)WNW"INHNIWHIWWllWlHHWl

   

mull1||l|”'"lll“llll Til 9~ 9 07 9" 7‘1

Iln
3 1293 00591 2484

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

LIBRARY
Michigan State

University

 

 

This is to certify that the

thesis entitled

REVISED BARRIERS AND ECP

presented by

Dae Hee Lee

has been accepted towards fulfillment
of the requirements for

Master's degreein Linguistics

    

Major professor

Date X/‘g/‘Io

0-7639 MS U is an Affirmative Action/Equal Opportunity Institution

 

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before date due.

DATEDUE DATE DUE DATE DUE N
39*" 1. '1 1'7 ll

 

 

 

 

 

 

 

 

 

 

 

 

1

ll 3] J

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

_JL

 

 

 

MSU Is An Affirmative Action/Equal Opportunity Institution

 

REVISED BARRIERS AND ECP

BY

Daehee Lee

A THESIS
Submitted to
Michigan State University

in partial fulfillment of the requirements

for the degree of
MASTER OF ARTS

Department of Linguistics

and Germanic, Slavic, Asian and African Languages

1990

ABSTRACT

REVISED BARRIERS AND ECP

BY

Daehee Lee

The goal of this paper is to compare certain variations
of the definitions of government, proper government, barriers,
and the ECP, and to explore the generalized definition of
proper government and barriers and their application to the

phenomena of the ECP related to Wh-movement and NP-movement.
Chomsky's (1986) concept of barriers will be redefined to
eliminate from the definition of barriers the stipulation
that, by assuming that.they are 'defectiveicategories', IPtand
I' are excluded from barrierhood.

Tot avoid. Lasnik/Saito's (1984) and. Chomsky's (1986)
inconsistent definition of the ECP, it will be redefined in
terms of 'consistent proper government'. In particular, the
algorithm of NP-index assignment will be introduced here to
explain why relative clauses lack the that-trace effect, and

will also be applied to a CNPC violation.

Copyright by
DAEHEE LEE

1990

Dedicated to my mother, wife, and son

iv

TABLE OF CONTENTS
Introduction
Two Types of Move-Alpha and Chains
The Nature of Government, Proper Government,
and Barriers
2.1 The Definition and Configuration of
Government
2.2 Barriers
2.3 Proper Government
The Empty Category Principle
3.1 Subject, Object and Adjunct asymmetries
3.2 Argument and Nonargument Asymmetries
3.2.1 That-Trace Effect
3.2.2 Multi-WH Phrases
3.3 NP-Movement
Summary and Conclusion

Reference

13

13

21

26

37

38

53

54

58

65

68

A. Introduction

Chomsky (1981) introduces the GB (Government-Binding)-
frame in which, as a central idea, he explores the Empty
Category Principle (ECP) underlying the notion of (proper)
government as a local condition, and tries to explain what
we call the ResNIC phenomena with this principle. Since
then, many studies have been developing the ECP; many
alternative solutions have been suggested the ECP has been
extended to a variety of grammatical phenomena, and the
consequences have had a wide-ranged effect on the theory.
The theory has been extended with Chomsky's (1986)
introduction of the Barriers framework.

The goal or direction of this paper is to compare
certain variations of the definitions of government, proper
government, barriers, and the ECP, and to explore the
generalized definition of proper government and its
application to the phenomena of the ECP in GB-theory. In
particular,I will take a close look at Wh-movement and NP-
movement. In this section, I will describe how this paper
is organized, and pose some questions about the ECP which I
will hopefully answer in this paper.

To determine the distribution of empty categories and
to explain and generalize the RESNIC phenomena using the
notion of government, Chomsky (1981) introduces the Empty
Category Principle. He defines the ECP in terms of proper

government, which is a more restricted condition than the

2
government itself (See Chomsky 1981, p.273-275, Lasnik &

Saito 1984, p.240):

(A1) ECP: [we] must be properly governed
(A2) X properly governs Y if and only if X governs Y,
and i. X is a lexical governor e.g. N, A, V, and P, or

ii. X is coindexed with Y.

As we see, the ECP is defined in terms of proper
government and given the two restrictions (1) and (ii)
Proper government is defined in terms of government but in a
more constrained way than government. Therefore, the
definition of government is crucial to the definition of
proper government, which will justify empty categories in a
sentence. It is clearly important how we define government
and restrict proper government.

In section 2 of this paper, I will reanalyze the.
concept of government and proper government and redefine
them in a properly generalized way. In addition, I will
explore Chomsky's (1986) concept of barriers which will
play a keyrole in the definition of government and proper
government.

In section 2.1, I will argue against Aoun/Sportiche's
definition of government (Aoun and Sportiche (1983)) which
has been widely accepted in GB-theory, and show why we need

the concept of a barrier as introduced by Chomsky (1986).

3

In section 2.2 I will redefine the concept of
barrierhood, which will be a very important concept for
government. In defining a barrier, Chomsky (1986) excludes
IP=Infl" or S, and I'=Infl' from barrierhood by assuming
that they are 'defective categories'. I will eliminate this
stipulation from the definition of barrier and try to
generalize barrierhood. In addition, unlike Chomsky's
extensive application of barrierhood to movementJ,II‘will
regard a barrier as a condition on government, not movement.
(For movement, I will preserve Chomsky 1977's bounding nodes
for Subjacency Principle.)

In section 2.3, first I will show why we need proper
government to explain proper movement which leaves traces.
Second, I will argue against Lasnik/Saito's and Chomsky's
definitions of proper government. Based upon
Aoun/Sportiche's concept of government, Lasnik/Saito (1984)
defines the proper government inconsistently (Lasnik &

Saito, p. 248):

(A3) X antecedent-governs Y if
a. x and y are coindexed
b. x c-commands y

c. there is no 2 (z = NP or S') such that x

 

Chomsky (1986) suggests that barriers should apply to both
government and movement. See chap.4 and 6 for detail.

4
c-commands z and z dominates y, unless y is the

head of 2.

Even though the proper government is a restricted
notion, or a subset of the case of, government, Lasnik/Saito
define proper government in terms of lexical government,
which is defined with the concept of government, and in
terms of antecedent-government, which is defined using the
concept of C-command not with the concept of government.

On the other hand, Chomsky (1986) introduces theta-
government to the definition of proper government, but
theta-government, which is defined in terms of theta-marking
also shows some problems. (see P.31-32 for detail.)

To avoid the problems with these approaches, I will
introduce the notion of proper C-command to proper
government: whereas government is defined in terms of C-
command, proper government is defined in terms of proper C-
command.

In section 3 I will explore the ECP phenomena with the
basic concepts of government and proper government defined
in section 2. In particular, I will apply the redefined
concept of proper government to Wh-movement and NP-movement.

In section 3.1, I will show what will make what we may
call adjunct/obj asymmetries and subj/obj asymmetries. In
this section, especially, I will redefine the ECP in terms

of consistent proper government, which will be applied to

5
all the phenomena of the ECP. In section 3.2, I will explore
why the subject behaves differently from adjuncts, referred
to as argument/nonargument asymmetries. In particular, I
will introduce here the algorithm of NP-index assignment to
explain why relative clauses lack the that-trace effect, and

will also apply it to a CNPC violation like.
(A4) * What do you believe the fact that Mary saw t?

I will suggest that the ungrammaticality of (A4) be
related to the violation of CP-agreement through index
assignment rather than to Subjacency. In section 3.3 I will
clarify how the ECP is related to NP-movement. In section
1 I will discuss two types of movement which will be related
to the difference between Wh-movement and NP-movement, and

later to proper government.
1. TWO TYPES OF MOVE-ALPHA AND CHAINS

Before considering the nature of government and proper
government, let us consider the types of movement which are
relevant to the ECP. In this section, I will briefly
discuss two types of move-alpha and their properties. First
of all, I will describe how the types of move-alpha are
different, and second, how and why they make NP-movement and

Wh-movement different. In particular, I will describe in

6

more detail the properties and structure of adjunction, one
of move-alpha, which is relevant to the Wh-movement and ECP.

In general, move-alpha is a rule which maps deep
structure (OS) on to surface structure (SS), constrained by
such principles as theta-theory, case-theory, scope, etc.
In particular, theta-theory and Case theory affect this
mapping, whereas scope? is more relevant to the mapping
between SS and LF. In addition, Subjacency and the ECP
constrain movement, which leaves traces. Moved elements and
traces form chains and obey the binding principles at SS and
the scope principle3 at LF. Furthermore, movement applies
only to a head or maximal category, since the X'-1evel is
assumed to be invisible to movement.

Following Chomsky (1986), Lasnik & Uriagereka (1988)
and others, there are, in general, two types of movement,
substitution and adjunction. Substitution is category-to-
category movement and obeys the structure-preserving
principle, and adjunction is movement which adjoins a

category to a category and doesn't obey the structure;

 

zFollowing Chomsky (1981), May (1985), and others, Wh-
in-situ and quantifiers will undergo movement at LP to
order to take scope for semantic interpretation, which is
called LF-movement. In particular to determine wider or
narrower scope among quantifiers and Wh-phrases, May
suggests a scope principle. See footnote 3 for detail.

3May (1985)introduces the scope principle which will
determine the scope of quantifiers and Wh-words at LF, being
independent of the ECP. That is, when an LF-representation
contains quantifiers, They will be interpreted in a
relatively ordered way if they do not govern one another.

7

preserving principle. In addition, adjunction is
optional. Therefore, if possible, move-alpha obeys the
structure-preserving principle and carries out substitution;
however, in case substitution violates some principle and
adjunction is necessary to preserve it, adjunction will be
triggered.

Substitution and adjunction cause a moved element and
its traces to form a chain. Here we will indicate the way
that they form a chain.

Let us consider the following:

(1) a. [SI wonder [s.who [sMary expects [$.t' [sPro to
see t]]]]]
b. [sJohn seems [st" to be expected [st' to
be killed t]]]

If a chain is A-bound, then it is called an A-chain,

and if it is A'-bound, then it is called an A'-chain‘.

 

‘ According to Chomsky (1981 P.184-185),

I. Alpha is X-bound by betta iff alpha and betta
are coindexed, betta c-commands alpha, and
betta is in an X-position.

II. Alpha is a variable iff
(a) alpha = [ e]
(b) alpha is Th an A-position (hence bears an
A-GF)
(c) there is a betta that is locally A'-binds
alpha.

A-positions are phrase structure positions in which
arguments may appear at D-structure, and A'-positions are
phrase structure positions in which any arguments cannot
appear at D-structure.

See Riemsdijk & Williams's (1986) sections 15.1, 15.2,

8
In (1a) "who” moves to the specifiemj in 8' through the
specifier in the embedded S' by way of substitution; in (1b)
"John" also moves by way of substitution. Thus,
substitution forms A- or A'-chains.

On the other hand, adjunction movement forms only
A'-chains, since, following Chomsky (1986), adjunction is
possible only to a nonargument maximal projection. In other
words, adjunction is possible only to an A'-position.
Movement to the spec of S' can optionally allow adjunction
movement‘ without violating the binding principles.
Adjunction cannot apply to (1b), since the traces in (1b)
should be according to the binding theory, and adjunction
forms an A'-chain. Examples (2a) and (2b) illustrate
substitution and adjunction movement, of 'who' respectively,
in (1a):

(2) a. I wonder [$.who [st' ' ' " [sMary [wt' ' "

[wexpects [s.t"' [st" [sPro [wt' [wto

 

and 17.1 for detailed discussion.

5Chomsky (1986) suggests that there are two positions
within S' (which he calls CP): a specifier and head. This
results from the assumption that S' or CF and S or IP are
both a headed projection. C(omp) is a head of CP or S', and
I(nf1) is a head of S or IP. On this assumption, the
subject NP is to be a specifier of IP, and CP also has a
specifier at DS.

6Here, too, substitution or structure-preserving
movement takes place prior to adjunction movement. As we
will see in section 3.3, the unmarked trace in the Spec of
the CP is substituted for by another movement rather than
being adjoined to that Spec.

see 11111111111]
b. I wonder [$.who [sMary [wt'H [Wexpects [s.t"

[sPro [wt' [wt—O see 111111111]

The difference between the representation of (2a) and the
representation of (2b) at SS is that in (2a) "who" is
adjoined to S and in (2b) it is not. Chomsky (1986) assumes
that WH-phrases cannot be adjoined to S, since they have
clausal scope over 8. Here I will accept his assumption that
only (2b) is correct.

As a representation of (1), (2b) obeys the ECP, which I

will take up later7.

 

7 This analysis raises a problem if we assume
VP-adjunction as in (2.b.), then move-alpha will not violate
Subjacency:

(I) I wonder [$.who [sMary [ t"' [wbelieves ["Pthe

fact [ ' that [ ogn [ t' [wsaw t]]]]]]]]
In (I) there is only one boun ing n e NP between t"' and
t", and the sentence in (I) will not violate Subjacency:
nevertheless, it is ungrammatical. (Chomsky (1986)
suggests that a maximal projection with an oblique is a
barrier to government and movement: in (I) the complement of
a noun has an oblique case, and 8' comes to bar the
government of the trace in S', which is said to violate the
ECP. He suggests that generalization of barriers for
movement as well as government. However, I will propose
defining barriers in a different way to avoid the stipulated
statement of the barriers. See section 2.2.)

For this problem we will have three possibilities: one
is to modify adjunction in a more restricted way to prevent
a moved element from being adjoined to a VP: the second is
to reformulate Subjacency to predict the ungrammaticality of
(I): the third is to reanalyze constructions like (I). To
take the first of these three choices, let us consider the
following examples:

II. a * Which paper do [ you [ VPbelieve [s [a th0

sections of t] to be‘guwll of mistakes1P]]]?
b Which paper do [syou [wt' [Wbelieve ["tho
sections of t]]]]?

10
To sum up, I will constrain the categories which

allow adjunction (also see Chomsky 1986, p. 6):

(3) i. maximal projection8
ii. nonargument

iii. phoneticized category’.

 

If we give up adjunction to VP altogether, then both
(II.a) and (II.b) will violate Subjacency, since there are
two bounding nodes between 'which paper' and its original
trace. If we allow a trace t' of 'which paper' to be
adjoined to a VP, however, (II.b) will no longer violate
Subjacency for the reason that there is only one bounding
node NP between t' and the original trace, and also only one
bounding node S between 'which paper' and t'. On the other
hand, (II.a) is still ungrammatical, since there are two
bounding nodes S and NP between an adjoined trace t' and the
original trace.

Now we encounter a dilemma: if we allow adjunction to a
VP, we will wrongly predict that (I) would be grammatical,
even though we explain the grammaticality of (II.b); if we
do not allow adjunction to a VP, we will wrongly predict
that (II.b) would be ungrammatical, although we can explain
the ungrammaticality of (I). As we shall see just below and
later in section 3, adjunction to a VP will have a wide
explanation of grammaticality, together with government,
proper government, and the ECP. In addition, this
adjunction structure prevents Subjacency from incorrectly
ruling out (II.b). For the serious problem of (I) I will,
on the other hand, take the way of reanalysis. As we shall
see in section 3, compared with a relative clause and the
ECP phenomenon, (I) will seem to violate another principle
rather than Subjacency. (See section 3 for detail.)

9A3 we have seen above in (2), Chomsky (1986) proposes
that Wh-phrases should not be adjoined to IPs, even though
IPs are maximal projections.

9Chomsky (1986) suggests the first two conditions on
adjunction; however, (3.III.) will be necessary: that is,
an empty category cannot be adjoined to. See section 3.3
for detail.

11
Finally, with respect to adjunction, let us consider
phrase structure and the domination relation that adjunction
will generate, which will have an important effect on

government, proper government, and the ECP:

(4) [xp...x...[YPZ[YP...]]]

XP
/ \
eeeXeee YP
/\
Z YP

What category will dominate 'z' in the adjunction
structure of (4)? According to May (1985) and Chomsky
(1986), '2' is dominated by 'xp', not by 'yp'. However, it
also should be dominated by 'yp', in order to count as a

constituent of 'yp'. To see this point, consider the

following examples:

(5) a. Some student admires every professor, but John

doesn't.

b. [8 some studentz [8 e2 [VP every professors
[vp admires e3]]]], but John doesn't [vp every
professor-3L,p admire e31]

c. * [8 every professor3 [8 some student2 [8

e2 hm admires e3]]]], but John doesn't

[Vp admire e3]

12

Arguing that VP, and S, are among the positions to
which quantified phrases can be adjoined at LP in the
context of VP-deletion, Sag (1976), Williams (1977) and May
(1985) state that sentences like "Some student admires
every professor" are themselves ambiguous, but in a VP-
deletion context, as in (5a), the ambiguous interpretation
is impossible, and only a specific construal is available.
The reason for this, in Williams's approach, is that such a
construction is well-formed only when a VP that contains the
every-phrase as in (5b) is reconstructed, classify
considering deletion, and recoverability at LP. The problem
with (So) is that the second conjunct contains a free
variable, since the scope of every professor is limited to
the initial conjunct. No such problem arises in (5b),
however, since the VP-adjoined phrase is reconstructed along
with the rest of VP, all variables occurring in this '
conjunct are properly bound.

As this example shows, the quantifier adjoined to the
VP works as a constituent of the VP, even though it goes
through LF-movement and attaches itself to the VP. If the
quantifier is dominated only by S, as defined by May (1985)
and Chomsky (1986), it cannot be a constituent of the VP.
Therefore, it should be dominated by the VP in a sense, too.

Here I will give the technical definition of
domination which will be used in the definitions of

government and proper government.

13

(6) i. X maximally dominates Y iff all segments10
of X dominate Y.
ii. X minimally dominates Y iff any segment

of X dominates Y.

In (5) 'xp' and 'YP' both dominate or minimally
dominate '2': however, 'XP' maximally dominates 'z', and
'yp' does not. In the following sections I will show
the relationship between adjunction, government, proper
government, and

the ECP.

2. The Nature of Government, Proper Government, and

Barriers

2.1 The definition and Configuration of Government

In this subsection, I will first compare two important
theories of government: those of Aoun/Sportiche (1983) and
Chomsky (1986). To compare these two theories properly, I
will discuss the concept of a barrier which is implicit in
Aoun/Sportiche's theory of government, but explicit Chomsky

(1986), who uses different definition. Furthermore, I

 

w The concept of a segment is issued for adjunction
structures. Adjunction to a maximal category generates the
same node itself. In other words, it is separated as two

segments.

14
will show the reason that we need Chomsky's barriers rather
than Aoun/Sportiche's. This will be an introduction to the
concept of barrier for subsections 2.2 and 2.3, and section
3..

Aoun and Sportiche (1983) suggest the following
definition of government as a conclusion of their comparison
of previous definitions (e.g. Chomsky 1979, Chomsky 1980,
Freidin and Lasnik 1979, Rouveret and Vergnaud 1980) (Aoun

and Sportiche 1983, example (8)):

(7) X governs Y iff all maximal projections which
dominate X also dominate Y, and all maximal

projections which dominate Y also dominate X.

This definition implies that the governor governs the
governee only within the maximal projections which dominate
both of them, and that a governor does not govern any
categories across a maximal projection. It means that the
maximal projection is an "absolute" barrier to government.
Hence if we rewrite (7) in terms of barriers, we get the

following:

(8) i. X governs Y iff all maximal projections which
dominate X also dominate Y, and there is
no barrier for Y.

ii. All maximal projections are barriers to

15

government.

The definition of government here was very widely
accepted in GB-theory. However, some modifications of (8)

are necessary, since X'-theory has been further developed in

some aspects.

First of all, (8) should be modified to account for
government in a configuration which includes an adjunction

such as in (9):

(9) ["3...x ...wa [w,...y ..z..]]]
XP
’,/’ ‘\\\
X... YP
/\Y
W P
./ \\\\

..OYOOCOOOZOOOO

If we assume that yp is a maximal projection or barrier
in (9), and following (8) above, x cannot govern w, y or z,
and w also cannot govern y and 2. As I will show later,
however, in an adjunction structure like (9), x must be able
to govern w, but not y and z, and w must be able to govern
x, too". In addition, w must be able to govern 2, which is

dominated by yp.

 

, "May (1985) shows that in a structure of adjunction
like (9), X and W should govern each other to permit the
correct scope relations at LF. See chapter 3. by May
(1985). '

16
To solve these problems, let us redefine (8) in terms

of domination, which was defined in (6):

(10) i. X governs Y iff all maximal projections which
maximally dominate X also maximally dominate Y,
and there is no barrier for Y:

ii. All maximal projections are barriers.

Here in adjunction structure (9), yp is divided into
two parts or segments, and since only one segment of yp
dominates w, yp does not maximally dominate w. Therefore, w
is maximally dominated only by xp but not yp. Hence x can
govern w according to the definition in (10). However, the
problem still remains, because w cannot govern 2 if wyis not
maximally dominated by yp which maximally dominates 2, since
yp is a maximal projection to blocks government.

For this problem, we should find a way of allowing the
adjoined maximal category to govern a category which is
dominated by the adjoined category. For this I will define
another concept "exclusion" for government, which is

introduced by Chomsky 1986, (p. 9):

(11) X excludes Y iff X does not dominate Y at all.

Now we can define government again as follows:

17
(12) X governs Y iff All maximal projections which
maximally dominate X also maximally dominate
Y, and there is no barrier for Y which excludes

X.

Now, in an adjunction structure such as (9), w governs
2, since w and z are both maximally dominated by a maximal
projection xp and the maximal projection yp does not exclude
X, and hence yp does not block government of 2 by w.

Even though we could modify the definition of
government in (8) we would still have problems with the

category S. Let us consider the following:

(13) John believed [.him to love Mary].

In (13) "him" receives Case from "believed". In order
to do so the verb "believed" should govern "him". To allow
the verb to govern "him", '8' should not be a maximal
projection, as Chomsky (1981) proposes. That is, S' is the
maximal projection of S, and S is not a maximal projection.
Then, as defined in (8), "believed" can govern "him" and
assign Case to it.

However, Stowell (1981) and Pesetsky (1982) argue that
S is a projection of Infl. Chomsky (1986) proposes that S
is a maximal projection of Infl and that S' is a maximal

projection of Comp. Therefore, there are two possible

18

configurations for (13):

(14) A. ("Comp [FNP Infl [w ...]]]
a'. John believes [phim [Ito] [wlove Mary]] ->
IP-deletion
B. [0Comp [WNP [“Infl Lw ...]]]]
b'. John believes [whim [“to love Mary]] ->

CP-deletion

If we assume that the structure of (13) is (14a), then
the definition in (8) can work, since the verb "believed"
governs "him" across I' after IP-deletion. However, it does
not work for (14b), since the verb cannot govern "him"
across the maximal projection IP, even though the CP is
deleted.

Before choosing one structure of the two in (14), let

us consider an example adapted from Kayne (1985):

(15) a. * I consider [5(3Mikie being unhappy] to be
unpleasant]
b. I consider [8[8Pro being unhappy] to be

unpleasant]

If S is not a maximal projection, then "consider"
governs "Mikie", and (4a) should be grammatical: "consider"

could also govern PRO and render the sentence ungrammatical,

19
since PRO cannot be governed, as Chomsky (1981 and 1982)
argues. Hence we will accept the argument that S is the
maximal projection of Infl and that S' is the maximal
projection of Comp. Even so if, S is a maximal projection,
IP as termed by Chomsky (1986), it cannot be a barrier, even
though it is a maximal projection.

To solve this problem, Chomsky (1986) proposes that a
barrier be determined not absolutely but contextually.
Hence, the same category is said to be sometimes a barrier
and sometimes not a barrier. Chomsky (1986) defines
government in terms of a barrier which is defined
relatively. Chomsky's definition of government is actually
of the same form as in (12) which includes adjunction
structures by modifying (8). The only difference from (12)
is that "barrier" is defined relatively (Chomsky 1986, p.

14):

(16) X is a barrier for Y iff
i. X is the first maximal projection that
maximally dominates a BC for Y.
ii. X=BC or blocking category for Y except for IP
or

(17) X is a BC for Y iff

i. X maximally dominates Y and

ii. X is a maximal projection that is not L-marked.

(18) X L-marks Y iff (Chomsky 1986, p. 24 (47))

20
i. X is a lexical head (i.e. N, V, A, P,) and
ii. (a) Y is a maximal projection that
X theta-marks, or
(b) Y is a head of a maximal projection that X
theta-marks, or
(c) Y is a SPEC of a maximal projection that X
theta-marks if X=IP or CF (=SPEC-Head
Agreement).
(19) X theta-marks Y iff X and Y are sisters in X'-

theory. (Chomsky 1986, p. 14)

Returning to (13), the maximal projection IP is no
longer a barrier for "him", since the verb L—marks IP, and
the verb "believed" governs it and assigns Case to it. In
(15) the lower IP is a barrier, since it is not L-marked at
all and is a BC and barrier, following (16) through (18).
Hence the verb cannot govern "Mikie" and cannot assign Case
to it, either.

Now we can include both (13) and (15) by defining
barrier in a relative way, which differs from Aoun and
Sportiche, who consider a maximal projection as an absolute
barrier to government.

In sum, even though the definition of barrier in (16)
stipulates IP and I' and is not generalized properly, it
seems to be desirable that a barrier be defined in a

relative way, since Comp and Infl are categorial heads, and

21
consequently, CP and IP are maximal projections. In the
next section, we look into the characteristics and problems
of the definition of barrier in (16), and redefine it

properly.

2.2 Barriers

Here, I will indicate problems with Chomsky's
definition of barrier, redefine barrierhood, and show the
adequacy in of this new definition application to the ECP”.

As discussed so far, there are two ways to define
barriers to make government a local condition. One is to
define a barrier as absolute, implicitly proposed by Aoun
and Sportiche (1983), and the other is to define a barrier
as relative, proposed by Chomsky (1986). As we have seen,
Aoun and Sportiche's approach encounters a problem with
exceptional Case assignment etc. by proposing that maximal
projections are absolute barriers to government. In
addition, it has a great effect on ECP phenomena. For

example, consider the following:

(20) [wwho [wt saw Mary]]?

As we will see later in detail, here the trace of 'who'

 

12Chomsky (1986) extends the notion of barrier to
movement by generalizing the barrier to apply to bounding
nodes for movement. However, I will apply the definition of
barrier given here to government only.

 

22
which is posited in the subject position does not have a
lexical governor which properly governs it: hence it needs
to be antecedent-governed to avoid violating the ECP.
However, since IP is a maximal projection and blocks ,
government, 'who' cannot antecedent-govern its trace.
Nevertheless, (20) is still grammatical.

As seen in (16) through (18), Chomsky (1986) defines a
blocking category in terms of L-marking first, and in turn
defines a barrier in terms of blocking categories (with one
exception for IP). However, ultimately he does not solve
the basic problem with IP. When he defines a barrier, IP is
a defective category as a barrier and considered
exceptional. For example, in (20) IP is not L-marked and so
should have been a barrier. But if IP were a barrier, then
the trace in the subject also could not be properly
governed, and the result would be the same as Aoun and
Sportiche predict, as seen above. Therefore, Chomsky (1986)
treats IP as a defective category with respect to
barrierhood.

In addition, Chomsky (1986) adds the following
minimality condition” in order to include the X'-level as a

barrier (Chomsky 1986, p. 42):

 

3 According to Chomsky (1986 P.43), minimality
condition prevents 'saw' from governing 'tom', since N' is a
barrier by virtue of minimality condition:

I. They saw [wBill's [mpicture of Tom]]

23
(21) X' is a barrier for a category Y if the head X

governs (or head-governs) Y.

In (22), I' is a barrier to government under the
minimality condition, and (22) should be ungrammatical.

However, (22) is grammatical.
(22) (”Why do [Wyou [Pinfl Lwleave t]]]]

That is, I' is a barrier and this blocks government, 'why'
cannot govern its trace, and the example (22) should be
ungrammatical.

If Chomsky 1986 (p. 48) regards IP and I' as
exceptional or 'defective', his proposal would be very weak
and, besides, there is almost no difference between
Aoun/Sportiche's and Chomsky's. As we have seen so far, the
big problem is related to IP, which is regarded as a maximal
projection. That is, IP is problematic under the absolute
concept of a barrier: however, it still poses problem for
the relative concept of barrier which Chomsky 1986 defines
for government. In addition, another concept of a barrier
which he calls the 'minimality condition' also excludes I'
from being a barrier by regarding it as a defective
category.

Here I will propose another definition of barrier in

order to eliminate this ad hoc exception.

 

24

Before defining a barrier directly, I will reconsider
the concept of 'L-marking', in terms of which I will define
a barrier. Chomsky 1986 defines 'L-mark' as in (18),
regarding 'L-mark' as lexical marking. However, I will
define 'L-mark' as indicating lexically inserted items at
DS. Hence we can include the complementizer 'that', 'for',
etc. as an L-marking category. However, Infl is a bunch of
features, not a lexical item, hence it is excluded as an L-
marking category.

Furthermore, let us assume that L—marking assigns [+1]
to a category which is L-marked, and assigns [-l] to a
category which is not L-marked. In addition, let us assume
that a head has the same features as its maximal projection
through percolation. Then, L-marking can be defined as

follows:

(23) X L-marks Y iff
i. X is a lexically inserted head (N, V, A, P, or
Complementizer), and
ii. a projection of X immediately maximally

dominates Y.

Now we can define a barrier in terms of L-marking.

(24) X is a barrier for Y iff

i. X immediately dominates Y, and Y has [+1], or

25
ii. X is a maximal projection and dominates Y, and
Y has [-l], or
iii. X is a maximal projection of a lexically

inserted head Z and Z has {-1}, and X dominates

Y, and Z.

Let us look at how these three parts will work for the
following configuration: (I will discuss how the barriers

defined here will work in section 3. in detail.)

(25) [@Spec [othat [WNP (“Infl [WV NP]]]]]

CP
1”, C
spec '
tHKt‘\\\
a IP
N; \I'
/ \
Infl VP
9” \\\
V NP

First of all, VP is a barrier for NP‘ by(I), since V L-
marks NP' and VP immediately dominates it. The condition
(I.) in (24) has the same effect as the minimality condition
Chomsky 1986 defines. Second, IP is a barrier for VP: I
cannot l-mark VP, since AGR of I is a bunch of features, not
a lexically inserted head. Hence, VP with [-l] is
dominated by IP, which is thus a barrier for VP by (II).

Third, C' is a barrier for IP, since 'that' L-marks IP and

26
IP obtains [+1], and C' immediately dominates IP. If C is
empty, then CP is a barrier instead of C', since an empty C
cannot l-mark IP and IP is assigned {-1}, and CP dominates
IP as a maximal projection.

As we see, we have treated IP and I' properly as
inheritant, not inherent barriers by extending and
redefining L-marking. In the next section I will show how
the concept of barrier defined here will also work for the

ECP.

2.3 Proper Government

Chomsky 1981 and 1982 introduces a stronger condition
than government, called proper government, in order to
constrain the distribution of traces. In this subsection,
will first show why we need such a more constrained
condition on government for traces, and, especially, I will
focus on why and how we define antecedent-government since
it is an inconsistent and debatable component of proper
government. Second, I will argue against previous
definitions of proper government, and redefine it properly
before discussion of the ECP in section 3. 1

Let us consider the empty categories in the following:

(26) a. * [cpwhoi is it probable [CP[IPei to go to
the party every day]]]?

b. * [1,,Tomi is probable [C',[wei to go to the

27
party every day]]].
c. [witi is probable [c,,[",ej to go to the
party every day]]].
(27) a. [cpwhoi does Tom believe [wei to go to the
party every day]]?
b. [wTomi is believed [wei to go to the
party every day]].

0. * [",Tomi believes [we] to go to the party

every day]].

In (26) the empty categories cannot be indexically
related to 'who' or 'Tom', (or cannot have coreferential
interpretations with them), whereas the empty categories
in (27) have such semantic interpretations. In other words,
movement is not possible in (26), and conversely, it must
occur as in (27), except for (27c). Chomsky 1981 and 1982
observes that empty categories occur in complementary
distribution and distinguishes between them by describing
their contextual environments, defined in terms of
government.

For example, Chomsky 1982 (p. 34-36):

(28) trace condition: traces must be governed.

(29) Pro condition: PRO may not be governed.

The empty category (e.g. EC) in (26) is not governed by

28
'likely', since CP blocks government: in (27) it is governed
by 'believe' by virtue of CP-deletion (or S'-deletion“).
Hence, the EC in (26) is a PRO, and the EC in (27) is a
trace. However the trace condition is too weak to explain

such examples as:

(30) * [cpwhq do you think [cpthat “Fe, saw Mary]]]?

Here the trace of 'who' is governed by Infl, and
satisfies the trace condition in (28): nevertheless, (30) is
ungrammatical. Therefore, according to Chomsky 1981, a
stricter condition is necessary in order for the trace

condition to rule out sentences such as (30):

(31) a nonpronominal empty category (e.g. a trace) must
be properly governed.

(32) a. X properly governs Y iff X lexically governs Y.
b. X lexically governs Y iff X governs Y, and X=a

lexical head (e.g. N, V, A, and P).

 

“ Bar-reduction such as s'-deletion seems to be

determined by the property of a particular lexical entry in
the lexicon. In (26), bar-reduction does not occur for
'probable': however, it is possible for 'believe' in (27).

I. John is likely [wt to win].

II. * It is likely ["Pro to win].

III. It is probable [”Pro to win].

IV. * John is probable [Pt to win].

'likely' and 'probable' have different lexical
properties with respect to bar-reduction. That is, a bar-
reduction is determined in lexicon.

29

(30) can now be ruled out by defining proper government in
terms of a more restricted nation of government called
lexical government, since the trace of 'who' is governed
but not lexically governed because Infl is not a lexical
head. However, this move gives the wrong result for the

following case:

(33) [CPWhoi do you think [cp[,,,e, saw Mary]]]?

According to (32), (33) should be ungrammatical, since
the trace of 'who' is not lexically governed: the verb
'think' cannot govern the trace, because CP is a maximal
projection which blocks government. However, (33) is
completely grammatical. Therefore, condition (32) is too
strong. To solve this problem, Chomsky (1981) needs to
redefine proper government. Let us assume that there is
some relationship which we will call antecedent-government
between the Wh-word and the trace in a relevant sense to

which we will return later.

(34) X properly governs Y iff
i. x lexically governs Y, or
ii. X antecedent-governs Y.
(35) X antecedent-governs Y iff X is coindexed

with Y in a relevant sense.

30

Now, (33) can be represented as follows:

(36) [cpwhoi do ("you think [CPti [tPti Infl saw

MarYJJJI?

The original trace is antecedent-governed by the
intermediate trace in Comp: the intermediate trace is also
antecedent-governed (or might be lexically governed) in a
sense. Then, we can predict the grammaticality of (36).

Note that proper government is defined disjunctively in
terms of lexical government and antecedent-government.
Hence, at least one of the two conditions (lexical
government and antecedent-government) should be satisfied,
in order not to violate the ECP.

Since Chomsky 1981 introduced the Empty Category
Principle in GB-theory, many studies have attempted to
construct the proper formulation of the ECP and to extend
the scope of the ECP. The original definition of the ECP is
as follows (see Chomsky 1981, p. 273-275, Lasnik and Saito

1984, p. 240):

(37) Empty Category Principle: A nonpronominal empty
category must be properly governed.
(38) X properly governs Y iff:

i. X lexically governs Y, or

31
ii. X antecedent-governs Y.
(39) X lexically governs Y iff X governs Y, and X is a
lexical item (N, V, A, and P).
(40) X antecedent-governs Y iff X is coindexed with Y,

and X and Y have some relevant relation.

Here let us take a close look at the configuration of
antecedent-government which is expressed as 'some relevant
relation' in (40).

Lasnik and Saito 1984 (p. 240 and p. 248) define proper

government as follows:

(41) X properly governs y iff
i. x lexically governs y, and

ii. x antecedent-governs y.

Lasnik and Saito 1984 define lexical government in the same
way as defined in (39), according to the notion of
government defined in (8). However, they propose that
antecedent-government consists in the coindexing and C-
commanding relations, independent of the concept of the

government (Lasnik and Saito 1984, p. 248):

(42) X antecedent-governs Y iff X is coindexed with Y,
and X C-commands Y, and there are no barriers CP

and NP such that X C-commands the barrier, and the

32
barrier dominates Y, unless y is the head of the

barriers.15

In addition, they assume that all maximal projections
are barriers to government, and that among them only CP and
NP are barriers to antecedent-government. By avoiding the
generalization of lexical government and antecedent-
government, they fail to define proper government as a more
restricted subset of government.

On the other hand, Chomsky 1986 (pp. 17, 19) defines
proper government in a different way from Lasnik and Saito

(1984), as follows:

(43) x properly governs y iff
i. x theta-governs y, or
ii. x antecedent-governs y.
(44) X theta-governs Y iff X governs Y, and X
theta-marks Y.
(45) X antecedent-governs Y iff X governs Y, and x is

coindexed with Y.

Chomsky 1986 defines antecedent-government in terms of

 

15Here Lasnik and Saito (1984) follows Aoun/Sportiche's
concept of government: hence they implicitly use a concept
of an absolute barrier which is a maximal projection. Among
all maximal projections, especially, CP and IP are
implicitly regarded as an absolute barrier for the
antecedent-government.

33
government, not in terms of C-command, and so defines proper
government in a more consistent way than Lasnik/Saito (1984)
do. However, he proposes that proper government be defined
in terms of theta-government instead of lexical government:
theta-government is in turn defined in terms of theta-
marking, as in (19). At this time, however, Theta-
government for proper government seems to result in some
problems. According to Chomsky (1986), VPs can be theta-
marked by Infl, and theta-governed (Chomsky 1986, p.20,

example (36)):

(46) [1,,[vpfix the car],, [,Pwonder (”whether [,Phe

[M1111 11,1111]

Here the VP 'fix the car' is adjoined to IP through
movement, and leaves a trace behind. Then, the trace should
obey the ECP and be properly governed. Since it cannot be
antecedent-governed because of the Wh-island, it should be
theta-governed: here Infl can theta-govern the trace of VP,
for it is a sister of a VP, and theta-marks it. If VP is
theta-governed by an Infl, it would be correct, considering
(46): however, it is strange for an Infl to theta-mark a VP,
because it then assigns a theta-role to the VP, and a VP can
therefore be an argument.

Furthermore, this analysis results in a problem with

VP-adjunction. Following Chomsky (1986), as in (3),

34

adjunction movement is possible only through a nonargument.
However, VP-adjunction is obligatory in Chomsky's analysis
in order to allow antecedent-government, even though the VP
is theta-marked and is therefore an argument.

In addition, Chomsky (1986) defines the theta-marker as
a head category in X'-theory. Therefore, we could also
assume that C also theta-governs IP because it is a sister
of IP. If this is correct, the following should be

grammatical:

(47) * [whphe saw Mary]I [,PJohn wonder [cpwhether

till]

Now let us redefine proper government to avoid these
problems and generalize lexical and antecedent-government.
But before redefining proper government, let us consider the
concept of C-command, in which terms I will define proper
government. Differing from Reinhart (1983), Aoun and
Sportiche (1983) and May (1985) base the definition of C-
command on the notion of a maximal projection (Aoun &

Sportiche 1983, example (42):

(48) X C-commands Y iff all maximal projections which
maximally dominate X also maximally dominate Y,

and X itself does not dominate Y.

35
This concept is the same as M-command, which is used by
Chomsky (1986 P.9) when he defines government. Here I will

accept this definition of C-command and redefine government

in these terms:

(49) X governs Y iff
i. X is a head or a maximal projection, and

ii. X C-commands Y, and there is no barrier for Y

that excludes X.

Actually, this definition has the same effect as in
(12). If we assume that proper government is a more
restricted subset of government, then we can define proper

government in terms of proper C-command, which likewise is a

restricted case of C-command.

(50) X properly C-commands Y iff a projection which
immediately minimally dominates X also dominates
Y, and x itself does not dominate y.
(51) X properly governs Y iff
i. X properly head-governs Y, or,
ii. X properly antecedent-governs Y.

(52) X properly head-governs Y iff

36

i. X is a lexical head m or a coindexed head
with Y, and

ii. X properly C-commands Y, and there is
no barrier for y that excludes x.

(53) X properly antecedent-governs Y iff
i. X is coindexed with Y, and
ii. X properly C-commands Y, and there is no

barrier for y that excludes x.

Here we will be able to have all lexical heads as
proper governors as well as Infl, assuming that Infl has an
index in agreement with the NP in subject position, and
indexed categories (e.g. heads or maximal projections) for
antecedent-government. We will exclude C as a proper
governor (at least at SS), however: because C is assumed to
have no index". In addition, proper C-command will work
for the ECP in the same way as Chomsky's theta—government.

In the following sections, I will show how this concept

 

w A 'lexical head' here will be assumed to be a head
which bears the characteristics of feature assignment. On
this assumption, a noun, a verb and a preposition are
'lexical': a noun assigns its index to its complement, and
verbs and prepositions assign their cases to their
complements. However, adjectives like 'likely' are not
lexical, since they have no characteristics of feature
assignment in English. With this concept of a 'lexical'
head we consider Infl to be a lexical head, too, since it
has an index. See section 3.2 for index assignment.

17the complementizer can be a proper governor at LF,
since it can obtain an index by virtue of CP-agreement at
LF. See section 3. for details.

37

will work for the ECP and other linguistic phenomena.

3. The EMPTY CATEGORY PRINCIPLE

We have assumed the existence of the empty categories
pro, PRO, empty operators, and traces in the theory of
grammar. PRO, and empty operators are, in general,
generated in D-Structure, and have the pronominal feature.
However, traces are generated in S-Structure and have the
anaphoric feature. In addition, traces occur where they are
governable, unlike other empty categories. For this reason
Lasnik and Saito (1984, p. 240) defined the Empty Category

Principle as follows:

(54) The nonpronominal empty categories (e.g. traces)

must be preperly governed.

I will initially assume that this definition is correct
and apply it in conjunction with the notions of the -
government18 proper government and barriers ,which we have
defined before. Furthermore, I will show how adjunction and
substitution will be incorporated into the ECP. In addition
to these definitions, for the ECP I will accept

Lasnik/Saito's 1984 (p. 257) gamma-feature assignment and

 

'3 I will later argue against (54), and substitute

consistent proper government for it.

38

Chomsky's 1986 (p. 24-26) CP-Spec-head agreement. (See
section 3.1 for details.) Following Lasnik and Saito
(1984), traces must be assigned +Gamma when they are
properly governed, and this gamma-feature which is assigned
to a trace is maintained at all levels of the grammar.
Furthermore, Gamma-marking occurs at different levels for
the traces of arguments and those of nonarguments: that is,
the trace of an argument must be g—marked at SS, and the
trace of a non argument must be g-marked at LF.

Lasnik and Saito 1984 assume the following filters at

LF:
(55) * t if it has {-9}

Then, the ECP predicts the grammaticality of a sentence
by checking whether traces are assigned [+g] or {-9}.
Using such assumptions I will start to explore the ECP
phenomena.

3.1 Subject, Object and Adjunct asymmetries

In this subsection, I will show how and why Wh-traces
show different properties according to their positions in a
sentence. First of all, I will explain the difference
between Wh-traces of adjuncts and objects, and further make
clear the difference between Wh-traces of subjects and
objects. For subj/obj asymmetries, I will argue against the

ECP as defined in (54), and redefine in terms of consistent

39
proper government. In particular, I will show how CP-
agreement will work to explain weak and strong violations,
rejecting Lasnik/Saito (1984)'s rule of Affect-alpha.

Huang (1982) and Lasnik and Saito (1984) using the ECP,
try to extensively explain the asymmetries which objects and
adjuncts show.

First of all, let us consider the different ways in
which adjuncts and objects will undergo movement, which will
show the properties which are relevant to proper government.

Consider two representations of the following example:

(56) a. [cpwho did [,Pyou [wsee t]]]
b. [cpwho did [,,,you [wt'[vpsee t]]]]

In (56a) the trace t is properly head-governed by the verb
"see", and obeys the ECP. Here two movements are
possible”. However, in (56b), trace adjoined to VP is
redundant, since the representation does not violate any
principle without the adjoined trace. However, let us now

consider adjunct movement, as in (57):

(57) a. [cpwhy did [,Pyou [wsee John]]]

b. [wwhy did (”you [wt' stee John t]]]]

 

‘wLater in a revised structure representation of the
Wh-movement of objects I will eliminate (56.b) for the
adjunction of the Wh-movement of objects, which will violate
the ECP.

40

Here the adjunct "why" cannot be properly governed by
any lexical or indexed head, since the verb 'see' does not
properly C-command "why": the structure of (57) at US is

represented as follows:

(58) [w[v.see john] why]

Therefore, the original trace of 'why' in (57) should be
properly antecedent-governed: for this, 'why' should undergo
adjunction to the VP. Unless there is a trace t' adjoined
to VP as in (57b), the trace t cannot be properly
antecedent-governed nor be assigned [+g], since IP dominates
the VP with [-l], and is a barrier for VP. Therefore, "why"
properly governs the elements within IP, but not within VP.

If there is an adjoined trace t', it can properly C-
command VP and also antecedent-govern and [+g]-mark the
trace t. Hence, adjunction is obligatory in case of adjunct
movement to avoid violating the ECP.

Under these structural assumptions, let us now
consider asymmetries of objects and adjuncts in multiple WH-

movement:

(59) a. [cpwhy did [lpyou [Wt [w[v.see what] t]]]]
-> SS

b. * [wwhat did [wyou [wlwsee t] why]]]?

41
In (59a), the noncomplement Wh-phrase moves to the CP,
rather than the complement Wh-phrase: otherwise, the
sentence is ungrammatical as in (59b), since its trace
cannot be properly governed. (59a, b) can be represented at

LF as (603, b), respectively;

(60) a'. [cp[spwhat, [spwhy,]], did [,Pyou [VPtI
[vp[v.see t,] 13,111]
C
// \\\‘\\‘
SP IP
9”], {\ /// \\
what, SP, NP '

\ \ /I\
VP

why, you I
t

. /’ ‘\\\
did t, VP
,/ \\\ i
V' PP
/ \ I
v NP t,
l I
see tJ
b'.* [cp[spwhy, [5,,what,]].l did [,Pyou
[th [v.[v.see t,] 11,1111
CP '
SP//’ ‘\\\‘IP
h/ 13.. ..p/ \I
w y '
I '1 ‘ /,‘\\\
what, you I VP
I /’ ‘\‘\

dld t, VP

v/ \p.
/ \ l

v NP t,

42

If we accept the assumption that all wh-phrases move to
the CP at LF in (60a), in order to represent scope
relations, then "what" is adjoined to the Spec at LF, since
'why' is already posited in the Spec at 88: here the trace
of 'what' is properly governed by the verb, and the trace of
'why' is properly governed by Spec which is coindexed with
'why' through percolation. On the other hand, in (60b) the
trace of 'why' cannot be properly governed by 'why' by
adjoining itself to the Spec at LF: in this case Spec is
branched, and not coindexed with 'why', and 'why' cannot
properly C-command and so properly govern its trace, and
hence will violate the ECP.

If we do not interpret the concept of domination as in
(7), we will predict that (60b) is grammatical, since,
following the previous definition of domination (Chomsky
(1986) and May (1985), the CP is a maximal projection, and
'why' governs its trace and antecedent-governs its trace.

Let us turn now to the following:

(61) a. [wWhat did ["you say [mt' that [whe saw

1111]]
b. [cpwhy did [,Pyou [wsay [cpt" [c.that
[,PJohn [wt' [w[v.1eft]t]]]]

In (61a) t is properly governed by the verb "saw", and

is [+g]-marked, and the intermediate trace in the CP is also

43

properly governed by 'say' and [+g]-markedah

However, there are some problems. In (61b) the trace of
the adjunct cannot be properly head-governed since it is a
noncomplement. Therefore it must be properly
antecedent-governed. The trace t is antecedent-governed by
t' which is adjoined to VP. Even though t" is properly
head-governed by 'say', however, the t' adjoined to VP
itself cannot be properly antecedent-governed by t" in CP,
since CP has a head, "that", and C' blocks government by
virtue of (241). To deal with this problem, Lasnik and
Saito (1984, sec. 3.2) assume that there is deletion of
nonsemantic elements at LF which they call affect-alpha. In
addition, the traces of arguments must be g-marked at
S-Structure, and the traces of nonarguments must be g-marked
at LF. Then, the complementizer "that" can be deleted at

LF, if we assume that a complementizer is a nonsemantic

 

1m According to Lasnik and Saito (1984) and Chomsky
(1986), the trace t' in CP cannot be lexically goVerned
(or theta-governed) by the verb "say". Hence it should be
antecedent-governed. Following Lasnik and Saito (1984), the
trace in CP is antecedent-governed by the WH-phrase in CF
of the matrix sentence.

Chomsky (1986) assumes that the trace adjoined to VP
antecedent-governs the trace within CP in the following
example:

I [c l,what did [, Pyou [wt" [wsay [cpt' that

[fine saw tmm
However, let us assume here that if the CP is not a barrier
the elements within the CP are accessible to a proper
governor, as it is accessible to a governor.

44
element: t" in C? can properly antecedent-govern t' at LF,
and [+g]-marks it, because adjuncts are not arguments.

If we can solve such problems without such an arbitrary
deletion rule at LP”, it will surely be a better solution.
First of all, I will also assume that the traces of
arguments like subject or object are g-marked at S-Structure
and that the traces of nonarguments are g-marked at LF,
following Lasnik and Saito (1984fn. Furthermore, I will
assume that there are spec-head agreements called
CP-spec-head agreement and IP-spec-head agreement, as
Chomsky (1986) has proposed. Just as Infl agrees with the
NP spec of IP, and shares some feature with it, the head of
CP agrees with the spec of CP and shares some feature with
it, including the index. Furthermore, I will assume that
IP-Spec-head agreement occurs at S-structure and that

CP-Spec-Head agreement occurs at LF.

 

21Hubert Haider (1986) discusses the problems of
Affect-alpha that Lasnik and Saito (1984) suggested.

:nNote that a trace in an argument position should be
g-marked at SS, not an argument chain and that a trace in a
nonargument position should be g-marked at LF, not a
nonargument. For example:

I [cpwho do [,Pyou [wt" [wthink [cpt' [wt saw

MarYJJJJJJ

The original trace t of 'who' should be g-marked at SS,
since it is in an argument position; however, the traces t'
in the embedded CP and t" adjoined to the VP should be g-
marked at LF, since they are both in nonargument positions.

45
Then, t" and "that" in (61b) can have the same index,
and the head of CP ”that" can properly antecedent-govern the
trace t' adjoined to VP at LF. If we assume spec-head
agreement also at LF, then we don't have to assume Lasnik
and Saito's deletion rule at LF.

CP-Spec-Head agreement is relevant in other cases, too.

(62) a. ? [cpwhat do [,Pyou wonder [cpt' [c.whether [,Phe
saw t]]]]]
b. * [cpwhy do [,Pyou [wt"' [wwonder [cpt"

[vwhether ("he [wt' [wleft t]]]]]]]]]

We have assumed that there is spec position in CP and
that WH-phrases move to the spec position. Then,
"whether" is the head of CP as a complementizer, and
therefore the spec position is empty and "who" or "why" can
move through it. In (62.a) t is properly head-governed,
and t' is also properly head-governed by "wonder". They
do not violate the ECP. Furthermore, WH-movement in this
case does not violate Subjacency. The selectional
restriction of the verb is also satisfied, since 'wonder' is
subcategorizes for a CP with the feature [+Wh], and the CP
in (62) has [+Wh] by virtue of sharing some feature with its
head 'whether'. Nevertheless, is ungrammatical. However,
if we assume CP—spec-head agreement at LF, then we can

easily solve this problem. Under spec-head agreement some

46

feature is shared. In the case of (62.a) the trace t' in
spec agrees with the head ”whether”. Following Lasnik and
Saito (1984), however, the trace has the [-WH] feature”.
In (62), then, CP-spec-head agreement will result in feature
conflict, since the trace in spec has the [-WH] feature and
the head "whether" has the [+WH] feature. Therefore,
sentence (62.a) violates spec-head agreement at LF and is
ungrammatical, even though the trace in CP is properly head-
governed and the sentence satisfies the ECP.

In the case of (62b) there is no spec-head agreement at
LF, and ”whether" cannot share its index with the trace t".
Therefore, t' cannot be antecedent-governed. For this
reason (62.b) violates the ECP and spec—head agreement at
LF. Spec-head agreement, along with the ECP, also explains
weak violations like (62.a) and strong violations like
(62.b).

Let us consider additional examples of weak violations

and strong violations:

(63.) a. ? [[mwhat do [Wyou wonder [mwho saw t]]]

 

23'Here is an example to show that a trace has [-
Wh] feature:
I. a. [wwhat do [Wyou wonder [at' [whe saw t]]]]
b. [moo ("you wonder [flyhat [whe saw t]]]]

The verb "wonder" should have [+WH] as its complement
because of the selectional restriction. (I.b) satisfies
this selectional restriction and is grammatical: however,
(I.a) is ungrammatical, since the trace in CP has [-WH] and
does not satisfy the selectional restriction.

47
b. * [cpwhy do [,Pyou [th [wwonder [cpwho [wt'

[vpleft 11111111]

In these examples, the WH-phrases cannot move through
the embedded spec in CP, since "who" occupies the spec
position. In (63.a) the trace t properly head-governed by
the verb, and does not violate the ECP. It violates only
Subjacency. However, in (63.b) t" cannot properly
antecedent-govern t', and it violates the ECP, and in
addition also violates Subjacency. Therefore sentence
(63.b) is worse than (63.a). As we have seen,
object-adjunct asymmetries are easily explained within the
ECP if we assume adjunction, a level of g-marking, and
spec-head agreements. In particular, spec-head agreement at
LF makes it possible for the elements in the embedded CF
to be lexically governed by the higher verb. In earlier
works, if the elements in CF are properly head-governed,
there is ungrammaticality. However, we can eliminate such
predictions of ungrammaticality by means of spec-head:
agreements as in (62).

Now that we have a problem with selectional
restrictions by assuming CP-agreement, let us take a close
look at selectional restrictions at SS, which will trigger
syntactic Wh-movement at SS in English. English-like
languages have syntactic Wh-movement to satisfy the

selectional restriction as in (63): otherwise such an

48
example would be ungrammatical, * "I wonder John saw what".
If we assume CP-Spec agreement at LF, then sentences like
(63) do not satisfy the selectional restriction: the CP in
(63) has an empty head and cannot have the feature [+Wh] at
SS, even though 'who' occupies the Spec. We cannot assume
either that CP-Spec agreement is possible at SS: To address
such problems, I will assume the following:
(64) a category with [+Wh] must contain a [+Wh]-feature
at SS.
(65) X contains Y iff X maximally dominates Y, and
there is no Z such that X dominates Z and Z
dominates Y.
Now we can predict the grammaticality in (63) by virtue
of (64) and (65).
Like adjuncts, the subjects of tensed clauses show

asymmetries with objects, as follows:

(66) a. ?? [mwhat do [Wyou wonder [at whether
[wJohn saw t]]]?
b. * [mwho do [Wyou wonder [wt [vwhether
[wt saw John]]]]?
(67) a. [wwhat do [Wyou say [wt that (”John saw
t1]]?
b. * [”who do [flyou say [wt [othat
[wt saw John]]]]?

c. [thy do [Wyou say [mt [vthat ("John Lw[t

49

mmumeinnun

These phenomena seem to result from the ECP, because
the subjects, like adjuncts, have no indexed or lexical head
to properly govern them: therefore they need to be properly
antecedent-governed to avoid violating the ECP.

As we saw earlier in example (62), the sentence in
(66.b.) is ungrammatical because it violates the ECP:
regardless of whether 'who' moves to the CP directly or
through the CP of the embedded clause, neither 'who' nor
its intermediate trace can properly govern the original
trace. Since the trace of 'what' is properly head-governed
by the verb 'saw', however, (66.a.) violates only Subjacency
or CP-Spec agreement, regardless of the ECP: it is a weak
violation as in example (62.a).

Returning to subj/obj asymmetries, now the fact that
the subject is antecedent-governed for the ECP will show

superiority as the adjunct does:

(68) a. [cpwho [,Pt saw what]]?

b. * [mwhat does [wwho see t]]?

Here 'who' is adjoined to the spec at LP in the case of
(68b), and does not properly C-command or properly govern
its trace: it violates the ECP. The LF-representations are

as follows:

50

(69) a'. [,:,,[s,,what.l [spwhod], [,Pt, saw t,]]

b'. * [cp[s,,who, [5,,what,]]J [,Pt, saw t,]]
Let us consider the following carefully, however:

(70) a. [cpwho did [,Pyou say [cpt' [,Pt [wsaw
Whatlllll (53)
b. * [cpwhat did [,Pyou say [c,,[,,,who saw t]]]] (SS)
c. * [c,,[s,,whoi [spwhatJJ did [,Pyou say

[cpt'i [[pti sawllll (LF)

In (70.a.) the trace t is properly antecedent-governed
by t', and t' is properly head-governed by "say": thus
(70.a) satisfies the ECP. On the other hand, 'who' (70.b)
is a Wh-in-situ at SS: therefore, at LF it undergoes LF-
movement, and generates structure (70.c.) at LF. Like in
(70.a), in (70.c.) the trace t.i is properly
antecedent-governed by the trace t', which is properly head-
governed by 'say' at LF: thus the sentence (70.b.) does not
violate any principle. Nevertheless, it is ungrammatical.

To solve this problem, we will redefine the ECP:

(71) Consistent Proper Government

i. all the traces in a chain must be properly-

governed, or

51
ii. all the traces in a chain must be properly

antecedent-governed.

That is, in (70.c.) the trace t is properly
antecedent-governed by t'. Therefore, the trace t' should
also be antecedent-governed because t' forms a chain with t
and 'who'. Even though the structure of (70.b.) is changed
as (72) at LF, it does not satisfy (71) for the ECP, and we

can predict the ungrammaticality of (70.b):

(72) [cp[s,,who, [spwhatfl] do [,Pyou [wt, [wsay [cpt,
[,pt [wsaw t,]]]]]]]]]

In (72), the trace adjoined to the VP of the matrix
clause cannot be properly antecedent-governed by 'who', even
though the intermediate trace in the embedded CP is properly
antecedent-governed by it.

By modifying the ECP as (71), we should also modify the
representations we have discussed before. First of all,

(70.a) should be changed as (70.a') to satisfy the ECP in
(71):

(70.a') [cpwho did [,Pyou “Pt" [wsay [cpt' [,Pt [wsaw
Whatlllllll

Second, let us consider how complement Wh-movement

52

works with (71).

(73) a. What did you see?
b. * [cpwhat did [,Pyou [wt [wsee t]]]]
c. [cpwhat did [,Pyou see t]]

Following the above analysis, the representation of
(73.b.) is ill-formed, since the chain [what, t', t] is not
consistently properly governed: t is lexically governed by
the verb 'see', and t' is antecedent-governed by 'what'.
Hence (73.c.) is assumed to be an acceptable representation.

Returning to (61.b.), the chain [why, t", t, t] is not
consistently properly governed, because t" is properly
head-governed by the verb 'say'. To avoid inconsistent

proper government, we should represent (61.b.) as follows:

(74) [cpwhy did [,Pyou [th' [wsay [cpt" [c.that [,Phe
[wt' [vp[vrleft] t]]]]lllll

Thus, the lack of a head-governor makes the subject
behave like an adjunct. If the subject is properly governed
by a head as in an exceptional case-marking structure,

however, it behaves in the same way as the object:

(75) 7? who do you wonder whether John expects t to see

Mary?

53

Here (75) does not violate the ECP: rather it violates
Subjacency or CP-Spec agreement. The original trace of
'who' is properly head-governed by 'expects'. Therefore
(75) is a weak violation.

Even though the adjunct and subject are both properly
antecedent-governed, the adjunct does not show the that-
trace effect. I will refer to these asymmetries as argument
and nonargument asymmetries, following Lasnik/Saito (1984).
In the next section I will discuss argument/nonargument

asymmetries in detail.

3.2 Argument and Nonargument Asymmetries

In this section, I will present the phenomena which are
related to the Wh-trace of subjects, and will show why the
wh-traces of subjects behave differently from adjuncts and
objects. First of all, I will explain why the that-trace
effect occurs only with the Wh-traces of subjects, and why
subject relative clauses do not show the that-trace effect.
In particular, I will introduce "unmarked trace
substitution" for multiple wh-traces of subjects and
adjuncts.

3.2.1 that-trace effectu’

 

2"Chomsky and Lasnik (1977) suggest that structure
[ ... that [wt ...]] should be ungrammatical, like (67.b):
1% is called that-trace effect.

54
Here we will consider why subjects show the that-trace
effect but objects and adjuncts do not. Furthermore, we
will consider why relative clauses show no that-trace
effects, even though their structure is similar to the
structure in which the subject WH-phrase moves to the spec

in CP as in (76):

(76) the dog [athat [wt bites John]]

Subjects have the properties of both Objects and
Adjuncts. In particular subjects are arguments, like
objects, and have no head-governor, like adjuncts. These
properties will make subjects behave differently from
objects and adjuncts.

Let us consider the following:

(77) a. * [wwho did [,Pyou [th [wsay [cpt' [c.that
[wt saw him]]]]]]]
b. [cpwhy did [,Pyou [wt"' [wsay [cpt" [c.that

[,PJohn [wt' [wsaw him ]]]]]]]]]t

Here (77a) involves movement of a subject ([...that
[t...]]), and (77b) movement of an adjunct. Why does the
Subject show the that-trace effect but not the adjunct?
This results from the fact that it is an argument. In

(77.b.) the trace t' is properly antecedent-governed at LF

55
after CP-spec-head agreement because "why" is a nonargument.
However, in (77.a.) t must be antecedent-governed at
S-structure, since it is an argument. At S-structure
CP-spec-head agreement does not occur and "that" cannot bear
the same index as the spec t', and cannot antecedent-govern
the trace t at S-structure. Therefore, t receives
[-g]-feature at S-structure, and this feature is maintained
at LF. Even though the head of CP agrees with and shares
its index with its spec at LF, the feature [-g] cannot be
changed, and the sentence is ungrammatical. However, 'why'
is a nonargument and hence CP-agreement makes 'that'
properly antecedent-govern the adjoined trace, and [+g]-mark
it at Ll"25 .

Along with CP-Spec-head agreement, the distinction
between the argument/nonargument g-marking levels
contributes to an explanation of a variety of ECP phenomena.
In addition, they contribute to presenting the properties of
subjects, objects, and adjuncts. These properties may be

summarized in the following table:

 

25In addition, we can explain the following examples
I.a [cpwho do [,Pyou [wt [Wthink [:9 t [c.that [,P he
[wt [ says [c.t [,pt saw Mary]]]im
b.* [new 0 do [,Pyou [wt [wthink [cpt [,Phe [wt
[..says [c.t [c.that [,t saw Maryiimmi
Here in (I.a.) the origina trace is [+g]-marked by the
intermediate trace in the lowest CP. However, other traces
are posited in nonargument position, and are g-marked at LF.
Hence this example does not show the that-trace effect.
However, in (I.b.) the original trace is [-g]-marked at
sS because of the existence of 'that', as we expected.

56

(78) Argument Complement G-marking level
Subj + - SS
Obj + + $8
Adjunct - - LF

Now let us consider why the relative clause in (76)
does not show the that trace effect. Chomsky 1986 (p. 27)
assumes that CP-spec-head agreement at S-structure enables
the head "that' to share an index with the empty operator in
Spec. Then, "that" can antecedent-govern the trace t,.
However how can we explain (77.a.), following Chomsky's
suggestion that CP-agreement occurs at LF? If the
CP-spec-head agreement occurs at S-structure, then t' and
"that" will have the same index at S-structure and "that"
will antecedent-govern t. This is the wrong result, as we
have seen so far.

To address this problem, I propose the following

feature assignment:

(79) X assigns its features to Y only once iff x

governs Y, and X is the nearest governor.

For example, in (80.a) the verb 'believe' governs the
IP and the NP within the IP; however, it assigns Case to the

IP, not to the NP subject.

57
(80) a. John believe [whim [“to [wlove Mary]]]
b. John consider [,PHer [,‘Inflo (”intelligentn]

If we assume that a maximal projection shares its
features with its head, and that IP—Spec agreement is
possible at SS, we can explain naturally how the NP can
have accusative Case. That is, the IP is assigned
accusative Case by the verb 'believe', and Infl can be
shared with its maximal projection IP through percolation.
Therefore Infl receives accusative Case from IP.
Furthermore, the NP in the Spec of IP will also have
accusative Case through IP-Spec-head agreement at 88.. Hence
it will not violate Case theory.

Returning to subject relative clauses, in (81a) the NP
governs the CP, and assigns its features (e.g. an index) to
it“.

(81) a. [upthe dog, [$0, [c.that, [,pt, bites John]]]]

b. * [upthe dog, [,:,,0.‘ [c.that, [1th bite John]]]]

Then, the head 'that' can have the same index as the NP
through percolation. Then, 'that' properly governs the

trace of the NP within the IP, and will satisfy the ECP.

 

2“’Let us assume that a verb has a case and assigns it
to its complement, and at the same time that a noun
originally has an index and assigns it to its complement.
Furthermore, the assigned feature i is assumed to be not
assigned to more than one category.

58
The empty operator and the NP should be coindexed (maybe at
08 or at SS): otherwise, the sentence would be ungrammatical
as in (81.b.).

As we have seen so far, CP-Spec-Head agreement should
be available only at LF for the ECP, and IP-Spec-agreement
should be available at least at 88 for Case theory. Also,
the distinctive levels at which g-marking is relevant are
also a good device to explain ECP-phenomena. }

In the next section, I will be going further for more
evidence of argument/nonargument asymmetries in the

structures of multi-Wh-phrases.

3.2.2 Multi-Wh Phrases

In the previous sections, I discussed subject/object
and adjunct/object asymmetries, which may be generalized as
Complement/noncomplement asymmetries, as in Huang (1982).
Among these phenomena, multi-Wh constructions formed by a
subject/object or adjunct/object pair have shown that the
subject or the adjunct moves prior to the object or
complement: we might refer to these as superiority effects,
as termed by Chomsky. Such asymmetries are determined by
whether a category is properly head-governed or properly
antecedent-governed. Furthermore, the distinction between
g-marking levels has been shown to have an effect on the
that-trace effect called subj/adjunct asymmetries or

argument/nonargument asymmetries.

59
First, let us look closely at the multi-Wh
constructions which consist only of properly antecedent-'

governed Wh-phrases.

(82) Who thinks that who saw Mary?

Now, we might doubt the grammaticality of (82): the
subjects 'who' in the main clause and in the embedded
sentence might be thought to be properly antecedent-
governed, as we have seen so far. Then, it should have been
ungrammatical, but actually, it is grammatical. In
addition, (82) does not show the that-trace effect.

However, we can explain the grammaticality of (82) by
virtue of consistent proper government as in (71), CP-
agreement at LF and indexed head-government. We have
assumed that the Wh-in-situ moves to the Spec of the CP at
LF for scope, in which case its target position should
contain [+Wh]. The structure of (82) could be represented
at LF in (83).

Among these representations at LF, only (83.a.) is
well-formed and obeys the ECP. That is, (83.d.) violates
the Extended Projection Principle, by Chomsky (1981)
proposed: that is, the verb 'think' subcategorizes not for a
[+Wh] but for a [-Wh] feature for CP. (83.b.) and (83.c.)

do not satisfy the ECP; however, (83a) does.

60

(83) a. [”[spwho, [spwho,]], [1th [,.Infl [Wthink

rcpt. [c.tha'ci It, saw Mary]]]im

/CP\
Spec] C'
/ \
who, specj 0 IP
I / \
who, NP, I'
l / \
tj I VP
I / \
Infl V CP\
think Spec. C'
I ' / \
t, c IP
' A.

that, t, saw Mary

b. [cp[s,,who, [spwho,]], [,Pt, [,.Infl “Pt, [vpthink

[cpti [crthati [Ipti 53W MarYJJJJJJJJ

CP
Spec, C'
/ \ /\
who, Specj 0 IP
I / \
who NP. 1'
J J
4 / \
I VP
J / \
Infl t, VP
/ \
V CP
thl k S/ \C'
an pec,
I / \
t. C IP

I

that, t, saw Mary

61

c. [cp[s,,who, [spwho,]], [,,,tj think that [,Pt, saw

MarYJJJ
CP
/ \
Spec, IP
z” \\ 2522;:7“-
who, Specj t , thank CP
/’ \\\\
who . C IP

.I

I
that t, saw Mary

d. [cp[who, [,,,tj thinks [cpwho, [c.that, [,Pt, saw

Mary]]]]]]
CP

’//
whoj Spec
A.
t, th1nks IP\
who, CP
that, C'

t, saw Mary IP

In (83a) the head 'that' of the embedded CP has, by
means of CP-agreement, the same index as its Spec, which
contains the trace of 'who'. Since the head 'that' can
head-govern the original trace of 'who,'. To preserve
consistent proper government, it moves directly to the Spec
without being adjoined to the main VP. The trace ti in the
embedded CP is also properly head-governed by the verb
'think', and therefore all the traces in the chain of 'whoi'

are consistently properly head-governed and satisfy the ECP.

62
However, in (83b.) the trace adjoined to the VP cannot be
properly governed: the original trace in (83c), too, is not
properly governed at all.

Compared with (82), (84) is clearly ungrammatical:

(84) * who does who think that saw Mary?

The structure at SS is as follows:

(85) * (”who, [,,,who [wt, [wthinks [cpt, [c.that [,Pt,

saw Mary]]]]]]]

Here the original trace.tq of 'whoi' is [-g]-marked at SS
since the existence of the head 'that' of the CP bars proper
government. Thus, the trace with [-g] violates the ECP,
even though the head 'that' agrees with the spec and obtains
an index. In addition, at LF, the Wh-in-situ does not
satisfy the ECP either, since the adjoined 'who' through LF-
movement cannot properly antecedent-govern the original

trace:

(86) * [,:,,[spwhoj [spwho,]] [,,,tj (wt, [Wthinks ...]]]]

(LP)

Now we still have another problem with this analysis.

Let us consider the following:

63

(87) a. * How do you think that who saw Mary?

b. * Who do you think that saw mary how?

(87b) is ungrammatical, since in S-structure (88) of
(87b) the original trace of 'who,' is [-g]-marked at SS
(e.g. a that-trace effect), and will violate the ECP at LF,
whatever the LF-representation of (87b) is after 'how'

moves at LF.

(88) [cpwho, do [,Pyou [wt, [wthink [cpt, [c.that [t,

[vp[vrsaw Mary] hOWJJJJJJJ (53)

Even though there is no complementizer 'that' in (87b),
(87b) is ungrammatical, since the trace of 'how is not
properly governed at all.

However, (87a) has the following structure at SS:

(89) [whow do [,,you [wtr [Wthink [cpt, [c.that [,Pwho

[vpt] [vp[vlsaw Mary] ttlllllllll

Here the traces of 'how' is not yet g-marked at 88,
since it is an adjunct: it is [g]-marked at LF. After the
LF-movement of Wh-in-situ 'who', it has the following
possible representations at LF:

(90) a. [cp[s,,who, [sphow,]]], [,Pyou [Wt, [Wthink

64
(mt, [c.that, [,Pt, [wt, [w[v.saw Mary]
t11111111]
b. [,:,,[s,,who,l [sphowd], [,Pyou [wt, [wthink [@t,
[cptx [1th [wt, [vp[vlsaw M317] t,]]]]]]]]
c. [t,,[spwho‘l [sphow,]], [,Pyou [wtr [th [wthink
[CPtl [c.that, [,,,t,l [wt, [w[v.saw Mary]
t]]]]IJ

(90b) shows one step movement of 'who' to the Spec;
however, the trace cannot be properly governed. In (90c)
'who' skips the embedded CP and adjoins itself to the main
VP; the trace of 'who' here is not properly governed either.
In (90a) the chain of 'who' is properly head-governed, and
satisfies the ECP: however, the chain of 'how' is not
properly governed this time. That is, when 'who' moves to
the Spec of the Main CP at LF, it substitutes for the trace
of 'how' in the Spec of the embedded CP. No g-marked trace
of 'how' allows such substitution movement of 'who' as in
(90a). Among the representations at LF, I will assume that
(90.a.) is correct, compared with the analysis of (82).27

On the other hand, a g-marked trace is assumed not to

 

27Let us also consider the example:
I. a.* who thinks that who said that who saw Mary?
b. [”who [wt thinks that [wwho said that
[,who saw Mary?]]]] (SS)

Compared with (82), one of the Wh-in-situ movements
will violate the ECP at LF, since cyclic movement will make
the later Wh-phrase substitute for the trace of the earlier
Wh-phrase in the embedded CP.

65
be subject to substitution at all. To justify this
assumption, let us reconsider (68) and (69). In (69b) 'who'
cannot be substituted for the trace of 'what' through LF-
movement, since the trace of 'what' in the embedded CP has
the [+g]-feature. Therefore, the original trace of 'who'
does not satisfy proper government. Without this
assumption, (69b) would be grammatical unexpectedly.
Therefore, substitution for unmarked traces correctly

predicts grammaticality of the relevant examplesat

3.3 NP-movement
So far we have discussed Wh-movement, which forms A'-
chains. At this point,let us consider NP-movement, which

forms A-chains, in terms of the ECP.

(91) a. John is likely to win.
b. [,PJohn, [,.Inf1, [wbe, [Aplikely, [,Pt, to
win.]]]]]]

 

:nThe empty category with g—marking is related to the
wanna-contraction at the PF:
I a. who do you wanna love?
b. £‘8Y??]do [,Pyou want [cpt' [,PPRO to drink
II a.* Who do you wanna go home?
b. [cpwho do [,Pyou want [cpt' [,,.t+9 [wto go
homelllil
Here the g-marked trace is visible, and blocks contraction;
however, the unmarked trace is invisible to contraction. In
accordance with footnote 22, only an argument position is
g-marked at ss, and a nonargument position is g-marked at
LF. Hence the trace in CF is not g-marked and thus does not
contain the g-marking feature at PF, either.

66

Here the adjective 'likely' is not lexical, since it
originally has no case or index to assign. Then, 'John'
should move to a case-assigned position, to preserve the
Case Filter, and its trace is properly head-governed for the
ECP, and forms an A-chain. Therefore, adjunction movement
is barred here because of the Binding Theory.

However, IP-agreement makes 'John' and Infl share their
features, including the index of the NP 'John'. Then, the
inflected verb 'be' obtains the index of Infl, and transfers
it to the AP. New 'likely' is indexed and coindexed with
'John' as well its trace. Then, 'likely' can properly head-

or antecedent-govern the trace of 'John', and it obeys the
ECP.

On the other hand, in (92), affixes transmit the index
of subject NP to the verbs to which they are attachednfl and
the verbs can obtain the same index from Infl. Then, the
passivized verbs properly coindex the head- or antecedent-

govern the trace.

(92) a. John is killed t.

b. The city has been destroyed t.

 

8' It is not clear how affixes hop: it was called
Affix Hopping in traditional transformational Grammar.
Chomsky (1986) proposes head movement for inflected verbs.
However, this proposal does not cover affixing hopping for
passivization, progressive, and perfective. Here I will
leave the question open.

67

Now let us look at cyclic NP-movement as in (93):

(93) a. John seems to be expected to be killed.
b. * John seems that it is expected to be killed.

(94) a [,PJohn, [,.Infl, [wseem, [,Pt, [,.Infl, [wbe,

[wexpected, [,Pt, [,.Ian, [wbe, [wkilled,
t]]]]]]]]]]]]

b. * [,PJohn, [,.Infl, [wseemI [cpthat
[,,,it.l [,.Infl, [WbeJ [WexpectedI [,Pt,

[1.Inf11 [wbe, [wkilled, tlilllilllllll

In (94a), which is the structure of (93a), the NP
'John' cyclically moves to the subject position in the main
clause through the embedded subject position. Such cyclic
NP-movement and IP-agreement at SS enables the Infls to
share the same indexes with the subjects. Furthermore, the
Infls assign their indexes to the verbs 'be', which will
transfer them (or carry out this process through affix
hopping) to the main verb which properly governs the traces.
However, in (94b), the representation of (93b), 'John' does
not move cyclically, and the second Infl cannot obtain the
index which 'John' has. The result is that 'expected'

cannot properly govern the trace of its complement, and it

violates the ECP.

68

4. SUMMARY AND CONCLUSION

In this paper we have discussed some important concepts
of government, proper government, and ECP, and applied them
to the ECP phenomena.

First of all, we have define barriers in a generalized
way, and eliminated Chomsky's concept that IP and I' are
defective 'category'.

Second, we have substituted 'C-command' for Chomsky's
(1986) 'm-command', and 'proper c-command' for Chomsky's
(1986) 'c-command': we have defined government in terms c-
command, and proper government in terms of proper c-command.

Third, we have showed that the ECP in (54) is not
proper to explain some grammatical phenomena, and thus, in a
more restricted way, defined the ECP as (71). As a
consequence, the type of complement Wh-movement is different
from noncomplement Wh-movement. That is, to obey the ECP
which is defined in terms of consistent proper government,
complement Wh-movements take substitution, and noncomplement
Wh-movements take adjunction.

Fourth, we have suggested NP-indexing algorithm and
showed the consequence: as we have seen in footnote 7), we
could keep Subjacency out of dilemma by virtue of NP-
indexing algorithm, and NP-indexing, together with CP-
agreement at LF, has effect on that-trace effect and the

ECP.

69

Finally, we have showed how NP-indexing and NP-movement
work together for the ECP.

However, the ECP has still left some questions open.
First of all, the ECP does not give a good answer to how
quantifier raising (QR) work together with the EC, even
though the ECP and QR apply at LF, they are sure to be
closely related, and the ECP also has effect on QR

(Hornstein 1984, p. 66):
(95) Someone believes that Franks likes everyone.

Here 'someone' is interpreted as having a wide scope
over 'everyone', but not vice versa, even though the ECP
allows 'everyone' to be extracted out of the embedded:
sentence.

Second, With the ECP we do not have a solution to

multiple Wh-questions yet (Chomsky 1984a, p. 238):

(96) a. John wonders what who bought where.

b. * John wonders what who bought at the store.

For further research we should extend the ECP to such

phenomena.

REFERENCES

Andrews, A. (1982) A Note on the Constituent Structure of
Adverbials and Auxiliaries. Linguistic Inguiny 13, 313-
317.

Aoun, J. (1986). Generalized Binding. Dordrecht: Foris.

Aoun, J., N. Hornstein, and D. Sportiche (1981) "Some
Aspects of Wide Scope Quantification," Journal of
Linguistic Research 1, 69-96.

Chomsky, N. (1972) Studies on Sennntics in Generative
Grammar, The Hague: Mouton.

Chomsky, N. (1977) On Wh-Movement Rules, in P. Culicover,
T. Wasow, and A. Akmajian, eds., Formal Syntax. New
York: Academic Press.

Chomsky, N. and H. Lasnik (1977) Filters and Control,
Linguistic Inguiny 8, 425-504.

Chomsky, N. (1981) Lec u o v ent a ndin
Dordrecht: Foris.

Chomsky, N. (1982) Some Congepts and Qunseguenges of the
Theorv of Govennment and Binding.Cambridge, MA: MIT
Press.

Chomsky, N. (1986) Barriens. Cambridge, MA: MIT Press.

Haider, Hubert (1986) Affect alpha: A Reply to Lasnik and
Saito on the Nature of Proper Government Linguistic

lnguiny 17, 113-126
Hornstein, N. (1984) Logic As Gramman, Cambridge, MA: MIT
Press.

Huang, C.—T. J. (1982) Logical Relations in Quinese and the
Tneony of Gnamman, Doctoral dissertation, MIT.

Cambridge, MA.

Johnson, K. (1988) Clausal Gerunds, the ECP, and Government
Linguistic Inguiny 19, 583-610.

70

Ii

Kayne, R. (1984) nec dness B' Bra c
Dordrecht: Foris.

Kayne, R. (1985) Principles of Particle Constructions, in
J. Gueron, H.-G. Obenauer, and J.-Y.Pollock, eds.,

Grammatical Representations. Dordrecht: Foris.

Lasnik, H., and M. Saito (1984) On the nature of proper
government. Linguistic Inguiny 15, 235-289.

Lasnik, H. Uriagereka (1988) A gnurse In Gb Syntax: Lecture
On Bindin And Em t Cate es. Cambridge, MA: MIT

Press .

May, R. (1985) Lngical Form. Cambridge, MA: MIT Press.

Pesetsky, D. (1982) Paths anu CateggriesI Doctoral
dissertation, MIT, Cambridge, MA.

Reinhart, T. (1983) Anaphora and Semantic Intenpretation.

London: Croom Helm.

Riemsdijk, H. Van and E. Williams (1986) Intnoduction no
the Tneony of Gnammar, Cambridge, MA: MIT Press.

Sag,I. (1976) Deletion and Logigui Form. Doctoral
dissertation, MIT, Cambridge, MA.

Stowell, T. (1981) Origins of Pnnasg Stnucturg, Doctoral
dissertation, MIT, Cambridge, MA.

Stowell, T. (1982) The Tense of Infinitives. Linguistic
129212213: 551-

Stowell, T. (1983) Subjects across categories.
Linguistic Review 2, 285-312.

Williams, E. (1977) Discourse and logical form. Linguistic
Inguigy 8, 101-139.

"lilllllllllllllli