, such as traces and
empty nodes. It should be emphasized that such devices are not adopted
within the context of realistic grammar as envisaged here. Thus, the in¬
terpretations discussed in the preceding pages rely on the functional struc¬
ture of lexical items for their internal parts. This is true of lexical functions
such as verbs and prepositions as well as determiners and specifiers.
There is one area, however, where an explicit account has not been
provided. This concerns the subject of binding constructions such as the
following.

(106) a. Who left?

b. the boy who left
c. I asked who left.
d. It was John who left.

Recall that who is interpreted ultimately as an operator, for example, as

Fq(An[si) in (106a). But we still require the presence of a subject An[s] if
binding is to apply to a composed sentential argument. We must therefore
provide for the subject argument and this can be accomplished by provid¬
ing a rule which will interpret An[sl in the appropriate position relative to
the interpretation tree.

(107) Subject Interpretation:

6 -> An[xl / Fop[y]_Fvp
54 Realistic Syntax

In (107) the symbol Fop[y] is designed to cover all operators of the F(Antxl)
type, for example, the question operator, relative operator, etc. (We as¬
sume Fop[y] = Fy(A), where y = q, rel, etc.) Notice that we have inserted a
general argument An[x] by rule (107). This causes no difficulty, for we need
rules to identify x of np arguments at any rate, as noted earlier. These
rules include the following.

(108) Let An[x] be the composition of some phonological string.

(i) Set x = s /_Fvp
(ii) Set x = o / Fv_or Fv An[x]_
(iii) Set x = dat / Fv_

By rule (108i), the x inserted by rule (109) will be 5.

Now, let us see how (106a) is interpreted. First, applying the interpre¬
tive rules to who left, we derive (109).

Fq(An[x]) : Fq(x) Fvp : f

An(x] : x Fv : f

who lift

Now rule (107) applies, assuming that Fop[y] is equivalent to Fq(An[x]) as

indicated above. (In fact Fop[y] could be included directly in (109).) This
gives (110).

(110) Fq(An[x]) : Fq(x) An[xl Fvp : f

An[x] . x Fv : f

who left

Now the composition rules and indexing procedure follow, giving the
partial interpretation provided in (111).

(111) Fq(An[xl)((Fvp(An[s])) : Fq(x)((f(An[s]))

Now OB can apply to identify An[s] with x and to index Fq. Analogous
steps are involved in the interpretation of the other examples listed in
(106).
 Realistic Grammar 55

Now this analysis can be utilized to solve another puzzle. The following
distribution of data will serve as an introduction.

(112) a. *Who did you give a book?

b. *John would be easy to give a book.
c. *This boy John gave a book.
d. *the boy who John gave a book
e. *It was John that Bill gave a book.
f. *1 asked who John gave a book.
g. *The boy, who John gave a book, is my friend.

It is interesting to note that the full range of binding constructions exhibit

deviance when a dative argument serves as the operator suffix. To ac¬
count for such examples we must first provide a lexical specification for
dative verbs such as give.

(113) give; F\ __(An[dat],An[o1) or _(An[0l,Ap[dirl)

According to (115), give is lexically specified to select a dative argument

plus object argument or else an object argument plus directional preposi¬
tional phrase argument.
Now all along we have been assuming that binding takes place when the
operator suffix is An[x] and the unspecified argument is An[0], Antsl, etc., as
in (111). We now insist that they be identical as indicated by the stipula¬
tion in OB that arguments be of the same type. To identify argument
suffixes as a specific type, we append the following case to rule (108).

(114) Set x = o, p[dir], p[loc], p[man], ..., s / (Fx_)A

Of importance here is the fact that rule (114) does not permit an identifica¬
tion ofx as dat. That is, it is not possible to identify the operator argument
as Antdat]. It then follows that an index standing for An[dat] cannot be
“quantified” into a binding context. From this it follows that examples
such as (112a-g) and all similar examples will not be completely inter¬
preted.
With this analysis, we have an explanation for another anomaly, first
brought to my attention by Ivonne Bordelois. It concerns the follow¬
ing contrast.

(115) a. *Who did John promise to sleep?

b. Who did John persuade to sleep.
56 Realistic Syntax

If we lexically specify promise to select a dative object plus a vp argu¬

ment, as opposed to persuade, which selects a direct object argument plus
a vp argument, among others, then we should expect not to be able to
completely interpret (115a). We should also expect the full range of judg¬
ments to hold.

(116) a. *John would be easy to promise to sleep,

b. John would be easy to persuade to sleep.
(117) a. *It was John who Bill promised to sleep,
b. It was John who Bill persuaded to sleep.

There are other approaches to these problems which must be explored

within the framework of realistic grammar. For example, we might con¬
sider the possibility of imposing restrictions on composition of complex
argument structures with Fv functions. This approach might be justified if
the following examples are judged deviant.

(118) a.Which book did you give John?

b.Which book would be easy to give John?
- c.the book which I gave John
d.This book I gave John.
e.It was this book that I gave John.
f.I asked which book you gave John.
g- the book, which you gave John, is on the table.

Whatever the case may be with the status of these examples, it appears
that the realistic grammar framework provides a ray of hope for express¬
ing the obvious generalization.

4.3 The Missing Argument Phenomenon

One of the most interesting of puzzles which has been discussed and
debated within the framework of transformational grammar involves the
status of the following distinction.

(119) a. Who did John say left?

b. *Who did John say that left?

Let us refer to the problem of accounting for the deviance of (119b) vs.
(119a) the missing argument problem . Interesting discussion of this prob-
 Realistic Grammar 57

lem together with a summary and criticism of previous work can be found
in Bresnan (1977). This problem has given rise to constraints such as the
Frozen Subject Constraint, cf. Bresnan (1972), indicating by the name
that the problem is that of inhibiting the subject from moving to the comp
position. Since we are operating within a framework which involves a
direct approach to binding, we must now account for why (119b) is unin¬
terpretable.
A new analysis was provided within the framework developed here and
presented at the Sloan Workshop at Stanford, cf. Brame (1979a). This
analysis involves an extension of rule (107), the rule of Subject Interpreta¬
tion.

(120) 0 Aytzl / Fx_Fvp

where x = op[w] or v

In other words, we allow the interpretation of a subject An[sl node, for

example, as in the earlier case (106), not only after operator functions and
before vp functions, but also after verbal functions before vp functions.
Thus, there is no problem interpreting (119a). By contrast, (119b) cannot
be provided a subject and therefore will be incompletely interpreted. Of
course, one can attempt to interpret that as the subject of left since that is
multiply specified. However, under this interpretation, binding will never
ensue so that the example will be incompletely interpreted. In short, there
is no way to completely interpret (119b).
Now rule (120) has been stated in a very general way. Thus, the inserted
element need not be An[sl. It can be As, instead. This will allow for the
interpretation of examples such as (121).

(121) That John left Bill said was obvious.

However, the following will not be interpretable.

(122) *That John left Bill said that_was obvious.

It should also be possible to insert prepositional phrase arguments. The

following example illustrates this case.

(123) In this garage John said could be found three cases of vintage
Martel.
58 Realistic Syntax

Clearly the prepositional phrase argument in this garage must be as¬

sociated with the post verbal position in (123), i.e. after said, as predicted
by rule (120).
Example (123) is essentially the same type of example adduced in Bres-
nan (1977, p. 179-80), repeated below.

(124) It’s in these villages that we all believe can be found the best
examples of this cuisine.

When I raised the missing argument problem at the recent Sloan Work¬
shop, Chomsky cited the recent unpublished research of Kayne, indicat¬
ing that its solution concerns the notion subject. Bresnan’s example,
however, as I pointed out, shows that it is not the notion subject that is at
issue. (In response, Chomsky claimed that in these villages in (124) is a
subject.) A similar set of deviant examples is observed in (125).

(125) a. *In this garage John said that could be found three cases of
vintage Martel.
b. *It’s in these villages that we all believe that can be found the
best examples of this cuisine. [Bresnan, p. 180]

To summarize this subsection, it seems to me once again that some

progress can be made toward resolving outstanding problems if we oper¬
ate within a realistic framework.

4.4 Chomsky-Lasnik Filters

In Chomsky and Lasnik (1977), a number of ad hoc filters are proposed

to rule out certain ungrammatical examples, including several involving
the missing argument phenomenon discussed above. In fact, all of these
filters can be dispensed with since they are simply ad hoc special cases of
incompleteness (or functional deviance), cf. Brame (1979).
For example, consider the following example discussed by Chomsky
and Lasnik.

(126) *The man left you is my friend.

To account for this example, they propose the following filter.

 Realistic Grammar 59

(127) *LPNP tense VP]

Notice that (126) can be generated within the Chomsky/Lasnik framework

by virtue of their rule of free comp deletion. Thus, they assume that who
can be freely deleted, even when there is no subject. By contrast, exam¬
ples such as (126) simply cannot be interpreted within the framework
advocated here. It is possible to interpret the man left you and is my
friend, but it is not possible to compose the two, since vp functions do not
select sentences such as the man left you, cf. Brame (1979).

4.5 Nonexistence of Ordering

Many syntacticians have been beating ordering over the head for a
number of years. Almost all such critics, however, replace ordering with
other ad hoc principles, for example, by principles which seek to establish
a universal ordering or by filters and other ad hoc devices. The first real
light to be shed on ordering can be found in Bresnan (1978). Bresnan has
shown that the ordering facts of transformational grammar follow from
realistic grammar. There is no need for special principles which simply
replace ordering as in other approaches and which have no other function
than to establish “universal” orders.

4.6 New Frontiers: Realistic Phonology

The approach I have been discussing paves the way for a more realistic
approach to phonology, which I have called lexical phonology. In particu¬
lar, the lexical entries provided in realistic grammar can be elaborated in
the following way. Let us take the perfective ed affix as an example. We
have been assuming that it is entered in the lexicon as (128) and that it
combines with verbal functions by a general lexical redundancy rule.

(128) ed; Fv, __(Av[perf])

However, it now seems feasible to take all phonetic realizations of the

perfective morpheme to be part of the lexical specification. Thus, (128)
will be replaced with (129).

(129) d or t or4d; Fv, _ (Av[perfl)

60 Realistic Syntax

Within a theory of lexical phonology, phonological rules are not dis¬

pensed with as they are in other approaches. Rather, they are interpreted
somewhat differently from the standard theory of generative phonology.
They can now be interpreted as instructions for combining morphemes by
lexical redundancy rules (or alternatively as filters).
For more discussion of this approach, see Brame (to appear), and Yil-
dirim’s work on Turkish phonology within such a framework. A goal to
shoot for within this theory is to show that ordering follows naturally, as it
does in syntax. Guerssel in forthcoming work will show that his important
constraints on phonological rules, the Adjacency-Identity Constraint and
the Constraint on Assimilation Rules do indeed follow from more general
considerations of a theory of lexical phonology. He also shows that the
phenomena are far more general than presented in his earlier work, cf.
Guerssel (1977, 1978).

4.7 Linguistic Problems

The thesis which I insist on emphasizing is this: linguists should be

investigating the significant problems that have been discovered over the
past two decades of work within the framework of transformational
grammar. In Brame (1979a), twenty three such problems are posed and
discussed. Several of these problems have been presented here along with
tentative solutions. As far as these problems are concerned, a number of
outstanding questions remain. Nevertheless, it seems to me that a direct
lexically based approach to such problems can provide some illumination
and further direction toward an understanding of how they are to be
resolved.

Notes

1 The distinction between topic operators and contrastive operators is being overlooked for
the purposes of this discussion, cf. Brame (1978b, fn. 3).
2 Of course there is an independent reason for the deviance of (10b) which relates to the
prohibition against sentential complements in embedded structures, cf. *Bill believes that
that Mary left surprised Harry.
3 There is a very interesting fact which will not be explored here. This is the fact that
examples (27a), (27b), (30a), and (30b) are in fact well-formed given appropriate pauses and
intonation which serve to destroy the natural sequencing of the phrase. Thus, the following
are fine.
(i) I saw him—after Mary pointed Bill out to me.
(ii) John thought she was tall—when he saw Mary
 Realistic Grammar 61

(iii) John has—if Bill has tasted the beans.

(iv) Sally can’t—if Bill can’t leave.
The fact that both the pronouns and auxiliaries function in this way is further evidence that
we are dealing with a unified phenomenon. It also seems to be true that the sequencing of a
discourse can be broken so that the appropriate relations can be established in the discourse
examples in (34) and (35). For example, the following would be fine.
(v) a. John saw him.
b. Oh yeah? Well I saw Bill too.
(vi) a. John has.
b. Oh yeah? Well, I've eaten beans too.
The proper treatment of these facts may well have a bearing on the whole question of the
proper generalization of ellipsis and pronominalization as anaphora.
4 Relevant studies include Bresnan (1978), Yildirim (1978), and Brame (1978a, 1978b). For an
analysis of binding very much in the spirit of my own, see Woolford (to appear).
5 In general all lexical specifications provided here are partial. Relevant lexical structure is
omitted where it does not affect the point at hand. Thus, for example, the lexical specifica¬
tion for met must be elaborated to permit examples such as the boys met, etc.
6 In fact, the general filtering device (53) can be dispensed with altogether if we define the
phrases of the language to include all those that receive a “complete” interpretation. It then
follows that incompletely interpreted strings are not phrases of the language. For conveni¬
ence, (53) is retained throughout this essay.
7 It should be noted that the following example is somewhat deviant.
(i) ?John, who I wanted to see and I did, is my friend.
However, this example clearly has a different feel from (103b). It is not the interaction
property that rings untrue of (i), but rather the head property, for (ii) seems to be equally
deviant.
(ii) ?John, who I wanted to see and I did see, is my friend.
It is generally the case that if one conjoined sentence is headed by some complementizer
element, then both must be so headed, as illustrated below.
(iii) a. ?I know that John ate udon and Mary drank kumiss.
b. ?John realizes that you are smart and Mary is not so smart.
c. ?Do you know whether he left or Mary arrived?
It should be noted, however, that there may well be speakers who accept such examples, the
point being, that such speakers reject (103b). Incidentally, some examples which violate the
head constraint strike me as more acceptable, including example (63) from Bresnan.

References

Brame, M. K. 1976. Conjectures and Refutations in Syntax and Seman¬

tics . Elsevier-North Holland Publishing Co., New York.
Brame, M. K. 1978a. Base Generated Syntax. Noit Amrofer Publishing
Co. P.O. Box 15176, Seattle, Wa. 98115.
Brame, M. K. 1978b. Discourse and Binding without Transformations.
Linguistic Analysis 4: 365^104.
62 Realistic Syntax

Brame, M. K. 1978c. The Base Hypothesis and the Spelling Prohibition.

Linguistic Analysis 4: 1-30.
Brame, M. K. 1979. Chomsky/Lasnik Filters Are Special Cases of Func¬
tional Deviance. Paper delivered at 1979 Sloan Foundation Workshop
at Stanford. In this volume.
Brame, M. K. 1979. (to appear a). Linguistic Problems (23 linguistic prob¬
lems discussed with proposed solutions)
Brame, M. K. 1978 (to appear b). Lexical Phonology.
Bresnan, J. 1972. Theory of Complementation in English Syntax. MIT
doctoral dissertation.
Bresnan, J. 1976. On the Form and Functioning of Transformations. Lin¬
guistic Inquiry 7: 3-40.
Bresnan, J. 1977. Variables in the Theory of Transformations. In Formal
Syntax, ed. by Culicover, Wasow, and Akmajian.
Bresnan, J. 1978. A Realistic Transformational Grammar. In Linguistic
Theory and Psychological Reality, ed. by Halle, Bresnan, and Miller.
Chomsky, N. 1957. Syntactic Structures. The Hague: Mouton.
Chomsky, N. 1968. Language and Mind. Harcourt, Brace, and World.
Chomsky, N. 1977. On WH-Movement. In Formal Syntax. Ed. by
Culicover, Wasow, and Akmajian.
Chomsky, N. and H. Lasnik 1977. Filters and Control. Linguistic Inquiry
8: 425-504.
DeCarrico, J. 1978. Anaphoric Relations of Indefinites. U. of Wa. unpub¬
lished.
Emonds, J. 1976. A Transformational Approach to English Syntax.
Academic Press.
Guerssel, M. 1977. Constraints on Phonological Rules. Linguistic
Analysis 3: 267-305.
Guerssel, M. 1978. A Condition on Assimilation Rules. Linguistic
Analysis 4: 225-254.
Lakoff, G. 1976. Pronouns and Reference. In Syntax and Semantics 7.
Lasnik, H. 1976. Remarks on Coreference. Linguistic Analysis 2: 1-22.
Lasnik, H. and R. Fiengo 1974. Complement Object Deletion. Linguistic
Inquiry 5: 535-571.
Reinhart, T. 1976. The Syntactic Domain of Anaphora. MIT doctoral
dissertation.
Ross, J. 1967. Constraints on Variables in Syntax. MIT doctoral disserta¬
tion.
Sag, I. 1976. Deletion and Logical Form. MIT doctoral dissertation.
Schachter, P. 1978. Propredicates. Linguistic Analysis 4:187-224.
Wasow, T. 1972. Anaphoric Relations in English. MIT doctoral disserta¬
tion.
 Realistic Grammar 63

Wasow, T. 1977. Review of Conjectures and Refutations in Syntax and

Semantics. Linguistic Analysis 3:377-395.
Williams, E. 1977. Discourse and Logical Form. Linguistic Inquiry
8: 101-139.
Woolford, E. (to appear) Topicalization and Clefting without WH-
Movement. Paper presented at NELS VIII.
Yildirim, C. 1978. A Functional Interpretive Approach to Turkish. Uni¬
versity of Washington doctoral dissertation.
 CHOMSK Y/LASNIK FILTERS

ARE SPECIAL CASES OF FUNCTIONAL DEVIANCE

Sensei, do they really miss this parallelism?

*It is unclear what John to do.

*It bothers me Bill to win.
*It came as a surprise to me John was here.

It is unclear what to do.

It bothers me for Bill to win.
It came as a surprise to me that John was here.

Yes, gakusei, they do.

 1. Introduction
2. Unexpressed Generalization
3. Functional Deviance
4. The Controversial Examples
5. A Hitherto Missed Syntactic Generalization
6. Lexical Integrity
7. Extraposition and Gapping as Anaphora
8. The Missing Subject Problem
9. for-for and for-to
10. Infinitival Constructions
11. Conclusion
Notes
References
 CHOMSKY/LASNIK FILTERS

ARE SPECIAL CASES OF FUNCTIONAL DEVIANCE

1. Introduction

Do the following examples really pose a dilemma for linguistic theory?

(1) a. *The man met you is my friend,

b. *He left is a surprise.

They do for Chomsky/Lasnik (1978). Before discussing the reasons be¬

hind the Chomsky/Lasnik concern for such examples, let us take a fresh
look at (la) and (lb) in the absence of theoretical bias. A gut response to
such cases runs as follows: It is possible to interpret the man met you in
(la) as a sentential proposition and it is possible to interpret is my friend in
the same example as a tensed predicate. But that is as far as we can go. It
is simply impossible to compose the two because sentences do not select
for tensed predicates and tensed predicates do not select for complemen¬
tizerless subjects. Nor does the predicate met select for a tensed predicate
in addition to its direct object argument. Hence, it is simply not possible
to provide an interpretation for (la).
Likewise, in the case of (lb), we can interpret he left as a sentential
proposition and is a surprise as a tensed predicate, but again we are at a
loss to compose the two.
It seems that this is precisely what a good theory of language should say
about these examples: they are deviant for failure to allow composibility
of their parts. Let us call this kind of devianc t functional deviance. Before
turning to a more explicit characterization of functional deviance, let us
first consider several additional examples.
Examples such as the following have been adduced in the literature, cf.
Chomsky (1971, 1973).

(2) a. *John is believed is hungry.

b. John is believed to be hungry.
67
68 Realistic Syntax

Again, common sense tells us that John is believed and is hungry in (2a)
can each be interpreted but cannot be composed. The predicate believed
simply does not select a tensed predicate as argument. By contrast, infini¬
tive phrases can function as arguments and in this case believed does
select for this type of argument. Hence the parts of (2b) are composable
and the example can be interpreted.
We see that there is a generalization to be expressed here in connection
with examples such as (la), (lb), and (2a). All should be out for the same
reason.
Indeed, all of the filters proposed by Chomsky and Lasnik are special
cases of functional deviance. But this is perhaps not so easy to see when
one is working within a framework such as transformational grammar as
opposed to a framework within which one works directly off the surface.
Now what is Chomsky and Lasnik’s account of (la) and (lb)? To ac¬
count for such examples, they propose the following filter.

(3) *tjpNP tense VP]

Filter (3) stipulates that all NP’s comprised of NP plus tensed predicates
are marked as deviant, where the deviance is indicated by an asterisk.
This account strikes the uninitiated reader as somewhat puzzling, for,
after all, the man met you in (la) and he left in (lb) are interpreted as S’s,
not NP's. However, the uneasiness may well recede if we could only
agree with an underlying assumption—that (la) and (lb) derive from a
more abstract underlying representation with complementizers such as
who and that, as indicated below.

(4) a. [the man who met you] is my friend

b. [that he left] is a surprise

If one could buy (4), then perhaps it would be possible to accept the filter
(3), for Chomsky and Lasnik claim that there is a rule which freely deletes
complementizer elements. The trouble here is that no evidence is pro¬
vided for a rule of complementizer deletion. Chomsky and Lasnik simply
assume the existence of such a rule and pursue its consequences.
Perhaps if we forego the latter assumption, the problems entailed by the
examples in (1) will go away. We could then do without the ad hoc filter.
After all, the rule of free complementizer deletion does violate the Spell¬
ing Prohibition which is supported by a good deal of evidence, cf. Brame
(1978a). But even if we do not adopt a rule of free complementizer
deletion, a problem arises nevertheless, in the form expressed from the
 Chomsky/Lasnik Filters 69

outset; namely, when the man met you and he left in (la) and (lb) are
bracketed with S, instead of NP, we must still account for the deviance of
the complete examples given.
It should even be granted that the filter (3) leads to a loss of generaliza¬
tion, for it fails to generalize to the situation just alluded to. A second
principle would be needed to cover such a situation and whether this
principle involves phrase structure rules, lexical insertion, a new filter, or
whatever, such a principle will differ from the filter (3).
It would not be unreasonable, therefore, to abandon the free deletion
rule once and for all and with it the filter given in (3), attempting instead to
provide a general account of (la) and (lb) in terms of functional deviance.
This conclusion becomes even more obvious when we turn to (2a)
where a simple comparison shows that there is indeed a generalization
underlying this example and (la)-(lb).
The problem of accounting for (2) was originally posed by Chomsky,
who considered the deviance of (2a) to be related to the passive transfor¬
mation. Thus, it was suggested that (2a) was derived from (5a), and that
(2b) was derived from (5b).

(5) a. Someone believes John is hungry.

b. Someone believes John to be hungry.

Chomsky’s question was this: Why does the passive transformation apply
to (5b) to yield (2b), but not to (5a) to yield (5b)? His answer was:
“. . . let us propose the principle that nothing can be extracted from a
tensed sentence” (1971, p. 34). This new principle became known as the
Tensed Sentence Condition [TSC] and later, after counterexamples began
to appear, as the Propositional Island Constraint [PIC].
Now it has been noted repeatedly that the deviance of (2a) has nothing to
do with the passive transformation per se, cf. Brame (1977, 1978b). For
example, it has been noted that the following examples are deviant where
they have nothing to do with passive.

(6) a. *John believes is hungry.

b. *Mary believed were hungry.

It is obvious that these examples have nothing to do with passive and that
therefore TSC cannot provide an account of their deviance. It is equally
obvious that a general account of these examples should extend to the
problematic example adduced by Chomsky in (2a). My contention is that
this account should also extend to all of the examples for which Chomsky
70 Realistic Syntax

and Lasnik propose a host of ad hoc filters. And this account is founded
on a more lexically based theory of syntax.

2. Unexpressed generalizations

Let us briefly review the alleged motivation for the Chomsky/Lasnik

approach to complementizers in English. They are intent on relating the
following examples.

(7) a. the man that I saw

b. the man who(m) I saw
c. the man I saw

All of these examples involve well-formed relative clause constructions.

In an attempt to relate such examples, Chomsky and Lasnik propose a
single underlying deep structure such as (8a), where that is spelled out, in
conjunction with a rule of wh-movement which gives rise to (8b).

(8) a. the man [that [I saw who(m)]]

b. the man [who(m) that [I saw t]]

To (8b) the following rule of free deletion is applicable.

(9) In the domain COMP, delete where a is an arbitrary category

and <p an arbitrary structure.

Rule (9) permits deletion of any element dominated by COMP. This rule,
then, provides for all three examples in (7). In the case of (7a), only
who(m) is deleted. In (7b), only that is deleted, and in (7c), both who(m)
and that have been deleted.
There is a problem with this analysis. Since the deletion rule (9) is
optional, it need not apply to (8b), which results in the following ungram¬
matical example.

(10) *the man who(m) that I saw

To compensate for this unwanted result, a special filter is proposed.

 Chomsky/Lasnik Filters 71

(11) *[compw/?-phrase tp], <p f e

Now we must ask: Is (11) simply an ad hoc consequence of a bad analysis

or is there in fact some independent motivation for it? The answer to this
question turns out to relate to the former possibility. To see this it will be
necessary to proceed to additional examples.
In fact, we have already seen that further ad hoc theoretical conse¬
quences inhere in the Chomsky/Lasnik approach. Thus, pursuing this
analysis, we have a deep structure such as (12a) below, which is con¬
verted to (12b) by wh-movement and subsequently to the ungrammatical
(12c) by rule (9).

(12) a. the man [that [who met you]] is my friend

b. the man [who that [t met you]] is my friend
c. *the man met you is my friend

As noted in section 1, this result makes additional theoretical elaboration

necessary, namely filter (3).
At this point I think it is necessary to seriously consider the possibility
that the two filters, (3) and (11), are simply ad hoc consequences of a
faulty analysis. Certainly a shadow will have been cast over the whole
undertaking if additional devices are a consequence of the analysis involv¬
ing deletion of material in COMP.
There is further evidence which does show that the proposed analysis is
faulty. Consider the following example.

(13) the man to whom John spoke

This example will derive from (14a) by wh -movement and deletion of that,
as indicated in (14b) and (14c).

(14) a. the man [that [John spoke to whom]]

b. the man [to whom that [John spoke]]
c. the man [to whom [John spoke]]

The problem that now arises is this. Since the material in COMP is freely
deletable by (9), we could delete whom in both (14b) and (14c) to derive
the ungrammatical examples presented below.

(15) a. *the man to that John spoke

b. *the man to John spoke
72 Realistic Syntax

Now what ad hoc device is summoned to fix up this consequence of the

analysis? Here Chomsky and Lasnik invoke A/A holding of the deletion
rule.

(16) “We assume that this is excluded by the A/A Condition holding of
[9], since these elements, taken as members of the category a, are
included in larger elements of some category (a being arbitrary).”
[p. 446]

It would be a mistake to concede that A/A is independently motivated

and that therefore its application to (14b) and (14c) involves no additional
complication in the grammar. In the first place, we have no idea of what
precisely A/A is. The principle has not been precisely formulated and it
has been unclear just what motivation is accorded A/A in light of Choms¬
ky’s adoption of subjacency, Specified Subject Condition, and Tensed
S Condition, etc. Until and unless the full range of evidence for and
against A/A is supplied, together with a precise formulation of the
principle, no “independent motivation” can be conceded. Moreover,
there is every reason to believe that A/A simply leads to a further loss of
generalization in this case. Thus, (15a) and (15b) are blocked by A/A
holding of deletion in COMP. But the following examples will be out for
totally different reasons.

(17) a. *the man to that John saw

b. *the man to John saw

Now it is not A/A that would rule out (17a) and (17b) within the
Chomsky/Lasnik theory. Rather it would be lexical subcategorization,
since the verb see does not select a prepositional phrase. Hence the
situation could not arise which would allow the prepositional phrase to be
moved to the COMP position which would in turn be susceptible to dele¬
tion by (9), blocked by A/A. In other words, one principle, A/A, is in¬
voked to rule out (15a) and (15b), while a second distinct principle, sub¬
categorization, is invoked to rule out (17a) and (17b). It would be nice,
however, if the same general device could be utilized to rule out both (15)
and (17). The theory which will be advocated here does in fact allow for
the generalization.
Are there yet further ad hoc consequences of the Chomsky/Lasnik
approach? There are. Since elements in COMP are freely deletable by (9),
we may apply it to the complete prepositional phrase to whom in (14b) and
(14c) and thereby derive the following examples.
 Chomsky/Lasnik Filters 73

(18) a. *the man that John spoke

b. *the man John spoke

Or consider example (19a). Rule (9) could apply to delete of whom to yield
ungrammatical (19b).

(19) a. the boy pictures of whom John took

b. *the boy pictures John took

What new principle must be invoked to avoid these bad results? Here
Chomsky and Lasnik invoke the alleged principle of recoverability of
deletion.

(20) “We assume that this possibility is excluded by the recoverability

principle for deletion ... items from the lexicon cannot be deleted
unless they are explicitly mentioned in the deletion rule. Since the
contents of the w/i-phrases [in (14b), (14c), (19a)] include lexical
material, these phrases cannot be deleted.” [pp. 446-7]

Again, an apparently old principle is drawn on to account for the un¬

wanted consequences of the proposed theory, a free ride. But once again,
“independent motivation” cannot be granted since no explicit formula¬
tion of the principle is ever provided. Moreover, as Chomsky and Lasnik
note, there is an inconsistency involved in (20). If lexical material cannot
be deleted by rule (9) since it is not explicitly mentioned in rule (9), then
who(m) should not be deletable in (7b) to ultimately yield (7c). To over¬
come this inconsistency, Chomsky and Lasnik remark:

(21) “There is a residual semantic content in relative pronouns, namely

the feature Human ... But this (or any) residual semantic content
is redundant, since it is expressed in the head of the construction.
We assume that an appropriate concept of recoverability will be
restricted to lexical items, ignoring such redundant features.”
[p. 447, fn. 46]

We see then that yet a further device is needed to account for an

additional unwanted consequence of the proposed theory. Let us now
tabulate these results. The following theoretical devices are proposed as a
consequence of very simple relative clause constructions.
74 Realistic Syntax

(22) a. Free deletion in COMP, rule (9)

b. Filter (3), an ad hoc consequence of (12c)
c. Filter (11), an ad hoc consequence of (10)
d. A/A, an ad hoc consequence of (15)
e. Recoverability of Deletion, an ad hoc consequence of (18), (19b)

Now it will be clear to unbiased observers that something is indeed missed

in connection with the theory which must summon this range of ad hoc
devices to account for unwanted results which are a consequence of the
theory. Perhaps we can do better and go on to account for the generaliza¬
tion underlying (15) and (17) and the generalization underlying the exam¬
ples discussed in section 1, repeated below in (23).

(23) a. *The man met you is my friend.

b. *John is believed is hungry.
c. *John believes is hungry.

Recall that the Chomsky/Lasnik approach involves the following devices

to account for (23).

(24) a. Filter (3)

b. Tensed S Condition
c. Phrase Structure, Lexicon, etc.

It would be desirable to express the generalization underlying (23 a-c) and

to be able to dispense with ad hoc devices such as the Tensed S Condi¬
tion, the proliferation of filters of the Chomsky/Lasnik type, Recov¬
erability of Deletion, A/A, and in fact, all of the constraints and conditions
proposed within the framework of transformational grammar. There is in
fact some hope that this objective can be realized. In this particular case,
all of (23 a-c), in addition to other examples discussed in this section and
many examples adduced by Chomsky and Lasnik, turn out to be examples
which are functionally deviant and incomplete in the sense to be eluci¬
dated in the following section.

3. Functional deviance

Before sketching out an analysis which can express the generalizations

uncovered in the previous sections, let us review the relevant examples
 Chomsky/Lasnik Filters 75

and provide a naive idea of what is to be accomplished. Relevant exam¬

ples include the following.

(25) a. *The man met you is my friend.

b. *John is believed is hungry.
c. *John believes is hungry.
d. *the man who(m) that I saw
e. *the man to that John spoke
f. *the boy pictures John took

The basic idea is that expressed in section 1. We might look at all these
examples as concatenated lexical words. We attempt to build up syntactic
interpretations, employing rules of composition, which ultimately hinge
on the functional structure of the lexical items in question. In all of the
examples listed in (25) something goes wrong and a functionally complete
interpretation cannot be derived.
Consider example (25a). Here we can build up a syntactic interpretation
for the man met you and is my friend, but the two cannot be composed.
Similarly, it is not possible to compose John is believed and is hungry in
(25b). Again, in the case of (25c), John believes and is hungry cannot be
composed, and in the case of (25d), who(m) and that I saw cannot be
composed. Similarly, there is no way to compose to, which selects an NP
and that John spoke, which is not an NP. And finally (25f) represents an
example for which it is not possible to compose the parts with pictures.
The idea to be pursued here is that when such a situation arises where
the various parts of a string cannot be provided a complete syntactic
interpretation, then that string is functionally deviant with respect to the
attempted interpretation. The string is not a phrase in the language. Let us
now turn to a sketch of such a theory.
To express the generalization mentioned above, it is necessary to
forego all transformations. Although it is possible to base generate most of
the constructions of interest, cf. Brame (1976, 1978b), Bresnan (1978),
here I will be working directly off the surface. It will no longer be possible
to simply gloss over lexical specifications, since, as we shall see, the
functional structure associated with lexical specifications plays an inti¬
mate role in the interpretive process.
By way of providing an introduction to the approach advocated here,
let us provide partial lexical specifications for words such as the, man,
met, you, and John.
 76 Realistic Syntax

(26) the; D,_(N)

(27) man; N
(28) met; V,_NP°
(29) you; NPX
(30) John; NPy

Lexical entries consist of two components, a phonological specification

comprising the initial component, and a functional or syntactic specifica¬
tion comprising the second component, which will be called the functional
structure of the word in question. Superscripts in lexical specifications
can be taken to indicate a range of information, including grammatical
function, number, gender, etc. Here I will simplify and take the
superscript to indicate grammatical function alone. Thus, in (29) and (30)
we see that you and John function as NP arguments which vary according
to their function as subject, object, etc. By contrast, an NP such as I, for
example, must be specified lexically as a subject NP, viz. NPS. The actual
function of you and John may be determined within a given syntactic
context and several rules to effect this identification will be provided
below.
It should be noted that lexical words exhibit two basic types, arguments
and functions, and an adequate notation should reflect this distinction.
Thus, arguments such as you and John and man have simple functional
structures, whereas functions such as the, met, etc. have complex func¬
tional structures, consisting of an initial segment and a context. The con¬
text represents the argument structure selected by the function word. It
should be noted that functions, like arguments, also reflect grammatical
function and should also be superscripted, but for simplicity these details
are omitted here. A more adequate (but still impoverished) notation is
provided in Brame (1978b, 1979).
The basic syntax is provided by a general concatenation operator,
which gives rise to strings of concatenated initial components of lexical
entries, as provided below.

(31) the man met you

(32) the met you John man

We now wish to provide a means of building up a complete interpretation

for (31), while at the same time failing to provide a complete interpretation
for (32). To this end the following rules of composition are proposed.
 Chomsky/Lasnik Filters 77

(33) Rules of Composition

a. For each phonological word a, compose a as some cateogory B.
b. Compose A BjBa ... Bn as the complete second component of
the lexical specification of A, where A is a function term and Bj
is an argument term, o-^i^n.

Applying (14a) to (12), we obtain the following.

D N V NPX
(34) || ||

the man met you

The variable superscript x, indicating the grammatical function of NP

arguments, can be identified by rules which make reference to the position
of the NP in the concatenated string. For example, subject and object can
be expressed by rules such as the following.

(35) NPX \ NPS / __V

NP° / V_

Thus, the x corresponding to you in (34) will be filled in as o, indicating the

you functions as an object noun phrase argument.
We now apply (33b) and obtain the following.

D(N) V(NP°)

the man met you

It is assumed that the following equivalences hold.

(37) D(N) = NPX

V(...) = VP
VP(NPS) = S

Thus, (36) is equivalent to the following, where s is supplied by (35).

78 Realistic Syntax

VP

met you

In general, examples such as (36) and (38) will be written as follows.

(39) VP = V(NP°)

I I
the man met you

Since VP is a function term, we now wish to compose it with its argument

NPS. This can be accomplished with the following rule; it is also assumed
that the equivalence VP(NPS) = S is to be added to (37).

(40) Compose NPS VP as VP(NPS)

Applying (40) to (39), we obtain the following.

(41) VP(NPS) = S

the man met you

We now wish to check (41) for well-formedness. This involves an indexing

procedure such as the following.

(42) a. Let T be a term dominating a lexical category a. If (a; T, ... )e

Lexicon, then assign T a distinct index /, written T : i.
b. Let T be a term directly dominating terms A Bx B2 ... Bn. Then
substitute the indices of the latter for corresponding terms in the
equivalence of T.
c. If T= A(Bj ... Bn) and every term in the latter is indexed, then
index T.
 Chomsky/Lasnik Filters 79

Applying the indexing procedure to (41), we obtain the following.

(43) S = VP(NPS):h(k) = p

NPS = D(N):f(i) = k VP = V(NPU): g (j ) = h

D:f N:i V:g NP°:j

I I I
the man met you

It can be seen that all of the relevant functional structure in (43) has been
indexed. We shall refer to the collection of all such index structures at the
root of a phrase as an interpretation of that phrase. Thus, an interpretation
of the phrase the man met you is the following.

(44) p = h(k) - (g(j))(k) = h((f(i)) = (g(j))(f(i)), where f = the, i = man,

etc.

With this background it is now possible to define incompleteness.

(45) Incompleteness
A string a1 ... an is incomplete with respect to an interpretation <p
provided there is a term T in <p corresponding to no index or pro¬
vided there is no index in corresponding to ax for some i,
1 ^ i ^ n.

A string is said to be completely interpreted with respect to an interpreta¬

tion <p provided the string is not incomplete. In the case of our example,
the man met you is complete with respect to the interpretation provided in
(44), Of course the same example may well be incomplete with respect to
a different interpretation, where the interpretive procedures were applied
in a way incompatible with the result given in (43). It is therefore impor¬
tant to bear in mind that incompleteness as provided in (45) is a relative
notion. We may now define functional deviance as follows.

(46) Functional Deviance

A string ax ... an is functionally deviant if it is incomplete under all
interpretations.

The example provided in (32) is functionally deviant in the sense of (46).

80 Realistic Syntax

There is simply no way to compose all its parts so as to be assigned a

single index p.

4. The CONTROVERSIAL EXAMPLES

Let us now return to examples such as (25a), repeated below as (47).

(47) *The man met you is my friend.

It should be clear that this example is deviant with respect to any at¬
tempted interpretation. A complete interpretation can be provided for the
man met you as well as for is my friend, but there are no provisions for
composing the resulting S and VP terms to yield a reading for the com¬
plete string in (47). Consequently, as a single phrase, (47) is incomplete.
This is also true of example (25b), repeated below as (48).

(48) *John is believed is hungry.

We can take believed to be a lexical word. Like many passive participles,

believed selects an optional agentive phrase argument, but never a VP
function. Hence, we are unable to compose John is believed and is hungry
so as to provide ultimately an index for the complete string in (48). It
follows that this example too is functionally deviant.
Precisely the same consequence holds for example (25c), repeated
below as (49).

(49) *John believes is hungry.

The phrase is hungry is not an argument term, but rather a function. The
lexical item believe is specified to select various types of arguments, for
example, NP arguments, sentential arguments, but never VP functions.
Hence, John believes and is hungry cannot be composed and con¬
sequently example (49) is incomplete.
We see that all three examples are generalized within the proposed
framework, whereas they are treated differently within the Chomsky/
Lasnik framework. Since example (47) is ruled out on grounds of incom¬
pleteness, it follows that filters such as (3) are not needed. Such a filter is
simply a special case of incompleteness, and not a very interesting one at
that. But this is also true of other examples. Thus, there is no reason for
adopting a filter to rule out (11), repeated here as (25d).
 Chomsky/Lasnik Filters 81

(50) *the man who(m) that I saw

The lexical item who selects a sentential argument S, not an S argument.

It thus follows that who(m) and that I saw in (50) will not be properly
composed to permit indexing to proceed in terms of Operator Binding, cf.
Brame (1978). As a consequence, (50) will be functionally deviant. The
net result is that the proposed filter (11) is unneeded. It is simply a special
ad hoc instance of functional deviance, and again not a very interesting
one.
Again, in the case of (15a) and (15b), repeated below as (51), we en¬
counter examples which are functionally deviant.

(51) a. *the man to that John spoke

b. *the man to John spoke

The lexical item to selects an object of preposition NP argument. In

neither of (51) will this lexical restriction be satisfied so that proper index¬
ing can be assigned to yield an interpretation for the complete strings.
Hence these examples too are incomplete.
Now let us return to the examples provided in (17), repeated below as
(52) .

(52) a. *the man to that John saw

b. *the man to John saw

Recall that these examples will be ruled out by completely different

mechanisms than the A/A principle ruling out (15a) and (15b). In section 2,
however, I asserted that they should be out for the same reason. Now it is
easy to see that in fact (52a) and (52b) are out for the same reason,
namely, to cannot be composed with that John saw or John saw to permit
proper indexing. Thus, the generalization can be expressed within the
proposed framework.
Now consider examples such as (18a) and (18b), repeated below as (53).

(53) a. *the man that John spoke

b. *the man John spoke

These examples, recall, would be derived under the Chomsky/Lasnik

analysis, which, as noted above, occasions a new principle, Recov¬
erability of Deletion. However, these examples are functionally deviant
82 Realistic Syntax

under the analysis I am advocating. To see this, it is necessary to provide

a lexical specification for the word that. It can be provided a lexical
specification with a multiple functional structure relating to its com¬
plementizer and relative functions.

(54) that; C, _(S) or NPX, Frel(__)(S)

According to (54), that can function as either a true complementizer in¬

troducing sentential arguments or else as a relative pronoun which is an
operator suffix of the relative operator Frel(NPx), as in Brame (1978b).
Let us now consider the following well-formed pair.

(55) a. the man that John met

b. the man John met

The following interpretations can be built up for the man and that John
met, respectively.

(56) NPX ; i where i = the man

(Frel(NPx))((V(NP°))(NPs)) : Frel(x)((f(NP°))(j))
where x = that, f = met, j = John

To (56) Operator Binding applies, identifying x with NP° and indexing the
prefix Frel, whereupon x is linked to the lexical head index i by a rule of
anaphora. The net result is that (55a) receives a complete interpretation.
(Note that there is no rule of extraposition from NP since such a rule is
replaced with the rule of anaphora linking the relative pronoun with the
index of the lexical head.)1
The same result holds for (55b) provided we posit a rule which will
ensure that Operator Binding applies. Such a rule would have the effect of
inserting a relative operator Frel(NPx) just in case an NP is adjacent to an
S, as follows.

(57) $ -> Frel(NPx) / NP_S

By anaphora the operator suffix NPX will again be identified with the
lexica] head index and Operator Binding will ensue, providing for a com¬
plete interpretation. This analysis automatically explains the following
range of data.
 Chomsky/Lasnik Filters 83

(58) a. the gun that I bought went off

b. the gun which I bought went off
c. the gun I bought went off
d. the gun went off that I bought
e. the gun went off which I bought
f. *the gun went off I bought

Example (58f) cannot be provided a single index because it is not possible

to link I bought with anything preceding it since rule (57) is inapplicable.
Also it is not possible to identify the object argument of bought since
Operator Binding will be inapplicable. Hence (58f) is functionally deviant.
This analysis will also explain the following distinctions.

(59) a. the man that I talked with

b. the man who(m) I talked with
c. the man I talked with
d. the man with whom I talked
e. *the man with that I talked

Because that is not lexically specified for the relevant complex functional
structure, it follows that (59e) will not receive a complete interpretation.
By contrast whom is provided the following partial lexical specification.

(60) whom; NPX, Frel((P(_)))(S)

Of course, for those speakers who employ the prescriptive dialect as

opposed to the formal dialect, i.e. allowing whom in (59b) ihstead of the
natural who, (60) must be complicated, cf. Brame (1978b, fn. 7). Others,
who speak the modern dialect do not include whom anywhere in their
lexicon, from which it follows that examples such as (59d) are not possi¬
ble.
Let us now return to the examples provided in (53). It is easy to see that
these examples can be provided variable indices for the relative pronoun
and it is equally clear that Operator Binding will not be applicable. This
follows from the fact that spoke selects a PP argument whereas that
functions as an NP argument, not as a PP argument. It follows that the
conditions associated with Operator Binding are not fulfilled, cf. Brame
(1978b), and consequently we are left with a functionally deviant string.
This implies that we can do without the recoverability of deletion princi¬
ple which Chomsky and Lasnik must employ. In sum, the generalization
 84 Realistic Syntax

can be expressed within the proposed analysis. Within the framework of

the filtering approach, it cannot.
All of the ad hoc devices listed in (22), and others besides, can be
eliminated when we view matters in the light of incompleteness. Each and
every phonological word has its associated functional specification. If
rules of composition apply so as to appropriately compose these words to
allow indexing of the root, we will derive a complete interpretation.
Otherwise we will be left with a functionally deviant example. All of the
deviant examples discussed by Chomsky and Lasnik which they claim
occasion a number of filters are treated in a straightforward manner in
terms of incompleteness. Indeed filters can be viewed as special cases of
functional deviance.

5. A HITHERTO MISSED SYNTACTIC GENERALIZATION

There is a general feature of the analysis advocated here which is not a

property of syntactic theories such as transformational grammar. Trans¬
formational grammar is a theory of sentence grammar and is, as such,
unnecessarily restrictive. This fact has important consequences, for it
gives rise to a new ad hoc filter. By contrast, the theory advocated here is
a theory of syntactic phrasal types, including sentences as only one of the
wide variety of syntactic types. The following are all well-formed syntac¬
tic types.

(61) a. met Mary

b. with a knife
c. John met Mary
d. that John saw
e. whether Mary ate the apple
f. John asked whether Mary ate the apple

The theory advocated here allows for the interpretation of finer syntactic
types than S’s. Thus, in the case of (61a), for example, met Mary can be
provided a complete interpretation qua VP, but, of course, not as S.
Similarly, with a knife in (61b) can be provided a complete interpretation
qua PP, given appropriate lexical specifications, but not as S. In (61c),
John met Mary can be provided a complete interpretation as S, but not as
VP or PP. Now, examples (61d) and (61e) are like (61a) and (61b) in that
they can be provided complete interpretations which are not S, but rather
 Chomsky/Lasnik Filters 85

finer syntactic types. But, clearly, (61d) and (61e) are not ungrammatical.
They are complete as S, as opposed to S.
We readily see that because transformational grammar views syntax in
a distorted way, special devices must be invented to rule out (61d) and
(61e). Chomsky and Lasnik propose the following filter.

(62) *fcCOMP NP ...], where S is a root sentence

Chomsky and Lasnik claim that (62) implies that deletion of that by rule
(62) is obligatory. In fact, the filter (62) is simply an ad hoc consequence of
failing to view grammar in the general way which has been suggested
above, i.e. as a theory of syntactic types, not as a theory of sentence
grammar. Examples such as (61 d) and (61e) are no more deviant than are
the phrasal types in (61a) and (61b), or, for that matter, than the sentences
in (61c) and (61 f).
Not only is it possible to formulate the theory to encompass finer syn¬
tactic types than S, we can go on to provide for coarser syntactic rela¬
tions, including, for example, the question/answer relation, the theme/
topic, relation, and others. Some preliminary work along these lines is
provided in Brame (1978c).
In conclusion, yet another filter can be dispensed with when syntax is
viewed in a nontransformational framework such as that advocated here.
This is again a welcome result since we are always interested in dispens¬
ing with excess theoretical machinery. Unlike the previous examples dis¬
cussed, however, the crucial examples here are not deviant, in spite of
claims to the contrary in terms of (62).

6. Lexical integrity

Let us now return to the Chomsky/Lasnik approach and in particular to

filter (3). Recall that a range of ad hoc devices, listed in (22), are required
to force through the analysis involving free deletion of elements in COMP.
There are, however, additional problems that accrue to the analysis.
Thus, the following examples are cited in the Chomsky/Lasnik paper.

(63) a. the fact that John was here surprised me

b. *the fact John was here surprised me
(64) a. it came as a surprise to me that John was here
b. *it came as a surprise to me John was here
86 Realistic Syntax

(65) a. it is unlikely that John is here

b. *it is unlikely John is here
(66) a. I pleaded with Bill that he shouldn’t hire her
b. *1 pleaded with Bill he shouldn’t hire her
(67) a. John believes (that) Mary saw Sam, and Bill, that Sue saw
Harry
b. *John believes (that) Mary saw Sam, and Bill, Sue saw Harry

The deviant (b) examples result from applying rule (9) within the
Chomsky/Lasnik framework. Once again, we see that the analysis raises
more questions than it answers. New complications are needed to cover
the (b) examples, and these complications amount to a redescription of
the problems themselves. The complication is incorporated into filter (3)
under the disguised assertion that filter (3) “is insufficiently general” [p.
484], as given below.

(68) *[aNP tense VP], unless a is adjacent to and in the domain of [+V],
that, or NP

I think it is clear that this “generalized” version of (3) is in fact a compli¬

cation. The true ad hoc nature of (68) becomes readily apparent when
Chomsky and Lasnik recognize that (68) predicts that examples which are
similar to their old example (la) are predicted to be grammatical. Such
examples differ from (la) only in that the complement is adjacent to [+V],
as indicated below.

(69) *1 saw the man met you <= I saw the man (who, that) met you

The “generalized” filter (68) predicts that (69) is grammatical since the
putative complement the man met you is adjacent to and in the domain of
a verb, i.e. [+V], after deletion of the complementizer elements. To cor¬
rect for this additional flaw, the more complex filter (70) is proposed.

(70) *[aNP tense VP], unless a NP and is adjacent to and in the

domain of [+F], that, or NP

There can be little doubt. Filter (70) is not a generalization in any natural
sense of that term. It indicates, once again, that something has been
missed.
 Chomsky/Lasnik Filters 87

There is an extremely telling criticism of the Chomsky/Lasnik approach

which goes to the heart of their dilemma. This involves their change of
[ + V] in (68) to [+F] in (70). What is the reason for this alteration? The
answer lies in examples such as the following.

(71) a. Bill wondered that Dennis was such an outstanding guard.

b. *Bill wondered Dennis was such an outstanding guard.
(72) a. I am sad that you don’t understand.
b. *1 am sad you don’t understand,
cf. I am glad you don’t understand.
(73) a. She demanded that he leave.
b. *She demanded he leave.

Chomsky and Lasnik recognize that many predicates do not allow free
deletion of that, although they claim that deletion of that is “fairly free’’
[p. 486]. Statistics are clearly irrelevant. We must ask just how they would
account for the deviance of the (b) examples in (71)—(73). What they do is
posit an abstract feature F, which they take to be a “subfeature of [+V]”
and include F in (70), assuming that F “characterizes verbs and adjectives
that permit the structure resulting from deletion of that."
This particular maneuver reveals quite clearly that Chomsky and Las¬
nik are attempting to transfer truly lexical phenomena to the domain of
filters, a move which is reminiscent of the generative semantics attempt to
force lexical phenomena into the domain of transformations, cf. Brame
(1976, Part I). As they note, whether or not a verb requires a that com¬
plementizer is to some extent “erratic”, but they fail to note that the
lexicon is traditionally the repository of erratic and idiosyncratic prop¬
erties of lexical items, as emphasized by Chomsky himself in earlier work.
Once this is realized, it becomes evident that the presence or absence of that
should in fact be expressed in the lexicon as part of the argument structure
associated with the verbs and adjectives in question, including perhaps
other items, such as modals, cf. Bresnan (1972). Thus, that itself should
be a lexical item and should be available for selection. The example listed
in (73b) is another example of an incompletely interpreted string, where
the argument structure of demand is not satisfied, implying that she de¬
manded and he leave cannot be composed so as to be appropriately in¬
dexed. Thus, demanded he leave is out for very much the same reason as
with he leave, as indicated in (74).

(74) a. *demanded he leave

b. *with he leave
88 Realistic Syntax

In both cases, we are left with incompletely interpreted results. The lexi¬
cal item demand selects an S argument, not an S argument. The lexical
item with selects an NP argument, not an S argument.
By failing to recognize that complementizers are lexical items on a par
with other function words, Chomsky and Lasnik fail to note generaliza¬
tions such as those indicated in (74) and completely miss the parallelism
between the following sets of examples.

(75) a. that John left assumes John left

b. *that to leave *assumes to leave

In (75) we see that the function words that and assumes are similar in that
both select an S argument, but not a VP argument. This suggests that that
is indeed a lexical item.
The fact that complementizers are lexical items and exhibit meaning
was recognized in Bresnan (1970) and discussed in great detail in Bresnan
(1972). Bresnan offers a range of cogent arguments to show that com¬
plementizers are not the meaningless elements that transformationalists
have assumed, beginning with Rosenbaum (1967). That complementizers
must be lexical items is a consequence of the Spelling Prohibition, cf.
Brame (1978a), although it must be noted that the Spelling Prohibition
itself follows from the theory advocated here. It seems that the basic point
should be emphasized again—words should not be abused. Each and
every one of them must be provided a phonological representation to¬
gether with a functional specification. None can be spelled out, deleted, or
altered by processes such as transformations.

7. Extraposition and gapping as anaphora

It remains to discuss and refute several additional filters proposed

within the framework of the theory under critique and to account for the
deviance of the (b) examples provided in the previous section under (63)-
(67), repeated below.

(76) a. *the fact John was here surprised me

b. *it came as a surprise to me John was here
c. *it is unlikely John is here
d. *1 pleaded with Bill he shouldn’t hire her
e. *John believes (that) Mary saw Sam, and Bill, Sue saw Harry
 Chomsky/Lasnik Filters 89

Of these examples (76d) is out for precisely the same reason as examples
such as the following.

(77) a. *She demanded he leave.

b. *The law requires he have a permit.
c. *1 insist she return promptly.

As noted in the previous section, lexical items such as demand, require,

and insist select S arguments in lexical representations, not S arguments.
Hence the examples listed in (77) are all functionally deviant and incom¬
plete. Now plead is another of the verbs which does not select S (although
it like many other verbs selects additional argument structure, e.g. NP, as
in John pleaded the case).
We might expect that the remaining examples in (76) are also function¬
ally deviant. In fact such an analysis is indeed what will be proposed here.
A cursory sketch of the idea behind an account of (76a)-(76c) is the
following. For adjuncts to lexical heads such as the fact, the idea, the
claim, etc., which are not relative clauses, we might take the index of the
complete S argument and set it equal to the index of the lexical head.
Thus, assume that that John was here is indexed by p and that the fact is
indexed by i. Then it is assumed that p is set equal to i. The operation
would constitute part and parcel of the operation of composing the S and
the lexical head. We could then provide a complete interpretation for the
fact that John was here. However, an example such as (76a) would
emerge as functionally deviant for failure to allow composition of John
was here and the lexical head the fact. This follows from the fact that John
was here is not 5, whereas S, not S, is required for proper indexing to
ensue.2
An analysis of (76b) and (76c) can be provided in the same spirit. Fol¬
lowing Emonds (1970) we assume that the “extraposed” clause does not
originate in the subject position. In fact, it could not since all transforma¬
tions are proscribed in the proposed theory. Now, as an alternative to
extraposition, or Emonds’ intraposition, we may simply identify the index
corresponding to the subject pronoun it with the adjunct S clause! How¬
ever this identification is allowed only in terms of 5, never of S. It follows
that examples such as (76b)-(76c) cannot be provided a single index and
hence be interpreted as a single phrasal type. Although it came as a
surprise to me and John was here can be interpreted individually, there is
no way to link the two, since the latter is not interpreted as 5, and hence
cannot be identified with the variable index corresponding to it. Although
it is unlikely and John is here can be interpreted individually, again they
90 Realistic Syntax

cannot be linked due to the fact that the latter is not 5, and hence cannot be
identified with the index corresponding to it.
The suggested analysis has much in common with the original idea
proposed in Emonds (1970), with the inverse analysis proposed in Brame
(1976), and with the suggestions found in Jayaseelan (1979). Thus,
Emonds assumed that the adjunct 5 was substituted for it providing they
are coindexed. Within the framework offered here, there is no need for a
rule of extraposition or intraposition. In the case of well-formed examples
such as it is obvious that John left, the necessary link will be effected by a
rule of anaphora identifying the index of it with that of that John left. In
the case of that John left is obvious, no identification is necessary. This
kind of analysis, incidentally, makes ad hoc constraints such as Ross’
Bounding or Chomsky’s Subjacency unnecessary, as will be shown in
forthcoming work.
Thus, we are left with example (76e), which should also be incomplete
under the assumptions of the proposed framework. Within the theory
advocated here, Gapping can be viewed in a more interpretive spirit. In
particular, we may assume that function terms such as V can be inserted
by rule, to be followed by the interpretive procedures and, crucially, a
rule of anaphora which associates the index of the left conjunct of the
gapping structure with the interpreted gap. The surface analogue of Gap¬
ping might be tentatively formulated as follows.

(78) </> —> V / Aj_A2, where A is an argument term j= S

Now let us consider a simple example such as (79).

(79) John eats spinach and Mary, kale.

We can proceed to derive a complete interpretation for (79) as follows.

(80) ((V(NP°))(NPS) Con NPX NPX by Composition

((f(i))(j) g k 1 and indexing
where f = eats, i = spinach, j = John, g = and, k = Mary, 1 = kale
(81) ((V(NP°))(NPS) Con NPS V NP° by (78) [and rule (35)]
(82) ((V(NP°))(NP0) Con ((V(NP°))(NP°) by further Composition
mm) g C(V(l))(k) and indexing
(83) mm) g «f(l))(k) by Anaphora
(84) Con(Sl5 S2) by further composition
h(p, q) and indexing
where p = John eats spinach, q = Mary eats kale, h = and
 Chomsky/Lasnik Filters 91

This analysis allows us to insert a verbal function between any two argu¬
ment types with the exception of S. That other argument types are in¬
deed relevant is indicated by the following examples.

(85) a. John ate in the kitchen and Bill, in the dining room. A2 = PP
b. John ate slowly and Bill, quickly. A2 = Adv
c. John wants to leave and Bill, to return. A2 = VP
d. John said that Bill left and Bill, that Mary returned. A2 =~S
e. In the kitchen are two spoons and in the dining room, two
knives. Ax = PP
f. That John failed is surprising and that Mary passed, incredi¬
ble. Ax = S
g. To err is human, to forgive, divine. Ax = VP
h. John ate spinach and Bill, kale. Ax = NP

It is interesting to speculate why S cannot be considered as A2 in (78),

which gives rise to the functional deviance of the Chomsky/Lasnik exam¬
ple (76e). Perhaps this indicates that S is not to be taken as an argument
term after all. but rather a special kind of term, which we might call a
proposition. (Complementizers could then be viewed as functions which
convert propositions into arguments.) If we then adopted the function/
argument notation of Brame (1978b, 1979), the conditions associated with
(78) could be eliminated.
There are many interesting problems and puzzles associated with gap¬
ping structures. My desire here is not that of providing a careful analysis
of the complete range of such examples, but rather that of suggesting that
a plausible alternative exists within the proposed interpretive framework
which can provide an account of the deviance of examples such as (76e).3

8. The missing subject problem

Let us now turn to the following examples discussed by Chomsky and

Lasnik.

(86) a. Who do you think — saw Bill?

b. *Who do you think that_saw Bill?

Chomsky and Lasnik incorporate into their framework an analysis which

is due to Perlmutter (1971), as reflected in the following statement.
92 Realistic Syntax

(87) Perlmutter’s Filter: Any sentence other than an Imperative in which

there is an S that does not contain a subject in surface structure is
ungrammatical.

According to Perlmutter, examples such as (86b) are ruled out by (87).

The obvious question to ask is how Perlmutter keeps (87) from applying
to rule out (86a) as well. His answer is this: “In order to account for
these examples, we must assume that when that is deleted, the S-node
above the embedded sentence is pruned away,” [p. 112].
Chomsky and Lasnik follow the lead of Perlmutter’s suggestion and
posit a filter within the framework of their approach. This filter is recorded
below.

(88) * [that [NPe]]

They note that (88) would lead to deviance in examples such as (89).

(89) a. the man [that [[NPe] saw Bill]

b. a book t arrived [5 that [NPe] may interest you] (t the trace of S)

As a consequence of such examples, they alter (88) as (90).

(90) *[g that [ nPc] ...], unless S or its trace is in the context. [^P NP_...]

They go on to suggest that (90) is part of universal grammar and that

therefore it holds of all languages (where that is suitably replaced, e.g. by
-WH).4
Now let us see how these examples can be treated within a framework
such as that advocated in the preceding pages. As one might suspect,
examples such as (86b) will not be provided a complete interpretation.
They will emerge as functionally deviant. Thus, we must propose an
interpretive rule which allows us to insert a subject argument in the case
of examples such as (86a), but not in the case of (86b). In this way, as we
shall see, examples such as (86a) will be provided a complete interpreta¬
tion. The relevant rule is provided in (91).5

(91) </> NPS / V__VP

With this rule, it is possible to provide complete interpretations for exam¬

ples such as (86a), but not for (86b). Thus, after composition, indexing,
and application of (91), we derive the following partial interpretation.
 Chomsky/Lasnik Filters 93

(92) (FQ(NPX))((V(VP(NPS)))(NPS)
(Fq(x))((f(g(NPs)))(i))
where x = who, f = think, g = saw Bill, i = you

Notice that NPS, supplied by rule (91) before the related composition of it
with its associated VP, is unindexed. Now it, as well as the operator prefix
Fq can be indexed if we apply the rule of Operator Binding proposed in
Brame (1978). Thus, NPS will be identified as x and Fq will receive some
index h, yielding a complete interpretation. Since rule (91) does not apply
to the partial interpretations related to example (86b), it follows that the
various parts of such examples cannot be composed so as to allow appro¬
priate indexing. It follows that such examples are functionally deviant.
We do not therefore need to appeal to filters such as (90) which make
subtle distinctions between traces and nothing.
Perlmutter noted that there are languages in which sentences analogous
to (86b) are in fact grammatical, citing examples from Spanish such as the
following.

(93) a. ^Quien dijiste que_salio temprano?

*Who did you say that_left early?
b. iQue dijiste que_paso?
*What did you say that_happened?

Unlike English, such examples in Spanish are well-formed. Perlmutter

went on to correlate these facts with the fact that languages such as
Spanish, in contrast to languages such as English and French, allow dele¬
tion of subject pronouns, as illustrated below.

(94) a. Salio temprano.

*Left early,
b. Paso anoche.
* Happened last night.

Perlmutter’s correlation is paraphrased in (95).

(95) Perlmutter’s Correlation: If a language does not allow deletion of

subject pronouns, it will not allow sentences such as (93).

To put the correlation contrapositively, if examples such as (93) are

grammatical in some language, then that language must allow deletion of
subject pronouns.6
94 Realistic Syntax

Chomsky and Lasnik accept this correlation and attempt to explain it in

the following way. They make a distinction between LPe], “a category
with a null content” and “nothing”, which is the “result of deletion” and
claim that those languages which allow free deletion of pronouns also
allow deletion of LPe] as part of the same process. They say that “the
rules of mental computation ‘see’ LP e]” in terms of the filter (90), but that
“this element is no longer ‘visible’” when deleted. The alleged explana¬
tion therefore rests on a very subtle distinction between trace and noth¬
ing.
Now if it is true that there is a correlation here, we must ask how it
could be expressed in interpretive theories such as that advocated here. In
fact, the correlation can be expressed, assuming that it exists. Thus, in
Brame (1978b) it was noted that it is no longer necessary to assume that
languages such as Spanish incorporate rules deleting subject pronouns.
Instead we posit an interpretive rule inserting NPS in the subject position
of such languages, whereupon the indexing of NPS will be determined in
part by discourse considerations (and person of the associated verb). Now
it would be this rule which could be utilized to insert the NPS argument in
the Spanish examples provided in (93), whereupon, once again, Operator
Binding could ensue to provide a complete interpretation. In this way, the
correlation is expressed.
In conclusion, we see that filter (90) is unneeded within the proposed
framework and we need not assume that traces in addition to subject
pronouns are “deleted”, a result which we would deduce in any case,
given the Spelling Prohibition, cf. Brame (1978a).

9. For-for and for-to

Let us now turn to the following pair of filters proposed by Chomsky

and Lasnik.

(96) * [for-for]

(97) * [for-to]

They posit filter (96) to account for examples such as the following.

(98) a. *We hoped for for John to win.

b. What we hoped for was for John to win.
 Chomsky/Lasnik Filters 95

They assume that in (98b) the subject derives from “hope for + for
clause” by Wh Movement. Hence filter (96) is invoked to account for
(98a) when movement does not take place.
One ad hoc consequence of this analysis is the fact that the trace t
purported to be left behind by movement rules plays no role in the ac¬
count of these examples, whereas it does in others, for example in filter
(90). Whether or not the trace has a residual effect then is simply coinci¬
dental within the Chomsky/Lasnik framework.
More to the point by way of criticism, however, is the fact that hope for
does not select a/br-complement, in spite of the Chomsky/Lasnik claim to
the contrary. Thus, hope for selects an NP, cf. John hoped for that result,
and the object of hope for in (98b) is not the/or-clause, but rather what,
which is in fact an NP. Now this NP’s index will be identified as the index
associated with for John to win. This identification should be effected by a
rule of anaphora.
This kind of analysis immediately explains the following facts in addi¬
tion to the facts listed in (98a) and (98b).

(99) a. What John thought about was that Bill left,

b. *John thought about that Bill left.
(100) a. What John argued against was that Harry should be present,
b. *John argued against that Harry should be present.

Here we see an obvious parallelism with the examples listed in (98).

Clearly, filter (96) has nothing to say about these examples, however,
indicating rather strongly that such an analysis misses the point. By con¬
trast, the anaphora approach should express the generalization.
This discussion indicates that there should be no rules deleting preposi¬
tions, contrary to common practice within the transformational
framework. Thus, contra Chomsky and Lasnik, examples such as (101a)
do not derive from (101b) by deletion of a preposition for.

(101) a. We hoped for John to win.

b. We hoped for for John to win.

Similarly, example (102a) does not derive from (102b), in spite of the
existence of examples such as (102c).

(102) a. We want Bill to win.

b. *We want for Bill to win.
c. We want very much for Bill to win.
96 Realistic Syntax

To my ear, (102b) is clearly deviant. Under the Chomsky/Lasnik analysis

one might expect it to be grammatical, given their analysis of examples
such as (98) and the fact that the following sentence is fine.

(103) What we want is for Bill to win.

Example (103) again shows that the object of want is what, an NP, which
is identified with the index offor Bill to win. Thus, the difference between
(102b) and (102c) must be expressed in the lexicon and reside in the fact
that in (102c) want selects an adverb plus complement, whereas in (102b),
there is a violation. In example (102a) want selects an NP followed by a
VP complement. Chomsky and Lasnik pursue the faulty sentential
analysis of such complements which has been shown to be incorrect, cf.
Brame (1976, 1978a) and Bresnan (1978). Thus, Chomsky and Lasnik
assume a structure resembling that of (104a) instead of the correct struc¬
ture (104b).

(104) a. We want [Bill to win]

b. We want [Bill] [to win]

Let us now turn to the filter provided in (97). This filter is designed to
rule out examples such as the following.

(105) *We want for to win.

Chomsky and Lasnik claim that examples such as (105) can in fact arise
since “NPs can appear freely in the subject position of the infinitival
complement ...” But this claim is simply false, since VP's need not have
subjects at all. Rather VP complements are lexically selected by many
verbs, including want.
We have, in fact, an immediate lexically based explanation for the
deviance of (105), namely, for, like that, does not select a VP argument.
Thus, proper interpretation cannot ensue in the case of the surface string
provided in (105). This, incidentally, allows us to provide a general lexical
account of a range of examples, including the following.

(106) a. *John thought about to leave.

b. *Bill argued against to buy chenille.
c. * Harry thought about to own a Volvo.
 Chomsky/Lasnik Filters 97

Notice, once again, the parallelism between the deviant examples in (106)
and example (105). This shows that for, like about and against, does not
select VP complements, not that a filter such as (97) is operative.
Chomsky and Lasnik lose sight of such lexical generalizations by forcing
restrictions into the domain of filters.
We see, then, that we can dispense with (96) and (97), which far from
solving problems of English syntax, lead to loss of linguistically significant
generalizations.7

10. Infinitival constructions

We are now left with only two filters. One of these is the following.

(107) *[V adjunct NP], NP lexical

This filter is designed to account for examples such as the following.

(108) *John believes sincerely Bill to be the best man.

Here Chomsky and Lasnik once again incorrectly take Bill to be the best
man to be a single constituent, when in fact Bill should be taken as an NP
complement and to be the best man as a separate VP argument, as noted
in the previous section, cf. Brame (1976) and Bresnan (1978).
A second set of examples treated by (107) is provided below.

(109) a. *1 believe sincerely John,

b. *1 like very much John.

Now in fact examples such as (109a) and (109b) can be interpreted when
there is sufficient pause between the adverb and the following constituent,
as is often required in the case of complex NP’s such as the following.

(110) a. I believe sincerely the girl who told me that she could drink
more jenever than anyone else.
b. I like very much the girl who saved the cat from the mouse.

In (110) the adverbs are provided a special intonation contour and are
followed by short pauses. The resulting examples are interpretable.
98 Realistic Syntax

Now it is interesting to note that a single pause associated with the

adverb is not sufficient when two distinct arguments follow the adverb as
shown below.

(111) a. *1 put quickly the car that everyone likes into the garage,
b. *1 called loudly the boy who you saw yesterday to dinner.

In fact, the only way that these examples can be interpreted is to provide a
second pause associated with the final element of the first argument fol¬
lowing the adverb. But this is precisely the situation that we encounter in
connection with the NP VP constructions as illuminated below.

(112) *John believes sincerely the boy who you saw yesterday to be the
best man.

This example, like those in (111) which contain two distinct arguments,
cannot be interpreted, unless, of course, as with (111a) and (111b), a
second pause is associated with yesterday. This seems to indicate that NP
VP constructions should indeed be viewed as two distinct arguments of
the verb and not as a single S argument as in previous transformational
work.
We might attempt to explain these data with the theory advocated here
by taking adverbs such as sincerely to be lexical functions applying to true
VP’s or to S’s, but not to arguments such as NP’s and VP’s. In examples
such as (109a) and (109b), without appropriate pauses, the adverbs will be
composed with the NP’s so that proper indexing cannot proceed. It is
necessary to have an appropriate intonation contour and pause available
for the adverb to be sensitive to the VP or S. In other words, it is neces¬
sary to break the link between the adverb and the following constituent in
order for the composition rules to be sensitive to this distinction. It ap¬
pears that we can do without filter (107), which resembles the type of
phenomenon discussed in Ross (1967).
We are now left with one remaining filter which has not been discussed.

(113) *[aNP to VP], unless a is adjacent to and in the domain of [-N] or

« = NP (where the feature [-N] holds of verbs and prepositions)

This filter is slightly revised after (107) is introduced but this alteration is
not crucial for the discussion which will follow. Filter (113) is designed to
account for the (b) examples of the following.
 Chomsky/Lasnik Filters 99

(114) a. his plan for Bill to win

b. *his plan Bill to win
(115) a. It bothers me for Bill to win
b. *It bothers me Bill to win
(116) a. It is illegal for Bill to take part
b. *It is illegal Bill to take part
(117) a. I want very much for Bill to win
b. *1 want very much Bill to win
(118) a. There is someone at the door for you to play with
b. *There is someone at the door you to play with

Now there is an obvious parallelism between many of these examples and

examples which were discussed earlier. For example, notice the similarity
between (63) and (114), repeated below as (119).

(119) a. the fact that John was here

his plan for Bill to win
b. *the fact John was here
*his plan Bill to win

This obvious parallelism goes unexpressed within the Chomsky/Lasnik

framework since the fact John was here is ruled out by one mechanism,
filter (70) and his plan Bill to win is ruled out by a second, filter (113). Now
earlier it was suggested that the fact John was here was out for failure to
permit composition of its parts, the fact WtfhJohn was here. Exactly the
same now can be said with respect to his plan and Bill to win. Thus, there
is no need for filter (113); the examples provided in (119b) are functionally
deviant.
A similar parallelism is obvious when we compare examples such as
(115) and (116) with the earlier (64) and (65), repeated below for clarity.

(120) a. It came as a surprise to me that John was here.

It bothers me for Bill to win.
b. *It came as a surprise to me John was here.
*It bothers me Bill to win.
(121) a. It is unlikely that John is here.
It is illegal for Bill to take part,
b. *It is unlikely John is here.
*It is illegal Bill to take part.

Once again Chomsky and Lasnik miss this parallelism by utilizing the two
100 Realistic Syntax

different filters to rule out the deviant examples. By contrast, within the
proposed framework, it follows that (120b) and (121b) are out for the same
reason; they are all functionally deviant. This deviance follows from the
fact that the anaphora analogue of extraposition applies to identify the
indices of arguments such as S (or VP) with the subject it, but not S
propositions or NP VP argument sequences!
There appears to be a parallelism also between the examples provided
in (117) and those provided earlier in (66), repeated below.

(122) a. I pleaded with Bill that he shouldn’t hire her.

I want very much for Bill to win.
b. *1 pleaded with Bill he shouldn’t hire her.
*1 want very much Bill to win.

As noted in section 6, plead must be lexically specified to take subjunctive

S complements. Similarly, in (122) want must be lexically specified to
select a/or-complement, (NP) VP complement, or Adv/or-complement,
but not an Adv NP VP complement sequence. Thus, the solution to (122)
turns on pure lexical considerations. Once again the examples listed in
(122b) will emerge as functionally deviant for failure to allow composition
which will permit appropriate indexing. In this way, we are able to ex¬
press the parallelism. In contrast to this, however, the filtering approach
advanced by Chomsky and Lasnik fails. A similar explanation holds for
(118).
Before going on to show that filter (113) leads to additional loss of
generalization, let us pause to take note, once again, of a fact which has
been stressed several times already. It is relevant to note that the “unless
...” clause associated with filter (113) is needed to permit the following
type of example.

(123) John wants you to leave.

We see that Chomsky and Lasnik mistakenly take the NP VP complement

to be an S complement. It is simply a coincidence that verbs are men¬
tioned in (113), as [-N], when verbs are typically function terms which
are specifically designated in the lexicon to select argument structure.
Within the framework advocated here, however, this is no coincidence.
Words such as want do select NP VP complements as one possible argu¬
ment structure and therefore the ad hoc statement need not be prolifer¬
ated to special filters. It is ad hoc in the appropriate sense, i.e. lexical.
It is now possible to proceed to demonstrate that filter (113) leads to
 Chomsky/Lasnik Filters 101

additional loss of significant generalization. Consider in this connection

the following examples.

(124) a. *It is unclear what John to do.

b. *1 told Bill what John to do.

Whereas these examples are ungrammatical, the following are fine.

(125) a. It is unclear what to do.

b. I told Bill what to do.

According to Chomsky and Lasnik, examples such as (124) are out for
failure to undergo an obligatory rule of control, whereas the examples
listed in (125) have underlying PRO following what and must undergo the
obligatory rule of control. This analysis is vague in several ways. First,
just what precisely is this rule of control? Chomsky and Lasnik do not
formulate it. And just how is (124a)—(124b) ruled out when this unstated
rule fails to apply? These crucial questions are not addressed.
One can readily see, in spite of the inexplicit character of the suggested
analys-is, that a significant generalization is being missed. Certainly we
should shoot for the goal of ruling out (124a) and (124b) by the same
general means we employ to rule out examples such as (120b), (121b), etc.
The parallelism can be clearly seen when we compare the relevant exam¬
ples below.

(126) a. *It is unclear what John to do.

b. *It bothers me Bill to win. [= (120b)]
*It came as a surprise to me John was here. [= (120b)]
(127) a. It is unclear what to do.
b. It bothers me for Bill to win. [= (120a)]
It came as a surprise to me that John was here. [= (120a)]

I submit that (126a) should be ruled out on grounds similar to those used
to rule out (126b) and that (127a) should be interpreted on grounds similar
to those used to interpret (127b). Recall that (126b) were ruled out on the
assumption that Bill to win cannot be identified with the index of it since
Bill to win is not the appropriate/or-complement. (It is not even a con¬
stituent.) Also John was here cannot be identified with the index of it
since it is not the appropriate S argument. By contrast in (127b) we find
the appropriate /or-complement and that-S complements so that iden-
102 Realistic Syntax

tification can ensue. Now to express the parallelism of the (126a) and
(127a) examples with the others, we must assume that what John to do
cannot be identified with it but that what to do can. This can be accom¬
plished by providing the requisite functional structure in connection with
wh-words such as what. For example, utilizing the general method of
Brame (1978b), we can assign the following partial argument structure in
the case of what.

(128) what; NPX, (Fq (__))(VP) or ...

The three dots in (128) indicate that what must be assigned additional
argument structure of the type indicated in Brame (1978b) for relative
clauses, direct questions, etc. Now if we take Fq(NPx)(VP) = S, we can
account for the relevant examples in exactly the same way that we ac¬
count for the other examples in (126) and (127). Thus, (126a) will emerge
as functionally deviant just like (126b), whereas (127a) will receive a com¬
plete interpretation, just like (127b). Notice the close link between this
solution and the lexicon. Just as what must be lexically specified to permit
only a certain type of argument structure, so must function words such as
for and that be specified to select only certain types of functional struc¬
ture. For example, the preposition for, in its complementizer function,
might be specified as follows.

(129) for; Pc, (_(NPp))(VP) or ...

Again, for has other functions, a range of prepositional functions other

than just the complementizer function indicated by the superscript c, cf.
Brame (1978b) for a discussion of some functions associated with preposi¬
tions. Again, we might assume Pc(NPp)(VP) = S and may go on to actu¬
ally distinguish the various S argument types in terms of superscripts, a
step which must be taken in any case. The point is that once this lexical
treatment is provided, it is possible to treat extraposition phenomena as
anaphora where S and VP argument indices are identified with the index
of it. This yields a unified analysis of the obviously parallel structures
discussed without recourse to ad hoc filters which are inherently incapa¬
ble of expressing the generalization. Once again we see that the
Chomsky/Lasnik analysis represents a move towards expressing idiosyn-
cracies in terms of filters, when they should be expressed in terms of
lexical structure. In this sense, the filter approach has very much in com¬
mon with generative semantics where many idiosyncracies are transferred
from the lexicon to the transformational component, often as exception
 Chomsky/Lasnik Filters 103

features, etc. Surprisingly, then, we find a good deal of similarity between

Chomsky/Lasnik filters and generative semantics.

11. Conclusion

It appears that there are ample grounds for disbelieving that filters
provide any significant advance in our understanding of the syntax of
English or natural language in general.8 Indeed, it has been argued that
each and every filter posited by Chomsky and Lasnik is unneeded when
we view matters in a more interpretive light.
In a sense, one might say that Chomsky/Lasnik filters are special cases
of incompleteness. After all, both say what is not a phrase in the language.
It should be emphasized, however, that this comparison is to some extent
an invidious one and that the analogy is perhaps too charitable. For one
can see that filtering within the Chomsky/Lasnik framework is simply one
kind of device which defines what constitutes a nonphrase in the lan¬
guage. In addition there are other considerations; for example, those
phrases which are not generated by their phrase structure rules and which
do not satisfy their lexical restrictions, etc. are not phrases in the lan¬
guage. However, functional deviance covers all cases within the theory
advanced here. It is a general way of defining the complement of the
language in question with respect to the concatenated vocabulary. We
could just as well define completeness and then define incompleteness
derivatively. The complete phrases are then those phrases which are rec¬
ognized by the grammar. This definition is general and is not ad hoc in the
sense of filters within the Chomsky/Lasnik framework. Completeness is
directly linked to lexical structure by way of composition and indexing.
Chomsky/Lasnik filters are not directly linked. They recapitulate other
obvious lexical restrictions. Thus, the following two examples are out for
different reasons within their framework.

(130) a. *John pleaded Bill leave,

b. *John slept Bill leave.

Example (130a) is out because that has been deleted, making (130a) sus¬
ceptible to filter (70). Example (130b) is out because the lexical restric¬
tions associated with slept are not satisfied, i.e. lexical insertion. But I
claim that both are out for the same reason. Both are incomplete for
failure to satisfy lexically based restrictions on the argument structure
associated with plead in the one case and slept in the other. Chomsky/
 104 Realistic Syntax

Lasnik filters are inherently defective in that such generalizations go un¬

expressed. We have seen many additional examples in the foregoing
pages.
As noted on several occasions, the postulation of filters represents a
move away from the lexicon by way of stating idiosyncratic properties
which should in fact be intimately bound up with the words themselves.
How is it possible that such obvious idiosyncratic properties could be
formulated in terms of ad hoc filters when they should in fact be related
directly to lexical structure? It is no doubt the belief that transformations
apply to more abstract underlying structure to yield surface strings that is
responsible. Such transformations include Wh-Movement and deletion in
COMP, rule (9). As we have seen, such rules give rise to insuperable
difficulties and provide a basis for many missed generalizations. It seems
therefore that we should forego transformations altogether and work di¬
rectly off the surface. When we do so, the future looms much brighter.
Many general properties begin to fall into place and many outstanding
puzzles admit of realistic solutions.

Notes

1 This rule of anaphora, which can be taken to be a rule of bound anaphora, must make
reference to specifiers since, as is well-known, determiners can have an effect on the gram-
maticality of the associated relative clause plus head. We must also distinguish vr/2-words
from relative that as indicated by contrasts such as the following.
(i) a. *1 saw many boys who Bill saw.
b. I saw many boys that Bill saw.
(ii) a. *He likes the same people who you like. [Emonds (1976, p. 190)]
b. He likes the same people that you like.
(iii) a. *the people who there are [examples (iii) and (iv) from Ross]
b. the people that there are
(iv) a. the legs which Sue has [inalienable legs]
b. the legs that Sue has [alienable or inalienable legs]
Additional relevant examples can be found in Carlson (1977).
Finally it should be noted that focus constructions such as the following must also involve
a rule of anaphora linking the variable index of the wft-word or that with the post-copular
NP, PP, or S.
(v) a. It was John that Bill saw.
b. It is Mary who Bill likes.
c. It was in the garage that Harry put the car.
d. It is simply that Harry is dishonest that Mary believes is true.
This analysis of clefts supersedes that proposed in Brame (1978b); however, Operator Bind¬
ing must nevertheless apply to the associated complement of focus constructions; hence the
examples cited there were quite relevant.
 Chomsky/Lasnik Filters 105

2 Here too specifiers are relevant for the proper functioning of this instance of bound
anaphora, since the following contrasts hold.
(i) a. the fact that John was here
b. *a fact that John was here
(ii) a. the idea that Mary is rich
b. *an idea that Mary is rich
3 For a good survey of many of the problems associated with gapping, see Sag (1976), which
draws on Lust (197?). See also Li (1979).
4 Maling and Zaenen (1978) note that the Chomsky/Lasnik filter “seems to miss syntactic
generalizations within English” in view of the fact that the ‘unless’ clause apparently re¬
capitulates the relativization process. It might be worthwhile to distinguish once again
binding and anaphora, as suggested in section 4. Thus, filter (90) apparently recapitulates
part of the general anaphora process which links the relative pronoun to the lexical head in
relative clauses, and thereby leads to a loss of generalization.
5 Rule (91) is generalized in Brame (1979) to account for the following examples adduced in
Bresnan (1977).
(i) It's in these villages that we all believe_can be found the best examples of this
cuisine.
(ii) *It’s in these villages that we all believe that_can be found the best examples of this
cuisine.
Such examples immediately refute proposals such as Perlmutter’s in terms of subject.
6 Maling and Zaenen (1978) argue that Perlmutter’s Correlation is not the correct one and
offer an alternative.
7 Chomsky and Lasnik draw on an alleged dialect of English to argue that filter (97) is
dialect-specific. If their data are factual, it would be possible to provide an analysis within
the framework proposed here. However, it is probably premature to jump to conclusions
concerning the grammaticality of the two examples cited, which follow.
(i) Who did you say you were going to try for to go to the church social with you?
(ii) Who are you going to try for to go to the church social with you?
Chomsky and Lasnik cite these two examples as grammatical in Ozark English. Before
anything can be granted, however, an analysis of Ozark English which draws on more than a
handful of examples is certainly required. One has reason to be skeptical, for notice how
involved and lengthy are these examples. Typically long and involved examples can often
misdirect the mind of native speakers with respect to grammaticality judgments.
Chomsky and Lasnik also cite the following sentence as grammatical in Dutch.
(iii) Wie heb je geprobeerd om te verslaan?
However, the judgment seems to be wrong in this case. Speakers of Dutch who I have
consulted say that (iii) sounds strange whem om is present. The speakers I have consulted
are not linguists, but rather naive theoretically unbiased native speakers.
8 Additional pertinent criticism of the Chomsky/Lasnik filters can be found in the interesting
paper by Iwakura, cf. Iwakura (1979). Iwakura’s criticism has not been repeated here;
however, we agree in concluding that deletion of material from COMP leads to undesirable
results. I differ from Iwakura in that unlike him, I have not chosen to provide new filters to
account for the relevant range of data.
 106 Realistic Syntax

References

1. Brame, M. K. (1976) Conjectures and Refutations in Syntax and

Semantics. New York: Elsevier North-Holland.
2. Brame, M. K. (1977) Alternatives to the Tensed S and Specified
Subject Conditions. Linguistics and Philosophy 1: 381-411.
3. Brame, M. K. (1978a) The Base Hypothesis and the Spelling Prohibi¬
tion. Linguistic Analysis 4: 1-30.
4. Brame, M. K. (1978b) Base Generated Syntax. Seattle: Noit Am-
rofer.
5. Brame, M. K. (1978c) Discourse and Binding without Transforma¬
tions. Linguistic Analysis 4: 365^104.
6. Brame, M. K. (1979) Realistic Grammar. Paper presented at the Con¬
ference on Current Approaches to Syntax. Milwaukee, Wisconsin:
March 1979. In this volume.
7. Bresnan, J. (1970) On Complementizers: Toward a Syntactic Theory
of Complement Types. Foundations of Language 6: 297-321.
8. Bresnan, J. (1972) Theory of Complementation in English Syntax.
MIT doctoral dissertation.
9. Bresnan, J. (1977) Variables in the Theory of Transformations. In P.
Culicover, et. al., eds., Formal Syntax. New York: Academic Press.
10. Bresnan, J. (1978) A Realistic Transformational Grammar. In M.
Halle, et. al., eds. Linguistic Theory and Psychological Reality.
Cambridge: MIT Press.
11. Carlson, G. (1977) Amount Relatives. Language 53: 520-542.
12. Chomsky, N. (1971) Problems of Knowledge and Freedom. New
York: Pantheon.
13. Chomsky, N. (1973) Conditions on Transformations. Reprinted in
Essays on Form and Interpretation. New York: Elsevier North-
Holland.
14. Chomsky, N. (1977) On WH-Movement. In P. Culicover, et. al., eds.
Formal Syntax. New York: Academic Press.
15. Chomsky, N. and H. Lasnik (1977) Filters and Control. Linguistic
Inquiry 8: 425-504
16. Emonds, J. (1970) Root and Structure-Preserving Transformations.
MIT doctoral dissertation.
17. Emonds, J. (1976) A Transformational Approach to English Syntax.
New York: Academic Press.
18. Iwakura, K. (1979) On Surface Filters and Deletion Rules. Linguistic
Analysis 5: 93-124.
19. Jayaseelan, K. (1979) On The Role of the Empty Node in the
Structure-Preserving Hypothesis. Linguistic Analysis 5: 247-92.
 Chomsky/Lasnik Filters 107

20. Li, E. (1979) Verb Gapping. Univ. of Wa. M.A. Thesis.

21. Lust, B. (197?) On Gapping. MIT ms.
22. Mailing, J. and A. Zaenen (1978) The Nonuniversality of a Surface
Filter. Linguistic Inquiry 9: 475^-97.
23. Rosenbaum, P. (1967) The Grammar of English Predicate Comple¬
ment Constructions. MIT Press.
24. Ross, J. R. (1967) Constraints on Variables in Syntax. MIT doctoral
dissertation.
25. Sag, I. (1976) Deletion and Logical Form. MIT doctoral dissertation.
 THE OPACITY CONDITION, THE NOMINATIVE
ISLAND CONDITION, AND THE MYTH OF CONTROL

Sensei, I love these examples, but control, is it really a myth?

John promised Mary his love, #her love.

John promised Mary to shave himself, #herself.
John promised Mary love.

John impressed Mary with his love, #her love.

John pressed Mary for #his love, her love.
John impressed Mary with love.
John pressed Mary for love.

Yes, gakusei, but only in their sense, for they miss the significance.
1. Introduction
2. Two Linguistic Puzzles
3. The Nominative Island Condition and the Opacity Condition
4. A More Realistic Solution
5. Chomsky’s Case Filter: Another Ad Hoc Principle
6. Extending the Data
7. The Myth of Control
References
 THE OPACITY CONDITION, THE NOMINATIVE

ISLAND CONDITION, AND THE MYTH OF CONTROL*

1. Introduction

In what follows, I will assume familiarity with the lexically based approach to
syntax advocated in [6], [7] and go on to compare it with trace theory with
respect to Chomsky’s most recent revision of his Specified Subject Condi¬
tion and Propositional Island Constraint [= Tensed S Condition] as the
Opacity Condition and the Nominative Island Condition respectively. It will
be shown that the revised conditions, very much like the earlier ones, serve
to describe rather than explain the relevant range of data investigated. In
contrast to trace theory, it will be seen that the data truly follow from the
lexically based theory without recourse to ad hoc theoretical elaboration of
the type required within the framework of trace theory. This result provides
support for the more lexically based view of grammar.
The plan of the essay is as follows. In section 2, several of Chomsky’s
“elementary properties of indirect questions’ ’ are reviewed and in section 3,
we take a look at how these properties are described by the new conditions.
Section 4 provides a more realistic solution to the puzzles unveiled in section
2 within the framework of the earlier references [6], [7]. A third puzzle is
discussed in section 5, where it is noted that yet another ad hoc device is
required within trace theory to handle it. By contrast, this puzzle is im¬
mediately resolved without prestidigitation within the proposed theory. It is
shown that additional generalizations go unexpressed within the framework
of trace theory, which are subsumed naturally within the lexically based
theory. In section 6, the claims for independent motivation of the trace
theoretic conditions are examined and found to be false. The extended data
do not provide a plausible argument for trace theory. Finally, in section 7,
two reasons are advanced for the failure of trace theory. These include a
fondness for lexical abuse and a mythical view of control.

*This paper was presented at the University of Washington during the spring of 1979.

Ill
112 Realistic Syntax

2. TWO LINGUISTIC PUZZLES

The following examples and associated structures are relevant to what

Chomsky claims to be the “truly explanatory function” of his Opacity
Condition and Nominative Island Condition.

0) a. ... [s who [g NP visited t ]]

a'. It is unclear who Bill visited,
a”. John asked who Bill visited.
b. ... [g who \s t visited NP ]]
b'. It is unclear who visited Bill,
b" John asked Bill who visited Mary.
c. ... [g who Is NPj to visit NP2 ]]
c'. It is unclear who to visit,
c”. John asked Bill who to visit.

According to Chomsky, the complement phrases in the examples are

analyzed as in the corresponding structural configurations, where t is the
trace left behind by application of the alleged rule of wh-Movement and
PRO is an abstract NP, realized as \nPe],e the identity element. It is to be
noted that this approach follows tradition and incorporates the otiose S
node in spite of the fact that there is considerable criticism of it in the
literature, cf. [2], [4], [8],
Now Chomsky provides us with what he considers to be several of the
fundamental properties of indirect questions. For example, the following
direct quotes provide us with a source of puzzles which must be resolved
within any explicit theory of grammar, apparently.

Puzzle #1
In case [la] and [lb], NP ^ PRO; that is construal is impossible. For
example, we cannot have such sentences as “John asked Bill who
visited,” meaning that John asked Bill which person he, John, visited or
which person visited him, John: even though ask assigns control by its
subject (John, in this case), as we see from [lc"]. [ll:p. 9]
Puzzle #2
In case [lc], NP2 cannot be PRO with NPX = trace. Thus, “it is unclear
who to visit” cannot mean that it is unclear who is to visit some unspec¬
ified person.” [ll:p. 9]

Let us now turn to Chomsky's solution for these puzzles in terms of his
Opacity Condition and Nominative Island Condition.
 The Myth of Control 113

3. The nominative island condition

AND THE OPACITY CONDITION

Within Chomsky’s framework, the two puzzles would be translated into

the following structures with respect to the examples mentioned.

(2) a. *Johnj asked Bill [§-whoj [s PROj visited tj ]]

b. *Johni asked Bill whoj [s tj visited PROj ]]
(3) *It is unclear [§■ whoj [s tj to visit PROarb]]

In these examples coindexing is assigned by an extremely general rule

termed COINDEX. In the case of the structures provided in (1), COIN¬
DEX applies in the following way.
COINDEX: an arbitrary occurrence of PRO in the embedded structure of
(1) is coindexed with some NP in ... or assigned the index
arb indicating arbitrary reference if there is no lexical NP in

Now Chomsky notes that COINDEX can be constrained by his condi¬

tions, SSC and PIC [= TSC]. Thus, PRO in both (2a) and (2b) could not be
coindexed with John since it is contained within a tensed sentence in both
cases. Moreover, it is assumed that assignment of / to PRO and John in
(2b) would constitute a violation of SSC and that assignment of arb in (3)
would violate SSC.
In [11] Chomsky proposes to revise PIC and SSC as the Nominative
Island Condition and the Opacity Condition respectively. These condi¬
tions are provided below.

Nominative Island Condition [NIC]: A nominative anaphor in S cannot be

free in S containing S.
Opacity Condition: If a is in the domain of the subject of/3, /3 minimal,
then a cannot be free in (3, /3 = NP or S.

According to Chomsky anaphors include PRO, trace, reciprocal pro¬

nouns, and reflexive pronouns. The theoretical terms bound and free are
defined as follows.
We say that an anaphor a is bound in (3 if there is a category
c-commanding it and coindexed with it in /3; otherwise, a is free in
(3. [ 11 :p- 13]
Now, NIC is designed to account for the deviance of examples such as
(2a) where PRO; is nominative by virtue of a special rule: “the subject of a
 114 Realistic Syntax

tensed-clause ... is assigned nominative case by a special rule, though the

phonetic effects, in English, are seen only with pronouns” [ll:p. 16]. In
(2a) we see that PRO is a nominative anaphor and free in S containing S.
Hence, this case is ruled out by NIC.
Example (2b)_ is ruled out by OC since PRO is in the domain of the
subject of/3 = S and is free in S. The same goes for (3), where Chomsky
takes PROarb to be free in general.
We see, then, that Chomsky would account for Puzzle #1 and Puzzle
#2 by invoking special conditions such as SSC and PIC, and, now, his
new revisions of these conditions as NIC and OC.
Let us now show that such ad hoc conditions are unnecessary when we
view matters in a more interpretive light which does not provide for
abstract underlying representations of the kind adopted within trace
theory and standard transformational grammar.

4. A MORE REALISTIC SOLUTION

It is assumed that the reader is familiar with the theory of syntax advo¬
cated in [6], [7], Within such a theory, we work directly off the surface
and attempt to provide a complete interpretation for the string in question.
If a complete interpretation cannot be provided, the string is not a syntac¬
tic phrase in the language; it is, in the terminology adopted, functionally
deviant.
Let us return to the crucial examples. Instead of (2) and (3), we will
attempt to provide an interpretation for the following strings, since we are
working directly off the surface.

(4) a. John asked Bill who visited,

b. John asked Bill who visited.
(5) It is unclear who to visit.

Now we must show that it is not possible to obtain an interpretation

analogous to (2a) in the case of (4a), to (2b) in the case of (4b), which
happens to be identical to (4a) since we are working directly off the
surface, and to (3) in the case of (5). Since we are working directly off the
surface, we do not commence with abstract devices such as PRO and
traces. The devices needed relate, rather, directly to the lexical repre¬
sentations or to general operations on functional structure, the latter in¬
cluding rules such as Operator Binding, cf. [5], [6], Anaphora, cf. [7], and
a few others.
 The Myth of Control 115

Now consider examples (4a) and (4b). In order to derive a complete

interpretation for any string containing a tensed verb, that verb must have
a subject argument to which the VP function applies. The relevant rule of
composition effecting this interpretation is provided in [6], [7]. When
subjects of tensed sentences are missing, we can attempt to interpret a
subject by applying the following rule, which is a special case of a rule
introduced in [6].

Subject Interpretation: 0 —» Antsl / Fx_Fvp, where x = op, v

This rule allows for the interpretation of a subject argument An[s] when
preceded by an operator function or verbal function and followed by a
specific type of function, the VP function, a tensed phrase, not an infinitive!
Infinitives are arguments, not functions, and are represented as Avp.
Let us now see how we can provide an interpretation for (4), concen¬
trating on the complement string who visited. We apply the rules of com¬
position and the indexing procedure of [6], [7] and derive the following
partial interpretations.

(6) Fll(An^Xb:F<l(y) F^ = FV(An[o]):f(An[o]

who visited

We have chosen to interpret visit as a transitive verb, which yields

Fv(An[o]) by the requisite rule of composition of [6] which selects func¬
tional structure on the basis of lexical specification, in part. We could just
as well have picked the intransitive lexical functional structure in which
case we would derive Fvp : /, a complete interpretation for the VP func¬
tion. Here we still have an incomplete VP function and an incomplete
question operator.
At this point interpretation cannot proceed further by application of
rules of composition and indexing. However, the rule of Subject Interpre¬
tation can now apply to yield the following.

(7) Fq(An[x]):Fq(y) Fvp = Fv(An[o]);f(An[o]

I
who visited
 116 Realistic Syntax

Now the rules of composition and indexing procedure can operate to yield
the following interpretation as the root of the composed structure.

(8) (Fq(An[x]))((f(An[0]))(An[s])) : Fq(y))((f(AnM))(An[s]))

where y = who ,f = visited

To (8), a second rule operating on functional structure can apply, the rule
of Operator Binding, cf. [5], [6]. This rule will bind the unspecified argu¬
ment An[s] or An[0] to y, depending on whether we take x to be o or 5. (It
will also supply an index for the operator prefix Fq.) In either case, we are
left with one uninterpreted argument, either An[s] or An[0]. (We are assum¬
ing that who can be objective in the dialect under discussion; if not, then it
could only be interpreted as subject, this difference being reflected
in the actual lexical specification of who.) We see, then, that the phrase is
not complete with respect to the attempted interpretation. This simply
means that there is no analogue to the interpretations provided in (2a) and
(2b) within Chomsky’s framework. But we accomplish this result without
recourse to special constraints such as NIC and OC.
Now consider the fact that an interpretation analogous to (3) is simply
impossible within the framework of lexically based syntax advocated
here. This follows from the fact that infinitive phrases are arguments, not
functions. Consequently, the rule of Subject Interpretation cannot apply
to supply a subject argument An[s] when we attempt to interpret the fol¬
lowing example.

(9) It is unclear who to visit.

We will derive a complete interpretation only when who is interpreted as

A ], whereupon Operator Binding will identify the index y of who with
the unspecified object argument of visit. A complete interpretation is de¬
rived by identification of the index of it with the index corresponding to
who to visit. This identification is the interpretive analogue of extraposi¬
tion, suggested in [7],
The theory advanced here is much more lexically based than trace
theory. Thus, who will include the following functional structure in its
lexical specification, among others.

(10) who; An[x], Fq((_))(Avp)

Abstract devices such as PRO are not part of underlying representations.

The surface is mediated directly by lexical specifications together with
general interpretive procedures such as those proposed in [6], [7].
 The Myth of Control 117

5. Chomsky's case filter: another ad hoc principle

It would be of some interest if additional puzzles followed directly from

the more interpretively based approach which did not follow from trace
theory without further elaboration. In fact, this is precisely what we find
to be the case, as will be shown in this section.
Chomsky provides us with yet another puzzle relating to indirect ques¬
tions. It is summarized in the following direct quote.

Puzzle #3
In case [lc], NPj must be PRO; that is, control is obligatory. There is
no such sentence as “it is unclear who Bill to visit” meaning that it is
unclear which person Bill is to visit, nor can NP2 be missing, even
though visit can be intransitive. [ 11 :p. 9]

The third puzzle, then, is to account for the ungrammaticality of the

following examples, which are represented in terms of the following
structures within trace theory.

(11) a. *It is unclear [§ whOj [s Bill to visit tj ]]

b. *It is unclear [s whoj [s C to visit ]]

Chomsky remarks that the “only property of indirect questions that does
not fall under SSC and PIC [and hence NIC and OC, MKB] is the last one
mentioned: the fact that control is obligatory in the case of an infinitive”
[ll:p. 9]. He summarizes his discussion, which I have included as puzzles
#1, #2, and #3 as follows.
These are the elementary properties of indirect questions. They
might be described in various ways ... In effect, the Specified
Subject Condition (henceforth, SSC) and the Propositional Is¬
land Condition (henceforth, PIC; alternatively, the “Tensed-S
condition”) of the references cited imply most of these prop¬
erties [but not the last, MKB]. To put it differently, some version
of SSC and PIC is independently motivated by these properties
of indirect questions, and is therefore available elsewhere in
grammar to serve a truly explanatory function. [ll:p. 9]

We have already seen that the properties expressed as Puzzle #1 and #2

can be explained without recourse to ad hoc conditions such as SSC and
PIC, or NIC and OC. This shows, I believe, that the claim to a “truly
explanatory function” is false. This demonstration carries all the more
force in view of the fact that Puzzle #3, which is not covered by Choms-
 118 Realistic Syntax

ky’s NIC and OC, is immediately explained without further elaboration

within the interpretive theory which is more lexically based than trace
theory. To see this, recall that we work directly off the surface. Hence,
both (11a) and (lib) must be interpreted in terms of the following string.

(12) It is unclear who Bill to visit.

Now there is simply no way to interpret (12) within the proposed theory.
This follows from the fact that who does not select NP VP argument
sequences. Although who may select a VP argument, as in the partial
lexical specification provided in (10), there is no provision for both NP
and VP, as noted earlier in [7]. Thus, the heart of the explanation, once
again, resides in the lexicon, not in ad hoc conditions.
Now the deviance of (12) truly follows without elaboration within the
lexically based theory, whereas new ad hoc devices are required within
trace theory. Thus, Chomsky introduces the following filter in conjunc¬
tion with the case rules provided in (14).

(13) Chomsky Case Filter: *N, where N has no Case.

(14) Chomsky’s Case Rules
a. NP is oblique when governed by P and certain marked verbs
b. NP is objective when governed by V
c. NP is nominative when governed by Tense

It is claimed that example (12), it is unclear who Bill to visit, is ruled out
because Bill is not assigned case by any of the case rules of (14). In other
words, after the case rules apply, filter (13) marks the resulting structure
as deviant since Bill has not been assigned case.
Clearly, such an analysis is artificial and contrived. It raises the discus¬
sion to an extremely abstract level where refutation seems impossible.
Thus, it is clear, for example, that he exhibits case in English, as does
him. The former is nominative, as opposed to him, which is objective (or
oblique). Yet both of the following examples are ill-formed.

(15) a. *It is unclear who he to visit,

b. *It is unclear who him to visit.

Clearly, he and him in (15a) and (15b) exhibit case! Yet, Chomsky must
now escape refutation by claiming that he and him derive from some more
abstract representation and that this representation is not assigned case!
 The Myth of Control 119

By contrast, within the theory which I have advocated, the integrity of

lexical items is respected. Both he and him are lexical items and in neither
case of (15) is a complete interpretation derived because who does not
select NP VP argument sequences in the lexicon. It follows that the in¬
terpretive procedures will never apply so as to compose who and he to
visit or who and him to visit. The Spelling Prohibition of [4] is satisfied and
we need no case filter such as (13).
The ad hoc nature of the trace theoretical approach is revealed with
even further clarity when we consider deviant strings such as the follow¬
ing.

(16) a. *It is unclear who John hit Bill to visit,

b. *It is unclear who among to visit.

These strings simply cannot be provided a complete interpretation within

the lexically based theory since who does not select a sentence such as
John hit Bill plus VP, nor does who select a preposition such as among
followed by VP. This follows from the lexical specification provided in
(10) in conjunction with the general interpretive procedures provided in
[6], [7]. But, notice that (16a) and (16b) are out for the same reason as are
(15a), (15b), and (12). In all cases, a complete interpretation cannot be
derived as a result of the lexical specification associated with who. By
contrast, trace theory must miss this generalization. Examples such as
(15a), (15b) and (12) will be out because of one device, Chomsky’s Case
Filter, and examples such as (16a) and (16b) will be out for a different
reason and accounted for by a second device, for example, phrase struc¬
ture. This clearly misses the underlying generalization.

6. Extending the data

The following ad hoc devices are required within the framework of

trace theory to describe the data of Puzzles #1, #2, and #3.

i. A condition such as the Nominative Island Condition

ii. A condition such as the Opacity Condition
iii. A filter such as Chomsky’s Case Filter

In contrast to trace theory, none of these devices is required within the

lexically based theory advocated in [6], [7]. The careful reader should
note, furthermore, that there is no counterpart to these ad hoc devices and
filters which trace theory adopts. For example, in [7], it is shown that the
 120 Realistic Syntax

effects of each and every one of the Chomsky/Lasnik filters proposed in

[12] follow without elaboration from the lexically based theory. The rele¬
vant data discussed here and in the latter paper follow from the proposed
theory without ad hoc elaboration and the reason is fairly simple. The
cooccurrence phenomena that we are investigating are genuinely linked to
lexical restrictions, one of the principal features of the proposed theory.
Lexical restrictions are indeed localized in terms of lexical items them¬
selves within the proposed theory, whereas they are elevated in terms of
unnatural and contrived devices such as filters and conditions like NIC,
OC, etc. within trace theory. When this latter approach is adopted with
more and more distance placed between the lexicon and the surface
strings, it follows, very much as with generative semantics, that new
devices will be required, time and again, to force through the unnatural
analysis.
Now let us pause for a moment to consider a wider range of data which
Chomsky claims motivates his conditions. He claims that anaphors in¬
clude such elements as reciprocal pronouns, reflexive pronouns, and
traces and that conditions such as NIC and OC generalize to construc¬
tions incorporating these elements. Since this is so, he argues, NIC and
OC represent genuine explanatory principles of grammar.
The fact of the matter, however, is that all the data discussed are
explained within a lexically based theory without such conditions. Con¬
sider, for example, the following cases involving reciprocals and re¬
flexives.

(17) a. The boySj said that each othe^ would win.

b. The boySj said that themselvesj would win.
(18) a. The boySj said that Sue hit each other^
b. The boySj said that Sue hit themselvesj.

The examples provided in (17) are allegedly accounted for by NIC and the
examples in (18), by OC. However, as has been noted elsewhere, cf. [3], a
conditions approach to such data is misleading. It leads to a loss of
generalization as shown by the following example.

(19) The boys said that them would win.

In other words, each other and themselves, like them, is lexically spec¬
ified as oblique arguments, written ob, which we may take to be a variable
tanging over all cases other than s, the subjective or nominative case.
Thus, each other, themselves, as well as them, will include at least this
 The Myth of Control 121

much functional information as part of their lexical specifications; this is

indicated below in (20).

(20) a. each other; An[obl

b. themselves; Antob]
c. them; An[obl

Of course, additional lexical information is required, such as number,

anaphoricity, etc., but the point here is that all three example types in (17)
and (19) are deviant in that a subject argument Ants] is required for compo¬
sition of the VP function with NP in accordance with the rules provided in
[6], [7]. Such an NP would be the following.

(21) they; An[s]

Composition could proceed if they replaced each other, themselves, or

them in (17) and (19). We see, once again, that special conditions such as
NIC and OC are not required to account for these data.
Let us now turn to (18a) and (18b). As noted quite explicitly in an earlier
reference, [3], the rule of anaphora requires that reflexive and reciprocal
pronouns be indexed with their subjects. (There is a special case such as
Mary talked to them about themselves.) When this indexing takes place in
(18a) and (18b), we have each other and themselves marked as anaphoric
with Sue and this constitutes a violation of number agreement. This
suggestion was put forward in [3], long before the appearance of “On
Binding”, yet we find no citation in the latter.
Let us now consider another type of example which allegedly motivates
NIC, etc. Such an example involves traces and is provided below.

(22) Maryj seems that t, likes beans.

Since movement of NPs such as Mary is completely free, (22) can be

derived within trace theory. But such examples are purported to fall under
NIC, since ts is free in its associated S. But, clearly, there is a gut-level
response to this example which surfaces upon examination of the bare
string.

(23) *Mary seems that likes beans.

The problem is simply that (23) lacks an NP which could serve as a subject
for complete interpretation of the string provided. Within the framework
122 Realistic Syntax

of lexically based syntax, then, (23) simply cannot be interpreted. This

example, is similar in many respects to the example provided in (12). To
see the parallelism, we repeat both examples below.

(24) a. It is unclear who—Bill to visit

b. Mary seems that—likes beans.

In (24a), recall, who cannot be composed with what follows. Just so, now,
in (24b), that cannot be composed with what follows. In both cases we
cannot provide a complete interpretation because there are no provisions
for composition. The lexical specifications of who and that do not include
NP VP or VP as potential arguments.
We see, in fact, that there are numerous generalizations that go unex¬
pressed within trace theory. For example, the following examples are all
deviant for reasons completely analogous to the deviance of (23).

(25) a. *Mary knows that likes beans.

b. *It seems that likes beans.
c. *Mary insists that likes beans.

All of these examples are out for the same reason within the theory of
lexically based syntax. By contrast, (23) is out because of NIC within
trace theory, whereas (23a-c) are out because of different reasons,
phrase structure or lexical insertion.
All of the data brought to bear in “On Binding” can be provided a
completely natural lexical explanation provided we adopt a surface
oriented approach. There is no independent motivation for conditions
such as NIC and OC, contrary to the claims made on the basis of re¬
flexives, reciprocals, traces, etc. Rather than belabor this point, let us now
attempt to pinpoint the fundamental drawback of trace theory.

7. The myth of control

There are at least two keys to the failure of trace theory; one has
already been touched upon above. On the one hand, trace theorists are
forced to propose devices such as NIC, OC, filters, Subjacency, Recov¬
erability of Deletion, etc. because of the unnatural remove they institute
between lexical items and surface strings. They abuse lexical items by
failing to recognize them as such, assuming abstract analyses, deriving he
 The Myth of Control 123

and him from some abstract source, never coming to grips with the sur¬
face data and never illustrating explicit derivations of the words in ques¬
tion. The actual phonetic detail is taken to be a trivial consequence of
low-level processes which are uninteresting. Yet we see that such as¬
sumptions cannot be granted. We are not free to assume that the distinc¬
tion between he and him, etc. is simply a trivial matter of detail. When we
adhere to the Spelling Prohibition, such ad hoc analyses as proposed by
trace theorists cannot be countenanced, cf. [4].
Besides lexical abuse, there is a second outstanding characteristic of
trace theory which is largely responsible for its failure. This is the fact that
trace theorists have been adamant in their reluctance to give up the
abstract sentential source for infinitives, in spite of the fact that there is
considerable evidence against the abstract source, cf. [2], [4], [8], On
Binding” is perhaps the quintessential expression of this pernicious
stance.
Now what is the reason for this continued bias against the obvious
nonclausal source for infinitives? First, Chomsky assumes that puzzles
such as those mentioned above follow naturally when sentential sources
are adopted in conjunction with his conditions. But we have seen that
such conditions and filters are in fact ad hoc devices that we can do
without. Second, Chomsky assumes that a sentential source for infinitives
yields a more general theory.

Notice that this approach maximizes the uniformity of the

lexicon and base system, just as in the case of the analogous
indirect questions. It is clear that the base rules permit structures
of the form V-S (“promise that ...”) and V-NP-S (“promise Bill
that ...,” “persuade Bill that ...”). Verbs that take^one or the
other structure typically allow infinitives in place of S ( promise
to ...,” “promise Bill to ...,” “persuade Bill to ...”) ... We may
therefore keep to what seems the simplest base theory, clauses
can be finite or infinitive ... and there is no nonclausal source for
infinitives. [ll:p. 11]

The reasoning revealed in this quote is faulty in several respects. In the

first place, it is simply not true that “verbs which take one or the other
structure typically allow infinitives in place of S. There are many verbs
which select S, for example, but do not “typically’ allow infinitives.

(24) a. John demanded that Mary leave,

b. *John demanded Mary to leave.
(25) a. John insisted that Mary leave,
b. *John insisted Mary to leave.
 124 Realistic Syntax

(26) a. John thinks that Mary left,

b. *John thinks Mary to leave.
(27) a. John denies that Mary left,
b. *John denies Mary to leave.

Nor do verbs which select infinitives “typically” permit sentential com¬

plements, as illustrated below.

(28) a. John tried to leave.

b. *John tried that he leave.
(29) a. John forced Mary to leave.
b. *John forced Mary that she leave.
c. *John forced that Mary leave.
(30) a. John telegrammed Mary to come.
b. *John telegrammed Mary that she come.
c. *John telegrammed that Mary come.

Citing two verbs which permit multiple complement types, such as prom¬
ise and persuade, as Chomsky does, does not constitute an argument for
an abstract sentential source of infinitive complements. Apparently, the
following criticism is still apropos even today.

Berman and Szamosi’s reasoning is not new to the field. It goes

something like this: Because some verbs take S complements, all
complements to those verbs must be S complements. By parity
of reasoning we could conclude that the NP direct object com¬
plement of expect in John expected Mary is really an S since we
do find S complements to expect, as in John expected that Mary
would come. Such reasoning, like that adopted by Berman and
Szamosi, is obviously unacceptable. A real argument would
necessarily show that NP patterns like S in a wide range of
contexts or that postulation of S provides an explanation for
otherwise puzzling phenomena. No such argument appears like¬
ly. [l:p. 198]

This criticism of several years past applies with equal force today to
advocates of the position advanced in Chomsky’s quote.
Next note that it is simply false that “this approach maximizes the
uniformity of the lexicon and base system, just as in the case of the
analogous indirect question.” It is false because proscribing VP as com¬
plements to verbs gives rise to lexical gaps and gaps in phrase structure,
as noted repeatedly, cf. [1], [2]. What “seems to be the simplest base
 The Myth of Control 125

theory” is simply an illusion, a hangover from initial work in transforma¬

tional grammar.
Are there other reasons which underlie the widespread reluctance to
give up the traditional transformational analysis of infinitives as sen¬
tences? It seems that the notion of control plays a central role in this
controversy. Thus, commencing with early work on reflexivization, Lees
and Klima [13] proposed to restrict reflexivization to simplex sentences,
as illustrated below.

(31) a. John said that Mary shaved herself,

b. *John said that Mary shaved himself.

Since reflexivization is apparently restricted to the embedded sentence in

(31) , reason was found for positing an S in examples such as the following.

(32) a. John persuaded Mary to shave herself,

b. *John persuaded Mary to shave himself.

Going hand in hand with this observation was the fact that subjects typi¬
cally function as antecedent for the reflexive pronoun in simple sentences.

(33) a. John shaved himself,

b. *John shaved herself.

Putting these facts together it was concluded, even as early as [9], that
infinitives derive from abstract sentential sources, with an abstract NP
subject.
The motivation for this leap to the sentential source is unjustified, how¬
ever, as noted in [2], where a review of the arguments is provided. There
is another view which differs from the traditional transformational view.
The alternative view is that there is a relation which we might call control,
but that this relation need not be expressed in terms of abstract subjects,
whether PRO or otherwise.
The alternative view can be illustrated in terms of examples such as the
following.

(34) a. John promised Mary to shave himself,

b. John persuaded Mary to shave herself.

Let us agree that John controls to shave himself, the latter being a VP
126 Realistic Syntax

argument, an infinitive which does not derive from a sentential source.

We might write this relation as follows.

(35) control(John, to shave himself)

We agree to call the initial argument of the control relation the controller
and the final argument the controllee. Now control is an interpreted rela¬
tion and anaphora may be formulated in terms of the controller relation,
making reference to the controller instead of the subject. Now clearly the
lexical item in question plays a role in determining the controller relation.
Thus, John does not control to shave herself in (34b).
The idea of deriving the controller relation as an interpreted relation is
not simply a notational variant of the standard theory of transformational
grammar or of trace theory. This follows from the fact that we may formu¬
late the interpreted relation in a more general light. Thus, consider the
following examples.

(36) a. John promised Mary a shave.

b. John promised Mary a ride to the store.

Here, once again, we can interpret a control relation, indicated as in (37).

(37) a. control(John, a shave)

b. control(John, a ride to the store)

Clearly, we do not wish to assume an abstract subject in the case of (36a)

and (36b). To this Chomsky would agree, following some remarks he
made on nominalization, cf. [10]. The point that is crucial for this discus¬
sion is that the interpreted relation in terms of obvious NPs in (37) can be
taken to be the same controller relation that holds between NP and infini¬
tive in the examples which are incorrectly analyzed by trace theorists.
Infinitives qua arguments relate to their controller just as noun phrases
relate to their controller. Thus, we encounter an analogous control rela¬
tion in the following two examples.

(38) a. John tried the pie.

b. John tried to leave.

We interpret the following control relations.

 The Myth of Control 127

(39) a. control(John, pie)

b. control(John, to leave)

The fact that control need not be formulated in terms of abstract

subjects—in fact it leads to a loss of generalization—shows that the
standard account of control is a myth. Once this is realized, we can get on
with the task of eliminating all transformations and proceed to develop
realistic models of syntax which operate directly off the surface without
ad hoc conditions and filters. This line of research, I conjecture, will
occupy the attention of serious syntacticians in the coming years.

References

1. Brame, M. K. (1975) On the Abstractness of Syntactic Structure: The

VP-Controversy. Linguistic Analysis 1: 191-203.
2. Brame, M. K. (1976) Conjectures and Refutations in Syntax and
Semantics. Elsevier North-Holland.
3. Brame, M. K. (1977) Alternatives to the Tensed S and Specified
Subject Conditions. Linguistics and Philosophy 1: 381^111.
4. Brame, M. K. (1978) The Base Hypothesis and the Spelling Prohibi¬
tion. Linguistic Analysis 4: 1-30.
5. Brame, M. K. (1978) Base Generated Syntax. Seattle: Noit Amrofer
Publishing Co., P.O. Box 15176.
6. Brame, M. K. (1979) Realistic Grammar. Paper presented at the Con¬
ference on Current Approaches to Syntax. Milwaukee, Wisconsin.
Published in Essays Toward Realistic Syntax, Noit Amrofer Pub¬
lishing Co., P.O. Box 15176, Seattle.
7. Brame, M. K. (1979) Chomsky/Lasnik Filters Are Special Cases of
Functional Deviance. Paper delivered at the Sloan Workshop at Stan¬
ford University in January 1979. Published in Essays Toward Realis¬
tic Syntax, Noit Amrofer Publishing Co., P.O. Box 15176, Seattle.
8. Bresnan, J. (1978) A Realistic Transformational Grammar. In M.
Halle, et al., eds. Linguistic Theory and Psychological Reality.
Cambridge: MIT Press.
9. Chomsky, N. (1955) The Logical Structure of Linguistic Theory.
MIT Fibrary.
10. Chomsky, N. (1970) Remarks on Nominalization. In Readings in
English Transformational Grammar, R. A. Jacobs and P. S. Rosen¬
baum, eds. Waltham, Mass.: Ginn.
128 Realistic Syntax

11. Chomsky, N. (1979) On Binding. Unpublished ms., MIT.

12. Chomsky, N. and H. Lasnik (1977) Filters and Control. Linguistic-
Inquiry 8: 425-504.
13. Lees, R. B. and E. Klima (1960) Rules for English Pronominalization.
Language 39.
 QUANTIFIERS, RECIPROCALS, AND RAISING

Sensei, is it true that Chomsky and Postal are both wrong?

Yes and no, gakusei. Chomsky is right to reject Raising and Postal is right
in adopting an object structure. But Postal goes wrong when he adopts
an underlying subject structure, and with it, Raising, while Chomsky is
wrong to adopt the subject structure twice over.
1. Introduction
2. Controversy over each other
2.1 Multiple Occurrences of Distributive Quantifiers
2.2 Conjunction
2.3 Distribution of each other
2.4 Universal Movement Constraints
3. New Arguments Against the Transformationalist Hypothesis
4. the men each and each of the men
5. expect NP VP versus expect S
References
 QUANTIFIERS, RECIPROCALS, AND RAISING*

1. Introduction

There is currently a controversy over the status of sentences such as

the following.

(1) a. John expected Mary to leave,

b. John believes Mary to be rich.

Transformationalists seem to agree that the underlying representations of

(1) contain an embedded S-complement as indicated in (2).

(2) a. John expected [Mary to leave]s

b. John believes [Mary to be rich]s

However, they differ on the status of the appropriate surface repre¬

sentations. One group contends that there is a rule of Raising, which
promotes the subject of the embedded sentence to the object position of
the matrix sentence, yielding (3), after Pruning.

(3) a. John expected [Mary]NP [to leave]Vp

b. John believes [Mary]NP [to be rich]VP

A second group favors (2) as the surface representation of (1). This group
denies the existence of Subject Raising.
Postal has argued for the existence of Raising in his extended study, [7],
whereas Chomsky has argued for the nonexistence of Raising in [2]. In
what follows, I will reconsider this question and draw the following con¬
clusions: (i) the structures in (2) are not the appropriate surface repre-

*This paper was written during the spring of 1975 and is cited in my "Alternatives to the
Tensed S and Specified Subject Conditions.” It is published here for the first time.

131
 132 Realistic Syntax

sentations and (ii) the structures in (2) are not the appropriate base struc¬
tures; (iii) hence, it follows that there is no rule of Subject Raising.
Chomsky is right in denying the existence of Raising and Postal is right in
denying (2) as the surface structures corresponding to (1). It follows that
Chomsky is mistaken in assuming (2) to be the appropriate surface repre¬
sentations of (1) and Postal is mistaken in assuming Raising to exist.
In arriving at these conclusions, I will have occasion to review the
arguments for transformationally deriving each other, a position adopted
by Dougherty [3] and Chomsky [2]. It will be shown that this position is
untenable and that each other should be considered a lexical pronoun,
squaring with arguments provided independently by Fiengo and Lasnik
[5]. It will also be argued that each of the men and the men each should
not be related transformationally, contrary to the claims of both
Dougherty [3] and Fiengo and Lasnik [5].

2. Controversy over each other

Sentences such as the following have been the focus of several recent
articles, cf. [4], [5].

(4) a. Each of the two candidates criticized the other.

b. The two candidates each criticized the other.
c. The two candidates criticized each other.

These sentences are, according to Dougherty [3], transformationally re¬

lated. Thus, Dougherty derives (4b) from (4a) by a rule he calls Quantifier
Postposition, and (4c) from (4b) by two rules he calls Quantifier Move¬
ment and each ot/zcr-Transformation. To Dougherty’s credit is the fact
that he provides an explicit statement of his rules.
Chomsky [2] follows Dougherty in assuming a transformational rela¬
tionship between (4b) and (4c). The latter derives from the former by a
rule Chomsky calls each-Movement.
In contrast to Dougherty and Chomsky, Fiengo and Lasnik propose
“that each other never arises from a transformation, but rather is gener¬
ated freely as a deep structure pronominal NP” [5:447],
For the purposes of this discussion, let us call the Doughtery/Chomsky
approach the transformationalist hypothesis and the Fiengo/Lasnik ap¬
proach the lexicalist hypothesis. In this section I will review the stronger
arguments for the transformationalist hypothesis and show that they in
fact strongly support the lexicalist hypothesis. Additional arguments will
 Raising 133

be provided against the transformationalist hypothesis during the course

of this evaluation. These will be seen to augment those already provided
by Fiengo and Lasnik [5] and thus serve to support the lexicalist
hypothesis.

2.1 Multiple Occurrences of Distributive Quantifiers

One alleged argument for the transformationalist hypothesis concerns

multiple occurrences of distributive quantifiers as illustrated in the (b) and
(c) examples that follow.

(5) a. The men talked to each other.

b. *Each of the men talked to each other.
c. *Both of the men talked to each other.

Under the transformationalist hypothesis, the deviance of (5b) and (5c)

follows since the each of each other originates with the antecedent and is
moved adjacent to other by transformations. Since both instances of each
in (5b) cannot be generated in the subject position, it follows that (5b) will
not be derived. Similarly, since both both and each cannot be generated in
the subject position, it follows that (5c) cannot arise. On the other
hand, (5a) does have an underlying source, namely, each of the men
talked to the other, and therefore can be derived.
This appears to be an argument for the transformationalist hypothesis
and Dougherty advances it as such. The evidence, however, strongly
supports the lexicalist hypothesis, as we shall see. In this regard, consider
the following examples.

(6) a. The men talked to one another.

b. *Each of the men talked to one another.
c. *Both of the men talked to one another.

Here there is no possibility of a transformational source for one another,

since one is not a distributive quantifier analogous to each, both, and all.
Yet there is a striking parallelism between the examples of (6) and those of
(5). Proponents of the transformational analysis must provide a non-
transformational account of the deviance of (6b) and (6c). But this leads to
a loss of generalization since (6b) and (6c) are ruled out by one means, and
(5b) and (5c) are ruled out by a different means. Clearly, the nontransfor-
134 Realistic Syntax

mational account of the deviance of (6b) and (6c) should extend to the
examples provided in (5b) and (5c). This extension cuts the ground out
from under the transformational argument.
It should also be noted that the presence of each other does not always
preclude the appearance of a distributive quantifier with the antecedent, a
fact which argues against the transformationalist hypothesis. Thus, the
following example is not deviant, although one should expect it to be
under the assumptions of the transformational hypothesis.

(7) All of the men talked to each other.

Linguists as far apart in theoretical orientation as Postal on the one hand

and Fiengo and Lasnik on the other agree on the status of such examples
and cite the following.

(8) a. All of the men will hit each other. [5:110]

b. All of them were hitting and biting each other. [7:72]

Dougherty attempts to blunt the force of such examples by postulating

two quantificational realizations of all: “ ... the quantifiers of amount
would include a quantifier all and the distributive quantifiers would in¬
clude a quantifier all" [4:42]. The all of examples (8a) and (8b) are, ac¬
cording to this suggestion, quantifiers of amount, not distributive quan¬
tifiers. In this way, Dougherty attempts to dismiss the apparent coun¬
terexamples to the transformationalist hypothesis. The suggestion, how¬
ever, appears to raise new questions, which may not have a satisfactory
answer, indicating that we are actually dealing with a single quantifier
type. Thus, the double source for all predicts ambiguities for examples
such as all of the men left, where all in one case is an amount quantifier
and in another is a distributive quantifier. Yet it is not clear that the
relevant ambiguity can be perceived. True, such examples can mean that
the men left together or separately, but this is a general property of exam¬
ples without quantifiers, such as the men left, where we clearly do not
wish to postulate a distinction.
Further evidence corroborating the incorrectness of the transfor¬
mationalist hypothesis with respect to multiple occurrence of distributive
quantifiers concerns the fact that there are examples which allow both to
be associated with the antecedent of each other. Thus, the following
example is quite natural.

(9) Of the five boys, only John and Bill both shouted at each other
simultaneously.
 Raising 135

Such examples should not be possible, given the transformational ap¬

proach.
We see, in conclusion, that multiple occurrences of distributive quan¬
tifiers do not provide evidence for the transformationalist hypothesis.
Rather, we have found reasons to reject it.

2.2 Conjunction

As a second argument for the transformationalist hypothesis,

Dougherty offers the following examples.

(10) a. John, Bill, and Tom each hate goats,

a' John, Bill, and Tom hated each other,
b. *John, Bill, or Tom each hate goats,
b' *John, Bill, or Tom hated each other.

If each other derives according to the transformationalist hypothesis, it

follows that when the distributive quantifier each permits a conjunction,
as in (10a), there will be a corresponding grammatical sentence with each
other, as in (10a'); when each disallows a particular conjunction, as in
(10b), it also follows that the corresponding example with each other
should be ungrammatical, as in (10b'). Since this distribution of data is
actually observed in (10), it appears to constitute evidence favoring the
transformationalist hypothesis.
Further investigation shows just the opposite. Relevant data include the
following examples.

(11) a. John, Bill and Tom each criticized the officials,

a' John, Bill, and Tom criticized one another,
b. *John, Bill, or Tom each criticized the officials,
b' *John, Bill, or Tom criticized one another.

These examples parallel those provided above in (10). In place of each

other, the reciprocal pronoun one another has been substituted. Since one
is not a distributive quantifier, it will not be moved into place by a trans¬
formation. Yet one another exhibits a similar distribution to each other in
these examples. Advocates of the transformationalist hypothesis will
need a separate device to rule out (1 lb'), while utilizing a transformational
device to explain (10b'). This amounts to a loss of generalization. The
same device should be used to rule out both examples, and this entails a
nontransformational device in view of (lib').
 136 Realistic Syntax

The deviance of these examples is not particularly mysterious. It ap¬

pears that each other and one another are anaphoric pronouns which are
plural in number. Yet disjunctions are singular, not plural. Thus, the
judgments relating to the following examples with reflexive pronouns
parallel those for the (a'), (b), and (b') examples in (10) and (11).

(12) a. John, Bill, and Tom hate themselves.

b. *John, Bill, or Tom each hate themselves.
c. *John, Bill, or Tom hated themselves.

Clearly there is a deeper generalization here that relates to number

agreement, not to each per se. This, once again, shows that the transfor¬
mational hypothesis is misguided.

2.3 Distribution ofeach other

As a third argument for the transformationalist hypothesis, Dougherty

notes that “no ad hoc rules need be postulated ... to account for the
correct distribution of the pronoun each other, since this is accounted for
by independently motivated rules ...” [3:884], Thus, the deviance of the
following example follows from the transformational analysis.

(13) *Each other will win.

Since the each of each other moves from its antecedent and since there is
no antecedent in (13) to provide a source for each, it follows that each
other simply could not arise.
This approach, however, leads, once again, to a loss of generalization.
The missed generalization becomes obvious when we consider examples
such as the following.

(14) a. *One another will win.

b. *Himself will win.
c. *Him will win.

Advocates of the transformationalist hypothesis must rule out the exam¬

ples in (14) by an independent nontransformational device, but surely this
device should be utilized to rule out (13). To fail to do so is to miss a
generalization. To do so, however, is to admit that examples such as (13)
provide no evidence for the transformationalist hypothesis.
 Raising 137

Again, these examples provide no mystery. All violate the requirement

that subjects of tensed phrases must be subjective pronouns. If we take
each other, one another, himself, and him all to be objective pronouns,
then all violate this requirement.
Let us also note that the transformationalist hypothesis does not in fact
predict the correct distribution of each other. Thus, Postal has noted that
this approach predicts deviance for examples with two occurrences of
each other, where there is only a single subject antecedent. Such exam¬
ples include the following, (15a) being due to Postal [6:723].

(15) a. The men talked to each other about each other.

b. Those two men saw each other and each other’s wives.

Such examples cannot be generated under the transformationalist

hypothesis and thus constitute evidence against it.

2.4 Universal Movement Constraints

Dougherty claims that “a fourth argument in support” of the transfor¬

mationalist hypothesis “involves a putative language universal.” This is
Chomsky’s Universal Movement Constraint which prohibits the introduc¬
tion of “morphological material into a configuration dominated by S once
the cycle of transformational rules has already completed its application
to this configuration” [3:885]. This alleged constraint, argues Dougherty,
accounts for the deviance of (16b).

(16) a. John, Bill, and Tom each thought that Mary had stopped seeing
the others.
b. *John, Bill, and Tom thought that Mary had stopped seeing each
other.

According to Dougherty, Chomsky’s constraint keeps each from moving

into the embedded S in (16b) from its matrix position in (16a). This result
follows, it is argued, only if each other derives from (16a) by a transforma¬
tion. Chomsky [2], of course, adopts Dougherty’s suggestion and revises
his constraint as the Tensed S and Specified Subject Conditions.
In fact, the alleged constraint does not account for (16b), nor do the new
conditions which Chomsky introduces, without loss of generalization. To
see this, note that the transformational hypothesis does not subsume the
following examples.
138 Realistic Syntax

(17) a. *John, Bill, and Tom thought that Mary had stopped seeing one
another.
b. *John, Bill, and Tom thought that Mary had stopped seeing
themselves.

What goes wrong with these examples is a violation of number concord

between anaphoric pronouns and the subject Mary. The pronouns one
another and themselves are plural, whereas Mary is singular. The same
goes for (16b), where each other is plural. If we formulate anaphora so as
to identify the anaphoric pronoun with its subject, then the generalization
can be expressed in terms of ultimate number concord violation. If we
attempt to account for (16b) by means of a transformation moving each,
we miss this generalization.

3. New arguments against the transformationalist

H YPOTHESIS

We have seen that the evidence adduced in support of the transfor¬

mationalist hypothesis actually favors the lexicalist hypothesis. Let us
now turn to additional facts bearing out this conclusion.
The foregoing discussion indicates that each other is a lexical pronoun.
This conclusion is supported by Fiengo and Lasnik's observation relating
to particles. It is well known that in the stress-neutral case, particles may
not precede pronouns. Thus, (18a) is fine, while (18b) is decidedly ill-
formed when up is given the particle interpretation.

(18) a. John looked her up.

b. *John looked up her.

This distinction does not exist in the case of nonpronominal NPs.

(19) a. John looked Mary up.

b. John looked up Mary.

Now it is interesting that the pronominal distribution holds of each other,

as illustrated below, where up is again interpreted as a particle.

(20) a. The two boys looked each other up.

b. *The two boys looked up each other.
 Raising 139

Fiengo and Lasnik’s observation shows that each other is a pronoun. It

can now be extended to argue against the transformational hypothesis,
since the other exhibits the nonpronominal NP distribution.

(21) a. The two boys each looked up the other,

b. The two boys each looked the other up.

Clearly it is no coincidence that the other functions as a nonpronominal

NP while each other functions as a pronoun. But this is sheer coincidence
under the transformationalist hypothesis. How is it that the movement
transformation just happens to convert the nonpronominal NP to a pro¬
nominal NP? By contrast, if each other is a lexical pronoun, then it is not
a coincidence that it functions like one another and other lexical pronouns
from which it not only follows that we find (21) above, but it also follows
that we obtain the following distinction.

(22) a. The two men each thought that the other would win.
b. *The two men thought that each other would win.

Example (22b) is deviant because each other, like other analogous

anaphoric pronouns is objective, as noted earlier. By contrast, the other is
not an objective pronoun and can function as subject. We thus find that
(22b) is out for the same reason as the following.

(23) a. *The two men thought that one another would win.
b. *The two men thought that themselves would win.

There are, in fact, plenty of examples showing that each other and the
other must be distinguished.

(24) a. Each of the two clowns pretended to be funnier than the other,
b. *The two clowns pretended to be funnier than each other.
(25) a. On successive days, each of the two boys is smarter than the
other.
b. *On successive days, the two boys are smarter than each other.

Perhaps it is wrong to star the (b) examples, but it is clear that a quite
distinct reading is perceived, the (b) examples involving a contradictory
reading.
There are, in fact, examples which are not treated by the alleged
movement constraint, or its modern progeny.
140 Realistic Syntax

(26) a. Each of the two girls left without the other,

b. *The two girls left without each other.

Such examples show that each other does not exhibit a distribution coex¬
tensive with the other, a somewhat embarrassing state of affairs for the
transformationalist hypothesis. Such examples show that the other may
appear in environments in which each other turns out to be ill-formed, per¬
haps semantically. The following examples illustrate the converse.

(27) a. *The two politicians each told lies about Harry and the other,
b. The two politicians told lies about Harry and each other.
(28) a. *In addition to the other, the two heroes each loved their wives,
b. In addition to each other, the two heroes loved their wives.
(29) a. *Each of these two toads, with the exception of the other, can be
said to love all other toads.
b. These two toads, with the exception of each other, can be said
to love all other toads.
(30) a. *It was the other that they each scolded,
b. It was each other that they scolded.
(31) a. *Sockeye and Coho salmon each does not spawn with the other,
b. Sockeye and Coho salmon do not spawn with each other.
(32) a. *Each kick the other!
b. Kick each other!
(33) a. *Chained to the other though they (each) were, they (each) still
escaped.
b. Chained to each other though they were, they still escaped.
(34) a. *If you two each help the other, you’ll manage better,
b. If you two help each other, you’ll manage better.
(35) a. *Remembering the other made them each forget their troubles,
b. Remembering each other made them forget their troubles.
(36) a. *1 asked which men had each stolen from the other,
b. I asked which men had stolen from each other.
(37) a. *Who here will each know the other?
b. Who here will know each other?

Such examples can easily be multiplied. They show that each other oc¬
cupies many environments disallowing each ...the other. This distribu¬
tion is unexpected under the transformationalist hypothesis.
The situation is even worse for proponents of the transformationalist
hypothesis, for there are examples for which there is no plausible source
for each!
 Raising 141

(38) a. To challenge each other would lead to war.

b. Challenging each other would ultimately amount to fighting each
other.
c. Imprisoned with each other for fifteen years, the two inmates
each grew tired of the other.

Consider, for example, (38b). What is the source of the two instances of
each under the transformationalist hypothesis? No acceptable answer
appears likely other than the obvious conclusion that each other is what it
is, a lexical pronoun, generated under the lexicalist hypothesis as such.

4. the men each and each of the men

Another argument might be advanced in support of the transfor¬

mationalist hypothesis. This argument might have to do with synonymy
and cooccurrence restrictions. Thus, it might be claimed that the follow¬
ing are synonymous.

(39) a. Each of the boys hit the other,

b. The boys hit each other.

If (39b) derives from (39a), by transformation, then both might be ex¬

pected to yield the same interpretation, since both derive from an identi¬
cal source. Second, it might be noted that similar cooccurrence restric¬
tions apply in many instances. Thus, when (39a) is grammatical, so is
(39b). And when an underlying alleged source is ungrammatical, as in
(40a) below, then so is the corresponding derived example in (40b).

(40) a. *Each of the man hit the other,

b. *The man hit each other.

The cooccurrence argument fails, as noted in the previous section. But it

does so even here in connection with examples such as (40). This follows
from the fact that the following examples illustrate a parallelism with (40),
but here there is no plausible analogous underlying source.

(41) a. *The man hit one another,

b. *The man hit themselves.
142 Realistic Syntax

Again, the transformationalist must lose a generalization, accounting for

the deviance of (40b) by one mechanism and (41) by another. Clearly,
(40b) should be ruled out by the same mechanism which rules out (41a)
and (41b), which does not relate to distributive quantifiers per se. All
involve a violation of number concord between the anaphoric pronoun
and its antecedent; this shows that each other, like one another and
themselves, is an anaphoric pronoun. There is no need to derive each
other from a more abstract source.
Dougherty makes an observation which is related to this discussion.
Thus, he cites the following examples.

(42) a. Each of the men will drink a beer,

b. The men each will drink a beer.

Dougherty claims that Quantifier Postposition “accounts for two facts,

the (a) and (b) sentences are synonymous” and “the quantifier in the (a)
sentence has the same selection restrictions as the quantifier in the (b)
sentence” [3:877], although he qualifies this comment, noting that
“synonymy and identity of selection restrictions are not motivation
enough to establish a transformational relation” [3:877],
Now let us examine Dougherty’s comments more carefully. His claim is
clearly that synonymy and selection restrictions are preserved under
transformation and that the results in (42) are consistent with this claim.
He indicates that the alleged transformation of Quantifier Postposition
“accounts” for such “facts”. But if synonymy and selection restrictions
are preserved under Quantifier Postposition, they should also be pre¬
served under the transformations which are alleged to give rise to each
other. Therefore, the transformational hypothesis would seem to predict
that the following are synonymous.

(43) a. Each of the two men slaughtered the other,

b. The two men slaughtered each other.

These examples are not synonymous. Two men can slaughter each other
by striking each other simultaneously with an ax, but (43a) would not be
appropriate in this case; rather (43b) would be used. On the other hand, it
would not be peculiar to say that each of the two men slaughtered the
other in successive games of chess, though in this sense (43b) would be
peculiar and inappropriate if the slaughtering occurred on the part of Sam
in the first game and on the part of Harry in the second. We see, then, that
the subject-predicate relation differs for (43a) and (43b) and that this dif-
 Raising 143

ference can be crucial. Thus, (43a) and (43b) are not synonymous, a fact
recognized by Fiengo and Lasnik [5].
We see, then, that synonymy and cooccurrence cannot be summoned
as an argument for the transformationalist hypothesis, in particular, for a
transformational derivation of each other. Can they, however, be ad¬
vanced as an argument for Quantifier Postposition, which is alleged to
relate each of the men and the men each. It is interesting to note that
Fiengo and Lasnik accept this relation as a transformational relation,
although, as noted, they reject the transformation(s) deriving each other. I
would like to suggest, however, that each of the men and the men each
are not transformationally derived, contrary to Fiengo, Lasnik, and
Dougherty. Crucial examples include the following.

(44) a. There are two boys each helping the other.

b. *There are each of two boys helping the other.
(45) a. There are two squirrels both eating peanuts.
b. *There are both of two squirrels eating peanuts.
(46) a. There are six thousand girls all singing. ^
b. *There are all of six thousand girls singing.

Now it has been argued elsewhere that there is no there-Insertion trans¬

formation, cf. [7], [1]. Rather, there should be base generated. And yet we
find that the (a) examples are grammatical. If the distributive quantifiers
each, both, and all derive from the subject position, then the (b) examples
should emerge as well-formed. This is not the case, indicating that the
relevant examples are not in fact transformationally related.

5. expect NP VP versus expect S

We have concluded that each other is a lexical pronoun and that it

should not be derived transformationally. It has also been suggested that
there is no transformational rule giving rise to the men each from each of
the men. We can now utilize this result to argue against the abstract S
structure which Chomsky posits as the surface structure of examples such
as John believes Harry to be stupid and John expected Harry to leave.
His proposal for surface structures is provided below.

(47) a. John believes [Harry to be stupid]s

b. John expected [Harry to leave]s
 144 Realistic Syntax

The usual assumption is that the following are the appropriate surface
structures, derived from (47) by a rule of Raising (and Pruning).

(48) a. John believes [Harry]NP [to be stupid]VP

b. John expected [Harry]NP [to leave]VP

Let us refer to the structures in (47) as subject structures and those in (48)
as the object structures. That the object structures should in fact be the
base structures was argued in [1], We therefore have three positions: the
traditional position, accepted by Postal, that (47) represents the base
structures and (48) the surface structures; Chomsky’s position, that (47)
represents both the base and surface structures; and my own, that (48)
represents both the base and surface structures.
Now each other cannot occupy the subject position in clear instances of
subjects, as noted earlier, cf. (13).

(49) a. *Each other are stupid,

b. *Each other leave.

Hence, we should conclude that each other is simply not the subject in
examples such as The boys believed each other to be stupid and The girls
expected each other to leave. The evidence, therefore, points to (48) as
the correct surface structures, as argued independently and at great length
by Postal [7],
Since Chomsky takes the subject structures to be the appropriate sur¬
face structures, we might pause to consider his reasons for doing so. In
[2], he provides three arguments for this position. Let us consider them in
turn. His first argument is recorded verbatim.

... consider The men were told to expect John to kill the other(s).
If John is raised on an internal cycle, it will no longer be a subject
when the matrix cycle is reached so that each-movement should
apply to give *The men were told to expect John to kill each
other. Similarly, we have We were told to expect John to kill me
but not *We were told to expect to kill me, the former inconsist¬
ent with subject raising ... [2:254, n. 33]

This argument is predicated on the existence of the Specified Subject

Condition, introduced in [2], Chomsky assumes that each cannot be
moved past a specified subject, in accordance with this constraint, which
is a revised version of the alleged universal constraint discussed earlier.
 Raising 145

Chomsky’s argument here is that if John is raised into object position (or
originates there as I claim), then we would expect *The men were told to
expect John to kill each other since the Specified Subject Condition would
not then be applicable. Since the result is ungrammatical, it follows, ar¬
gues Chomsky, that John is the subject dominated by S.
This argument goes wrong in several ways. First, it has been shown
above that there is no rule of eac/7-movement. Second, suppose that there
is a relevant notion of subject. It does not follow that subject need be
defined as that NP directly dominated by S. In fact, we could take John to
be a subject in the relevant example and still adopt the obvious object
structure. Third, there is really no evidence for the Specified Subject
Condition as formulated in [2].
What then is the reason for the deviance of *The men were told to
expect John to kill each other. Following Emonds’ suggestion, cf. [1], let
us assume that each other must be marked anaphoric with some NP which
stands in some grammatical relation to the verb to which the anaphoric
pronoun itself relates. If John is taken to be the subject of kill in an
extended sense of subject, not the traditional Aspects definition, then
each other will be marked as anaphoric with John. But here, as with
previous examples discussed earlier, we end up with a violation of
number agreement. The pronoun is plural, while John is singular.
The example involving disjoint reference can be treated in an analogous
way. Two NPs must stand in a grammatical relation to the same verb in
order to qualify for disjoint reference. In We were told to expect John to
kill me, John and me are disjoint in the intended sense and in *We were
told to expect to kill me, we and me are marked as disjoint, which provides
a violation of what is already stipulated about first person pronouns,
namely that first person pronouns intersect in reference.
Let us now turn to Chomsky’s second argument for the subject struc¬
tures. Again, a direct quote, follows.

Another problem is suggested by some observations of John

Kimball, who points out that from It was easy for Jones to force
Smith to recover we can derive Smith was easy for Jones to force
to recover, but from It was easy for Jones to expect Smith to
recover, we cannot form *Smith was easy for Jones to expect to
recover. Assuming subject raising [or underlying object struc¬
ture, MKB], the two sentences are identical at the point where
//-Replacement takes place. If there is no subject raising, the rule
of //-Replacement can make the required distinction by permit¬
ting the NP moved to be followed by S (as in Bill is easy to
persuade that the moon is made of green cheese). [254:n. 33]
 146 Realistic Syntax

The data mentioned in this quote should, I believe, be attributed to Kis-

seberth, who mentioned them in unpublished work around ’67. We can
see quite clearly that Chomsky’s account does not argue for the distinc¬
tion. Thus, it does not appear to be true that there must be an S following
an NP for //-Replacement to be applicable. For example, in the case of it is
easy to please John, there is no following S, and yet we derive John is
easy to please, assuming that //-Replacement exists (which I do not, cf.
[1]). Thus, Chomsky’s proposal does not work. Second, even when a
following S is assumed to exist with Chomsky’s framework, we some¬
times end up with examples which appear to be equally deviant as the
expect example. Thus, consider the following parallelism.

(50) a. *Mary is easy to expect to participate.

b. *Mary is easy to promise to be allowed to participate.

I Although it is easy to promise Mary to be allowed to participate is fine,

(50b) seems to be strange, certainly as strange as (50a). In fact, we seem to
be dealing here with special properties of the predicates themselves, for
notice the following data.

(51) a. It is easy to promise Mary (but not to deliver on the promise).

b. It is easy to expect Mary (but not to see her show up).
(52) a. *Mary is easy to promise,
b. *Mary is easy to expect.

3 These observations show that Chomsky’s suggestion has nothing to do

with what is in fact going on here. Thus, we find the following correlation.

(53) a. It is easy to persuade Mary,

b. Mary is easy to persuade.

When the example without the extra VP complement is grammatical, as in

(53), then apparently the same example is fine with a VP-compiement.
When it is deviant, as in (52), then the corresponding example with VP-
complement is also deviant, as in (50). This correlation shows that
Chomsky’s suggested solution is not sufficiently general.
Finally, Chomsky offers a third argument for the abstract subject struc¬
ture with abstract S. This argument relates to the following example.

(54) *Who do you expect stories about to terrify John.

 Raising 147

The argument runs as follows, where the number reference has been
altered to agree with (54) above.

If there were such a rule [Raising], wh-Movement should apply,

giving [54], analogous to Who did you see pictures of last night,
Who did you tell stories about at the campfire. These forms,
though hardly elegant, seem to me much more acceptable than
[54], as we would expect on the assumption that there is no rule
of subject raising to object position. [250: n. 31]

Here the argument once again turns on the alleged Specified Subject
Condition. Movement of the wh-word to yield (54) involves a violation of
the putative condition, provided, the argument goes, stories about who is
taken to be the specified subject dominated by S. This argument collapses
when we once again observe that we are actually dealing with a more
general phenomenon. Consider the following examples.

(55) a. John sent stories about someone to Mary.

b. *Who did John send stories about to Mary?
(56) a. John told stories about someone to his best friend.
b. *Who did John tell stories about to his best friend.

Clearly, these examples are relevant to this discussion. The (b) examples
show that there is a deeper generalization to be expressed which does not
relate to the notion subject.
We see, then, that none of Chomsky’s arguments holds up upon
scrutiny. There is simply no evidence that the subject structures should be
favored as surface structures, nor is there evidence that the object struc¬
tures should not be taken to be the appropriate surface structures. In fact,
Postal has provided a wealth of evidence to support the claim that the
object structures are the appropriate surface structures, cf. [7].
It is appropriate to ask whether there is any evidence that the subject
structures are the appropriate base structures, or, alternatively, whether
there is any evidence that the object structures are not the appropriate
base structures. In [1] it has been argued that the object structures are in
fact the appropriate base structures. The traditional arguments simply do
not hold up. Thus, all of the following can be taken to be the appropriate
base structures, as well as surface structures.

(57) a. John expected [Mary]NP [to leave]Vp

b. John believes [Mary]NP [to be rich]Vp
c. John persuaded [Mary]NP [to leave]Vp
d. John promised [Mary]NP [to leave]Vp
148 Realistic Syntax

This conclusion in turn has important consequences for transformational

grammar, as shown in [1]. A number of classical transformations must be
abandoned and replaced by other devices.
We thus arrive at a position which is in a sense intermediate between
that of Postal on the one hand and Chomsky on the other. With Chomsky
we agree that there is no rule of Subject Raising, a rule which Postal
accepts, and with Postal we agree that object structures are the appropri¬
ate surface structures, which Chomsky denies; at the same time, we dis¬
agree with both Chomsky and Postal in accepting subject structures,
whether in deep structure, as with both Chomsky and Postal, or in surface
structure, as with Chomsky. The claim that object structures are appro¬
priate both as surface and base structures is a position which has far-
reaching consequences for the theory of grammar, consequences which
are, however, highly desirable.

References

1. Brame, M. K. (1976) Conjectures and Refutations in Syntax and

Semantics. New York: Elsevier North-Holland.
2. Chomsky, N. (1973) Conditions on Transformations. In A Festschrift
for Morris Halle, S. R. Anderson and P. Kiparsky, eds. New York:
Holt, Rinehart, and Winston.
3. Dougherty, R. C. (1970) A Grammar of Coordinate Conjunction, Part
I. Language 46: 850-98.
4. Dougherty, R. C. (1974) The Syntax and Semantics of each other
Constructions. Foundations of Language 12: 1-47.
5. Fiengo, R. and H. Lasnik (1973) The Logical Structure of Reciprocal
Sentences in English. Foundations of Language 9: 447^168.
6. Jenkins, L. (1972) Modality in English Syntax. Doctoral dissertation,
MIT. Distributed by Indiana University Linguistics Club.
7. Postal, P. M. (1974) On Raising: One Rule of English Grammar and
Its Theoretical Implications. Cambridge: MIT Press.
THE BASE HYPOTHESIS AND THE SPELLING
PROHIBITION

Sensei, the Spelling Prohibition goes far in spite of its simplicity.

Yes, gakusei, it even has implications for phonology, but it is for us alone.
Those who abuse our friends, the words, cannot open their minds, for
they know too much.
Introductory Remarks
1. The Base Hypothesis
2. Affix-Hopping and there -Insertion
3. Extraposition and Sentential Subjects
4. The Spelling Prohibition
5. Lexicalism vs. The Inverse Cycle
References
 The Base Hypothesis and the Spelling Prohibition*

In the first section of this essay several arguments are recapitulated

which appeared in an earlier publication [2]. These arguments turn on
the nonexistence of Equi, a rule which has been posited by transfor¬
mationalists working within the framework of the standard theory.
The consequences of abandoning Equi are also reviewed and shown
to culminate in what is called the “base hypothesis.”
Section 2 is given over to new motivation for the base hypothesis.
The basic argument concerns Affix-Hopping and existentials employ¬
ing the expletive there. It is argued that a basic generalization can be
expressed if Affix-Hopping and there-Insertion are supplanted with
base generated structures. It is also suggested that VP gaps be
generated directly by base rules. These results lead once again to the
conclusions drawn in the first section. Thus, section 2 can be viewed
as additional support for the base hypothesis.
Section 3 undertakes a brief reanalysis of sentential subjects and
related phenomena. It is shown that a wide range of data previously
described by constraints such as Ross’s Sentential Subject Constraint
and Island Internal S Constraint, Emonds’ prohibition against multi¬
ple application of root transformations, Horn’s NP Constraint and
global Pruning, etc., can in fact be explained by a straightforward
elaboration of the base. It is therefore suggested that Extraposition is
not a transformation in the classical sense, a result consistent with the
base hypothesis.
In section 4 a general constraint on the functioning of transforma¬
tions is advanced. The effect of this constraint, called the Spelling
Prohibition, is to choose grammars consistent with the arguments
advanced in the first three sections. Thus, the Spelling Prohibition
requires that grammars be formulated so as to be consistent with the
informal base hypothesis sketched out in section 1.
The question of alternatives to the standard theory, including the

* This paper was delivered at the Eighth Annual Colloquium on Directions in

Linguistics held at the University of Calgary on March 21, 1977.

151
152 Michael K. Brame

inverse cycle approach introduced in [2] and Bresnan’s lexicalist

model, is touched on in section 5. Some evidence is introduced in this
section to empirically distinguish the two approaches vis-a-vis the
question of passives.
Finally, section 6 is an all too brief excursus into the transforma¬
tional component. In this section a powerful constraint on the form of
transformations is conjectured and alternative models are mentioned,
including a hybrid of the lexicalist model and that introduced in [2].

1. THE BASE HYPOTHESIS

Within the standard theory of transformational grammar, a problem

of overgeneration emerges. This problem has not really been con¬
fronted. It concerns the following examples:

(1) a. John tried to avoid the issue.

b. *John tried for Mary to avoid the issue.

(2) a. I will persuade you to recognize the superiority of an Elna.

b. *1 will persuade you for John to recognize the superiority of an
Elna.

(3) a. Mary wants us to provide genever for everyone.

b. *Mary wants us for Harry to provide genever for everyone.

The problem is to avoid generating examples such as (lb), (2b), and

(3b), while allowing for (la), (2a), and (3a). Various constraints have
been proposed to accomplish this, but none of these constraints
touches the heart of the problem. They simply amount to redescrip¬
tions of the basic problem.1 Rather than proposing new constraints
which serve no independent purpose, we might alternatively question
some of the basic assumptions which contribute to the generation of
the ill-formed examples. Thus, taking (1) as a representative example,
we might ask why (lb) should arise in the first place as a problem for
the standard theory. The answer in this case is quite obvious.
Example (lb) arises because (la) is derived from an underlying
structure incorporating a sentential complement, as illustrated in (4).

1 To take just one example, consider Lakoffs absolute exception mechanism

advocated in [28], Lakoff marks predicates such as try with ad hoc absolute exception
features so that they will obligatorily meet the structural description of Equi. In other
words, try is syntactically irregular within Lakoffs framework for appearing in the
structures in which it appears. For summary and criticism of proposals for dealing with
examples such as (lb), (2b), and (3b), see Brame [2: 94-97, 103-107],
 The Base Hypothesis and the Spelling Prohibition 153

(4) a. John tried [s for John [VP to avoid the issue]Vp ]s

b. John tried [s for PRO [VP to avoid the issue]VP Is

Whether one adopts (4a), as in the standard theory, or (4b), as in the

extended standard theory, the very fact that try can be lexically
subcategorized so as to select an S-complement gives rise to new
problems connected with ill-formed sentences such as (lb), and
likewise for (2) and (3) with persuade and want. However, if one
rejects this basic assumption and adopts instead what is recognized to
exist on the surface, i.e., a VP-complement as the underlying subcate¬
gorization associated with try, then the question of the ungrammati¬
cally of (lb) does not even arise. Likewise, if persuade, want, etc.,
are subcategorized so as to select NP VP-complements in the case of
(2a) and (3a), then the question of the ungrammaticality of (2b) and
(3b) does not even arise. Rather, (2b) and (3b) are out for precisely
the same reason as the following examples are out.

(5) a. *1 saw Mary for John to leave.

b. *Mary pinched me for Harry to speak up.
c. *They all recognized Bill for us to forget our grudge.

Thus, persuade and want, like see, pinch, and recognize, do not
select NP S-complements, although the former do select NP VP-
complements.
We see then that the standard and extended standard theories not
only must contend with examples such as (lb), (2b), and (3b), but that
the means that these theories employ to avoid such examples
amounts to a loss of generalization inasmuch as the constraints
proposed to rule out such examples do not generalize with mecha¬
nisms proposed to rule out examples such as (5a—c). By contrast, the
approach that I have advocated, cf. Brame [2], avoids the ill-formed
examples in a general, straightforward way. Predicates such as try
select VP-complements and predicates such as want and persuade
select NP VP-complements.
Such an approach has further consequences. Consider the well-
known problem of the following examples:

(6) a. Who do you want_to see John?

b. *Who do you wanna see John?

As has been pointed out often, reduction and contraction are prohib¬
ited in the presence of a removal site as in (6b). Now, if there is
154 Michael K. Brame

indeed a rule of Equi, we would expect (7b) to be ill-formed just as in

(6b).

(7) a. Do you want_to sleep?

b. Do you wanna sleep?

But (7b) is fully grammatical, suggesting that no equi-NP was in fact

present in the structure underlying (7b). Consequently, we are led to
conclude that there is no rule Equi. It then follows without theoretical
elaboration that (7b) is possible, whereas (6b) is not, since there is no
removal site in the case of (7b), as opposed to (6b), which does
exhibit a removal site. This explanation strikes me as natural and
straightforward. It requires no theoretical subtlety. It demands no
theoretical elaboration to write off (7b) as counterevidence. Things
follow without employment of theoretical prophylactic. It is the
difference between an explanation and a description.
There are some important consequences of our conclusion that
Equi does not exist. Consider the following examples:

(8) a. Mary was examined by the doctor.

b. Mary tried to be examined by the doctor.

(9) a. They expected Jimmy to rip us off.

b. Jimmy was expected by them to rip us off.

It is generally assumed that (8a) derives from its active counterpart

via a transformation called Passive. But consider the derivation of
(8b).

(10) a. Mary tried [s the doctor to examine Mary]s =$> Passive

b. Mary tried [s Mary to be examined by the doctor]s => Equi

c. Mary tried_to be examined by the doctor

To derive (8b), utilizing the Passive transformation, we arrive at stage

(b) of (10), which requires Equi to derive the desired results. Thus, if
Passive exists, then Equi exists. But we have concluded that Equi
does not exist. Therefore, we must conclude that Passive does not
exist. It follows, then, that passive structures, like VP structures,
should be generated directly. This is not at all an undesirable
consequence. In fact, if passives are generated directly, there is no
need for an agent-deletion transformation since agents need not be
generated at all by base structures, cf. [2: 133].
Let us now turn to (9a) and (9b). Examples such as (9a) are
 The Base Hypothesis and the Spelling Prohibition 155

generally derived by a transformation that raises the NP subject of the

complement into the object position of the matrix S. Let us call this
putative rule, Raising to Object Position, or ROP for short. Notice
that ROP is also utilized in deriving (9b). The requisite derivation is
given in (11).

(11) They expected [s Jimmy to rip us off]s ROP

They expected Jimmy [s to rip us off]s ==> Passive

Jimmy was expected by them to rip us off

Thus, if we assume that there is a rule of ROP, then we need Passive

to yield the desired results in (11). But there is no rule of Passive
because there is no Equi. Consequently there is no ROP.
Similar conclusions can be reached for other transformations that
do not make essential use of variables, what I will call local
transformations. Thus, consider the following examples:

(12) a. Mary tried to appear to be calm.

b. Every child was given a dime.
c. He tried to be easy to please.
d. There was believed to be jelly between his toes.

By reasoning completely analogous to the foregoing, it is possible to

conclude on the basis of such examples that there are no rules such as
Raising to Subject Position (hereafter RSP), Dative, and the re¬
insertion. Thus, consider (12a), on the assumption that RSP applies in
the course of its derivation.

(13) a. Mary tried [s it to appear [s Mary to be calm] ] => RSP

b. Mary tried [s Mary to appear [s to be calm] ] => Equi
c. Mary tried_to appear to be calm

Again, we see, quite clearly, that Equi is a consequence of RSP for

such examples. Thus, if we are to maintain our assumption that there
is no rule of Equi, then we must also give up RSP and search for an
alternative account of the range of data that it explains. Or, turning to
(12b), under the assumption that Dative moves an NP leftward, we
obtain the following derivation:

(14) a. A gave a dime to every child =$ Dative

b. A gave every child a dime => Passive (and A-deletion)
c. Every child was given a dime.
156 Michael K. Brame

Here we see that if Dative applies to (14a) to yield (14b), then Passive
must subsequently apply so as to derive (14c). But we previously
concluded that Passive does not exist since Equi does not exist.
Hence if Dative is contingent on Passive, there can be no Dative in
the form assumed for this derivation. And likewise for Object-Shift.
Thus, consider the derivation associated with (12c).

(15) a. He4 tried [s it to be easy [s A to please he,] => Object-Shift

b. He* tried [s he* to be easy [s A to please] =$ Equi
c. He tried_to be easy to please

After Object-Shift has applied to yield (15b), Equi is needed to

provide for the desired surface structure. Thus, Object-Shift must be
eliminated also if Equi is given up. And, finally, it is possible to arrive
at a similar conclusion with respect to there-Insertion, as indicated in
the derivation of (12d).

(16) a. A believed [s jelly to be between his toes] => there-Inser¬

tion
b. A believed [s there to be jelly between his toes] =} ROP
c. A believed there [s to be jelly between his toes] ==> Passive
d. There was believed to be jelly between his toes.

Derivation (16) illustrates a contingency of t/icrc-Insertion on ROP

and Passive, which, recall, do not exist if Equi does not exist.
Proceeding on the assumption that Equi does not exist, we conclude
that there-Insertion does not either.
Thus, we see that the consequences of giving up Equi are some¬
what drastic. One could perhaps take these conclusions as a reductio
ad absurdum argument against abandoning Equi, but I will accept the
correct alternative hypothesis and assume that all structures previ¬
ously derived by local transformations must in fact be generated as
base structures. Let us call this approach the base hypothesis.

Base Hypothesis:
All structures previously derived by local transformations are in
fact generated by phrase structure rules.

The base hypothesis was first advanced in the course of my Fulbright-

Hays lectures at the Rijksuniversiteit te Utrecht in the fall of 1973 and
later published as [2]. If one accepts the base hypothesis, the
interesting question of just what is to replace the traditional local
transformations arises. I will have more to say concerning this
 The Base Hypothesis and the Spelling Prohibition 157

question below. First, however, I wish to provide several additional

arguments to support the base hypothesis.

2. AFFIX-HOPPING AND the re-Insertion

In this section new evidence will be adduced to support the base

hypothesis. First, let us consider the putative rule known as Affix-
Hopping, which was initially proposed and formulated in Chomsky
[10],[11]. The key fact is that Affix-Hopping is a local transformation
in our sense. That is, it does not make essential use of variables and,
therefore, according to the base hypothesis, is not a transformation.
There is, in fact, some evidence indicating that auxiliaries can be
generated in English directly by base rules and that direct generation
is desirable. This approach makes use of a more sophisticated theory
of the lexicon, one promising approach being that developed in Hust
[21],[22].
By way of presenting the evidence, let us note first that the
standard theory misses a significant generalization. Consider, for
example, the following ungrammatical cases:

(17) a. *John is being drinking beer.

b. *John’s being drinking beer all day surprises me.
c. *Being drinking beer all day, John is slowly going under.

Consider first (17a). How is this example prohibited within the

framework of the standard theory? It is blocked by virtue of the
phrase structure rules expanding Aux, as in (18).

(18) Aux -> tense (M) (have en) (be ing)

Since two instances of be ing cannot be generated by rule (18),

example (17a) will never arise subsequent to Affix-Hopping. Thus,
one mechanism, namely, phrase structure, is utilized to block (17a).
However, a totally different mechanism must be utilized to block
(17b) and (17c), since presumably the gerundive and participial
phrases in such examples are derived via transformation from senten¬
tial sources in the standard theory. In fact, Chomsky quite explicitly
remarks that examples such as (17b) are blocked by a special
constraint:

(19) Forms such as *John’s being reading the book (but not John’s
having been reading the book) are blocked by a restriction
against certain -ing -ing sequences (compare *John s stopping
reading, John’s having stopped reading, etc.) [14:16, note 6]
158 Michael K. Brame

Thus, examples (17b) and (17c) are ruled out by a special restriction
on -ing -ing sequences, constituting one mechanism, whereas exam¬
ple (17a) is ruled out by a totally different mechanism, phrase
structure. This amounts to a loss of generalization. The significant
generalization for all the examples appears to be that being does not
select -mg-complements. This generalization could be expressed if
being were a lexical item, but it goes unexpressed under the tradi¬
tional approach utilizing Affix-Hopping.
I will sketch briefly and tentatively one possible lexical approach
below. But first let us turn to a second set of examples which bears
on the base hypothesis.

(20) a. There was a boy being arrested.

b. *There was being a boy arrested.
c. There were to be three boys arrested.

Examples such as these pose critical problems for the standard theory
of transformational grammar, where expletive there is inserted via
transformation. Within such an approach adopting t/icrc-Insertion,
special conditions and complications must be associated with this rule
in order to keep it from applying so as to provide (20b). It might be
asked how such restrictions could correctly prohibit (20b) and yet
allow (20c). If we assume that such restrictions are, in fact, formula-
ble so as to account for the data in (20), we must still conclude that
such an account leads to a loss of significant generalization inasmuch
as it fails to provide a general account of the examples in (20) and
those in (21).

(21) a. It was a boy being arrested.

b. *It was being a boy arrested.
c. It was to be a boy arrested.

Apparently (21b) must be ruled out by a mechanism completely

independent of t/zere-Insertion, if there is inserted transformationally
in (20). However, the correct approach would utilize the same
mechanism to rule out (21b) and (20b). This mechanism entails
generating existential sentences directly. It thereby supports the base
hypothesis which provides for direct generation of existentials.
A tentative account of the data can be sketched by making use of
Must’s theory of the lexicon along with a version of Emonds’
approach to auxiliaries in English. Tentative phrase structure rules
are provided in (22).
 The Base Hypothesis and the Spelling Prohibition 159

(22) a. S -■» Comp S

b. S -» NP Aux VP
c. VP-> V VP . . .
d. Aux -» tense (M)

Lexical entries are represented as trees and lexical items are taken to
be individual leaves of lexical trees. By Hust’s precipitation conven¬
tion, “all features of a node A in a lexical entry are assigned to all
nodes dominated by A” [21: 51]. Thus, a simplified representation for
refuse ~ refusal is, according to Hust, the following:

(23) refuse

+_NP
+_VP

(features peculiar to the (features peculiar to

verb refuse) the noun refusal)

Hust supplements his lexicon with a set of redundancy rules, as

suggested by Chomsky in earlier work. Thus, for example, +N in (23)
can be filled in by redundancy rules on the basis of the presence of
the feature -l-Det_It is to be emphasized that individual lexical
items are leaves of trees such as (23). Together with the precipitation
convention, then, a lexical entry will be a constellation of syntactic
and phonological (and perhaps semantic) features. It is to be noted
that morphological features such as +--al in (23) are considered
to be part of the lexical entry itself after feature precipitation (as
opposed to a context for lexical insertion). Thus, the lexical entry for
refusal, after precipitation, is [+-NP, +-VP,
-[-human]_, +N, +Det_] together with the phonological
sequence refusal.
A similar approach can now be taken in the case of auxiliary
elements in English. It will be assumed that active participles such as
eating, having, being, etc., are dominated by the category V, that
past participles such as eaten, had, been, etc., are dominated by the
 160 Michael K. Brame

category V, and that nonderived verbs such as eat, have, be, etc., are
dominated by V. This elaboration entails a slight modification of rule
(22c).

V
(24) VP-> V • VP . . .
V

The lexical entry for eat and its related participles can now be
tentatively given as follows:

(25) eat

+_NP
+_#
+ [+human]

Actually, (25) is redundant in various ways. For example, the

participial endings can be omitted provided we supplement the
lexicon with the following redundancy rules2:

(26) [+V]-> [+-#ing]

[+V]-> [+-#en]

Thus, eat, eating, and eaten are all considered to be individual lexical
items, although the fact that all three select complements is expressed
in terms of a single lexical entry. That is, the precipitation convention
will associate the relevant contextual features of the root node of (25)
with its leaves.
Now let us consider how the correct order of auxiliaries is
predicted within the approach sketched here. In this case, be can be
given the tentative representation in (27).

2 Many details and redundant properties of lexical items are being glossed over here
in order to avoid unduly burdening the exposition. For example, irregular past
participles are listed ad hoc and cannot be predicted, whereas the feature +V for
present participles is itself redundant. A detailed discussion of these and other
properties of the auxiliary system in English is provided in Brame [4],
 The Base Hypothesis and the Spelling Prohibition 161

(27) be

+ _Pred
— V X
+ to VP

Of interest here is the feature [—_V X], This feature will

prohibit sequences such as John was eat the beans but will allow
John was eating the beans and John was seen. Thus, be can be
followed by active and passive participles. This is also true of been,
as predicted by feature precipitation.

(28) a. *John has been eat the beans.

b. John has been eating the beans.
c. John has been seen.

Likewise, by feature precipitation we expect a similar state of affairs

to hold for being.

(29) a. *John is being eat the beans.

b. *John is being eating the beans.
c. John is being taken.

Of course, the fact that (29b) is ungrammatical is just the problem

described earlier in terms of (17a). But, now, we may account for
(29b)-(17a), as well as (17b) and (17c) with one general redundancy
rule.
(30) [+V] -> [-_V X]

This rule may be taken as a formalization of Chomsky’s restriction

noted in (19); however, by giving up Affix-Hopping and by treating
the distribution of auxiliaries lexically, we succeed in expressing all of
the facts in (17) by the same means, namely, by (31), whereas the
standard theory, it will be recalled, requires two separate mecha¬
nisms.3
3 The so-called doubl-ing facts have been discussed by Ross [35], who proposes a
global rule to account for the relevant data. Both Emonds [17] and Milsark [30] have
demonstrated that a nonglobal account is superior. Pullum [32] has adduced new
evidence. In Brame [4] the controversy is reviewed and it is shown that the doubl-ing
facts all follow naturally from an approach similar to that advanced here, i.e., from a
more sophisticated theory of the lexicon together with redundancy rules.
162 Michael K. Brame

Let us now also return to (20b) and (21b). Recall that the standard
theory must rule out (20b) by constraining there-Insertion without,
however, blocking (20c), and it must rule out (20b) by a different
mechanism since there-Insertion is irrelevant in this case. It can now
be shown that both of these examples, (20b) and (21b), can be ruled
out by a natural extension of the redundancy rule (30) and that,
consequently, there is indeed a deeper generalization missed by the
standard theory.
Let us assume that Pred is expanded according to rule (31).4

VP
PP
(31) Pred (NP) <
AP
ASd

Let us assume further that expletive there is generated directly in

subject position by base rules and that there is no rule of there-
Insertion. Then (20a) and (21a) are allowed by virtue of the lexical
entry for be in (27). The same distribution should hold for being and
been. However, in the case of being, as in (20b) and (21b), the
presence of Pred leads to ungrammaticality. But now we may simply
extend the redundancy rule (30) in the following way:

(32) [ + V] -* [-_V X, -_Pred]

Thus, by (32), active participles, in particular being, will select

neither active participles nor predicate phrases as their complements.
This analysis thus succeeds in ruling out (20b) in precisely the same
way that it rules out (21b), a generalization missed in the standard
theory. This mechanism, it will be recalled, also accounts for the
examples laid out in (17).
Additional examples, brought to my attention by Robin Cooper,
make it necessary to further revise (32). These examples include the
following:

(33) a. I do not approve of there being two knives on the table,

b. There being nobody at home, we left.

Such examples show that the prohibition against predicate comple-

4 Rule (31) is tentative and by no means exhausts the predicate phrase, which has
been neglected in transformational studies. The inclusion of (NP) within the Pred
phrase may be incorrect, as assumed in [31].
 The Base Hypothesis and the Spelling Prohibition 163

ments is not absolute in the case of active participles, but rather

contingent on the presence of an overt auxiliary element such as tense
or modal. Therefore, it is necessary to alter (32) so as to express this
fact.

(34) [+V] -> [-_V X, -NP Aux_Pred]

It thus appears that a number of facts can be handled adequately

within the framework advanced here. The generalizations that
emerge, however, depend crucially on existential sentences being
base generated. This is a conclusion reached on independent grounds
in section 1, a conclusion which is consistent with the base hypothesis
and argued independently in Jenkins [24] and Brame [2].5 The data
adduced in this section therefore support this hypothesis.
Before turning to additional evidence for base generated structures
previously derived by transformation, let us consider the following
examples mentioned in Akmajian and Wasow [1] and elaborated in
Iwakura [23].

5 It was Emonds [15] who first noted and attempted to come to grips with examples
such as the following:

(i) a. Some boys were doctors.

b. *There were some boys doctors.
(ii) a. An elephant was small.
b. *There was an elephant small.
Emonds’ explanation was based on structural considerations and the structure-
preserving hypothesis. To create the ungrammatical (b) examples required a violation
of the structure-preserving hypothesis, which is possible only in the case of root
transformations (and minor movement rules) within Emonds’ framework. Jenkins [25]
attempted to account for the same phenomena utilizing Emonds basic observation
concerning phrase structure. However, for Jenkins, expletive there is base generated
with no rule of the re-Insertion. Thus, the (b) examples could not arise because the
phrase structure rules could not give rise to the relevant structures. Recently, Milsark
[31] has, perceptively, noted that some analogous sentences are well-formed.

(iii) a. (All over the world) there are people hungry.

b. (After the banquet) there were several people sick.

Milsark indicates that such examples tend to argue against base generated existentials;
however, such a conclusion does not appear to me to follow. Milsark divides the
adjectives into two classes: those that permit there as in (m), which hecalls state-
descriptive (SD), and those that do not as in (ii), which he calls property (?) predicates.
This distribution is reflected in rule (31) of the text by the inclusion of Asd, signifying
the class of state-descriptive adjectives. However, Milsark considers the distinction to
be semantic in nature. If it is, there is no reason why the semantic explanation cannot
be expressed in terms of the base generated structures in the absence of a rule of ^re¬
insertion. Thus, I can see no way in which such examples argue against base generated
existentials.
 164 Michael K. Brame

(35) a. Sam was being examined by a psychiatrist and Bill was

_too.
b. *Sam was being examined by a psychiatrist and Bill was being
_too.

Akmajian and Wasow fail to express a basic generalization concern¬

ing the full range of relevant examples involving VP phenomena, as
Iwakura [23] shows. However, Iwakura also fails to express the
generalization since he requires at least two mechanisms to rule out
the structures which would otherwise be generated. Thus, for Iwak¬
ura (35b) is prohibited because examined by a psychiatrist is not a
single VP constituent within his framework, an assumption which can
be questioned, and (36) is prohibited because of a special condition
which he associates with VP-Deletion.

(36) *Which bothers you more: John’s having taken a drug, or Bill’s
having_?

If we consider VP gaps to be base generated as such, both (35b) and

(36) can be ruled out by the same means. We need only revise the
redundancy rule (34) so that being will not select the empty context.

(37) [+V]-> [-_V X, -NP Aux_Pred, -_]

If this approach is correct, it shows that VP’s should not be deleted,

as assumed in early studies. Rather, VP phenomena must be treated
interpretively as argued in some recent work. Moreover, this conclu¬
sion has consequences which bear on the base hypothesis, for
consider the examples listed in (38).

(38) a. They said that Mary appears to be sick, and she does_
b. Joe was examined by a quack and Bill was_too.
c. They said there was an elephant in the yard and there was

d. Mary was given a message and Sue was_too.

According to the approach advocated here, the gaps in (38a-d) are

base generated. But if this is indeed the case, then Raising to Subject
Position, Passive, t/iere-Insertion, and Dative cannot account for the
data exhibited in the right conjunct of each of these examples. Thus,
the relevant structures must be based generated, a conclusion arrived
at on independent grounds in section 1.
 The Base Hypothesis and the Spelling Prohibition 165

3. EXTRAPOSITION AND SENTENTIAL SUBJECTS

In the foregoing it has been suggested that structures previously

derived by local transformations are in fact base generated. By local
transformations is meant transformations which do not make essential
use of variables. Now what about the putative rule of Extraposition?
This transformation has been stated with variables in the literature of
transformational grammar. However, there is an important difference
between Extraposition and other rules stated with variables, such as
Question Formation, Relative Clause Formation, Topicalization, etc.
For, the variable in Extraposition could in principle be replaced by a
disjunction of constant terms, although such is not the case for
Question Formation, Relative Clause Formation, etc. To put it
differently, in Ross’s terms, Extraposition is bounded, whereas the
other rules mentioned are not.
Now, if we take essential variable to mean unbounded essential
variable of the type found in Question Formation, Relative Clause
Formation, etc., then Extraposition is a local transformation in the
relevant sense and, therefore, accepting the base hypothesis, should
not be considered a transformation. Rather, extraposed S’s should be
base generated and sentential subject S’s should also be base gener¬
ated. This position is taken in Brame [2: 142].
Let us now turn to some evidence which supports the base
generation hypothesis in this case. First, it is important to observe
that S’s and VP’s are not NP’s, a position defended in Emonds
[15],[16] on the basis of examples such as the following6:

(39) a. *They told that everything would turn out for the best to the
children.
b. They told the children that everything would turn out for the
best.
c. They told a story to the children.
d. They told the children a story.

(40) a. *He blamed it on that Bill was too strict.

b. He blamed it on Bill’s strictness.

(41) a. *It was that John refused to see the light that bothered us.
b. It was John’s refusal that bothered us.

6 Emonds’ correct assumptions concerning the non-NP nature of sentential comple¬

ments are questioned in Higgins [19] and Horn [20], The relevant arguments are refuted
in Brame [4]. Emonds himself abandoned his analysis of the relevant examples in [ ]
for an inferior analysis. See [4] for discussion and arguments against Emonds more
recent analysis.
 166 Michael K. Brame

(42) a. *You promised to be quiet to Mary.

b. You promised Mary to be quiet.
c. You promised a new hat to Mary.
d. You promised Mary a new hat.

These data indicate that S’s and VP’s are not NP’s since they would
be expected to exhibit the identical distribution of NP’s if they were.
Thus, in (39c), the true NP a story can occupy the immediate
postverbal slot in the sentence, but this is not so in the case of (39a),
indicating that the italicized phrase is not an NP, etc.
Let us accept Emonds’ straightforward explanation for this range of
data and proceed now to the crucial examples.

(43) a. That John was sick bothered Mary,

b. *Did that John was sick bother Mary?
(44) a. That John dislikes artichokes surprised you.
b. *Which artichokes did that John dislikes surprise you?
(45) a. That it bothers the teacher for John to smoke is quite
possible.
b. *That for John to smoke bothers the teacher is quite possible.
(46) a. That he forgot his lesson proves that John was not serious,
b. *It proves that John was not serious that he forgot his lesson.

These examples have been discussed and treated by grammarians

working within the transformational framework. The grammarians
include Ross [35], Emonds [15],[18], Higgins [19], Kuno [27], Horn
[20], and Iwakura [24], It seems to me that all of the analyses
advanced have failed in one way or another to express the basic
generalization underlying the full range of data.7 Accepting the base
hypothesis, we can attempt to express the generalization in terms of
phrase structure.
Let us recall that in Emonds’ earlier work [15: 69], it was argued
that sentential subjects are not dominated by NP in surface structure.
Thus, Emonds argued that an example such as For the house to be
painted would irritate him should be represented in surface structure
as (47).

It should be noted that although I disagree with Iwakura’s ingenious analysis in

[24], he was the first to recognize that there is indeed a deeper generalization to be
expressed. Iwakura also provides much penetrating criticism of Horn [20],
 The Base Hypothesis and the Spelling Prohibition 167

(47) S

s Aux VP

V NP

for the house would irritate him

to be painted

According to Emonds [15], (47) is derived by a rule which is

essentially the inverse of Extraposition, where S is substituted for the
NP subject. However, adopting the base hypothesis, let us now
assume that Emonds’ surface structure is in fact the base generated
structure.8 That is, let us assume that S’s and VP’s can be generated
directly as subjects where the relevant categories are not dominated
by NP. To resolve the difficulties that arise in connection with (43)-
(46), however, let us introduce the category S, which will play a
crucial role in explaining the relevant data.

b. S -> Comp S
c. S -> NP Aux VP
S
S
VP
d. VP -* V < > —

Now let us see how this reanalysis explains all of the data tabulated in
(43)_(46) without recourse to special constraints such as Ross’s
Sentential Subject Constraint and Island Internal S Constraint,
Emonds’ prohibition against multiple application of root transforma¬
tions, Horn’s NP Constraint and global Pruning, etc.

8 Alfredo Hurtado has informed me that he also has made this assumption for the
grammar of Spanish.
168 Michael K. Brame

Consider first (43b). Since that John was sick is a sentential

subject, Comp is unavailable for the root S which dominates it, and
consequently a question cannot be formed. For exactly the same
reason, a w/z-question cannot be formed and thus (44b) cannot be
derived. Example (45b) also cannot be generated since the sentential
subject S cannot be expanded to yield a second sentential subject by
rule (48b). Finally, (46b) is avoided by virtue of the fact that VP
cannot be expanded by rule (48d) so as to provide two sentential
complements. In sum, we see that all of the relevant examples are
explained quite naturally by phrase structure. This analysis should be
compared with the existing proposals of the literature mentioned
above.
Additional support for the position outlined here is gleaned from
the following examples.

(49) a. I imagine for one to work that hard would require a lot of
effort.
b. I suppose to do that would require a good deal of manual
dexterity.
c. I guess that the night is black demonstrates that the universe
is expanding.
d. I imagine that Bill works hard shows that he is German.

Many speakers apparently accept some sentences such as these with

embedded sentential and VP subjects, although there is variation from
speaker to speaker. Other examples are clearly ungrammatical for all
speakers.

(50) a. *1 resent for you to leave would bother her.

b. *It is surprising for you to say that would annoy him.
c. *It bothers me that John left proves that he was not serious.

The fact that we do obtain different results for different predicates,

say imagine vs. resent, indicates that we are dealing with a lexical
phenomenon, i.e., with subcategorization. But the reanalysis
sketched out in the foregoing makes just such an approach possible
since rule (48d) has been introduced to provide for S. Thus, imagine,
guess, etc., but not resent, etc. can be subcategorized so as to select
S, providing for (49a-d), but not for (50a-c).
The new analysis makes a further prediction which is borne out. It
predicts a contrast between the examples laid out in (51) and those
listed in (49).
 The Base Hypothesis and the Spelling Prohibition 169

(51) a. *1 imagine that for one to work that hard would require a lot
of effort.
b. *1 suppose that to do that would require a good deal of manual
dexterity.
c. *1 guess that that the night is black demonstrates that the
universe is expanding.
d. *1 imagine that that Bill works hard shows that he is German.

These data follow from the fact that S does not directly dominate a
Comp node as does S. Hence, the complementizers such as that and
for in (49) are all associated with the sentential S which is dominated
by S. This explanation appears to be quite natural and straightfor¬
ward. If correct, it should lend credence to the analysis advanced in
this section.

4. THE SPELLING PROHIBITION

In sections 3 and 4, new evidence has been provided to support

base generated structures which were previously derived via transfor¬
mation. In this section a general constraint will be suggested which
has the effect of favoring the analyses outlined above. The Spelling
Prohibition will, in fact, have the effect of prohibiting the previous
transformational analyses which have been impugned by the evidence
adduced in the previous discussion. This constraint can be initially
formulated as follows.

Spelling Prohibition (weak form):

Transformations cannot spell out morphological material (or
effect the spelling out).

The effect of the Spelling Prohibition is to ensure that all morphol¬

ogical material is present after lexical insertion. Thus, the presence of
morphological material such as the expletive there and the passive
be en is contingent on phrase structure and the lexicon, a result
which is consistent with the arguments of previous sections. Conse¬
quently, the Spelling Prohibition makes classical transformations such
as there-Insertion and Passive unavailable to the theory, a desirable
result.
The Spelling Prohibition has consequences for many other transfor¬
mations proposed in the literature of transformational grammar. For
one, it disallows a rule switching some to any as suggested in earlier
170 Michael K. Brame

work such as Klima [26]. Rather, any must be introduced directly into
phrase structure. That there is no rule turning some to any has been
argued in Lakoff [29].9 Another proposed transformation which is
prohibited by the Spelling Prohibition is Comp-Placement, which was
originally proposed in Rosenbaum [34], The fact that Bresnan [5],[7]
has shown that complementizers such as that, for, and wh must be
present in phrase structure shows that the Spelling Prohibition
(hereafter SP) makes a correct prediction in this case. SP also rules
out the possibility of deriving NP’s such as John’s picture from
relative clause constructions such as the picture that John has as
proposed in early work and accepted in Chomsky [14: 40 passim] and
suggests base generation as per the base generated possessives
involving intrinsic connection discussed by Chomsky. This follows
from the fact that the morphological material ’5 would otherwise have
to be spelled out by the transformation. Other elements are also
prohibited as transformationally inserted elements, such as the of in
several of John’s proofs of the theorem, cf. Chomsky [14: 53].
Bresnan has in fact suggested that of should not be inserted by
transformation; cf. [8]. Again we see that SP makes the correct
prediction. SP likewise prohibits deriving who, what, etc., from wh-
someone, wh-something, etc., as argued in Chomsky [12]. Chomsky
has shown that who and someone, what and something, etc., bear a
close relationship in that they exhibit many distributional similarities,
one illustration of which is given in (52).

(52) a. Harry will meet with someone else.

b. Who else will Harry meet with?
c. *Harry will meet with a girl else.
d. *Harry will meet with the adviser else.

In (52) we see that else can co-occur with wh-words and with some-
words. This fact could be explained if who, what, etc., derived from
wh-someone, wh-something, etc. However, SP disallows such a
derivation, and therefore we might expect to find a difference in
distributional properties. We do indeed find a distinguishing environ¬
ment.

(53) a. Who the hell broke my typewriter?

b. *Someone the hell broke my typewriter.

9 This point holds for the general process of Negative Incorporation proposed by
Klima, Ross, and others. For a discussion of how examples such as John has any
money are to be blocked, see [3] and [4],
 The Base Hypothesis and the Spelling Prohibition 171

(54) a. What in tarnation are you talking about?

b. *You are talking about something in tarnation.

(55) a. Where in the name of God did Sam buy that Okito box? (^)
b. *Sam bought that Okito box somewhere in the name of God.

Such examples again show that SP makes correct predictions. Pro¬

ceeding, we see that SP also disallows c/o-Support as a transformation
in any of the variant formulations in which do is inserted by
transformation. Rather, according to SP, do must be introduced in
base structures. Rules such as Case-Marking, Subject-Verb Agree¬
ment, and other agreement rules are also prohibited by SP. Rather,
case violations, agreement violations, and the like should apparently
be treated by a checking or filtering mechanism, as suggested in
Bresnan [12: note 8] and Brame [3]. The classical approach to
pronominalization and reflexivization is also disallowed by SP. This
accords with much recent research which shows that pronouns,
reflexives, and reciprocals should be base generated. Base generation
of such pronouns and anaphoric pronouns, including so, is a direct
consequence of SP, thus lending further credence to its correctness.
Finally, SP has the virtue of prohibiting lexical decomposition, which
plays a prominent role in generative semantics. Thus, for example,
Seuren writes that generative semanticists “posit a great deal of
transformational syntax inside, or behind, lexical items, especially
verbs” [37: 5]; advocates of this lexical decomposition approach
derive the (b) examples of (56)-(59) from structures akin to the
corresponding (a) examples.

(56) a. John struck Bill as being similar to a gorilla,

b. Bill reminded John of a gorilla.

(57) a. The sauce came to be thick,

b. The sauce thickened.

(58) a. John caused Harry to die.

b. John killed Harry.

(59) a. Mary persuaded Bill not to go.

b. Mary dissuaded Bill from going.

Considerable language-internal evidence has been adduced in [2:

Chap. 1] to demonstrate that lexical decomposition cannot be main¬
tained without loss of significant generalization. The fact that SP is
consistent with these language-specific independent arguments further
supports it as a basic constraint on the functioning of grammars.
172 Michael K. Brame

It is possible, and quite conceivably correct, to strengthen SP in the

following way:

Spelling Prohibition (strong form):

Transformations cannot spell out or alter morphological material.

The strong form of SP has the added effect of prohibiting deletion of

morphological material as well as substitution for morphological
material. Thus, substitution of a sentential complement for it, accord¬
ing to one version of Raising to Subject Position and Raising to Object
Position is prohibited, a result consistent with the conclusions of
section 1. Similarly, Object-Shift, which substitutes an object NP for
it so as to derive sentences such as John is easy to please from it is
easy to please John is disallowed, a result which again coincides with
the conclusions of section 1. Of course, the deletion approach to
Object-Shift phenomena, as proposed in [30], is also prohibited by the
strong form of SP; cf. [2] for discussion. The strong form of SP also
disallows Extraposition, albeit somewhat indirectly, since /7-Deletion
is prohibited. Alternative approaches involving spelling out of it, as in
[18], are ruled out by the weak form of SP. Dative is also disallowed
since deletion of to or for is ruled out, as are alternative approaches
which spell out to and for. Finally, the deletion approach to VP gaps
is also ruled out by the strong form of SP.
We see, then, that SP has consequences consistent with the
language-internal arguments advanced in [2] and in the preceding
sections. This convergence provides considerable motivation for SP
and, therefore, it is advanced as a universal constraint on the
functioning of transformations.

5. LEXICALISM VS. THE INVERSE CYCLE

In view of the fact that a large number of transformations must be

abandoned in favor of base generated structures, we might ask just
how the explanations provided by the classical transformations are to
be expressed. Before attempting an answer to this question, it is
important to recognize just what it is that the classical transforma¬
tions account for since there is a good deal of misunderstanding on
this point. Let us take Passive as the basis of discussion in this
section.
It is often asserted that transformational rules account for para¬
phrase relations or synonymy relations which hold between the
sentences related by transformation. The truth is that paraphrase and
synonymy remain as vague today as they were before the advent of
 The Base Hypothesis and the Spelling Prohibition 173

transformational grammar. Transformations provide no insights into

and shed no light on putative relations such as paraphrase and
synonymy. Others have argued that grammatical relations are pre¬
served under transformation, yet such advocates fail to indicate to
what extent such relations yield insight into any interesting questions
or how they resolve outstanding issues. It seems, therefore, that we
must recognize that classical transformations have served one and
only one basic function, and that is the syntactic function of account¬
ing for distributional generalizations by avoiding repetitive statements
involving co-occurrence phenomena.10 For example, in the case of
Passive, it has been assumed that if an active verb selects an NP as
direct object, its passive counterpart will select that NP as subject.
And if an active verb does not select an NP as object or fails to select
a specific kind of NP as object, then the same facts will hold for the
subject of the passive counterpart. Thus, because (60a) allows the
beans as object of eat, (60b) allows the beans as subject of be eaten.
And, equally important, because (61 a) does not allow sincerity as
object of sleep, sincerity is not allowed as subject of be slept.

(60) a. John ate the beans.

b. The beans were eaten by John.

(61) a. John slept sincerity.

b. Sincerity was slept by John.

Since the foregoing is a distributional argument, it is quite obviously a

syntactic argument. Now, if passives are derived from actives, then
the convergence of distributional facts follows as a consequence and
need be stated only once in the grammar, namely, before the Passive
rule applies. However, if Passive is abandoned in its classical
formulation, then how can the syntactic generalizations be expressed?
One answer to this question was provided in Brame [2], where both
actives and passives are base generated in conformity with the base
hypothesis of section 1. In lieu of the classical rule of Passive, its
inverse was proposed which had the effect of converting base
generated passives into structures identical to actives. It was at this
deep level of derived actives that the co-occurrence restrictions were
checked, thus expressing the basic generalization inherent in (60)—
(61).
Another possibility was suggested in earlier work: “Granted that
passives are generated directly, it remains to relate them to corre¬
sponding actives. One approach would be to formulate a lexical
redundancy rule taking a form quite similar to the Passive transforma-

10 Excellent discussion of this point can be found in Hust [21].

174 Michael K. Brame

tion itself” [2: 132]n. Recently Bresnan [9] has provided an explicit
statement of such a redundancy rule and has suggested that all local
transformations be supplanted by lexical redundancy rules, thus
accepting the base hypothesis. Let us identify the redundancy rule
approach with the lexicalist hypothesis which was initiated in Chom¬
sky [14]. This position can be informally stated in a strong form as
follows:

Lexicalist Hypothesis:
All co-occurrence generalizations previously expressed by local
transformations are expressed by lexical redundancy rules.

The inverse cycle and the lexicalist hypothesis are contrasted sche¬
matically in (62).
(62) INVERSE CYCLE

deep structures
A

inverse T s (= genotype
rules)
co-occurrence checked
inverse-cyclically
(active-passive related
here)
base rules
> base structures
lexicon
T’s (= phenotype rules)
v
surface structures

LEXICALIST MODEL

base rules base structures

lexicon

redundancy T’s
rules relating
active-passive,
etc.

surface structures

11 Several classical transformations were formulated as redundancy rules during the

course of my Fulbright-Hays lectures at the Rijksuniversiteit te Utrecht in 1973;
however, the project was terminated by my failure to work out the relevant facts
concerning Equi.
 The Base Hypothesis and the Spelling Prohibition 175

There is a certain naturalness to the lexicalist model depicted in

(62). Local transformations are typically local by virtue of the fact
that a particular lexical item, usually a predicate, serves as the locus
of the transformation. It is quite natural to express the locality of such
phenomena in that area of the grammar where lexical items are
represented and where unboundedness cannot be expressed, namely,
in the lexicon. Then, it is no longer a mystery why certain lexical
items typically serve as the locus of local transformations.
Now, however, we must ask if there are any empirical arguments
favoring one approach over the other. In the case of Passive, there
appears to be some evidence favoring the lexicalist model. To see
this, let us return to the lexical entry for eat as (25) in section 2. By
feature precipitation, the past participle eaten will be positively
specified for +_NP, whereas the NP object must assume the
subject position in this case, in fact. Otherwise, a grammar incorpo¬
rating (25), along with base generated past participles, will give rise to
ungrammatical strings such as NP was eaten the bananas. To
overcome this difficulty, just as we posited lexical redundancy rule
(34) to account for additional distributional characteristics of V, we
may now do likewise for V.12

(63) [+V] -> [«[/]-] / +_NP

a-[/]

This redundancy rule is of course the lexical analog of the transforma¬

tional rule known as NP-Preposing or the inverse rule known as NP-
Postposing. By accepting the lexical approach to participles outlined
in section 2, we are thereby forced to choose the lexicalist model
depicted in (62).
There is further evidence that this conclusion is not only desirable,
but that it allows us to express a much deeper generalization. Note
first that adjectives in -able which are derived from verbs exhibit one
characteristic feature of passives. When the V of the adjective
sequence selects NP as object, the related adjective itself selects this
NP as subject. Chomsky noted this redundancy and proposed a
lexical approach,
insofar as a subregularity exists regarding selectional rules in the
case of -able, it can be formulated as a lexical rule that assigns the
feature [X_] to a lexical item [N-able] where V has the intrinsic
selectional feature [-X]. [14: 56]

12 Details concerning Aux and other material are omitted here. See [4], Also note that
Extraposition can be replaced with a redundancy rule so as to relate S and S
complements discussed in section 3 to a given predicate such as surprise NP, be
possible, etc.
176 Michael K. Brame

Chomsky’s rule has been formulated by Hust [20: 67] as approxi¬

mately the following13:

(64) [+_#able\ * [«[/] ] / +_NP

a-if]

Hust has gone on to show that unpassive adjectives exhibit identical

selectional properties. Thus, he was able to express the two sets of
co-occurrence restriction by one general rule, repeated here as (65).

[+_#able] |
(65) * [«L/1_] / +_NP
[+un#_#ed] J
a-[/I
Of relevance for the present discussion is the fact that rule (63) can
now be collapsed with Hust’s rule (65), thus expressing a deeper
generalization.

[+V]
(66) ' [+_#able] [olf\_] / +_NP
[ +un #_#ed\ «-if)

The fact that a deeper generalization is naturally expressed within the

lexical framework provides some justification for favoring it, at least
in the case of passives.

6. THE TRANSFORMATIONAL COMPONENT

If local transformations are to be replaced with lexical redundancy

rules, one might ask what transformations remain. A rather natural
answer to this question is embodied in the following constraint on the
form of transformations:

Variable Constraint:
All transformations make essential use of unbounded variables.

Thus, rules such as Question Formation, Relative Clause Formation,

13 By convention it is assumed that all features present in the context of redundancy

rules employing alpha notation are relevant, including +_NP itself. That is,
although +-NP appears in the context of (64), it also is covered by the alpha
feature of the context matrix and thus +NP_ is associated with adjectives in
-able, etc.
 The Base Hypothesis and the Spelling Prohibition 177

Topicalization, and the like, all of which make essential use of

variables which are unbounded, are the remaining candidates for the
transformational component. This result is natural when contrasted
with the observations of the preceding sections. Thus, unlike local
transformations, no single lexical item serves as the locus of the long
distance rules. Hence, long distance rules could not in principle be
expressed as lexical redundancy rules. By contrast, given the Varia¬
ble Constraint as a constraint on the form of transformations, local
transformations no longer qualify as transformational rules.
These conclusions may prove too strong and additional analyses
with precise formulations of the rules in question are needed before a
final determination can be made. For example, there are some root
transformations such as Subject-Aux Inversion which do not qualify
as transformations by the Variable Constraint. Perhaps it would not
be pointless to consider the feasibility of base generating all relevant
structures of a local character by elaborating the base, as was done in
section 3 with respect to sentential subjects and extraposed S’s.
There is furthermore a model of grammar which is also worthy of
serious consideration. This model can be thought of as a hybrid of the
two models depicted in (62) and is laid out in picture form in (67).

(67) INVERSE LEXICAL MODEL

deep structures (= base structures of

lexical model)

inverse T’s

base rules
surface structures

lexicon

redundancy rules
relating active-
passive, etc.

If the Variable Constraint can be maintained, then the inverse

lexical model is appealing for the following reason. By treating long
distance rules in an inverse fashion, we succeed in expressing a
generalization which cannot be expressed in the standard theory, as
argued in Brame [2]. To see this, consider the following ungrammati-
178 Michael K. Brame

cal strings:

(68) a. *John saw the boy (who) Mary likes the girl.
b. *What did they see the parade?
c. *It was the car that I parked the truck in the garage.
d. *Harry I saw Sue yesterday.
e. *That she would do such a thing I believe Harry to be
impossible.
f. *They said she was examined by Dr. Qu, and examined by
Dr. Qu she was arrested by the police.

Against the examples in (68), the following grammatical sentences

should be contrasted.

(69) a. John saw the boy (who) Mary likes.

b. What did they see?
c. It was the car that I parked in the garage.
d. Harry I saw yesterday.
e. That she would do such a thing I believe to be impossible.
f. They said she was examined by Dr. Qu, and examined by Dr.
Qu she was.

Now how does the standard theory rule out the examples in (68)?
According to Chomsky [13], (68a) is ruled out by the filtering function
of transformations. Thus, special boundary symbols suffice to ensure
identity in relative clauses and if the boundary symbol is not erased
by application of Relative Clause Formation, which itself requires
identity, then the internal occurrence of the boundary symbols will be
sensitive to a filtering device, thus blocking the example. However,
other examples in (68) will be ruled out by different mechanisms.
Thus, (68b) will be blocked by virtue of the fact that what originates
as the object of the verb see, which does not allow two objects.
Hence by purely lexical considerations, (68b) is ruled out, as are
some of the other examples in (68). But this surely constitutes a loss
of generalization. An adequate theory would rule out all of (68) by the
same basic mechanism. If we generate the examples in (69) directly,
that is, if the who in (69a) is generated in Comp position, what in
Comp position in (69b), the car in focus position in (69c), Harry in the
topicalization position in (69d), etc., then it is possible to rule out all
of (68) by purely lexical considerations, as advocated in [2],
 The Base Hypothesis and the Spelling Prohibition 179

REFERENCES

1. Akmajian, A., and Wasow, T. 1975. The constituent structure of VP and

Aux and the position of the verb be. Linguistic Analysis I: 205-45.
2. Brame, M. K. 1976. Conjectures and Refutations in Syntax and Seman¬
tics. New York: Elsevier North-Holland.
3. -. 1977. Alternatives to the Tensed-S and Specified Subject Condi¬
tions. Linguistics and Philosophy 1.
4. -. Forthcoming. Base Generated Syntax.
5. Bresnan, J. W. 1970. On complementizers: toward a syntactic theory of
complement types. Foundations of Language 6: 297-321.
6. _. 1971. Sentence stress and syntactic transformations. Language
47:257-81.
7. _. 1972. Theory of complementation in English syntax. Doctoral
dissertation, MIT.
8. _. 1973. Syntax of the comparative clause construction in English.
Linguistic Inquiry 4: 275-343.
9 _. 1977. Toward a realistic model of transformational grammar.
Paper presented at the Bell Telephone Convocation at MIT.
10. Chomsky, N. 1975. The Logical Structure of Linguistic Theory. Plenum
Press, New York. (Condensation of 1955 manuscript.)
11. -. 1957. Syntactic Structures. The Hague: Mouton.
12. _. 1964. Current Issues in Linguistic Theory. The Hague: Mouton.
13. _. 1965. Aspects of the Theory of Syntax. Cambridge, Mass.: MIT
Press.
14. _. 1970. Remarks on nominalization. In Readings in English Trans¬
formational Grammar, R. A. Jacobs and P. S. Rosenbaum, eds. Wal¬
tham, Mass.: Ginn. Page references to reprinted version in Studies on
Semantics in Generative Grammar, N. Chomsky, The Hague: Mouton.
15. Emonds, J. E. 1970. Root and structure-preserving transformations.
Doctoral dissertation, MIT.
16 _. 1972. A reformulation of certain syntactic transformations. In
Goals of Linguistic Theory, S. Peters, ed. Englewood Cliffs, N. J.:
Prentice-Hall.
17 _. 1973. Alternatives to global constraints. Glossa 7: 39-62.
18. _. 1976. A Transformational Approach to English Syntax. New
York: Academic Press.
19. Higgins, F. R. 1973. On J. Emonds’s analysis of extraposition. In Syntax
and Semantics 2, J. P. Kimball, ed. New York: Seminar Press.
20. Horn, G. M. 1975. On the nonsentential nature of the POSS-ING
construction in English. Linguistic Analysis 1: 333—87.
21. Hust, J. R. 1976. A lexical approach to the unpassive construction in
English. Doctoral dissertation, University of Washington, Seattle.
22. _. 1977. Lexical redundancy rules and the unpassive construction.
Linguistic Analysis 3: 137-76.
180 Michael K. Brame

23. Iwakura, K. 1977. The Auxiliary System in English. Linguistic Analysis

3: 101-36.
24. -. 1977. The syntax of complement sentences in English. Linguistic
Analysis 3: 000-00.
25. Jenkins, L. 1972. Modality in English syntax. Doctoral dissertation, MIT.
26. Klima, E. S. 1964. Negation in English. In The Structure of Language:
Readings in the Philosophy of Language, J. A. Fodor and J. J. Katz, eds.
Englewood Cliffs, N. J.: Prentice-Hall.
27. Kuno, S. 1973. Constraints on internal clauses and sentential subjects.
Linguistic Inquiry 4: 363-85.
28. Lakoff, G. 1970. Irregularity in Syntax. New York: Holt, Rinehart and
Winston.
29. Lakoff, R. T. 1969. Some reasons why there can’t be any some-any rule.
Language 45: 608-15.
30. Lasnik, H. and Fiengo, R. 1974. Complement object deletion. Linguistic
Inquiry 5: 535-71.
31. Milsark, G. 1972. Re: Doubl-ing. Linguistic Inquiry 3: 542-49.
32. -. 1977. Toward an explanation of certain peculiarities of the
existential construction in English. Linguistic Analysis 3: 1-29.
33. Pullum, G. K. 1974. Restating Doubl-ing. Glossa 8: 109-20.
34. Rosenbaum, P. S. 1967. The Grammar of English Predicate Complement
Constructions. Cambridge, Mass.: MIT Press.
35. Ross, J. R. 1967. Constraints on variables in syntax. Doctoral disserta¬
tion, MIT.
36. -. 1972. Doubl-ing. Linguistic Inquiry 3: 61-86.
37. Seuren, P. A. M., ed. 1974. Semantic Syntax. London: Oxford Univer¬
sity Press.
 ALTERNATIVES TO THE TENSED S AND
SPECIFIED SUBJECT CONDITIONS

Sensei, why do we lose the right wing interpretation?

It is not good for Communists to work in defense plants.
Defense plants are not good for Communists to work in.

Don’t let them fool you, gakusei! It’s really a matter of interpretation, not
movement.
0. General Introduction

1. Introduction to TSC and SSC

2. Motivation for SSC and TSC: A Critique

2.1 Object Shift
2.2 each-Movement and TSC
2.3 each-Movement and SSC
2.4 Unlike Person Constraint and Disjoint Reference
2.5 Vk/7-Movement
2.6 Scope of Negation
2.7 Passive
2.8 Naturalness

3. Counterexamples to SSC, TSC, and Subjacency

3.1 Introduction to Subjacency and Successive Cyclicity
3.2 Counterexamples
3.2.1 Clefting
3.2.2 Comparative Subdeletion
3.2.3 w/?-Movement
3.2.4 each-Movement
3.2.5 Object Shift
3.2.6 Superiority
4. Conclusion

Notes

References
 ALTERNATIVES TO THE TENSED S AND

SPECIFIED SUBJECT CONDITIONS*

0. General introduction

In his contribution to the Halle Festschrift, Chomsky provides a battery of

arguments for his Specified Subject and Tensed S Conditions. Chomsky
argues that these conditions provide a unified account of a wide range of
disparate facts. More recently, this earlier work has led to the development
of trace theory, with the accompanying claim that all aspects of semantic
interpretation are determined by surface structure.1 It is my opinion that this
kind of enterprise - a detailed discussion of a wide range of data, together
with an explicit theory advanced to explain the data — is precisely the kind of
enterprise that linguists must undertake if advances are to be made in syntax
and neighboring disciplines. For only when commitments are made to
detailed analyses is the productive potential of the critical method fully
realized in the face of analyses and theories which do not reflect the true
nature of things.
In the following pages I will undertake a critical reappraisal of the evidence
advanced to support the arguments for the Specified Subject and Tensed S
Conditions in Chomsky’s original paper [12].2 A careful look at the evidence
has convinced me that there are plausible alternatives for dealing with the
data provided as motivation for these conditions. In fact, it can be shown, I
will argue, that to treat the data by means of these conditions leads in many
instances to a loss of generalization; in other instances, the conditions
approach raises more problems than it solves.

* A preliminary draft of this paper was written in the winter of 1975 and constituted the subject
matter of a seminar given at the University of Washington during the fall of 1975.

183
184 Realistic Syntax

After discussing and criticizing the Specified Subject and Tensed S

Conditions, I will turn to a brief discussion and critique of Chomsky’s
Subjacency and Superiority Conditions. These too can be criticized with
some force, I believe.

1. Introduction to tsc and ssc

Let us first consider SSC. This condition can be illustrated by considering the
well-known rule of Object Shift ( = Tough Movement). It is often claimed
that this rule relates pairs such as the following.

(1) a. It is not easy for a novice to faro a deck of cards,

b. A deck of cards is not easy for a novice to faro.
(2) a. It is hard to deal deceptive seconds without arousing suspicion,
b. Deceptive seconds are hard to deal without arousing suspicion.

The traditional analysis has it that Object Shift moves the object of the
complement, in these exampls, a deck of cards in (la) and deceptive seconds
in (2a), and substitutes it for it in the matrix. Now a problem immediately
arises in connection with (3a).

(3) a. It is pleasant for the rich for poor immigrants to do the hard
work.
b. *The hard work is pleasant for the rich for poor immigrants to do.

The problem is why the object the hard work in (3a) cannot be shifted by
Object Shift to yield (3b). According to Chomsky, movement is blocked in
(3a) by a constraint on transformations. This constraint prohibits the move¬
ment, deletion, etc. of a category across a specified subject, where specified
subject amounts to, roughly, a lexical subject (although the revised defini¬
tion of specified subject will be discussed below). A simplified formulation of
this constraint is given in (4).

(4) SSC: No rule can involve X, Y in the structure

... X ... [a ... Z ... —WYV ...]...

where Z is the specified subject of WYV in a, a cyclic category.

The effects of this constraint are illustrated diagrammatically in (5).

(5)X

It is pleasant for the rich [s for poor immigrants to do the hard work]
t-x-1
 Tensed S and Specified Subject 185

Since the poor immigrants functions as a specified subject, the hard work
cannot cross to substitute for it.
Let us now turn to additional examples illustrating a prohibition against
Object Shift.

(6) a. It is good that C. C. Wei invented the precision bidding system,

b. *The precision bidding system is good that C. C. Wei invented.

Again we see that Object Shift cannot apply to (6a) to give (6b). The
prohibition of course follows from SSC, but there is a second condition
which also serves to block (6b). This condition prohibits movement or
deletion from a tensed sentence and is repeated in a somewhat simplified
form as (7).

(7) TSC: No rule can involve X, Y in the structure

... X ...[a ... Y...]...
where a is a tensed sentence.

Since the embedded complement of (6a) is a tensed sentence, by TSC it is

not possible to extract the precision bidding system by Object Shift.

2. Motivation for ssc and tsc: a critique

In this section I will examine Chomsky’s motivation for SSC and TSC and
provide plausible alternatives to account for the relevant range of data. I will
show that these alternatives are independently justified inasmuch as they
account for facts which are not treated by SSC and TSC. Thus, it appears that
these conditions may be artifacts.

2.1 Object Shift

Since SSC and TSC have been introduced by way of Object Shift in section 1,
let us now consider an alternative and historically antecedent account of the
relevant range of data. In [7] Bresnan made the following observation.

It has been proposed ... by Lees 1960 ... that certain infinitival complements should be
derived from deep-structure VP’s rather than S’s. Suppose this proposal is applied to the
analysis of certain adjective plus complement constructions. [7:263]

According to Bresnan’s proposal, following Lees, some adjectives, such as

obvious, clear, etc. are lexically subcategorized to select S’s, but not VP s,
while others, such as tough are subcategorized to select VP’s or PP +VP but
not S’s. This situation is illustrated in (8).

(8) a. It is obvious [s that Mahler composed that symphony]

b. It will be tough [PP for some students][Vp to solve that problem]
 186 Realistic Syntax

Bresnan provides a wealth of evidence to show that predicates such as tough

should not be analyzed as selecting an S complement. Here I will repeat only
one of those arguments.

Several facts argue against the sentential analysis: first, if there were an underlying sentence,
one would normally expect a sentence-cyclic transformation such as There-Insertion to take
place. But though one can say It will be tough for at least some students to be in class on time, one
cannot say *It will be tough for there to be at least some students in class on time. [7:264]

Bresnan contrasts this situation with predicates such as eager which do select
S complements and consequently do permit There-Insertion: The adminis¬
tration is eager for there to be at least some students in class on time.
Now, in connection with (8), the crucial observation is that Object Shift
correlates with presence of VP or PP + VP complement and is not allowed in
the case of S or PP + S complement.

(9) a. *That symphony is obvious that Mahler composed.

b. That problem will be tough for some students to solve.

The obvious conclusion is that Object Shift does not apply to S comple¬
ments. In other words, the rule of Object Shift should be formulated so as to
be sensitive to VP or PP + VP complements. Bresnan’s tentative formula¬
tion along these lines is given in (10).4

(10) Object Shift: [SA Pred (PP) [VP V* (P) NP]]

t___j

This analysis provides an immediate explanation for the ungrammaticality

of Chomsky’s (3b) and for example (6) cited above in section 1.

(11) a. *The hard work is pleasant for the rich [s for poor immigrants to
do]
b. *The precision bidding system is good [s that C. C. Wei invented]

The explicit formulation of Object Shift as (10) does not allow extraction of
an NP from an S complement. Therefore (11a) and (lib) cannot be
generated. Thus, SSC and TSC are not supported by movement under
Object Shift.
This discussion highlights a rather important point relating to all syntactic
research, namely, that serious discussion can ensue only when explicit
formulations of the rules in question are provided. In this case Bresnan’s
formulation leads to a solution for an outstanding puzzle. The puzzle
involves certain ambiguous constructions noticed by traditional gram¬
marians such as Jespersen and later by Lees.

There is in fact a class of adjectives permitting both S and PP + VP complements - namely, the
class including good, bad, sweet, pleasant, and appropriate; the ambiguity in such cases is noted
 Tensed S and Specified Subject 187

by Lees. The sentence It is good for John to leave may mean either ‘For John to leave is good — It
is good (for John to leave)’ or ‘To leave is good for John = It is good for John (to leave.)’ Lees
maintains a clear distinction between the ambiguous class (good; his type 7) and the unambigu¬
ous class (hard, tough-, his type 8). [7:266-7],

Now the puzzle is this: whereas (12a) is ambiguous, the result of applying
Object Shift in (12b) is not ambiguous.

(12) a. It is good for John to learn such things,

b. Such things are good for John to learn.

Example (12b) receives the interpretation ‘To learn such things is good for
John’ but not ‘For John to learn such things is good’. What is the explanation
for the absence of ambiguity under transformation? Obviously Bresnan’s
rule of Object Shift predicts that the object NP such things will be shifted
from the VP complement only. Therefore, it follows that (12b) will receive
the PP + VP interpretation and not the S interpretation for the complement
structure. This is accomplished without theoretical elaboration. Let us
consider briefly how Chomsky’s constraints would express these facts.
First, consider example (9b). According to Chomsky, the underlying
representation is roughly (13).

(13) It will be tough [PP for some students] [s for PRO to solve that
problem]

Chomsky attempts to preserve the abstract S by postulating an abstract

subject PRO. This PRO is taken to be an unspecified subject, and therefore
SSC will not block Object Shift, which allows extraction of an NP from S
provided SSC and TSC are not violated. Now, turning to (12), Chomsky
would propose two underlying sources for (12a) corresponding to the two
interpretations.

(14) a. It is good [s for John to learn such things]

b. It is good [PP for John] [s for PRO to learn such things]

In the case of (14a) Object Shift will be inapplicable because of SSC.

However, Object Shift will apply to (14b), just as in (13), yielding (12b).
Thus, by recourse to SSC, Chomsky is also able to predict the nonambiguity
of (12b).
In comparing the two alternatives it is important to bear in mind that
Bresnan does not employ ad hoc extensions (of what we agree appears on
the surface) such as the abstract S and PRO. Now such extensions are not
objectionable in themselves, but if no independent motivation can be given
to support them, the theory which avoids such elaborations is to be favored.
In this case, no evidence is offered to support the abstract S or the PRO. In
 188 Realistic Syntax

fact, it would appear that new problems entailing further ad hoc excesses are
a consequence of the proposed elaborations.
Notice first that preservation of the abstract S creates a gap in phrase
structure as well as in the lexicon in terms of lexical subcategorization. Thus,
if all VP-like complements are derived from S complements, there will be no
VP complements expressed in the phrase structure rules or in lexical
subcatorization. But certainly one expects to find VP complements, since
one finds all the other major categories such as S, NP, and AP. This gap is
closed if VP complements are in fact VP’s in underlying representations.
Second, notice that the availability of PRO now raises the question of
distributional generality. Since true NP’s can occupy all the NP positions
provided by the phrase structure rules, e.g. the object position, object of
preposition position, subject position, etc., one would expect PRO, if it
really exists, to occupy the full range of NP positions provided by the phrase
structure rules. But no evidence has been advanced to indicate that PRO
occupies anything but the subject position. Surely this limited distribution
should lead us to question the existence of PRO. Third, the availability of the
abstract S raises problems of the following sort. How are we to block
underlying complement structures from being associated with tough when
they are of the following type?

(15) a. It is tough [s for John to learn such things]

b. It is tough [s for us to yield to such demands]

According to Lees and Bresnan, adjectives such as tough do not select S

complements, but rather only PP + VP or VP complements. And this claim is
borne out, at least for me, by the fact that It is tough for John to learn such
things cannot receive an interpretation ‘For John to lean such things is
tough’ but rather must be interpreted as ‘To learn such things is tough for
John’ and likewise for (15b). Notice that it will not do to block (15) by simply
marking tough to obligatorily select a PP, i.e. a PP + S complement. This is so
in view of examples such as (16).

(16) It is tough to solve such problems.

If (16) derives from It is tough [s for PRO to solve such problems], then one
should expect (15) to be a legitimate structure; but it is not, at least for
speakers such as Lees, Bresnan, and myself. On the other hand, if (16)
derives from a source such as It is tough [PP for PRO] [s for PRO to solve such
problems], then new difficulties arise. First, a new rule will have to be
postulated to delete the PP for PRO. This rule is postulated for no other
reason than to undo the results the abstractions entail and in this sense is ad
hoc. Second, what now is to block It is tough for PRO [for John to solve such
 Tensed S and Specified Subject 189

problems]? This in turn raises the general problem of how the presence of
PRO in the abstract S subject position is to be insured when PP is present.
Thus, what is to block It is tough for John [s for Mary to leave]!
It is certainly remarkable that none of the problems even arise within the
VP framework. All other things being equal, I think the conclusion is
inescapable. The VP framework is to be favored over the analysis adopting
the abstract S, PRO, and constraints such as SSC and TSC. Until the
problems raised here can be satisfactorily resolved and independent justifi¬
cation provided for the theoretical elaborations entailed by the framework
adopted by Chomsky, Object Shift phenomena cannot be taken as evidence
for SSC and TSC.5
2.2. Each-Movementand TSC

Chomsky [12] claims to follow Dougherty [16] in assuming a rule which

moves each so as to relate pairs such as the following.6

(17) a. The candidates each hated the other,

b. The candidates hated each other.
Chomsky calls this rule each -Movement (and occasionally each-Insertion
[12:239]) and he draws on it throughout his study as evidence to support
SSC and TSC. Thus, consider the difference between (18) and (19).

(18) a. The candidates each expected the other to win.

b. The candidates expected each other to win.
(19) a. The candidates each expected that the other would win.
b. *The candidates expected that each other would win.

It is claimed that (18b) derives from (18a) via each -Movement (hereafter
EM), whereas (19b) does not derive from (19a) by virtue of the fact that EM
is blocked by TSC as illustrated in (20).
(20) X Y
I I
The candidates each expected [s that the other would win]

Since that the other would win is a tensed sentence, application of EM to

(19a) would constitute a violation of TSC. EM is not prohibited from
applying to (18a), however, since the other to win is not a tensed sentence.
Now it can be shown that EM is not a movement rule. Rather, each other
should be generated as such rn the base.7 The question therefore shifts from
that of explaining why EM cannot apply to (19a) to that of why each other
cannot be generated directly in (19b), or, alternatively, why (19b) is filtered
in the event that it is generated. Chomsky recognizes the latter possibility
 190 Realistic Syntax

when he remarks: “Notice that if one were to accept the alternative analysis
of Jackendoff (1969), principle (20) [ = TSC, MKB] would again apply-in
this case, not to a movement rule but to a rule of interpretation” [12:238, fn.
17]. Thus, again we see that TSC is instrumental in accounting for the
deviance of (19b).
There is some reason to believe, however, that TSC has nothing to do with
the deviance of (19b). Consider by way of example the following cases.

(21) a. *Each other will win.

b. *The boys and each other will win.

The point is simply this. Under the conditions approach, TSC is invoked in
ruling out (19b), whereas it plays no role whatsoever in ruling out (21a) and
(21b). But, clearly, the mechanism that is independently needed to rule out
(21a) and (21b) can be utilized to rule out (19b), in which case (19b) cannot
be said to constitute motivation for TSC. We might do this by constraining
lexical insertion so that only subject pronouns are inserted into the subject
position of tensed S’s. Reciprocals such as each other, being object pronouns,
would not be inserted into the subject position, as desired. Formally, this
might be accomplished by enriching the phrase structure rules or by expand¬
ing the complex symbol associated with Pronoun to yield [-objective] when
followed by Aux. Subject pronouns would then be marked with this feature,
as opposed to objective pronouns such as each other, himself, him, etc.,
which are [-(-objective]. Such an account would insure that each other could
not be inserted into the subject position of (21a), (21b), and also (19b). This
proposal, incidentally, is simply a lexical analogue of Klima’s case marking
rule, which marks the subject of a tensed S as nominative. Such an account
immediately generalizes to account for the following data.

(22) a. *Himself will win.

b. *Him will win.
c. He will win.

It will automatically allow for the following distinctions.

(23) a. *We believe that each other are incorrect.
b. We believe that each other’s theories are incorrect.

It also provide? for the following distribution of data.

(24) a. The candidates knew that pictures of each other would be on

sale.
b. The candidates knew that pictures of themselves would be on
sale.
c. The candidates knew that pictures of them would be on sale.
 Tensed S and Specified Subject 191

The constraint against the appearance of each other in the subject position of
a tensed S is thus seen to be a special case of a more general lexical or base
rule constraint against object pronouns in this position. TSC is incapable in
principle of accounting for this range of data. Rather, it appears to lead
to a loss of generalization in this area.
There is a further consideration which is worthy of comment. This
concerns languages with an underlying order of constituents which differs
from the English SVO order. Joan Bresnan has informed me that my idea
that each other and other instances of ‘bound anaphora’ must agree with
their subjects predicts that reciprocals and reflexives could precede their
antecedents in VOS languages, whereas Chomsky’s assumption that bound
anaphors agree with arbitrary antecedents, subject to SSC and conditions on
proper binding, predicts that this could happen only if NP-movements are
subject to the same conditions, namely rightward movement as opposed to
leftward. If reciprocals and reflexives in Malgasy are bound anaphora,
Bresnan points out that such a language would appear to support my
assumption and disconfirm Chomsky’s, because reflexives do precede their
subjects, whereas NP-movements are leftward.
Now let us turn to the following examples.

(25) a. The boys each think that Bill will fool the other,
b. *The boys think that Bill will fool each other.
(26) a. Bill thinks that the boys each will fool the other,
b. Bill thinks that the boys will fool each other.

Under the movement approach to each other, (25b) cannot derive from
(25a) because of TSC, whereas (26b) can derive from (26a) without incident.
But, as noted above, the movement approach is misguided.8 Therefore,
according to Chomsky’s interpretive analogue, each other and the boys in
(25b) cannot be related by the interpretive anaphora rule because of TSC
(and also by virtue of SSC).
But TSC again has nothing to do with the deviance of (25b). To see this
consider the following examples.

(27) a. *Bill will fool each other.

b. The boys will fool each other.

The most obvious and straightforward account of (27a) is simply that each
other and Bill are anaphoric, but since anaphorically related NP’s must agree
in number and gender, (27a) constitutes a violation of concord, Bill being
singular and each other plural. Such an account generalizes to other
anaphoric pronouns.
192 Realistic Syntax

(28) a. *Bill likes herself.

b. *Bill likes themselves.

This account also generalizes to rule out (25b), granted that anaphoric
pronouns such as each other are marked as anaphoric with their subject
antecedents.
In summary, there are two assumptions. First, I assume a rule of anaphora
which marks anaphoric pronouns such as reciprocals and reflexives as
anaphoric to their subject antecedents.9 Let us write this interpreted relation
as (NPI( NPj), to be read NP, is anaphoric to NPy, where NPj is an anaphoric
pronoun and NPj is its subject antecedent. Second, I assume that there is a
filter which marks those sentences in which the anaphorically related NP’s
differ in concord as ill-formed. Such a filter might be formulated in the
following way.

(29) Gender/Number Filter:

[s... NPj... NPj... ]-► #
ctF -aF
where (NPj, NPi) and F is gender or number

Now it is true that the interpretive rule which provides us with the anaphoric
relations (NPj, NPj) must itself be constrained so that it does not apply across
tensed S’s; however, it is shown in [6] that TSC receives no support from the
general rule of anaphora.
The point is again this. Even Chomsky must develop an independent
mechanism for marking examples such as (27a) and (28) as deviant. But this
independently needed mechanism, which I have formulated as (29), should
in fact be utilized to account for the deviance of (25b). Once this indepen¬
dent device is brought to bear, examples such as (25b) cease to count as
evidence for TSC.

2.3. EM and SSC

On quite analogous grounds it can be shown that EM does not constitute

evidence for SSC. Consider the argument for SSC based on the following
examples.

(30) a. The men each expect Bill to see the other,

b. *The men expect Bill to see each other.

According to the conditions approach, EM is blocked by SSC from applying

to (30a) to yield (30b). Or, looking at it from the interpretive perspective,
 Tensed S and Specified Subject 193

each other cannot be marked as anaphoric to the men in (30b) because of

SSC, as illustrated below.

(31) X Z Y
I I I
the men expect [s Bill to see each other]

I-X-1

However, here, as in the case of (25b), special conditions such as SSC are
superfluous. Completely independent mechanisms such as the rule of
anaphora will mark Bill and each other as standing in an anaphoric relation¬
ship, i.e. (each other, Bill), since Bill is the antecedent subject. This in turn
constitutes a violation of number agreement, so that (30b) is filtered by (29).
This explanation derives additional support from the fact that the following
must be filtered in a completely analogous fashion.

(32) a. *John’s pictures of each other are on the table.

b. *John’s pictures of themselves are on the table.
c. *The men’s pictures of himself are on the table.
d. *John’s pictures of one another are on the table.

On the other hand, the following examples are not filtered by (29).

(33) a. The men’s pictures of each other are on the table.

b. John’s pictures of himself are on the table.
c. The men’s pictures of one another are on the table.

In (32) the subject and anaphoric pronoun must again stand in an anaphoric
relationship, but this time the relevant phrase is the cyclic category NP as
opposed to S. These anaphoric relations, however, give rise to concord
violations. Quite apart from SSC, some extra mechanism is required to mark
the examples of (32) as deviant. The claim made here is that precisely this
mechanism should constitute the explanation for the deviance of (30b). This
conclusion shows that SSC derives no support from examples such as (30b).

2.4. Unlike Person Constraint and Disjoint Reference

Chomsky adduces the following examples with associated grammaticality

judgments.

(34) a. *1 saw me.

b. *1 watched us leaving.
c. *We watched me leaving.
d. *You (all) noticed you standing there.
194 Realistic Syntax

Following Postal, Chomsky claims

... the point seems to be that a rule of interpretation RI applying to the structure NP-V-NP
(among others) seeks to interpret the two NPs as nonintersecting in reference, and where this is
impossible (as in the case of first and second person pronouns...), it assigns ‘strangeness’,
marking the sentence with *. [12:241]

Chomsky goes on to cite the following examples.

(35) a. We expect them to visit me\ I expect them to visit us (me).

b. * We expect me to visit them; */ expect us (me) to visit them.
c. *We expect me to be visited by them; */ expect us (me) to be
visited by them.
d. We believe I may still win; / believe we (I) may still win.

In connection with (35a) and (35d), it is claimed that SSC and TSC block RI,
and thus these examples are not starred. “Therefore in these sentences the
pair of italicized NPs may intersect in reference; the sentences are not
marked with * by RI” [12:241].
Let us attempt to be more explicit about what is asserted in the foregoing
quote. First, the reference sets denoted by NPs are interpreted as non¬
intersecting, i.e. their intersection is null, when SSC and TSC do not function
to block this interpretation. This claim is represented by the following rule,
where Ref (X) indicates the reference set denoted by X.10

(36) RL: Ref(NPO D Ref(NPj) = <t> if [s ... NR... NPj...]

Second, some sentences are actually starred when the NPs involved are both
first person or both second person pronouns. This claim is represented by the
following rule.

(37) RI2: [s ... NPj... NPj ...]-»*

where NP; and NPj are both first or second person pronouns

Now there is some reason to suspect that reference, intersecting or

otherwise, is not crucial to this discussion. First of all, note that (35b) and
(35c) are in fact perfectly grammatical. This fact could be accounted for by
incorporating a further restriction into (37), namely, that NPj and NPj agree
in number as well as person (exclusive of third person). However, there is
another fact, and this is simply that (34a-d) are also perfectly grammatical in
spite of the claim to the contrary. What is remarkable about (34a), for
example, is not that it is ungrammatical, but, rather, that it receives a special
interpretation, which, to borrow a term from Lewis, we might call the
counterpart interpretation. To see that such examples are indeed grammati¬
cal, we might consider examples involving more contextual material, such as
the examples listed under (38).
 Tensed S and Specified Subject 195

(38) a. While watching TV the other day, I saw (us, me) leaving.
b. Why don’t you come up to my place and we’ll watch (us, me)
leaving on TV.
c. If I were you, I would kiss me.
d. I dreamed I was you and that I kissed me.
Now the crucial point here is not that rules such as (36) and (37) are blocked
by conditions such as SSC and TSC, but rather that certain pronouns are
interpreted as counterparts under structural conditions which turn out to be
quite analogous to those required for anaphora. Thus, again representing
the anaphora relation with parentheses as in section 2.2 and representing the
counterpart relation with angled brackets, we observe the following.

(39) a. I saw myself (myself, I)

b. I saw me. (me, I)
(40) a. I know you saw myself. ~ (myself, I)
b. I know you saw me. ~<me, I)
(41) a. I told myself to leave. (myself, I)
b. I told me to leave. (me, I)
(42) a. I told you to kiss myself. ~ (myself, I)
b. I told you to kiss me. ~ (me, I)

This parallelism is expressed in a general way in the theory outlined in [6],

where counterpart interpretation, like anaphora, is an interpreted syntactic
relation. Though coreference must take such syntactic relations into
account, neither anaphora nor counterpart itself can be considered a seman¬
tic relation on a par with pronominalization. In conclusiop, the relevant
examples cannot be considered as evidence for SSC or TSC.
There is, incidentally, evidence of another sort, brought to my attention
by Joan Bresnan, which suggests that intersecting reference is not crucial to
the correct account of the relevant examples. Thus, consider the following
examples.

(43) a. John and Bill like him.

b. John and Bill like them.
Now him in (43a) and them in (43b), under the approach being criticized
here, exhibit reference sets which do not intersect with the reference of John
and Bill, which is presumably the set of John and Bill. But if reference is
indeed involved, then an NP denoting the complement set of John and Bill
should allow pronominal reference to John and Bill. However, this predic¬
tion is not borne out in the relevant examples listed in (44).

(44) a. Neither John nor Bill like him.

b. Neither John nor Bill like them.
 196 Realistic Syntax

In (44), just as in (43), the reference of him and them cannot intersect with
the set denoted by the subject, the set of John and Bill in the case of (43) and
the complement set in the case of (44). This suggests that it is not reference,
but rather the counterpart relation that we are dealing with here since in
both (43) and (44) the pronouns can be interpreted as counterparts to their
subjects. This conclusion is supported by the fact that reflexive pronouns are
also possible under analogous conditions.

(45) a. John and Bill like themselves.

b. Neither John nor Bill like themselves (likes himself).

The correlation between anaphora and counterpart is thus seen to hold in

these cases as well. This shows that such relations are syntactic and do not
depend on reference, in spite of the fact that any theory of reference will take
such relations into account.

2.5 Wh-Movement

The following examples are offered as further motivation for SSC.

(46) a. Who did you see pictures of?
b. *Who did you see John’s pictures of?

In (46b) John's is a specified subject, whereas (46a) has no subject

associated with pictures.12 Thus, it is argued, (35b) constitutes a violation of
SSC, accounting for its deviance.
Some speakers have disagreed with the grammaticality judgments indi¬
cated in (46), maintaining that (46b) is perfectly grammatical. Since (46a)
and (46b) are quite similar structurally (they do differ, as will be noted
below), it seems reasonable that some speakers will provide a parsing of
(46b) on analogy with (46a). This false parsing should be due to performance
factors and can be dismissed in view of many additional clear-cut examples,
including the following.

(47) a. Bill saw paintings by a famous artist,

b. Which artist did Bill see paintings by?
(48) a. Bill saw Mary’s paintings by a famous artist,
b. *Which artist did Bill see Mary’s paintings by?

However, additional examples show that it is not a fact about subjects, i.e.
SSC, that accounts for the deviance of (46b) and (48b).

(49) a. Bill saw those paintings by Rembrandt.

b. *Which artist did Bill see those paintings by?
(50) a. Mary saw those pictures of Harry.
b. *Who did Mary see those pictures of?
 Tensed S and Specified Subject 197

Chomsky recognizes that determiners like those, these, the, etc. lead to
violations and suggests that SSC might be refined to accommodate “the
feature [+definite] as well as the property of lexical specification” [12:239,
fn. 19]. He goes on to consider accounting for “a three-way gradation of
acceptability” with examples such as (46b) apparently being less accept¬
able that (49b) and (50b). Although there may indeed be recognizable
degrees of acceptability in these examples, it seems that these acceptability
judgments ought to be construed as a relatively clear case of a performance
phenomenon. Most speakers, would react even more strongly against (51),
for example, than (46b) because the subject is even more complex.

(51) *Who did Bill see John and Mary’s pictures of?

It seems better to assume that all such example are ungrammatical, their
deviance being due not to the presence of subjects, but rather to the
presence of constructions bearing heads. I have in mind here a structural
difference between pictures ofNP on the one hand and Det pictures of NP on
the other, where Det may dominate articles or possessive NPs. One proposal
for such a distinction is provided in (52).

(52) a. [n pictures [PP of N]pp]n

b. [n Det [n pictures [PP of N]pp]n]n

Given such a structural difference one might propose a general constraint on

the scope of rules applying to structures such as (52b). In fact, such a
constraint has been proposed by a number of authors. A recent formulation
of the Head Constraint by van Riemsdijk [25] is the following.

(53) Head Constraint:

No rule may involve X and Y in the structure
... X ... [Hn • • • [h1 •.. H ... Y ... Jh1 •.. ]Hn • • •
where H is the phonologically specified head and Hn is the
maximal projection of H

The Head Constraint is an interesting alternative to SSC. Together with

(52a, b), it correctly predicts that (46a) and (47b) are grammatical, while it
predicts that (46b), (48b), (49b), and (50b) are not. Moreover, it does not
block the following, given rational assumptions about the structure of such
examples.

(54) a. Who do you expect John to talk to?

b. Who do you want Bill to see?

Such examples as these are counterexamples to SSC and occasion new

elaborations such as successive cyclicity with special COMP nodes. The
Head Constraint does not entail these devices.
 198 Realistic Syntax

There is, however, evidence that the Head Constraint is not the correct
account in such cases. Consider, for example, the important fact that the
following examples exhibit a range of grammaticality completely analogous
to the foregoing examples.13

(55) a. Who saw pictures of who(m)?

b. *Who saw John’s pictures of who(m)?
c. *Who saw those pictures of who(m)?
(56) a. Who saw pictures by who(m)?
b. *Who saw Mary’s pictures by who(m)?
c. *Who saw those pictures by who(m)?

Examples such as these should be read without echo intonation. They

demonstrate, I believe, that the constraints approach to such phenomena is
fundamentally misguided. Rather, the underlying structural difference laid
out in (52) (or some such structural difference) should be crucial to filtering
(or lexical insertion) of scope words such as who, what, etc. Such an account,
elaborated in [6] and to some extent in the following subsection, does not
make reference to SSC.

2.6. Scope of Negation

A further argument for SSC is attributed to Lasnik and is based on examples

such as the following.

(57) a. I didn’t see many of the pictures.

b. I didn’t see pictures of many of the children.
c. I didn’t see John’s pictures of many of the children.

Chomsky claims that (57a) and (57b) are ambiguous. On one reading, the
negative element can be associated with many; on the other reading, it is
associated with see. On the former reading of (57a), few pictures were seen,
while on the latter reading, many or few may have been seen, but many were
not seen. A similar ambiguity is said to hold of (57b). However, it is claimed
that (57c) is not ambiguous. It can bear only the interpretation associating
the negative element with the verb see. The negative element cannot be
associated with many in (57c). The reason for this is that SSC blocks
interpretation across the specified subject John's.
These judgments are subtle and I find difficulty sorting out the various
interpretations, but they are not implausible. However, if scope of negation
can be used to argue for SSC and TSC, certainly it can be used to argue
against SSC and TSC if relevant examples can be found. In fact, it is well
known that negation in matrix structures can affect material in complement
 Tensed S and Specified Subject 199

structures. Thus, to take a familiar example from Chomsky [11], we find the
following facts concerning until phrases.

(58) a. *Bill will leave until midnight.

b. Bill will not leave until midnight.

Thus, until midnight must fall within the scope of negation in order for the
sentence to exhibit full grammaticality. But now consider the fact that the
following is fully grammatical.

(59) Mary didn’t expect Bill to leave until midnight.

Now, according to the approach being criticized here, Bill is a specified

subject in (59), and yet the scope of negation is able to filter past Bill so as to
permit the until phrase. Such examples provide strong counterevidence to
SSC, and analogous examples can be multiplied. Thus, idioms such as give a
damn, cut any ice, and others behave like the until phrase in exhibiting full
grammaticality in the presence of negation.14

(60) a. *We give a damn whether we pass or fail.

We don’t give a damn whether we pass or fail,
b. *It cuts any ice whether you pass or fail.
It doesn’t cut any ice whether you pass or fail.

Again, the scope of negation is able to filter past the specified subject as
exhibited in the following sentences.

(61) a. John doesn’t expect us to give a damn whether we pass or fail,

b. John doesn’t expect it to cut any ice whether you pass or fail.

Thus, the facts attributed to Lasnik do not support SSC.

Assuming that Lasnik’s and Chomsky’s judgments concerning the
ambiguity of (57a, b) and the nonambiguity of (57c) are correct, it remains to
provide an alternative explanation for these facts. In this connection let us
note the following examples.

(62) a. I didn’t see any of the children.

b. I didn’t see pictures of any of the children.
c. *1 didn’t see John’s pictures of any of the children.
(63) a. I didn’t see any of the artists.
b. I didn’t see pictures by any of the artists.
c. *1 didn’t see John’s pictures by any of the artists.

These examples are quite analogous to those cited in (57), repeated here to
bring out the parallelism.
 200 Realistic Syntax

(64) a. I didn’t see many of the pictures.

b. I didn’t see pictures of many of the children.
c. I didn’t see John’s pictures of many of the children.
(65) a. I didn’t see many of the artists.
b. I didn’t see pictures by many of the artists.
c. I didn’t see John’s pictures by many of the artists.

There is indeed an analogy here. Many may or may not receive negative
scope, but in (64c) and (65c) it cannot, so these examples are unambiguous.
Any, on the other hand, requires negative scope,15 but does not have it in
(62c) and (63c), so these sentences are deviant. Thus, there is a similarity,
and the difference in grammaticality can be correlated with the following
difference.

(66) a. *1 saw any of the pictures,

b. I saw many of the pictures.

That the generalization cuts deeper still can be seen by comparing (62)-(65)
with the facts about reflexives to which we now return.

(67) a. The boys saw each other.

b. The boys saw pictures of each other.
c. *The boys saw John’s pictures of each other.
(68) a. I saw myself.
b. I saw pictures of myself.
c. *1 saw John’s pictures of myself.

Clearly, SSC is the wrong way to go about generalizing the negation and
anaphora cases. Further, the Head Constraint fails to generalize to the
examples discussed in the previous subsection (repeated as (69)) which
should be compared with (62)-(68).

(69) a. Who saw who(m)?

b. Who saw pictures of who(m)?
c. *Who saw John’s pictures of who(m)?

Therefore, instead of approaching the problem negatively, i.e. by specifying

what scope cannot do, let us approach the matter positively and stipulate
what scope must do.
I have already suggested a means of attacking the facts about anaphora in
subsection 2.2. It now remains to generalize such an account. In what follows
I will sketch a general treatment for N; a more general account covering S as
well can be found in [6].
 Tensed S and Specified Subject 201

Let us stipulate that scope items such as who, what, any, ever, etc. as well
as anaphoric pronouns such as each other, myself, etc. be obligatorily related
to their associated heads in the presence of N. This was the account provided
for anaphoric relations earlier where it was assumed that himself is marked
as anaphoric to John in (33b) and as anaphoric to the men in (32c). The latter
example constitutes part of the motivation for a concord filter such as (29).
Suppose now we do the same for the other scope phenomena. Pictorially, we
might formulate the scope rule as follows.

(70) Scope Rule: Interpret R(S, Det) in [nH[ ... S .. .]]n where S is a
scope word such as wh, any, myself, H is the lexical head of N,
and R is the scope relation

A specific instance of the scope relation R is anaphora which was stated

above as (NP;, NPj). Now in (62c) we get R (any, John), but this does not
satisfy the constraint that an affective item must be the second coordinate of
R when any is the first coordinate. Therefore, (62c) is filtered in a way
analogous to examples of concord disagreement in anaphora.
Within the framework I am advocating here, the key point is that scope
items are obligatorily interpreted as related to their heads. Within the
conditions framework, the affective contexts are blocked from having scope
which crosses a head. My framework positively specifies its scope relations;
the other allows arbitrary scope constrained by conditions such as SSC or the
Head Constraint. In a sense, it says what cannot constitute a legitimate
scope relation, rather than what can.
The theory advocated here should be able to express the following
parallelism.

(71) a. We believe that each other’s theories are incorrect.

b. We don’t believe that anyone’s theories are incorrect.
(72) a. *We believe that John’s pictures of each other are on the table.
b. *We don’t believe that John’s pictures of anybody are on the
table.
(73) a. We believe that pictures of each other are on the table.
b. We don’t believe that pictures of anybody are on the table.
Examples such as (71) are discussed in [6]. Here, let us be content to note
that relative clauses exhibit a similar parallelism and are treated by the
Scope Rule (70).

(74) a. I saw the boy who hit himself

I_I
b. *1 saw the boy who hit myself
 202 Realistic Syntax

(75) a. I never met anyone who ever won at baccarat.

b. *1 never met the boy who ever won at baccarat.

(76) a. John is the only boy who saw anything.

b. *John is the boy who saw anything.

Notice that the relation is well-formed and hence is not filtered when there is
no concord disagreement in the case of anaphora and when an affective
element is present in the head in the case of a negative scope item such as
ever or anything. Such an account can be generalized beyond N as in [6], This
account correctly predicts the facts concerning scope of negation in (62) and
(63), while allowing (59) and (61), which are counter examples to SSC.

2.7. Passive

Another argument offered as motivation for TSC concerns the following

examples.

(77) a. I believe the dog is hungry.

b. *The dog is believed is hungry by me.

According to Chomsky, there is no way to block the Passive transformation

from applying to (77a) to yield the ungrammatical (77b) within the standard
theory since nodes that dominate no terminal string play no role “in
factoring terminal strings for transformation” [12:273, n55]. Thus, even
though COMP intervenes between the verb believe and the NP the dog in the
relevant deep structure (78), Passive will still apply.

(78) I believe [s COMP the dog is hungry]

However, as Chomsky notes, “there are other conventions” that might be

conceived, and one obvious proposal in this case is that nodes that dominate
no terminal strings, such as COMP, do play a role in factoring terminal
strings for transformation. If the rule of Passive is formulated so as to require
that the object immediately follow the associated verb, then (77b) will be
avoided under this alternative convention.
However, it seems to me that there is another even more plausible
explanation for the ungrammaticality of (77b). This explanation rests on
some results argued for in [3] and depends on base structures which are
 Tensed S and Specified Subject 203

considerably less abstract than assumed in the standard theory. In [3] and [5]
I argue that VP complements are generated directly, along with passives,
expletive there sentences, raising sentences, etc. If this is the case, then it is
necessary to explain why the following examples are ungrammatical.

(79) a. I persuaded Mary [Vp is present]

b. I expected Mary [Vp is present]

One answer is to adopt Emonds’ proposal that no Aux node be generated as

an expansion of VP. Rather Aux is reserved for S and gives rise to tense (M).
If Emonds’ proposal is adopted, then (79a, b) cannot arise. Alternatively,
Aux might be generated in the VP, but expanded only as to and possibly ing
when dominated by VP. In either case, such restrictions on VP complements
will allow both of the examples listed in (80) while prohibiting the examples
in (81).

(80) a. Mary persuaded John to be quiet.

b. John was persuaded by Mary to be quiet.
(81) a. *Mary persuaded John is quiet.
b. *John was persuaded by Mary is quiet.

But clearly (81b) is analogous to (77b). Precisely the same restrictions on

VP complements can be used to prohibit (77b) where is hungry is taken to
be VP. Passive of course cannot apply to (77a) because it is not a trans¬
formation. Rather, passive structures are generated directly as argued in [3]
and [5].

2.8. Naturalness

As a final argument for SSC, Chomsky notes that SSC “has a certain
naturalness” in having “the effect of reducing ambiguity, or, to put it
differently, of increasing the reliability of a reasonable perceptual strategy
that seeks the nearest NP to a verb (or the head noun of a nominal phrase) as
its subject” [12:270]. Thus, in the example the men expected the police to
arrest each other, SSC reduces ambiguity by requiring the each of each other
to be associated with the police and not the men. But, clearly, within the
approach outlined in subsections 2.2,2.3 and 2.6, this result is also obtained,
and in a positive manner. Moreover, there is little to be gained by resorting
to plausibility arguments based on very abstract ideas such as naturalness,
given the poverty of our present understanding of what constitutes a natural
syntactic process. Indeed languages are quite rampant with ambiguity
and one might even argue that it is natural that languages are blessed with
such richness. Whatever the case, naturalness is no substitute for internal
204 Realistic Syntax

arguments, and I do not see that naturalness provides any support for SSC
over the alternatives sketched out here.

3. Counterexamples to ssc, tsc, and subjacency

I think that the foregoing discussion demonstrates that there is little motiva¬
tion for SSC and TSC. Moreover, many of the facts brought to bear in
section 2 argue against these conditions. In the present section I want to
advance additional arguments against SSC, TSC, and a further condition,
subjacency. Preparatory to presenting these arguments, let us review subja¬
cency and its consequence, successive cyclicity.

3.1. Introduction to Subjacency and Successive Cyclicity

Chomsky notes examples such as those listed in (82).

(82) a. What did you tell me Bill saw?

b. Who did you say left?

He points out that (82a) appears to violate both SSC and TSC, while
(82b) appears to violate TSC. Therefore Chomsky reanalyses wh -fronting,
utilizing Bresnan’s COMP node, so as to move the w/i-word leftward in a
successive cyclic fashion as illustrated in (83).

(83) a. COMP you told me [s COMP Bill saw what]

t_It_|
b. COMP you said [s COMP who left]
t_It_I
A more complicated example is (84a), which is derived in a successive cyclic
fashion according to (84b).

(84) a. Who did John say that Bill wanted Harry to investigate?
b.
COMP John said [s COMP Bill wanted [§ COMP Harry to investigate who]]
t_It_It_I
This analysis is apparently required given the new constraint Chomsky
proposes.

(85) Subjacency: No rule can affect material in two cyclic domains

unless these domains are adjacent.
 Tensed S and Specified Subject 205

This condition has the effect of prohibiting w/i-fronting in one sweep as in

the traditional approach. Further, to avoid violations of TSC and SSC, these
conditions are reformulated and collapsed to give the following.

(86) No rule can involve X, Y in the structure

.’. .X...[a ... Z ... -WYV ...]...
where (a) Z is the specified subject of WYV
or (b) Y is in COMP and X is not in COMP
or (c) Y is not in COMP and a is a tensed S

(86a) is SSC and (86c) is TSC revised so as to allow extraction when Y, in the
above case the w/i-NP, is in COMP. Condition (86b) prohibits movement
from COMP unless the relevant trigger is itself in COMP. The net effect is
that wfc-fronting proceeds from COMP to COMP.
Before turning to the counterevidence, let us consider one further revision
of SSC. This revision is occasioned by examples such as the following.

(87) a. We each persuaded Bill to kill the other,

b. *We persuaded Bill to kill each other.

According to (86), EM (or the interpretive analogue) should not be blocked

in (87a). Therefore (87b) should be grammatical. This follows from the fact
that the embedded clause contains PRO in Chomsky’s framework, which is
an unspecified subject under the previous definition. To obtain the situation
illustrated in (88), Chomsky revises SSC as in (89).

(88) We each persuaded Bill [COMP PRO to kill the other]

(89) SSC: No rule can involve, X, Y in the structure

... X ...[... Z ... -WYV ...]...
where Z is the subject of WYV and is not controlled by a
category containing X.

Now in (88) it is Bill, not X (we each), which controls PRO. Therefore (89)
applies, and EM will be inapplicable to (88). On the other hand, SSC will not
apply in the case of (90), so that EM will be allowed.

(90) We each promised Bill [COMP PRO to kill the other]

Thus, we each promised Bill to kill the other can be converted to we promised
Bill to kill each other.
 206 Realistic Syntax

3.2. Counterexamples

Let us now turn to some problems that arise in connection with TSC, SSC,
and Subjacency.

3.2.1. Clef ting

Clefting provides a new source of counterevidence against successive cycli¬
city in conjunction with constraint (86). The relevant example is due to
Joseph Emonds (personal communication).
(91) It was in the garage that John put the car.

Note that (91) apparently must involve a transformational operation since

the following is ungrammatical.

(92) *John put the car.

Therefore, (92) apparently derives from a source containing the focus

prepositional phrase in the f/uzf-complement. Plausible deep structures
include the following.

(93) a. It was in the garage [s that John put the car in the garage]
b. It was A [s that John put the car in the garage]

Now if (93a) is adopted as the underlying source of (91), a deletion

transformation will be required, but deletion under identity will violate
subjacency, SSC, and TSC in the following example.

(94) It was in the garage that Mary said that John put the car.

On the other hand, if (93b) is adopted as the underlying source for (91), a
movement transformation will be required. Clearly, movement in one sweep
from the nonfocus position to A will again constitute a violation of subja¬
cency, SSC, and TSC. Consequently, movement must proceed in a succes¬
sive cyclic fashion. This is illustrated in (95).
(95)

It was A [s COMP Mary said [s COMP John put the car [PP in the garage]]]
t_It_It |

But this derivation also leads to a violation, since the final movement from
the COMP to the focus position entails a violation of (86b). This is so since
the PP moves from a COMP position to a nonCOMP position.

3.2.2. Comparative Subdeletion

Bresnan [9] has provided additional examples which pose problems for
 Tensed S and Specified Subject 207

Subjacency. She argues for a rule of Comparative Subdeletion which deletes

underlying measure phrases in examples such as the following.

(96) a. Why were there more women on TV than there were_men?

b. There weren’t as many men on TV as there were_women.

This deletion, she argues, violates Subjacency in examples such as the

following.

(97) Therefore, they can hire more women than the Administration
would allow them to hire_men.

Bresnan advances arguments against a movement analysis for such

examples. If her arguments against movement are correct, her examples
constitute counterevidence to Subjacency.

3.2.3. wh-Movement
Chomsky adduces the following examples and assigns them the underlying
structures provided in (99).

(98) a. They will obey any request to kill each other,

b. *They will okay any request to kill each other.
(99) a. They each will obey [any request [COMP PRO to kill the other]]
b. They each will okay [any request [COMP A to kill the other]]

It is assumed that they each controls PRO in (99a), but not A in (99b). Given
the revised SSC (89), PRO will not function as a specified subject in (99a).
Therefore EM will be applicable, giving rise to (98a). However, A will
function as a specified subject under (89) in (99b). Hence EM will not be
applicable and (98b) will be blocked.16
In this connection consider the following examples in which, following the
above reasoning, we would conclude that A is present in the structures
underlying the examples.

(100) a.*They will okay any proposal to kill each other.

b. Which Cubans did the White House okay a proposal to kill?
c. It was Castro that the White House okayed a proposal to
assassinate.
Since presumably (lOOa-c) all contain A in their underlying structures, that
A should function as a specified subject given the revised version of the
constraint (89). This leads to the false prediction that (100b, c) are ungram¬
matical.
 208 Realistic Syntax

3.2.4. Each-Movement
Let us return now to EM in connection with examples such as the following.

(101) a. It would be pleasant for the rich for the poor immigrants to do
the hard work.
b. It would be pleasant for each of the men for the poor immigrants
to do the hard work.
c. It would be pleasant for each of the men for the others to do the
hard work.
c. *It would be pleasant for the men for each other to do the hard
work.

All of these examples involve a PP + S complement, and it is clear from

(101c) that quantified phrases can occupy the PP slot with the others
functioning as the subject of the embedded S. Chomsky’s approach predicts
that EM should be applicable to (101c) to yield (lOld) since there is no
intervening specified subject and no tensed S. Yet (lOld) is not grammatical.
If, alternatively, EM is viewed as interpretive, as suggested in section 2,
there will be nothing to block the relevant interpretation in the case of
(lOld) since SSC and TSC are not applicable. Examples such as (lOld),
therefore, appear to constitute evidence against the conditions approach. By
contrast, (lOld), can be ruled out in a straightforward manner according to
the suggestion of section 2.2 that object pronouns not be generated in the
subject position of S’s.17

3.2.5. Object Shift

Consider the following examples.

(102) a. It is pleasant for the rich to do the hard work,

b. The hard work is pleasant for the rich to do.

Example (102b) is perfectly grammatical and yet the deep structure

Chomsky assigns to it is the following.

(103) It is pleasant for the rich [COMP PRO to do the hard work]

Now Chomsky notes that Object Shift should not be applicable given the
revised version of SSC as (86). This follows since it functions as X, and PRO
is not controlled by a category containing X. It is because of this new
difficulty that Chomsky reanalyses Object Shift as two rules -PRO-
Replacement and It-Replacement - and incorporates traces into his
framework. A representative derivation is given in (104).
 Tensed S and Specified Subject 209

(104) It is easy [for PRO to please John] PRO-Replacement

It is easy [for John to please] It-Replacement
Johni is easy [for ti to please]

Chomsky argues that this reanalysis does not constitute a complication since
It-Replacement is needed in any event to get John seems to be a nice fellow
according to (105).

(105) It seems [COMP John to be a nice fellow] It-Replacement

John seems to be a nice fellow

Now the claim that such a reanalysis of Object Shift does not constitute a
complication in the grammar of English clearly rests on the truth of the claim
that It-Replacement can be utilized in the derivation (105). But Chomsky
himself notes that this entails complications. Thus, It-Replacement must
somehow be blocked in (106).

(106) It is pleasant for the rich [for the workers to do the work]
*The workers are pleasant for the rich to do the work

Since the workers are pleasant for the rich to do the work is ungrammatical,
new machinery is necessary to avoid the bad result. Chomsky’s analysis
involves stating PRO-Replacement so as to ‘combine’ the moved object of
the complement with PRO and stating It-Replacement so that it is obliga¬
tory when (i) PRO is part of the subject complement and when (ii) PRO is
not present in structures such as that evidenced in (105). This proposal is
contrived inasmuch as PRO-Replacement is the only such rule to ‘combine’
constituents and inasmuch as the ‘independent’ motivation for It-
Replacement is repossessed in view of the difference in obligatoriness with
respect to the related structures and presence of PRO. It should also be
added that none of the relevant rules are explicitly formulated.
Thus, the conclusion would seem to be inescapable, namely that simple
Object Shift examples such as (102) are counterevidence to SSC. In this
connection, it should be remarked that none of the difficulties discussed here
arise in the VP framework discussed in section 2.1.
Before closing this subsection, I would like to offer one final argument
against the foregoing approach to Object Shift. It is well known that
arbitrarily long strings of verbs do not block application of this rule. Thus,
Bresnan has cited examples such as the following.

(107) John is hard for Bill to even begin to try to please.

Now, if Chomsky’s underlying structures are adopted, then (108) will

underlie (107).
210 Realistic Syntax

(108) It is hard for Bill [for PRO to even begin [for PRO to try [for
PRO to please John]]]

Now if John is moved directly and substituted for it, not only will this violate
SSC, as noted above, but also it will violate subjacency. On the other hand, if
the PRO-Replacement analysis is adopted, John will ‘combine’ with the
PRO of the innermost embedded S. But now It-Replacement applied to this
result will violate subjacency! Thus, new elaborations are needed to derive
(107). Presumably PRO-Replacement must apply in a successive cyclic
fashion. But apparently this rule ‘combines’ an object NP with PRO,
whereas after one application, it would have to ‘combine’ a subject instance
of an NP/PRO combination with PRO. Without explicit rules, it is difficult
to envision what such an analysis would look like, and whether, in fact, it
would even be formulable.

3.2.7. Superiority
Chomsky attempts to account for the ungrammaticality of the following
examples.

(109) a.* John knows what who saw.

b.*To whom does John know what books to give?

These examples are to be read without echo intonation, for when echo
intonation is present the examples are fine. The account of (109) offered in
[12] involves a new condition which is called Superiority.

(110) Superiority: No rule can involve X, Y in the structure

... X ...[... Z ... —WYV ...]...
where the rule applies ambiguously to Z and Y and Z is superior
to Y.

Consider how this condition predicts the ungrammaticality of (109b). Here

the COMP of the underlying matrix clause functions as the X of (86). The
rule of wh -fronting would apply ambiguously to what books, which is Z, and
to to whom, which is Y. But Z is superior to Y, so (109b) constitutes a
violation of (110). By contrast, What books does John know to give to whom,
in which case Z is moved, does not constitute a violation and is indeed
grammatical.
The following examples are counterexamples to Superiority.

(111) a. What books does John want which student to read?

b. What books does John want who else to read?

To my ear such examples are quite natural when read without echo
intonation.
 Tensed S and Specified Subject 211

4. Conclusion

In section 2, the evidence adduced to support TSC and SSC has been
assessed, and alternatives have been suggested which adequately account
for the relevant range of data. In some cases, it has been argued that the
alternatives provide a more general account than the constraints. Additional
counterexamples to SSC and TSC have been provided in section 3, and
Subjacency and Superiority were also criticized.
In this paper, I have confined my remarks to Chomsky’s original paper
[12]. In recent works [13, 14], traces have assumed a more prominent role
than in the original essay. Constraints such as TSC and SSC are now
interpreted as constraints on surface structures, a move which is made
possible by the fact that traces serve to code into surface structures all
relevant aspects of the earlier transformational history of moved phrases.
Now clearly, trace theory and constraints such as TSC and SSC are logically
independent. Nevertheless, a good deal of the plausibility for trace theory is
intimately bound up with the plausibility of TSC and SSC, since, in practice,
it is constraints such as these that are now taken to be “conditions on an
enriched surface structure involving traces, instead of conditions on the
application of rules” [14:317], In light of the criticism that can be leveled
against TSC and SSC, it may be appropriate now to rethink the feasibility of
trace theory itself.

NOTES

1 See, for example, [14].

2 A number of relevant papers have appeared too late for discussion here: critical essays by
Yagi [26], Ikeuchi [20], and Bach/Horn [1] for example, as well as Chomsky’s most recent
contribution [15], which further refines and revises the proposals appraised here.
3 Actually the treatment of Object Shift presented here reflects Chomsky’s initial formulation.
At a later point in the development of his conditions he reanalyzes Object Shift as two
processes, thus obviating some of the initial development. The reanalysis is discussed below in
subsection 3.2.5.
4 Rule (10) can be extended in the obvious way to allow this violin would be easy to play the
Kreuzer sonata on.
5 A somewhat different arialysis of Object Shift cases which is consistent with Bresnan’s VP
hypothesis is offered in [3]. Cf. also the exchange between Bresnan and Berman/Szamosi,
which is discussed in [2].
6 Chomsky and Dougherty differ in important respects. Dougherty converts each of the
candidates hated the other into the candidates hated each other by means of three distinct
transformations - Quantifier Postposition, Quantifier Movement, and each other-
Transformation. Dougherty provides an explicit formulation of all three rules. Chomsky
apparently accepts a version of Dougherty’s Quantifier Postposition while the distinction
between Quantifier Movement and each of/ier-Transformation is blurred. These rules are not
stated, and it is therefore difficult to determine precisely to what extent Dougherty’s analysis is
followed.
7 One basic objection to the movement approach to each other has been voiced by Postal, who
212 Realistic Syntax

cited the men talked to each other about each other as underivable under the movement
assumption since the subject could give rise to only a single instance of each, cf. [24:72].
Additional arguments against movement are provided in Fiengo and Lasnik [18] and Brame
[4], In the latter it is shown that the putative arguments for movement are in fact strong
arguments against it. Additional arguments against movement are also provided, including
arguments against the Quantifier Postposition transformation that is accepted by Fiengo and
Lasnik.
8 An obvious argument against movement here is the fact that reciprocals such as one another
behave similarly, viz. *the boys think that Bill will fool one another, where there is no candidate
for movement such as each. See Brame [4] for additional examples.
9 Apparent counterexamples are discussed in [6],
10 Here, as elsewhere in this paper, subscripts are utilized as a formal device to distinguish
distinct occurrences of syntactic units, not as an indication of intended reference.
11 For more on the distinction, cf. Hust and Brame [19] and Lasnik [22].
12 I am following Chomsky [10] in assuming that John’s is the subject of the NP construction.
13 A similar point is made in a somewhat different context by Bresnan [8:46].
14 The idioms differ from the until phrase in exhibiting full grammaticality in ‘affective’
contexts.
(i) a. Do you give a damn about your exam?
b. Does it cut any ice whether you pass?
c. *Will you leave until midnight?
(ii) a. If you give a damn, you’ll pass that exam.
b. If it cuts any ice, he’ll tell us.
c. *If he leaves until midnight, he’ll tell us.
15 Excluding the “... at all” reading which is often possible in nonaffective contexts.
16 Throughout his paper, Chomsky assumes a notion of control which is never made explicit,
citing Jackendoff [21] for clarification. But Jackendoff here provides no explicit analysis of how
control is assigned; cf. [19] for discussion. It seems that some of the problems raised might
resolve themselves automatically if control were made explicit. Control appears to be simply a
name for a problem; its resolution involves taking seriously a program for working out Equi
within an interpretive framework. See [3] for one proposal.
17 In earlier sections, I spoke of the restriction as one related to tensed S’s, but the more general
formulation is needed in view of examples such as (10Id). If correct, this shows that reciprocal
pronouns such as each other are not part of S complements in examples such as They expect each
other to win. Thus, the evidence argues against Chomsky’s anti-raising position; cf. [12],
However, as I have argued in [4], there is no rule of Raising as classically conceived. Thus, I take
a position intermediate between Chomsky on the one hand, who argues against Raising but fails
to adopt the correct surface structures, and Postal on the other, who adopts the correct surface
structures, but fails to recognize that such structures are identical to the base structures, i.e. that
Raising does not exist. Cf. [4] for detailed discussion.

REFERENCES

[1] Bach, E. and Horn, G. M.: 1976, ‘Remarks on “Conditions on Transformations”’,

Linguistic Inquiry 7, 265-299.
[2] Brame, M. K.: 1975, ‘The VP Controversy’, Linguistic Analysis 1, 191-203.
[3] Brame, M. K.: 1976, Conjectures and Refutations in Syntax and Semantics, North-
Holland Publishing Co.
[4] Brame, M. K.: 1976, ‘Quantifiers, Reciprocals, and Raising’, unpublished ms.
[5] Brame, M. K.: 1977, ‘The Base Hypothesis and the Spelling Prohibition’, unpublished
paper delivered at the Eighth Annual Colloquium on Directions in Linguistics, University
of Calgary, March 21, 1977.
 Tensed S and Specified Subject 213

[6] Brame, M. K.: 1977, ‘Scope Phenomena’, to appear.

[7] Bresnan, J. W. 1971. ‘Sentence Stress and Syntactic Transformations’, Language 47,
257-81. Reprinted in Contributions to Generative Phonology, M. K. Brame, ed. Austin,
Tex.: University of Texas Press.
[8] Bresnan, J.: 1975, ‘Comparative Deletion and Constraints on Transformations’, Linguis¬
tic Analysis 1, 25-74.
[9] Bresnan, J.: 1976, ‘Evidence for a Theory of Unbounded Transformations’, Linguistic
Analysis 2, 353-393.
[10] Chomsky, N.: 1970, ‘Remarks on Nominalization’, in Readings in English Transforma¬
tional Grammar, R. A. Jacobs and P. S. Rosenbaum, eds. Waltham, Mass.: Ginn.
Reprinted in Studies on Semantics in Generative Grammar, N. Chomsky, 1972. The
Hague: Mouton.
[11] Chomsky, N.: 1972, ‘Some Empirical Issues in the Theory of Transformational Gram¬
mar’, In Goals of Linguistic Theory, S. Peters, ed., 1972. Englewood Cliffs, N.J.:
Prentice-Hall. Reprinted in Studies on Semantics in Generative Grammar, N. Chomsky,
1972. The Hague: Mouton.
[12] Chomsky, N.: 1973, ‘Conditions on Transformations’, in A Festschrift for Morris Halle, S.
R. Anderson and P. Kiparsky, eds. New York: Holt, Rinehart, and Winston.
[13] Chomsky, N.: 1975, Reflections on Language. New York: Pantheon Books.
[14] Chomsky, N.: 1976, ‘Conditions on Rules of Grammar’, Linguistic Analysis 2,303-351.
[15] Chomsky, N.: 1976, ‘On wh-Movement’, paper presented at the Conference on Formal
Syntax, June 1976, University of California at Irvine.
[16] Dougherty, R. C.: 1970, ‘A Grammar of Coordinate Conjoined Structures I’, Language
46,850-98.
[17] Emonds, J. E.: 1976, A Transformational Approach to English Syntax : Root, Structure-
Preserving, and Local Transformations. New York: Academic Press, Inc.
[18] Fiengo, R. and Lasnik, H.: (1973), ‘The Logical Structure of Reciprocal Sentences in
English’, Foundations of Language 9, 447-468.
[19] Hust, J. R. and Brame, M. K. 1976, ‘Jackendoff on Interpretive Semantics: A Review of
Semantic Interpretation in Generative Grammar by R. Jackendoff’, Linguistic Analysis 2,
243-277.
[20] Ikeuchi, N.: 1976, ‘Notes on the New Subjacency Condition’, Studies in English Linguis¬
tics 4, 139-142.
[21] Jackendoff, R. S.: 1972, Semantic Interpretation in Generative Grammar. Cambridge,
Mass.: MIT Press.
[22] Lasnik, H.: 1976, ‘Remarks on Coreference’, Linguistic Analysis 2; 1-22.
[23] Lees, R. B.: 1960, ‘A Multiply Ambiguous Adjective Construction in English’, Language
36, 204-21.
[24] Postal, P. M.: 1974, On Raising. Cambridge, Mass.: MIT Press.
[25] van Riemsdijk, H. 1975. ‘Extraction from PP and the Head Constraint’, to appear in
Linguistic Analysis.
[26] Yagi, T.: 1976, ‘Review of “Conditions on Transformations” by N. Chomsky’, Studies in
English Linguistics 4, 112-137.