THE ATB THEORY OF PARASITIC GAPS EDWIN WILLIAMS I would

The Linguistic Review 6: 265-279
THE ATB THEORY OF PARASITIC GAPS
EDWIN WILLIAMS
I would like to explore some considerations that suggest that all parasitic
gaps arise äs a consequence of the principle of across-the-board (ATB)
rule application. This proposal was first hinted at in ROSS (1967); it was
adopted for Dutch by Huybregts and van Riemsdijk (1984), but rejected
by them for English, which they assume to have 'Chomsky style' parasitic
gaps along the lines of Chomsky (1982, 1986). I want to show that there
is an ATB theory which accounts for the basic properties of all parasitic
gaps. An account similar to the present one in several respects, but
independently arrived at, is that of Haik (1985). In the course of outlining
my ideas I will point out comparisons with hers. The principle differences
between her proposals and mine are that in her view, parasitic gaps are
not traces in s-structure, whereas they are in mine, arising äs any ATB
gap arises; and in her view, the coordination that licences the parasitic
gaps arises through a restructuring of s-structure (to derive LF) whereas
in mine parasitic gaps are licenced by s-structure coordination like all
ATB gaps.
Consider the following typical parasitic gap (pg):
(1)
Who would you warn / before striking/?#
Suppose that (1) is structured in the following way:
(2)
Who [[would you warn /]s [before striking pg]$\s
If the juncture between the two interior Ss were nominated "coordinating"
instead of "subordinating", then the so-called parasitic gap would arise
äs a normal second gap in a coordinate structure. I will postpone until
section 7 the question of what it means to call a pair of sisters "coordinative".
I will use an abstract marker "COORD" to mark a juncture that is construed
äs coordinate; thus (2) will be marked äs follows:
(3)
Who [[would you warn r]s COORD [before striking pg]s]s
Clearly the juncture in (2) must be optionally coordinative, eise only (2)
would be derived, and not (4):
(4)
Who [[would you warn f]s [before striking him]s]s
äs this would otherwise violate the Coordinate Structure Constraint, äs
the extraction is not ATB.
The Linguistic Revier 6 (1989-1990) 265-279. 0.167-6318/88/500732X7$ 02.75
Copyright © 1990 by Foris Publications, Dordrecht, Holland. U.S.A.
266
In fact, parasitic gaps are always optional, and so the junctures that
give rise to them are optionally coordinative. The only difference then
between such junctures and true coordinative junctures, such, äs those
marked by and, is that the latter are obligatorily construed äs coordinative.
For the conjunction before it is a reasonable Stretch to call it ambiguously
"coordinating" or "subordinating". Other cases strain the meaning of the
term however. Consider the cases in which the parasitic gap is in a relative
clause:
(5)
Who does everyone who meets pg like t
It is not so obvious what could be construed äs coordinated with what
here; but it is somewhat clearer in the following, where the relative is
extraposed:
(6)
Who does [[everyone like t] [who meets pg]]
It might be said that in (6) the relative is optionally construed äs coordinated
with the matrix S, and if it is, then extraction is from both conjuncts
in the ATB fashion. We might say of (5) that the derivation of the pg
is based on (6) in some manner. We will return to pgs in relative clauses
in section 4.
Even more difficult to reconcile with ordinary notions of coordination
are the parasitic gaps that appear in simple subjects:
(7)
Who would [pictures of pg] [impress /]
Here, we would be required to say that the juncture between the subject
and the VP was optionally construed äs coordinating.
1. COORDINATION
What dpes it mean to refer to some juncture äs "coordinating" if it does
not contain a conjunction? I will defer a füll discussion of this question
until the end of the paper, but a few remarks are in order here. A true
coordination, one based on a conjunction, consists of three parts: two
identical nodes, and a governing conjunction:
(8)
conj: X X
If we lax the condition that the two nodes must be identical, and the
condition that the governor must be a conjunction, we can generate the
füll set of environments in which pgs occur:
(9)
a.
before:
S'S'
267
(9)
b. INFL:
c. verb:
d. the:
NP VP
NP NP
S' S' (see section 4)
A simple indication that the structures just discussed can have a coordinative
character is that they admit Right Node Raising extractions, which are
ordinarily found only in coordinate structures:
(10)
I talked to /, without actually meeting / all of the members who
voted against Hinkly
Even the relative clauses cases are surprisingly amenable to this kind of
extraction:
(11)
Anyone who meets / really comes to like / any of our salespeople
The exception is the NP VP cases:
(12)
*Full-sized pictures of / really upset t any of the vampires you
• can find these days
But these are the most marginal cases for parasitic gaps in the first place,
the reason being, if any of this is right, that the NP VP juncture is most
resistant to being construed äs coordinating.
In fact, the cases which have been analyzed in the literature äs pgs
licenced by Heavy NP Shift (HNPS) are actually cases of Right Node
Raising (RNR) applying in ATB fashion, äs it always does:
(13)
I read t without filing / all of these books
A couple of observations support this. First, Heavy NP Shift cannot remove
an N P from a PP:
(14)
* I talked to t yesterday all of your stupid friends
But RNR can remove an NP from a PP:
(15)
I talked to and intimidated all of your stupid friends
And, äs we have seen, the pgs in such cases äs (10) are licenced by traces
in PPs - so these could not have arisen via HNPS. Furthermore, when
the moved NP does not cross the adjunct, no pg is licenced:
(16)
* I säw / yesterday all your friends without talking to t
So, Heavy NP Shift does not licence parasitic gaps; rather, the gaps thought
268
to arise from HNPS are due to ATB application of RNR. Of course, there
is nothing in principle to prevent HNPS from applying in ATB fashion
äs well; but, since HNPS does not perform extractions of any sort, äs
(14) shows, it could never give rise to ATB gaps.
In sum, then, I am suggesting that a number of junctures that are not
strictly coordinative be optionally construable äs coordinating, and that
this is the source of all parasitic gaps - they arise äs normal gaps through
the mechanisms of the ATB formalism.
2. THE C-COMMAND CONDITION
The c-command condition on pgs follows from the ATB theory with some
ordinary assumptions, in fact ones already made in the literature.
A subject trace cannot licence a pg that it commands (p. 149 from
Chomsky 1987):
(17)
*Who / warned the men that they were about t o arrest pg
Under the ATB theory, this example would have the following structure:
(18)
Who [[/] COORD [warned the men that they were about to arrest
pg]]
But it is a general property of ATB constructions that it is impossible
to remove the entirety of one conjunct (a theorem in fact of the ATB
formalism of Williams 1978):
(19)
*Who did John see t and pictures of /
But in the case of object main gaps, the entirety prohibition would seem
to be violated:
(20)
Who did you warn t that you would kill /
We might assume, äs Chomsky (1986) following Stowell (1981) does, that
the complement clause is extraposed; then the structure of (20) will be
(21)
Who did you [warn /]VP COORD [that you would kill pg]s]v?
Here, the inner VP is COORDinated with the complement S, and the
entirety prohibition is respected. Note that the extraposition must be to
the edge of VP, and not to the edge of S, eise subject gaps would be
licenced.
However, it may be an illusion that object traces can licence parasitic
gaps. The illusion is based on the fact that the direct object of certain
verbs is optional, or implicit:
269
(22)
I warned that I would talk to Bill
When the direct object is implicit, it is pragmatically inferred. I think
that the direct object in (21) is not a trace, but is an implicit argument
which is interpreted äs (optionally) related to the Wh-word, and the trace
in the complement is the primary and only extraction. Under this view,
verbs which take obligatory direct objects pgs in the complement clauses
should be worse, and this seems to be the case:
(23)
*John, who I V'ed / that Bill was looking for t
a. verb = *tell, *persuade, *convince, *inform
b. verb = warn, promise
The (a) verbs, but not the (b) verbs, require direct objects. It is reasonable
to conclude that the pg is systematically excluded in the complement if
there is a gap.
Why the exclusion, if it is so? In the previous accounts, such äs Chomsky
(1982) and Engdahl (1983), these pgs are excluded because of the antic-command condition. In the present context, they are excluded because
there is no juncture that could licence to configuration of the gaps. The
direct object could be COORDinated with the complement, but, äs we
have seen, the ATB principle does not permit either conjunct to be totally
evacuated, and in this case the direct object is totally evacuated. Assuming
that the object forms a small VP with the verb will not help either, if
we require there to be a governing COORDinator: there is no candidate
for this in the juncture of the small VP and the complement S.
Of course it will be possible to extract a pari of the direct object and
part of the complement:
(24)
John, who I warned friends of t that Bill was looking for i
Here the juncture of the direct object and the complement is governed
by the verb, which is taken to be the governing COORDinator. This is
an awkward juncture to construe äs COORDinated, and so the result is
on the low end of the scale for pg constructions, but it is possible.
3. CONNECTEDNESS
A rather strong argument for the ATB derivation of parasitic gaps is based
on the observation that Kayne's results about the "connectedness" of
structures with parasitic gaps will follow. A similar account of the relation
between connectedness and the ATB account is given in Haik (1985).
Kayne observed, contra Chomsky (1982), that there were "movement"like restrictions pn parasitic gaps, äs illustrated in the following, where
270
the pg seems to violate something like the condition against extracting
from subjects, even though in other circumstances, pgs inside subject
Position are alowed (25b):
(25)
a.
b.
* Which stars do people that think pictures of pg are attractive
annoy /
Which stars do pictures of pg annoy t
Apparently, if a pg is to occur in a subject, the true gap must be close
by.
Kayne further observed that pgs such äs the offending one in (25b)
bould be "saved" by the insertion of further pgs close by; in other words,
a pg could licence a pg:
(26)
Which stars do people that think pictures of pg impress pg annoy
Kayne proposed that the second pg served äs a bridge to properly "connect"
the first, otherwise illicit, pg to its antecedent Wh-phrase; this connecting,
the primary focus of his paper, was accomplished by definitions that I
will not explore here.
Rather, I will be satisfied to show that the type of examples that Kayne
uses to make his point are accounted for by the ATB principle.
This will consist in showing, for example, that there is a consistent
assignment of the "COORD" marker to the nodes of (25b) and of (26),
but not of (25a). (25b) is marked äs follows:
(27)
Which stars do [pictures of /] COORD [annoy t]
That is, the subject and VP are COORDinated, and extraction applies
ATB. Since the assignment of COORD is optional, other possibilities are:
(28)
a.
b.
c.
* Which stars do [pictures of /] [annoy t]
* Which stars do [pictures of /] COORD [annoy us]
* Which stars do [pictures of t] [annoy us]
(28a) has ATB rule application but not to a coordinate structure, so i t
counts äs a subject condition violation. (28b) violates the ATB principle.
(28c) is a violation of the subject condition.
Now, consider (25a = 20), and recall that relative clauses must appear
in extraposition:
(29)
* Which stars do [people annoy /] Xi that think [pictures of pg]
X2 [are attractive]
XI and X2 mark two junctures in this sentence. Xi must be COORD,
since there are two gaps, one in each conjunct of XI. Now, suppose that
271
X2 is COORD. Then (29) violates the ATB principle, äs the conjuncts
of X2 do not both contain gaps. On the other band, suppose that X2
is not COORD; then (29) is a simple subject extraction violation, just
like (28c). So there is no assignment of COORD that will allow this.
But for (26 = 21) there is an assignment of COORD that results in a
valid structure, namely:
(30)
Which stars do [people annoy /] COORD [that think [pictures of
pg] COORD [impress/?£]]
The relative must be COORD with the matrix S, and the embedded subject
with its VP; in that case, all gaps are ATB gaps.
I believe that not only for these cases, but for all possible cases, the
ATB principle will construct only structures "connected" in the sense of
Kayne.
In the preceding examples I have used "pg" to mark parasitic gaps;
however it should be clear that the "pg" in these structures is not really
parasitic on the "real" gap; in fact, both gaps have equal Status, each
occurring in a conjunct marked with COORD. The sense that one gap
is depejident on the other derive from the fact that junctures not marked
COORD but otherwise identical to these permit extraction from only one
term of the juncture.
Chomsky (1987), citing Kearney, argue that the parasitic gap is truly
parasitic, from the following examples (Chomsky 1987, from Kearney 1983):
(31)
a.
b.
Which books about himself did John file / before Mary read
e
* Which books about herseif did John file / before Mary read
e
The "real", but not the "parasitic" gap Supports reconstruction, suggesting
that it has a different Status from the "real" gap. But the effect these
excamples illustrate is actually a "proximity" effect, äs it shows up in
coordinate structures äs well, weher both gaps are "real":
(32)
a.
b.
Which books about himself did John file t and Mary read
e
* Which books about herseif did John file / and Mary read
e
In fact, it is possible to show that parasitic gaps do show reconstruction
effects entirely parallel to real gaps:
(33)
Whose mother did we warn t before he arrested pg
He is understood äs not linked to whose, presumably because he c-commands
272
the trace (pg) that whose mother binds, a typical reconstruction effect.
Hence, there is every reason to consider the two gaps on a par.
4. PARASITIC GAPS IN RELATIVE CLAUSES
The pgs in relative clauses present a problem for the theory of subjacent
movement, äs they appear to violate Ross's (1967) Complex NP Constraint
(CNPC). And äs Kayne (1984) and Chomsky (1987) have shown, any
embedding of a pg in a complex NP other than a relative at the same
"level" äs the real gap always leads to ungrammaticality. Chomsky solves
the problem of the relative clause by positing an abstract operator movement
in the derivation of pgs; in the case of relative clause pgs, this yields structures
of the following kind (ex. 136a from Chomsky 1987):
(34)
He is a man that [everyone [CP G [that gives presents to B]]] likes
A
Presumably in the Spec of CP G is close enough to the chain of the real
gap to be Consolidated with it in the manner Chomsky goes on to describe.
But, according to Chomsky, only a very special kind of relative will
permit this, namely one which is formed by "vacuous" movement of the
topmost subject - otherwise the Spec of COMP will be occupied, and
G cannot move there. In support of this Chomsky cites the following:
(35)
*This is a book that any man to whom we'll give e t will like
e
This follows, he says, because to whom occupies Spec of CP, preventing
G from moving there. However, this example can be ruled out on other
grounds, namely that it violates the Nested Dependency Condition, äs
book binds e and to whom binds t. And, in fact, if the binders are reversed,
(35) improves markedly:
(36)
This is a man that any book we give e i please /
And besides this example, there are other examples of relative clauses
with pgs that could not have been formed by vacious movement, but which
do obey the Nested Dependency Constraint:
(37)
This is the book which anyone who I persuade /' to read pg likes
Here, who binds /'. So it appears that the vacuous movement hypothesis
(VMH) along with the operator G cannot provide for the füll ränge of
pgs in relative clauses, so the problem of why pgs in relatives do not
count äs severe CNPC violations remains.
273
We might first consider relative clauses that are extraposed; such relatives
might then be construed äs COORDinated with the matrix S:
(38)
He is a man that [everyone likes t] COORD [that gives presents
to t]
The extraction from this structure is still a violation of subjacency, but
perhaps one on a par with a Wh-island extraction, that is, on the edge
of acceptability, and no longer a CNPC violation. It is not at all clear
that (38) should be accorded higher Status than that; it is certainly better
than an extraction from a complex NP extraction, but not perhaps perfect.
On the other hand, a primary (nonparasitic) extraction from a relative
clause is worse than primary extraction from Wh-island, even when the
relative is extraposed:
(39)
He is a man that everyone dies who meets t
This would seem to make the analysis in (38) suspect. But of course a
primary extraction does not involve COORDination. In the next section
we will look at the process of Subordination or demotion of adjuncts,
which permits weakly grammatical primary extractions from them, putting
them on a par with Wh-islands; if relative clauses are not subject to this
Subordination, then there will be a reason in addition to subjacency for
the ungrammaticality of a primary extraction, something like Huang's
(1982) CED.
Now, what about parasitic gaps in unextraposed relatives? We might,
äs Haik (1985) did, consider deriving an extraposed structure, and sanctioning the parasitic gaps in that structure, just äs above. Apart from
the fact that it works, there are some further considerations which suggest
this. First, there is the fact that relatives do extrapose in s-structure. Second,
there are languages in which relatives are always extraposed (Hittite,
Bangla); these "correlatives" in fact resemble adjunct conditional clauses.
There is a sense in which relatives even in English, especially ones with
indefinite heads, resemble adjunct clauses; note the close parallel between
these two:
(40)
a.
b.
Anyone can stay who has not finished
Anyone can stay if he has not finished
And, in fact, it is the relatives with indefinite heads (the one most easily
paraphrased äs adjuncts) that admit parasitic gaps:
(41)
a. ?? What did that friend of yours that bought pg break /
b.
What did anyone who bought pg break t
Finally, sentential complements to nouns are in the same structural
274
configuration äs relative clauses in all relevant respects, but they do not
bear an adjunctual relation to the matrix; rather, they are a part of the
theta structure of the NP in which they appear. And they absolutely do
not admit parasitic gaps, äs observed independently by Haik (1985):
(42)
* Who did the idea that Bill might kill pg occur to /
But it is perhaps unnecessary to invoke a literal extraposition, and the
concomitant commitment to an LF representation distinct from s-structure.
If we return to the idea that a coordinate structure consists of a governing
COORDinator and two phrases, we might consider identifying these items
in s-structure without restructuring it. The obvious candidate for the
COORDinator is the determiner. In a sense, the relates the relative clause
to the matrix. This relation is expressed in the traditional logical translation
of the determiner:
(43)
The man who won died
(44)
the ((x win) & (x died))
there is a unique x such that x won and x died
In this rendition, the relative and the matrix turn out to be parallel
constituents under the operator the, much äs two conjuncts are constituents
under the moperator and. Perhaps it is this notion of constituency - parallel
arguments of a bivalent operator - that permits ATB extraction.
Another bivalent operator is than, and here too we find parasitic gaps,
even when the subject is not extraposed, äs in (a) below, and even, äs
in (b), when it could not be extraposed without changing the meaning:
(45)
a.
b.
Who did more people [than we had expected to V /] see /
What can you eat more of than your father did [V f] without
flinching
The coordinate structure constraint would then read:
(46)
If an item is extracted from one argument of a bi- (or multi-) valent
operator, then an extraction must be made from every argument
of that operator.
5. ADJUNCT ASSYMETRIES
The following suggests that there is a second way, besides COORD, in
which an adjunct may be "adjusted" in its relation to the rest of the sentence:
(47)
a.
Which boy would you warn / before striking /
275
(47)
b.
c.
Which boy would you warn t before striking him
* Which boy would you warn him before striking /
It is ordinarily assumed that adjuncts cannot be extracted from (by a primary
extraction). But (47c) cannot be ruled out simply by prohibiting extraction
from adjuncts, because it contrasts sharply with the following:
(48)
Which boy would you warn Mary before striking t
To generate (48), extraction from an adjunct must be permitted. Perhaps
(48) still has a marginal Status; we might suppose that in order to extract
from an adjunct, it is necessary to put the adjunct in the VP so that it
is governed, and that this "demotion" of an adjunct from S domination
to VP domination is what is responsible for the marginality:
(49)
Which boy would you [warn Mary before striking t]
Suppose that the adjunct is demoted, and in effect taken to be an argument
of the verb. Then the direct object will c-command it. This will account
for the contrast between (47c) and (37): in (47c) the adjunct is demoted,
so the pronoun him will command it, and the trace it contains, giving
a condition C violation. In (48), no demotion takes place.
The other thing that can happen is that the adjunct can be construed
äs COORDinated with the matrix; in that case, we get ATB extraction.
This accounts for the paradigm in (47):
demotion
COORD
a
b
c
* by C
OKbyATB
OK by ATB
* by ATB
* by C
* by ATB
There is no need in this account of parasitic gaps for a principle of the
kind proposed by Safir (1984) (the PCOB), which says that if an operator
binds two variables, they must be either both empty categories, or both
not. The fact that the parasitic gap cases always result in two gaps follows
from the ATB principle, which is always involved when there are two
gaps. Safir's principle has scope beyond parasitic gap structures which
l will not explore here, and so still may be valid.
The different roles of demotion and COORDination in licencing extraction can be seen in extraction from preposed adjuncts:
(50)
a.
b.c.
This is the boy that I gave a warning to Bill before I shot
at /
This is the boy that [before I shot at t1] l gave a warning
to
b.
r'
c. * Bill
276
(50) shows that the adjunct in question can be demoted and can contain
a primary extraction. But when the adjunct is preposed, äs in (50b), it
cannot be demoted to VP domination, and so t1 äs a primary extraction
is impossible; however, the adjunct can still be COORDinated with the
matrix S, making i1 one of a pair of ATB extractions (50c).
A further argument for demotion comes from the following paradigm:
(51)
a.
b.
c.
d.
Who did you warn Mary without telling t
I warned her before striking Mary
Who did you warn Mary before striking her without telling
t
* Who did you warn her before striking Mary without telling
(51a) shows that the without phrase can be demoted, and extracted from.
(51b) shows that the before phrase is optionally demoted, for if it were
demoted in (b), a condition C violation should result. (51c) plausibly
involves the demotion of both the before and the without phrases, the
demotion of the latter forcing the demotion of the former; if, for example,
the demotion involves putting it into the VP, then this would be the result.
And (51d) shows that this is in fact the case: the demotion of the second
adjunct for purposes of extraction forces the demotion of the first adjunct,
which leads to a condition C violation.
The same point can be made with the following more succinct examples:
in the (a) example, the subordinate character of the adjunct is guaranteed
by the extraction from it; the NP which it contains should therefore be
strictly not coreferential with the matrix object, which c-commands the
demoted position:
(52)
a.
b.
c.
d.
e.
f.
* Who did we arrest herj before Maryj could shoot /
We arrested he^ before Maryj could shoot Bill
Who did we arrest Bill before Mary could shoot
* Who should I shoot herj if Maryj talks to
You should shoot her if Mary talks to Bill
Who should I shoot Bill if Mary talks to
Though more succinct than the previous examples, these are perhaps no
easier to judge.
6. DOUBLE EXTRACTIONS
A further argument for the ATB account can be derived from consideration
of examples derived from the following model:
277
(53)
Which book do you wonder who
[Bill told - that Mary bought -] COORD
[before Sam persuaded - that Mary wanted -]
The juncture marked COORD is obligatorily so marked, äs each Whphrase binds two gaps, one in each of the conjuncts of COORD.
The striking generalization that holds of (53) is that if one of the parasitic
gaps is replaced by a pronoun, then the other must be äs well:
(54)
Which book do you wonder who
[Bill told - that Mary bought - ] COORD
[before Sam persuaded - that Mary wanted -]
(55)
Which book do you wonder who
[Bill told - that Mary bought - ]
[before Sam persuaded him that Mary wanted it]
(56)
* Which book do you wonder who
[Bill told - that Mary bought - ]
• (before Sam persuaded him that Mary wanted - ]
(57)
* Which book do you wonder who
[Bill told - that Mary bought - ]
[before Sam persuaded — that Mary wanted it]
The explanation is straightforward: if one gap is replaced by a pronoun,
then its mate is no longer a member of an ATB pair of gaps: the unaffected
pair of ATB gaps nevertheless requires the presence of the COORD marking,
but with that marking, all gaps must be ATB, and so the mate of the
pronoun is in violation.
7. WHAT IS A COORDINATE STRUCTURE?
What does it mean to say that the NP and VP of a sentence can be "construed
to be coordinate"? What is a coordinate structure?
A true coordinate structure is a bivalent operator governing two perfectly
symmetrical arguments. The relevant points of symmetry are syntactic and
semantic. Syntactic symmetry involves categorial identity; semantic identity
involves identity of semantic function.
Structures which show one kind of symmetry, but lack another, will
be quasi-coordinate. Often, for example, and involves causal or temporal
sequencing, äs in
(58)
a.
b.
John went to town and bought a book
John was wounded and died
278
When the coordination is not perfectly symmetrical, exceptions to the ATB
principle creep in:
(59)
a.
b.
What did John go to town and buy t
* What does John have a pipe and want
The temporal sequencing in (59a) makes the coordination nonsymmetrical,
and the non-ATB extraction is tolerable; in (59b) it is not, because there
is no temporal sequencing, hence the nodes remain symmetrical in every
respect.
Interestingly, when a conjoined structure goes asymmetrical, it is the
right conjunct alone that can be extracted from:
(60)
a.
b.
Who did Tom buy a gun and shoot
* What kind of gun did Tom buy and shoot Bill
This is in fact the general case in English. Extraction is from the rightmost
of two sisters, perhaps because government is to the right, though how
to apply the notion of government to these cases is not clear.
In sum then, we might regard "coordinate" äs a property that a node
dominating two sisters can have to a degree.
At the opposite extreme of these examples are the many cases of "parasitic
gaps", where, according to my view, the property "coordinate" is applied
to pairs of terms that have barely any symmetry:
(61)
who do [NP
pictures of /
COORD(INFL) VP]
upset
/
The coordinate label does not sit well with such structures, because the
two terms exhibit almost no symmetry, syntactic or semantic, and that
is in fact the reason for the marginality of the parasitic gaps.
In fact, the more symmetry exhibited, the better the parasitic gap; so,
compared to the fairly awkward subject-predicate COORDination, we have
the relatively good cases of parasitic gaps in adjuncts:
(62)
Who did [John warn /] COORD [before he struck /]
This is better because the two terms joined by COORD are at least
syntactically similar, äs they are both Ss.
There is a wide grade of acceptability in parasitic gap environments,
and this calls for some explanation. So, for example, the following is a
more or less fully ordered list of cases, from most acceptable to least
(on the right I have listed the two COORDinated items in each case, prefixed
with the governor of the junction):
279
Who did you meet t and dislike
and: S S
What did you füe before reading t
before: S S
The man who people who meet t like /
the: S S
Who would pictures of pg upset /
INFL: NP VP
Who did you promise friends of to try to find
/
V: NP S
There is an intuitive sense in which the juncture gets less and less coordinatelike äs one descends the list.
Differences between languages in their use of parasitic gaps may
correspond to this grading of COORDinatability äs well. Dutch has parasitic
gaps only in adjuncts, äs Huybregts and van Riemsdijk (1984) have shown,
thus taking only the top of the list. English is simply more liberal.
REFERENCES
Chomsky, N. (1982) Concepts and Consequences, MIT Press.
Chomsky, N. (1986) Barriers, MIT Press.
Engdahl, E. (1983) "Parasitic gaps", Linguistics and Philosophy 6, 5-34.
Haik, I. (1985) "The Syntax of Operators", Doctoral Dissertation, MIT.
Huang, J. (1982) "Local Relations in Chinese and the Theory of Grammar", Doctoral
Dissertation, MIT.
Huybregts, R. and H. van Riemsdijk (1984) "Parasitic gaps and ATB", in NELS 15.
Kayne, R. (1984) "Connectedness and Binary Branching", MIT Press, Cambridge.
Kearney, K. (1983) "Governing categories", ms., University of Connecticut.
ROSS, J. (1967) "Constraints on Variables in Syntax", Doctoral dissertation, MIT.
Safir, K. (1984) "Multiple variable binding", Linguistic Inquiry 15, 603-637.
Stoweil, T. (1981) "Origins of Phrase Structure", Doctoral dissertation, MIT.
Williams, E. (1978) "ATB rule application", Linguistic Inquiry 9, 31-43.
Linguistics
Princeton University
Princeton, NJ 08540