What Differentiates VP Ellipsis from Sluicing in Island Repair

advertisement
What Differentiates VP Ellipsis from Sluicing in Island Repair
Jun Abe
Abstract: In this paper, I argue for the in-situ approach to sluicing, proposed by
Kimura (2010), by examining what differentiates sluicing from VPE in so-called
island repair. I argue that the crucial key to the relevant distinction lies in whether a
given elliptic construction involves movement of the remannt wh-phrase or not rather
than island repairability in terms of the size of the elliptic sites involved, advocated
by Fox and Lasnik (2003). Thus, sluicing is capable of “island repair” since no
movement is involved, whereas VPE is not capable of “island repair” since the
remnant wh-phrase undergoes overt movement.
Keywords: VP ellipsis, sluicing, island repair, in-situ approach, fragment answers
1 Introduction
This paper aims to address the question of what differentiates VP ellipsis (henceforce,
VPE) from sluicing regarding the possibility of island repair. It has been well known
since Chung, Ladusaw and McCloskey (1995) (henceforce, CLM) that while sluicing
allows island repair, VPE does not, as witnessed by such a constrast as the following,
provided by Merchant (2001):
(1) They want to hire someone who speaks a Balkan language, but I don’t remember
which (*they do).
(Merchant 2001:4-5)
There are at least three approaches to account for such a contrast. One is what
1 may be called the copying vs. deletion approach, taken by CLM, according to which
VPE involves PF deletion while sluicing involves LF copying. Under this approach,
the sluiced wh-phrase which in (1) is base-generated in Spec-CP, and the empty TP is
supplied with its content by LF copying of the relevant part in the antecedent clause.
Since no movement is involved in this derivation, no island effects show up. In the
VPE case, on the other hand, the surface form is derived by first applying overt
movement to which from its base-position, i.e., the complement position of speaks in
the relative clause island and then deleting the relevant VP at PF. Since overt
movement is involved in this derivation, an island effect ensues.
The second approach may be called the island repair approach, taken by Fox
and Lasnik (2003), which crucially assumes, following Ross (1969), Chomsky (1971)
and Merchant (2001, 2004, 2008), among others, that deletion is capable of island
repair. Under this approach, the contrast shown in (1) is accounted for in terms of the
difference with respect to the size of the elliptic sites involved. Since sluicing
involves TP deletion, any node that will induce a violation of an island condition is
erased by this deletion operation, hence no island effect. In VPE, on the other hand,
there exists an intervening barrier in the sense of Chomsky (1986), TP or Asp(ectual)P,
that remains after the relevant VP is deleted, which thus gives rise to an island effect.
The third approach, which is less familiar and yet will be supported in this
paper, may be called the in-situ approach, taken by Kimura (2010), according to
which the remnant wh-phrase in sluicing does not undergo overt movement but rather
stays in situ. Under this approach, the contrast shown in (1) follows immediately, as
2 Kimura claims; no overt movement is involved in sluicing whereas VPE involves
such movement. This approach shares with the copying vs. deletion approach the idea
that the contrast in question is derived by whether overt movement is involved or not
and that there is no such thing as “island repair” by deletion. Despite this similarity, it
overcomes the most serious problem of the latter approach: no unified operation is
involved in the two elliptic constructions.1 This said, the paper concentrates on
comparing the island repair approach and the in-situ approach, arguing for the latter.
The paper is organized as follows: Section 2 outlines Fox and Lasnik’s (2003)
account of the contrast shown in (1), and points out some problems. Section 3
introduces the in-situ analysis of sluicing, proposed by Kimura (2010), and her
approach to the contrast in question. I extend this analysis to capture a further
difference between sluicing and VPE in terms of a LF identity condition. Section 4
discusses a consequence of the proposed analysis for those cases of sluicing and VPE
that involve adjunct wh-remnants. Section 5 discusses fragment answers as an
extension of the present analysis, and argues that they should be analyzed in the same
way as sluicing under the in-situ approach. Section 6 concludes with a summary.
2 The Island Repair Approach: Fox and Lasnik (2003)
As briefly mentioned above, Fox and Lasnik (2003) derive the contrast shown in (1)
in terms of the size of the elliptic sites that are involved in sluicing and VPE. On the
assumption that deletion makes island repair possible, they claim that “such repair is
possible in sluicing since every intermediate projection is deleted,” and that “in VPE
3 a smaller constituent is deleted, leaving one (or more) of the islands pronounced and
consequently unrepaired.” (p. 149) It should be noted here that by “island repair” they
mean that “barriers” in the sense of Chomsky (1986) are erased rather than “islands”
in its normal sense, hence becoming innocuous. This is based upon their observation
that “failure of repair” in VPE occurs even when there is no island involved, as shown
below:2
(2) a. They said they heard about a Balkan language, but I don’t know which Balkan
language (they said they heard about).
b. *They said they heard about a Balkan language, but I don’t know which Balkan
language they did.
(3) a. They heard a lecture about a Balkan language, but I don’t know which Balkan
language (they heard a lecture about).
b. *They heard a lecture about a Balkan language, but I don’t know which Balkan
language they did.
(Fox and Lasnik 2003:148)
Fox and Lasnik make crucial use of the following assumption:
(4) In the relevant environments, the parallelism conditions on deletion (Parallelism)
make intermediate landing sites unavailable.
(ibid.:149)
Note that in the elliptic constructions under consideration, the antecedent clauses
include indefinte DPs (i.e., what CLM call inner antecedents) that correspond to the
wh-remnants in the elliptic parts. They assume that such indefinites are licensed in
situ by something like an existential closure à la Heim (1982) or a choice function à la
Reinhart (1997). This then demands, according to Parallelism, that wh-movement
4 takes place in one fell swoop in the elliptic parts, as stated in (4). This requirement
does no harm to the cases of sluicing, since all intermediate projections are deleted.
On the other hand, it causes a problem with VPE. Fox and Lasnik assume that “VPE
deletes VP and leaves Tense and Aspect pronounced.” (p. 151) Thus, the elliptic part
of (3b), for instance, has the following structure:
(5) which Balkan language [TP they Past [AspP do [VP hear a lecture about t]]]
They claim that “the unacceptability of VPE follows if we assume that (at least) one
of the two maximal projections is an island that must be circumvented by an escape
hatch or deletion.” (p. 151)
2.1 Problems
There are mainly two problems that come to mind. One has to do with the unclarity of
the system of barriers Fox and Lasnik (2003) have in mind. They assume that either
TP or AspP constitutes a barrier, but do not provide any definition of what constitutes
a barrier. One conceivable way of making sense of their assumption will be to assume,
following Takahashi (1994), that every maximal category constitutes a barrier. But it
is not at all clear whether such an assumption is well-motivated under the current
theory of movement. Note in paricular that when Takahashi makes such a proposal,
he adopts the operation of Form-Chain, proposed by Chomsky (1993), so that each
step of movement that is to void barrierhood of a maximal category needs no
triggering feature as long as the whole chain serves to check such a feature. The most
important motivation for Form-Chain is to resolve a paradox that arises from
5 incorporating two natural economy conditions: fewest steps vs. shortest steps. If a
derivation tries to minimize steps, then each step will become longer, and vice versa.
Such a case is instantiated by successive-cyclic movement of a wh-phrase, as
illustrated below:
(6) Which girl do you think that Fred said that I talked to t?
Given these two economy conditions, how can we ensure that a wh-phrase undergoes
successive-cyclic movement in such a case as (6)? Form-Chain is a key notion to
answer this question: the fewest steps condition applies to Form-Chain as a whole
while the shortest steps applies to each step of movement. Given the phase theory
proposed by Chomsky (2000), however, we are obliged to abandon exploiting
Form-Chain to guarantee successive-cyclic movement, since an operation must apply
phase by phase. Thus, to the extent that the phase-based theory of movement is
well-motivated in the current minimalist theory, there will be no place in which such
a notion as Form-Chain can play any role. This in turn undermines Takahashi’s
(1994) theory of movement that tries to derive locality effects on the assumption that
every maximal category is a barrier.3
The second problem has to do with the assumption that intermediate traces of
wh-movement must remain at the LF output, hence relevant for Parallelism. Recall
that this assumption is crucial for the wh-remnant in sluicing and VPE to be forced to
undergo one-fell-swoop movement; if intermediate traces could be deleted at LF,
Parallelsim would be satisfied even if the wh-remnant underwent sucessive-cyclic
movement and hence would be immnue from island effects. Thus, (6), for instance,
6 would have the following LF output, on the assumption made by Fox and Lasnik
(2003) that choice function is involved in the semantics of a wh-chain:4
(7) which g girl λg’ you think [g’ λg” that Fred said [g” λg”’ that I talked to
g”’(girl)]]
However, it is dubious that such a LF output as (7) is a legitimate one, since the
functions postulated in the intermediate Spec-CPs do not seem to play any role.
Rather, it will make more sense to assume, following Chomsky (1993), that since
intermediate traces of wh-movement do not contribute to the interpretation of the LF
output, they are simply deleted at LF,5 so that the LF output of (6) will be represented
as follows:
(8) which g girl λg’ you think [that Fred said [that I talked to g’(girl)]]
Once this assumption is established, Parallelism will not be violated even if the
wh-remnant of sluicing and VPE undergo successive-cyclic movement. Hence Fox
and Lasnik’s account for the contrast between sluicing and VPE with respect to the
possibility of island repair will not hold.6
2.2 The Contrast Type of VP Ellipsis
Despite the defects pointed out in the previous subsection, Fox and Lasnik’s (2003)
account for the impossibility of island repair in VPE gives rise to an interesting
prediction: if the antecedent clause involves movement in a way parrallel to that of
the wh-remnant in VPE, the resulting sentence should be possible unless no real
island intervenes in the movements involved. Fox and Lasnik claim that this is in fact
7 borne out by the following contrast:
(9) a. I know that John said that Mary read a certain book, but I don’t know which
one.
b. *I know that John said that Mary read a certain book, but I don’t know which
one he did.
(10) a. I know which book John said that Mary read, but YOU don’t know which
one.
b.??I know which book John said that Mary read, but YOU don’t know which
one he did.
(Fox and Lasnik 2003:151)
(9) illustrates the point we have already seen: VPE does not allow wh-movement out
of the elliptic site even if no island intervenes. Fox and Lasnik observe that “the
contrast observed in [9] is largely absent in [10].” (p. 151)7 The grammaticality of
(10b) follows under their system, since application of wh-movement in a
successive-cyclic fashion in the antecedent clause allows parallel application of
wh-movement in the elliptic part, which thus frees both applications of movement
from locality violations.
Notice that (10a), a sluicing counterpart of (10b), differs from (9a) in whether
the sluiced wh-phrase has an inner antecedent in the preceding clause: in (9a), a
certain book serves as the inner antecedent of the sluiced wh-phrase which one, thus
this construction belonging to what CLM call the merger type, whereas no such inner
antecedent exists in (10a). Rather, the latter example belongs to a different type, what
may be called the contrast type, as indicated by the contrastive stress on YOU. In fact,
8 Schuyler (2001) observes that such a VPE case as (10b) that involves contrastive
focus allows wh-movement out of the VP elliptic site rather freely unless the
movement in question induces an island violation.8 Compare (i) in fn. 2 and (3b),
which are both reproduced below in (11), with the examples in (12), cited from
Schuyler (2001):
(11) a.??They studied a Balkan language, but I don’t know which Balkan language
they did.
b. *They heard a lecture about a Balkan language, but I don’t know which
Balkan language they did
(12) a. I don’t know which puppy you SHOULD adopt, but I know which one you
SHOULDN’T.
b. I know which woman HOLLY will discuss a report about, but I don’t know
which woman YOU will.
Further, as Fox and Lasnik (2003) observe, this holds true even when the antecedent
clause does not involve overt movement, as shown below:
(13) a. He likes ABBY, but I don’t know who else (?he does).
b. He said he likes ABBY, but I don’t know who else (??he did).
They comment on these examples as follows: “to our ears they are similar in status to
[10b].” (p. 153) They suggest that the degraded status of these examples should be
attributed to an independent condition such as MaxElide. This said, they capture the
grammaticality of these examples by assuming that covert movement applies to the
element in the antecedent clause that carries contrastive focus. Thus, in (13a, b),
9 ABBY undergoes covert movement to the top of the antecedent clauses, hence giving
rise to the LF structures that satisfy Parallelism.
We have then reached the following conclusion:
(14) VPE tolerates wh-extraction only under contrast readings.
In the next section, I propose an alternative to Fox and Lasnik’s (2003) island repair
approach while maintaining their insight on Parallelism in such a way that this
requirement is satisfied most straightforwardly when VPE involves contrastive focus.
3 Alternative: The In-Situ Approach
3.1 The In-Situ Approach of Sluicing: Kimura (2010)
In this subsection, I outline the in-situ approach of sluicing, proposed by Kimura
(2010), according to which the sluiced wh-phrase in this construction stays in situ.
Kimura adopts Agbayani’s (2006) theory of movement, in which this operation is
factored into two sub-operations, following Chomsky (1995): Move-F and Pied-Pipe.
Agbayani follows Chomsky in assuming that Pied-Pipe functions as a repair
operation that recovers isolated features created by Move-F. Further, he proposes a
new condition as to how such isolated features are recovered, which Kimura calls PF
Adjacency Condition, as stated below:
(15) PF Adjacency Condition
Features isolated by movement and the remnant wh-category must be
phonologically adjacent.
This PF condition does a special job when extraction of a subject wh-phrase is
10 involved. Consider sentence (16), whose derivation is given in (17).
(16) Who has fixed the car?
(17) a. CQ [who has fixed the car]
b. wh+CQ [who has fixed the car]
At the stage of (17b), the wh-feature of who umdergoes Move-F to be checked upon
Q in the C head, and yet the remnant category of who that is deprived of its
wh-feature does not have to undergo Pied-Pipe to meet condition (15) since it is
already adjacent to the isolated wh-feature without such an operation. In this way,
Agbayani’s system of movement derives the vacuous movement generalization
defended by George (1980) and Chomsky (1986).9
At the same time, this system of movement solves what Agbayani calls VMH
paradox; that is, considerations of selection and scope seem to demand that even
wh-subjects that do not undergo Pied-Pipe behave as if they occupied Spec-CP. Such
a fact can be naturally attributed to the Move-F part of the whole operation. Thus, in
(17b), even though the wh-subject stays in Spec-TP, the fact that the whole sentence
is interpreted as interrogative as well as the fact that the wh-subject takes scope over
the whole sentence is indicated by the movement of the wh-feature to the C head.
Given the above mechanism of movement, Kimura (2010) claims that not only
Pied-Pipe but also Delete can be exploited to meet the PF condition (15), and that
sluicing is exactly a case where this condition is satisfied by deletion. Thus, the sluice
of (18), for instance, has the derivation given in (19):
(18) She’s reading something, but I can’t imagine what.
11 (CLM:241)
(19) a. I can’t imagine CQ [she’s reading what]
b. I can’t imagine wh+CQ [she’s reading what]
c. I can’t imagine wh+CQ [she’s reading what]
At the stage of (19b), what must undergo Pied-Pipe so as to be adjacent to the isolated
wh-feature if the intervening material is not deleted. On the other hand, if it is, what
does not have to undergo Pied-Pipe, as indicated in (19c), since it is adjacent to the
isolated wh-feature even if it stays in its original position.10 This is how the in-situ
analysis of sluicing works under the mechanism of movement advocated by Agbayani
(2006).11
Notice that the above analysis crucially relies on the assumption that
non-constituent deletion is in principle possible. In order to address the question of
what makes this option possible in sluicing, let us first note that what survives
deletion under the in-situ approach is exactly a wh-phrase that can undergo Pied-Pipe
in Agbayani’s (2006) sense. Ross (1969) observes that a possible wh-remnant of a
sluice corresponds with a phrase that can undergo wh-movement in regular
wh-question formation; compare the possible sluices in (20) with the legitimate
wh-phrases in (21):
(20) I know he has a picture of somebody, but I don’t know who/of whom/*a picture
of whom.
(21) I don’t know who he has a picture of/of whom he has a picture/*a picture of
whom he has.
(Ross 1969:262)
Suppose, following Abe (2005), that a wh-phrase needs to be licensed in the
12 following way:
(22) a. It carries or includes a wh-feature.
b. It is assigned a [Focus]-feature.
c. The wh-feature is merged into the [Focus] feature via percolation.
A wh-feature is inherently carried by a wh-item such as who, what, which, etc.
whereas a [Focus]-feature is not the kind of feature that is carried by a lexical item
but rather is assigned to a phrase. (22c) then requires that a wh-feature must be
percolated up to a phrase that a [Focus]-feature is assigned to. With these assumptions,
the acceptability of of whom and a picture of whom in the regular wh-question in (21)
is accounted for in the following way: in of whom, the wh-feature of whom is
successfully percolated up to the whole phrase of of whom, to which a [Focus]-feature
is assigned, whereas in a picture of whom, the percolation of the wh-feature carried by
whom up to the whole phrase fails. Given this mechanism of licensing a wh-phrase,
we can formulate the way deletion applies in sluicing as follows:
(23) Deletion applies to a given E(llipsis)-site except the phrase carrying [Focus]
where an E-site is a target for deletion.
What is intended with this formulation is that a target of deletion is a constituent and
yet an actual deletion operation applies to it in such a way that a phrase carrying
[Focus] evades such an operation. Let us then consider the derivation of the sluice of
(20) with of whom as its wh-remnant. Under the present assumptions, it will have the
following derivation:
13 (24) a. I don’t know CQ [he has a picture of whom]
↓assignment of [Focus]
b. I don’t know CQ [he has a picture [Focus of whom]]
↓percolation of the wh-feature
c. I don’t know CQ [he has a picture [Focus+wh of whom]]
↓movement of the wh-feature
d. I don’t know wh+CQ [he has a picture [Focus of whom]]
↓deletion
e. I don’t know wh+CQ [he has a picture [Focus of whom]]
The unacceptability of (20) with a picture of whom as its remnant is due to the fact
that the percolation of the wh-feature carried by whom up to the whole phrase a
picture of whom fails.
Along the lines of Lobeck (1990) and Saito and Murasugi (1990), I assume that
the E-site of the sluicing construction must be licensed by a [+wh] C that enters into
an “agreement” relation with a wh-phrase in a broad sense; in this case, Move-F can
establish such an “agreement” relation. Thus, in (24), the E-site TP is licensed by CQ,
which is in “agreement” with of whom.12 Note that under the present system of
licensing, for a phrase to be licensed as an E-site does not entail that the whole phrase
undergoes deletion, though the target of deletion must be confined to an E-site. Thus,
in (24), the TP complement of the CQ head is licensed as an E-site and deletion
applies to this site, leaving the wh-phrase of whom intact, since the latter carries the
[Focus] feature.
14 This in-situ analysis straightforwardly explains the island insensitivity of the
type of sluicing under consideration, as illustrated below:
(25) a. They want to hire someone who speaks a Balkan language, but I don’t
remember which language.
(Merchant 2001:87)
b. Ben will be mad if Abby talks to one of the teachers, but she couldn’t
remember which.
(ibid.:88)
Since the wh-remnants of the sluices in these examples stay in situ, thanks to deletion
applying to the material intervening between the isolated wh-features and these
wh-remnants, they are free from island conditions, just like in-situ wh-arguments in
English multiple wh-questions, as illustrated below:
(26) a. Who wants to hire someone who speaks which language?
b. Who will be mad if Abby talks to which teacher?
One of the significant predictions of this in-situ analysis is that unlike
wh-arguments, wh-adjuncts should exhibit island sensitivity when they function as
remnant wh-phrases in sluicing. This is because it has been standardly observed since
Huang (1982) that in-situ wh-adjuncts differ from in-situ wh-arguments in showing
island sensitivity, as illustrated by the following contrast from Chinese:13
(27) a. [shei xie de shu] zui youqui?
who write DE book most interesting
‘lit. Books that who wrote are the most interesting?’
b. *[ta weisheme xie de shu] zui youqui?
he why
write DE book most interesting
15 (Huang 1982:526)
‘lit. Books that he wrote why are the most interesting?’
(ibid.:527)
(27a) shows that the in-situ wh-argument shei ‘who’ is insensitive to the relative
clause island, whereas (27b) shows that the in-situ wh-adjunct weisheme ‘why’ is
sensitive to this island. The type of sluicing under consideration also exhibits island
sensitivity when the remnant wh-phrase is an adjunct, as predicted:14
(28) a. Sandy is very anxious to see if the students will be able to solve the
homework problem in a particular way, but she won’t tell us (in) which
(way)/*how.
(Sauerland 1996:303)
b. He wants to interview someone who works at the soup kitchen for a certain
reason, but he won’t reveal yet ?what reason/*why.
(Merchant 2001:129)
These facts will be captured under the assumption, made by Abe (1993) and Tsai
(1994), that while argument wh-phrases in situ can be licensed by way of binding,
hence not exhibiting island effects, adjunct wh-phrases in situ do not have access to
such a licensing and instead must undergo covert movement to Spec-CP to be
licensed. Thus, the sluice of (28a) with the remnant wh-phrase how, for instance, will
have roughly the following derivation:
(29) a. she won’t tell us [CP CQ [TP she is very anxious to see if the students will be
able to solve the homework problem how]]]
b. she won’t tell us [CP <how> CQ [TP she is very anxious to see if the students
will be able to solve the homework problem how]]]
(29b) is derived from the underlying structure (29a) by (i) deleting the embedded TP
except how and (ii) applying covert movement to this wh-phrase (here the head of the
16 how-chain is marked with angled brackets to indicate that the movement in question
is covert). This covert movement induces a violation of the wh-island constraint;
hence the ungrammaticality of (28a) with the remnant wh-phrase how. On the other
hand, when (in) which (way) is involved in (28a) instead of how, the argument
wh-phrase whcih (way) is licensed by way of binding rather than movement, hence
immune to the wh-island constraint. The correctness of this line of reasoning is
confirmed by the observation that such a PP phrase as in what way can stay in situ in
multiple wh-questions in English, unlike such a true adjunct as how, as shown
below:15
(30) Who is very anxious to see if the students will be able to solve the homework
problem in which way.
Another significant prediction of the in-situ analysis is, Kimura (2010) claims,
that though the type of sluicing under consideration shows island insensitivity,16 the
VPE counterpart should show island sensitivity, since in such a case, the PF
adjacency condition given in (15) is not met when the remnant wh-phrase sits in situ,
hence requiring it to undergo overt movement. This is borne out by such an example
as (1), reproduced below:
(31) They want to hire someone who speaks a Balkan language, but I don’t know
which (*they do).
The VPE case will have the following derivation:
(32) a. I don’t know [CP CQ [TP they do [VP want to hire someone who speaks which]]]
b. I don’t know [CP wh+CQ [TP they do [VP want to hire someone who speaks
17 which]]]
c. I don’t know [CP which wh+CQ [TP they do [VP want to hire someone who
speaks <which>]]]
At the stage of (32c), which needs to move overtly to the embedded Spec-CP even
though the material in VP gets deleted; otherwise, it cannot satisfy the PF adjacency
condition. Since this movement crosses the relative clause island, it induces
ungrammaticality.17
Though this account is very attractive, it does not cover the whole story on the
difference between sluicing and VPE with respect to the possibility of island repair
Fox and Lasnik (2003) dealt with. We need to address the question of why the
generalization given in (14), reproduced below, holds true.
(33) VPE tolerates wh-extraction only under contrast readings.
We now turn to this question in the following subsection.
3.2 Proposal
We have seen that VPE does not allow wh-extraction when it is of the merger type,
i.e., it has an inner antecedent (cf. (2b), (3b), (11)). To address the question of why
this is so, let us first consider how the elliptic site of this type of sluicing is identified
with the corresponding part of its antecedent clause. I first outline CLM’s theory of
identification for this type of sluicing and then discuss how it is adapted under the
in-situ analysis of sluicing.
CLM adopt the LF copy theory for the derivation of sluicing. Thus, the
18 underlying structure of the sluice in (18), for instance, which is reproduced below, is
something like (35):
(34) She’s reading something, but I can’t imagine what.
(35) I can’t imagine what [TP e]
They claim that from this underlying structure, a LF representation appropriate for
interpretation is derivable simply by copying (or recycling, in their terms) the
antecedent TP. Thus, the final LF representation looks like the following:
(36) I can’t imagine [CP what [TP she’s reading something]]
In order to get a proper interpretation from this LF representation, it is necessary for
something to be taken as a variable of the operator what. CLM then propose a process
called merger, which combines the indefinite part of a wh-phrase with the indefinite
in its original position to make them serve together as a variable bound by the
wh-operator. In a simple case like (34), the indefinite part of what, which amounts to
something, is totally merged into the inner antecedent by this process, and hence the
domain of the wh-operator is unchanged. However, in a more complex case such as
the following:
(37) She’s reading a book by Chomsky, but I can’t imagine what.
the merger of what with its inner antecedent a book by Chomsky after recycling yields
the interpretation in which the domain of the wh-operator is restricted to the books by
Chomsky. CLM demonstrate that in order for merger to succeed, inner antecedents
must be indefinites, providing such examples as the following:18
(38) a.?*I know that Meg’s attracted to Harry, but they don’t know who.
19 b.?*Since Jill said Joe had invited Sue, we didn’t have to ask who.
(39) a. *She said she had spoken to everybody, but he wasn’t sure who.
b. *She’s read most books, but we’re not sure what/which.
The ungrammaticality of these sentences is attributed, according to their theory, to the
fact that the inner antecedents of the remnant wh-phrases are not indefinites (proper
names in (38) and generalized quantifiers in (39)) and hence merger fails.
In order to capture such a restriction on inner antecedents under the in-situ
analysis of sluicing, we need to adapt CLM’s theory of merger into one that assumes
deletion rather than recycling as a process involved in ellipsis. Abe (2008) proposes
the following as an identity requirement on the merger type of sluicing:
(40) An E-site is identified with its antecedent only if the remnant phrase and its
inner antecedent can undergo the process of merger.
Thus, the TP E-sites of (34) and (37) satisfy this requirement since something and a
book by Chomsky can successfully undergo merger with the indefinite part of what,
while those of (38) and (39) do not, due to the failure of this process.
Given this, the fact that VPE does not allow wh-extraction when it is of the
merger type is naturally attributed to the failure of merger in the case in which the
wh-movement involved is overt. Thus, I hypothesize the following:
(41) An inner antecedent cannot undergo merger with the trace of the corresponding
wh-phrase.
Let us consider (3) for illustration, which is repeated below:
(42) a. They heard a lecture about a Balkan language, but I don’t know which Balkan
20 language (they heard a lecture about).
b. *They heard a lecture about a Balkan language, but I don’t know which
Balkan language they did.
The elliptic parts of these examples have the following representations under the
present assumptions:
(43) a. I don’t know [CP wh+CQ [TP they did [VP hear a lecture about which Balkan
language]]]
b. *I don’t know [CP which Balkan languagei wh+CQ [TP they did [VP hear a lecture
about ti]]]
In (43a), which Balkan language does not need to undergo Pied-Pipe to satisfy the PF
adjacency condition thanks to deletion applying to the intervening material. In (43b),
on the other hand, which Balkan language does need to undergo Pied-Pipe to satisfy
this PF condition, since it is not adjacent to the wh-feature attached to CQ when in its
original position. (43a) satisfies the LF identity condition given in (40) since which
Balkan language is able to undergo merger with its inner antecedent a Balkan
language. (43b), on the other hand, cannot satisfy this identity condition due to the
restriction on merger given in (41), since a Balkan language is not able to undergo
merger with the trace of which Balkan language.19
Let us now turn to the contrast type of sluicing and VPE and consider how such
cases satisfy a LF identity condition. The relevant notion in this case is Parallelism in
the sense of Fox and Lasnik (2003). Consider (12a), reproduced below:
(44) I don’t know which puppy you SHOULD adopt, but I know which one you
21 SHOULDN’T.
Under the present assumptions, (44) will have the following representation:
(45) I don’t know which puppyi you SHOULD [VP adopt ti], but I know which onej
you SHOULDN’T [VP adopt tj]
Given that traces are regarded as free variables when they lack their antecedents in
given domains, hence non-distinct with each other for the purpose of LF
identification, the antecedent VP is taken to be identical to the elliptic VP in (45). We
have observed those cases of the contrast type of sluicing and VPE in which the
antecedent clauses do not involve overt movement, as shown in (13), reproduced
below:
(46) a. He likes ABBY, but I don’t know who else (?he does).
b. He said he likes ABBY, but I don’t know who else (??he did).
For these cases, Fox and Lasnik (2003) propose that the focused phrases in the
antecedent clauses undergo covert movement to satisfy Parallelism. Thus, the VPE
case of (46a) will have the following representation, on the assumption that the
focused phrase ABBY is adjoined to TP (the angled brackets put in ABBY indicate that
the movement involved is covert):
(47) [TP <ABBYi> [TP he Pres [VP like ti]]], but I don’t know [CP who elsej [TP he does
[VP like tj]]]
Thanks to covert focus movement applying to ABBY, the antecedent VP is now
regarded as identical to the elided VP, since the traces involved function as
“alphabetical variants” of free variables.
22 As for the sluicing cases in (46), we are naturally led to the claim that the
remnant wh-phrases undergo movement to Spec-CP to satisfy the LF identity
condition. Thus, the sluicing case of (46a) will have the following representation:
(48) [TP <ABBYi> [TP he Pres [VP like ti]]], but I don’t know [CP who elsej [TP he Pres
[VP like tj]]]
Here, the antecedent TP is identical to the elided TP, hence satisfying the identity
condition. Recall that we have advocated the in-situ analysis of sluicing proposed by
Kimura (2010) but that her arguments for it are based exclusively on the merger-type
of sluicing. That the contrast type of sluicing involves movement, unlike the merger
type, is supported by the fact that the former type is island-sensitive, unlike the latter,
as shown by Fox and Lasnik (2003) and Merchant (2008):
(49) a. *The detective ruled out the possibility that Fred killed ABBY, but I don’t
know who elsei [the detective ruled out the possibility that Fred killed ti].
(Fox and Lasnik 2003:152)
b. *Abby wants to hire someone who speaks GREEK, but I don’t remember
what other languagesi [she wants to hire someone who speaks ti].
(Merchant 2008:148)
The ungrammaticality of these sentences is attributed to a violation of island
conditions caused by the covert focus movement of ABBY and GREEK and/or the
overt movement of the corresponding wh-remnants.20
To sum up, the fact that the distribution of the merger type of VPE is more
restricted than is expected if it is constrained solely by island conditions is attributed
23 to the failure of merger in CLM’s sense in the case where the remnant wh-phrase
undergoes overt movement. In this way, the generalization given in (33) falls into
place. Given the in-situ analysis of the merger type of sluicing, the standard
observation that sluicing is capable of island repair whereas VPE is not is only
apparent. Rather the crucial distinction between sluicing and VPE lies in the fact that
the wh-remnant in the merger type of sluicing sits in situ while that of VPE undergoes
overt movement, as originally claimed by Kimura (2010).
4 A Consequence: Adjunct Wh-Remnants
Recall that we saw in Section 3.1 that the in-situ analysis of sluicing brings a nice
consequence with respect to those cases of the merger type that involve adjunct
wh-remnants; that is, such cases show island sensitivity, as illustrated in (28),
reproduced below:
(50) a. *Sandy is very anxious to see if the students will be able to solve the
homework problem in a particular way, but she won’t tell us how.
b. *He wants to interview someone who works at the soup kitchen for a certain
reason, but he won’t reveal yet why.
We captured these facts on the assumption that adjunct wh-phrases in situ must
undergo covert movement to Spec-CP to be licensed. Thus, the sluice of (50a), for
instance, has roughly the following derivation:
(51) a. she won’t tell us [CP CQ [TP she is very anxious to see if the students will be
able to solve the homework problem how]]]
24 b. she won’t tell us [CP <howi> CQ [TP she is very anxious to see if the students
will be able to solve the homework problem ti]]]
This immediately raises the question of how such simple cases of sluicing as
illustrated below satisfy the LF identity condition stated in (41), which is repeated as
(53) for ease of reference:
(52) a. A student solved the problem somehow, but I am not sure exactly how.
b. John left for some reason, but I don’t know exactly why.
(Lasnik 2008:23)
(53) An inner antecedent cannot undergo merger with a trace of the corresponding
wh-phrase.
Under the present assumptions, the sluice of (52a), for instance, will have the
following representation:
(54) I am not sure exactly [CP <howi> CQ [TP a student solved the problem ti]]
This representation clearly violates the LF identity condition in question since the
inner antecedent somehow cannot undergo merger with the trace of the sluiced
wh-phrase how. What makes such cases as in (52) possible, then?
I propose, along the lines of Rizzi (1990), that in such cases, the adjunct
wh-remnants are base-generated in Spec-CP. Thus, the sluice of (52a) has the
following representation instead:
(55) I am not sure exactly [CP how CQ [TP a student solved the problem]]
Thus, such cases do not really involve merger for satisfaction of identity but rather
involves “non-distinctness” in the sense of Chomsky (1965); in the embedded TP of
(55), the means by which a student solved the problem is simply unidentified, hence
25 the content of that TP is non-distinct with that of the antecedent clause. Under this
analysis, then, it will be more appropriate to treat such cases as in (52) as special
species of what CLM call the sprouting type, which lacks the antecedents
corresponding to the sluiced wh-phrases, as shown below:
(56) a. A student solved the problem, but I am not sure exactly how.
b. John left, but I don’t know exactly why.
(Lasnik 2008:23)
Given the option in which such wh-adjuncts as how and why are base-generated in
Spec-CP, these cases may involve strict identity between the elided TPs and their
antecedent TPs.
Under this analysis, it is predicted that the VPE counterparts of the wh-adjunct
remnant cases of sluicing should also be possible since they do not involve merger,
hence immune to the identity condition given in (53). That this is in fact the case is
shown by the fairly acceptable status of the relevant example, given below:
(57) a. ?A student solved the problem (somehow), but I’m not sure exactly how
he/she did.
b. ?John left (for some reason), but I don’t know exactly why he did.
Given that such adjuncts as how and why can be base-generated in Spec-CP, the LF
identity condition in question is met with these examples in exactly the same way as
with (52) and (56).21
Notice that such cases of the sprouting type as in (56) and (57) have derivations
other than the ones that involve base-generation of the wh-remnants in Spec-CP, that
is, those that involve movement of the wh-remnants to Spec-CP. Thus, the sluice of
26 (56a) can have a representation like (54), which is reproduced below:
(58) I am not sure exactly [CP <howi> CQ [TP a student solved the problem ti]]
The condition in (53) is irrelevant here, since no inner antecedent is involved in this
type of sluicing and VPE. The LF identity condition is met with the notion of
non-distinctness, as in the case of (55), but in the opposite direction in the present
case: in the antecedent TP for (58), the means by which a student solved the problem
is simply unidentified, whereas the trace of how in (58) marks the slot that specifies
the means in question. Such a way of satisfaction of the LF identity condition is
independently necessary for explaining the grammatically of such a typical case of
the sprouting type of sluicing:
(59) She’s reading. I can’t imagine what (she’s reading).
In this case, the object of reading is added in the elided TP.
So far we have seen that the merger type of sluicing and VPE allows the
wh-adjunct remnant to be only base-generated in Spec-CP due to the prohibition on
merger stated in (53), and that the sprouting type allows it to move up to Spec-CP as
well. Now a prediction: in the merger-type, the wh-adjunct remnant should allow only
a “local” reading and prohibit “long-distance” readings. This is exactly the
observation reported by Lasnik (2008) for the sluicing case, who attributes it to
Benjamin Bruening:
(60) a.?*Mary claimed that John left for some reason, but I don’t know exactly why
Mary claimed [that John left t].
b.?*Bob thinks that Mary fixed the car somehow, but I don’t know exactly how
27 Bob thinks that [Mary fixed the car t].
(Lasnik 2008:23)
In these cases, the wh-adjuncts cannot be interpreted as modifying into the embedded
clauses. The same thing holds true for the VPE case:
(61) a. *Mary claimed that John left for some reason, but I don’t know exactly why
she did claim [that John left t].
b. *Bob thinks that Mary fixed the car somehow, but I don’t know exactly how
he does think that [Mary fixed the car t].
How about the sprouting type? What Lasnik (2008) reports on (60) is in fact
more than I just stated: the data given in (60) hold the same even if the inner
antecedents for some reason and somehow are dropped. This comes down to the
question of what prohibits why and how from being base-generated in the lower
clause and moving to the higher Spec-CP in the sprouting varieties of (60a, b). I
suggest, along the lines of Nakao and Yoshida (2007) and Nakao (2009), that such a
derivation is excluded in terms of Parallelism à la Fox and Lasnik (2003); that is, in
order to observe Parallelism, why and how cannot move in a successive-cyclic fashion,
which thus induces an “ECP violation.” Although this account leaves it open exactly
what conditions are responsible for guaranteeing the locality of wh-adjuncts, it
suggests that Fox and Lasnik’s (2003) account in terms of Parallelism is valid for the
wh-adjunct remnant cases of sluicing and VPE.
5 A Further Extension: Fragment Answers
Merchant (2004) raises the question why fragment answers such as (62) below cannot
28 repair island violations, unlike sluicing:22
(62) a. Who did she see?
b. John.
Merchant notes that with a language such as English that demands an overt
wh-movement, “testing island sensitivities in fragment answers is not simple,
however, since the simple questions that would test for them are themselves island
violations.” (p. 687) As one strategy of avoiding this difficulty, he exploits fragment
answers to “implicit salient questions”, such as the following:
(63) a. Does Abby speak Greek fluently?
b. No, Albanian.
He claims that in (63), “the answer can take it that the questioner may be interested in
the answer to the question What language(s) does Abby speak?, in addition to the
narrower answer to the yes-no question.” (p. 687-688) Under this assumption, it is the
implicit constituent question that serves as the antecedent clause of such a fragment
answer as (63b). That said, Merchant demonstrates that such a pair as in (63) exhibits
island sensitivity, but not clause-boundedness, as shown below:
(64) a. Did Abby think Ben wrote the letter?
b. No, Charlie.
(65) a. Does Abby speak the same Balkan language that Ben speaks?
b. * No, Charlie.
(66) a. Did Ben leave the party because Abby wouldn’t dance with him?
b. *No, Beth.
29 From this observation, he concludes that fragment answers are island-sensitive. He
then tries to answer the question raised at the beginning of this section, basically in a
manner similar to Fox and Lasnik’s (2003) way of capturing the difference between
sluicing and VPE with respect to island repair. On the assumption that
island-inducing features are attributed not to barriers but rather to all the traces of a
phrase that crosses an island, Merchant claims that while in sluicing, all the traces that
carry island-inducing features are deleted, hence inducing no island effects, in
fragment answers, one in an intermediate projection that is on the path of movement
of a fragment phrase remains, hence causing island effects. He assumes the following
structure for a fragment answer:
(67) [FP fragment XPi [CP ti [TP … ti …]]]
In this structure, XP is moved from its original position to the Spec-FP and crucially it
is assumed to pass through the Spec-CP on its way. With the unmotivated assumption
that it is TP rather than CP that undergoes deletion, Merchant claims that when XP
crosses an island, all its traces bear island-inducing features and the one in Spec-CP
survives deletion, hence inducing island effects.
It is obvious that Merchant’s account of the island sensitivity of fragment
answers is based upon rather ad hoc assumptions, hence far from a real explanation.
Furthermore, the claim that fragment answers are island-sensitive is doubtful.
Nishigauchi and Fujii (2006) observe that fragment answers in Japanese, a wh-in-situ
language, show island insensitivity:
30 (68) a. Minna-ga [Akira-ga doko-de totta] syasin-o
mita ka osiete.
everyone-NOM -NOM where took picture-ACC saw Q tell
‘Lit. Tell me Q everyone saw a picture [Akira had taken where].’
b. Tokyo-de desu.
-in be
‘It is in Tokyo.’
(Nishigauchi and Fujii 2006:4)
(69) a. Hanako-wa [Akira-ga nani-o
-TOP
nusunda kara]
-NOM what-ACC stole
okotteiru no?
because be-angry Q
‘Lit. Hanako is angry [because Akira stole what]?
b. Tokei-o
desu.
watch-ACC be
‘It is a watch.’
Notice that in the questions given in (68a) and (69a), the wh-phrases are embedded
within islands, a relative clause island in (68a) and an adjunct island in (69a), and
these sentences are fine. Then, the acceptability of the fragment answers given in
(68b) and (69b) clearly demonstrate that this construction is island-insensitive.
This fact will be accommodated by the in-situ analysis of fragment answers in
the same way as the island insensitivity of sluicing is accommodated by such an
analysis. Thus, under this analysis, (62b), for instance, will have the following
derivation:
(70) a. [FP [TP she saw JOHN]]
↓deletion of TP except the focused phrase JOHN
31 b. [FP [TP she saw JOHN]]
This will make sense if we consider what kind of identity requirement is involved in
fragment answers. Recall that we have adopted (40), repeated below, as an identity
requirement on the merger type of sluicing:
(71) An E-site is identified with its antecedent only if the remnant phrase and its
inner antecedent can undergo the process of merger.
In this case, the process of merger is successfully undergone on the condition that the
antecedent clause serves as a presupposition of the sluiced clause. Thus, in the
following sentence:
(72) She saw someone, but I don’t know who.
the sluice has the following representation:
(73) [CP CQ [TP she saw WHO]]
Here the first clause of (72), i.e., the antecedent clause of the sluice, serves as a
presupposition of what is represented in (73). In this way, the E-site of the sluice is
identified with its antecedent clause. In the case of fragment answers, it is obvious
that the remnant phrase must be a possible answer to the corresponding question; in
particular, it must be among the set of possible answers presupposed by the
corresponding wh-phrase. Then, it may be said that the relationship between the
antecedent clause and the E-site in fragment answers is opposite, in a sense, to that in
sluicing: fragment answers serve as satisfiers of the truth conditions of their
corresponding questions. We can then extend the notion of merger in the following
way:
32 (74) α and β undergo the process of merger if (i) α constitutes a presupposition of β
or (ii) β is a satisfier of the truth condition of α.
With this characterization, fragment answers and the merger type of sluicing can be
seen to constitute a natural class regarding the identity requirement involved, hence
amenable to the in-situ analysis.
Having established that fragment answers are island-insensitive, we need to go
back to Merchant’s (2004) cases of fragment answers to “implicit salient questions”,
whose example is repeated below:
(75) a. Does Abby speak Greek fluently?
b. No, Albanian.
As one reviewer suggests, it is more plausible to analyze such a pair on a par with the
so-called stripping construction, whose example is given below:
(76) Abby speaks Albanian fluently, not Greek.
As Reinhart (1991) observes, this construction exhibits island sensitivity rather than
clause-boundedness, as shown below:
(77) a. Lucie will admit that she stole the diamonds if you press her, but not the car.
b. *We have interrogated the burglar who stole the car already, but not the
diamonds.
(Reinhart 1991:374)
Based upon such a fact, Abe and Hoshi (1997) propose that this construction involves
leftward movement of the remnant phrase in the elliptic site as well as of its
corresponding phrase in the antecedent clause. Thus, (76) will have the following
representation:
33 (78) [FP ALBANIAN [TP Abby speaks <Albanian> fluently]], not [FP GREEK
[TP Abby speaks <Greek> fluently]]
Since leftward movement is involved in this derivation, it is correctly predicted that
stripping is sensitive to island conditions, but not clause-bound. Note that in this case,
the identity requirement to license deletion of the TP of the second clause cannot be
the one in terms of the merger given in (74). Rather, this construction involves
contrastive focus, and hence the identity requirement in question is more
appropriately characterized in terms of Parallelism.
Given this characterization of stripping, it is quite natural to take such a
fragment answer as (75b) as having the following representation:
(79) No, [FP ALBANIAN [TP Abby speaks <Albanian> fluently]]
Here, Albanian bears contrasted focus along with Greek, hence undergoing leftward
movement to create the semantic formula [λx. Abby speaks x fluently]. By applying
the same operation to the structure of (75a), thereby obtaining the same semantic
formula, we are entitled to delete the TP of (79) under the identity requirement in
terms of Parallelism. Then the paradigm of (64)-(66) falls into place under the present
analysis. Importantly, we are then led to the conclusion that the type of fragment
answers given in (75) is not the same species as those standard fragment answers that
can be dealt with under the in-situ approach. Thus, the strategy under consideration
adopted by Merchant (2004) to test island sensitivity of fragment answers seems to be
ill-advised.23
34 6 Conclusion
In this paper, I argued for the in-situ approach to sluicing, proposed by Kimura (2010),
by examining what differentiates sluicing from VPE in so-called island repair. I
argued, following Kimura’s original insights, that the crucial key to the relevant
distinction lies in whether a given elliptic construction involves movement of the
remannt wh-phrase or not rather than island repairability in terms of the size of the
elliptic sites involved, advocated by Fox and Lasnik (2003). Thus, sluicing is capable
of “island repair” since no movement is involved, whereas VPE is not capable of
“island repair” since the remnant wh-phrase undergoes overt movement. In order to
capture the fact that the merger type of VPE does not allow wh-extraction out of the
elliptic site, I proposed a LF identity condition to the effect that inner antecedents
cannot undergo merger with the traces of the corresponding wh-remnants. The latter
condition also accommodates the fact that the contrast type of VPE does allow
wh-extraction out of the elliptic site, since the relevant notion for identity is not
merger but rather Parallelism. Further, I have demonstrated that when wh-adjuncts are
involved as remnants, sluicing and VPE behave alike. I claimed that this is because in
such cases, merger is irrelevant, but rather such constructions are regarded as special
species of the sprouting type. Finally, I have argued that fragment answers are
amenable to the in-situ analysis, hence showing island insensitivity. From this
perspective, the type of construction Merchant (2004) deals with as a fragment
answer to an “implicit salient question” is reanalyzed as a species of stripping.
35 References
Abe, Jun. 1993. Binding conditions and scrambling without A/A' distinction. Doctoral
dissertation, University of Connecticut, Storrs, CT.
Abe, Jun. 2005. An economy condition on scrambling. Ms., Tohoku Gakuin
University.
Abe, Jun. 2008. Embedded sluicing in Japanese. In Pragmatic functions and syntactic
theory: In view of Japanese main clauses, a report for Grants-in-Aid for
Scientific Research, 121-174. Kanda University of International Studies.
Abe, Jun. 2009. How to probe expletives. Ms., Tohoku Gakuin University.
Abe, Jun. 2010. Oblique vs. remnant VP movement: A case of Japanese multiple
sluicing. Ms., Tohoku Gakuin University.
Abe, Jun and Hiroto Hoshi. 1997. Gapping and P-stranding. Journal of East Asian
Linguistics 6:101-136.
Abe, Jun and Norbert Hornstein. to appear. ‘Lasnik-effects’ and string-vacuous ATB
movement. In Ways of structure building, ed. by Myriam Uribe-Etxebarria and
Vidal Valmala. Oxford: Oxford University Press.
Agbayani, Brian. 2006. Pied-piping, feature movement, and wh-subjects. In
Wh-movement: Moving on, ed. by Lisa Lai-Shen Cheng and Norbert Corver,
71-93. Cambridge, MA: MIT Press.
Agüero-Bautista, Calixto. Diagnosing cyclicity in sluicing. Lingusitic Inquiry
38:413-443.
Bošković, Željko. 2002. A-movement and the EPP. Syntax 5:167-218.
36 Chomsky, Noam. 1965. Aspects of the theory of syntax. Cambridge, MA: MIT Press.
Chomsky, Noam. 1971. Some empirical issues in the theory of transformational
grammar. In Goals of linguistic theory, ed. by Stanley Peters, 63-130.
Englewood Cliffs, NJ: Prentice-Hall.
Chomsky, Noam. 1986. Barriers. Cambridge, MA: MIT Press.
Chomsky, Noam. 1993. A minimalist program for linguistic theory. In The view from
Building 20: Essays in linguistics in honor of Sylvain Bromberger, ed. by Ken
Hale and Samuel Jay Keyser, 1-52. Cambridge, MA: MIT Press.
Chomsky, Noam. 1995. The minimalist program. Cambridge, MA: MIT Press.
Chomsky, Noam. 2000. Minimalist inquiries: The framework. In Step by step: Essays
on minimalist syntax in honor of Howard Lasnik, ed. by Roger Martin, David
Michaels and Juan Uriagereka, 89-155. Cambridge, MA: MIT Press.
Chung, Sandra, William A. Ladusaw, and James McCloskey. 1995. Sluicing and
Logical Form. Natural Language Semantics 3:239-282.
Fox, Danny and Howard Lasnik. 2003. Successive-cyclic movement and
island-repair: The difference between sluicing and VP-ellipsis. Linguistic
Inquiry 34:143-154.
George, Leland. 1980. Analogical generalizations of natural language syntax.
Doctoral dissertation, MIT, Cambridge, MA.
Heim, Irene. 1982. The semantics of definite and indefinite noun phrases. Doctoral
dissertation, University of Massachusetts, Amherst. Published in 1989 by
Garland, New York.
37 Huang, C.-T. James. 1982. Logical relations in Chinese and the theory of grammar.
Doctoral dissertation, MIT, Cambridge, MA.
Kimura, Hiroko. 2010. A wh-in-situ strategy for sluicing. English Linguistics 27:
43-59.
Lasnik, Howard. 2001. When can you save a structure by destroying it? In
Proceedings of Northeast Linguistic Society 31, ed. by Minjoo Kim and Uri
Strauss, 301-320. Amherst, MA: GLSA, University of Massachusetts.
Lasnik, Howard. 2006. Multiple sluicing in English? Ms., University of Maryland. Lasnik, Howard. 2008. Repair by ellipsis revisited. lectures given at Nanzan
University.
Lobeck, Ann. 1990. Functional heads as proper governors. In Proceedings of North
East Linguistic Society 20, ed. by Juli Carter, 348-362. Amherst, MA: GLSA,
University of Massachusetts.
Merchant, Jason. 2001. The syntax of silence: Sluicing, islands, and the theory of
ellipsis. Oxford: Oxford University Press.
Merchant, Jason. 2004. Fragments and ellipsis. Linguistics and Philosophy 27:
661-738.
Merchant, Jason. 2008. Variable island repair under ellipsis. In Topics in ellipsis, ed.
by Kyle Johnson, 132-153. Cambridge: Cambridge University Press.
Nakao, Chizuru. 2009. Island repair and non-repair by PF strategies. Doctoral
dissertation, University of Maryland, College Park, MD.
Nakao, Chizuru, and Masaya Yoshida. 2007. ‘Not-so-propositional’ islands and their
38 implications for swiping. In Proceedings of WECOL 2006, ed. by Erin
Bainbridge and Brian Agbayani, 322-333. University of California, Fresno.
Nishigauchi, Taisuke and Tomohiro Fujii. 2006. Short answers: Ellipsis, connectivity,
and island repair. Ms., Kobe Shoin Graduate School and University of
Maryland.
Reinhart, Tanya. 1991. Elliptic conjunctions − Non-quantificational LF. In The
Chomskyan turn, ed. by Asa Kasher, 360-384. Cambridge, MA: Blackwell.
Reinhart, Tanya. 1997. Quantifier-scope: How labor is divided between QR and
choice functions. Linguistics and Philosophy 20: 335-397.
Rizzi, Luigi. 1990. Relativized minimality. Cambridge, MA: MIT Press.
Ross, John R. 1969. Guess who? In Papers from the Fifth Regional Meeting of
Chicago Linguistic Society, ed. by Robert I. Binnick, Alice L. Davison, Georgia
M. Green, and Jerry L. Morgan, 252-286. University of Chicago.
Saito, Mamoru, and Keiko Murasugi. 1990. N’-deletion in Japanese. University of
Connecticut Working Papers in Linguistics 3:87-107.
Sauerland, Uli. 1996 Guess how? In Proceedings of CONSOLE 4, ed. by J. Costa, J.
Goedemans, and R. van Vijver, 297-309. SOLE, Leiden.
Schuyler, Tamara. 2001. Wh-movement out of the site of VP ellipsis. Ms., University
of California, Santa Cruz.
Stepanov, Arthur, and Wei-Tien Dylan Tsai. 2008. Cartography and licensing of
wh-adjuncts: A cross-linguistic perspective. Natural Language and Linguistic
Theory 26:589-638.
39 Szabolcsi, Anna, and Frans Zwarts. 1993. Weak islands and an algebraic semantics
for scope taking. Natural Language Semantics 1:235-284.
Takahashi, Daiko. 1994. Minimality of movement. Doctoral dissertation, University
of Connecticut, Storrs, CT.
Tsai, W.-T. Dylan. 1994. On economizing the theory of A'-dependencies. Doctoral
dissertation, MIT, Cambridge, MA.
Department of English
Faculty of Letters
Tohoku Gakuin University
1-3-1 Tsuchitoi, Aoba-ku
Sendai 980-8511 Japan
jabe@tscc.tohoku-gakuin.ac.jp
40 Footnotes
1
See Fox and Lasnik (2003) for the argument that sluicing and VPE should be
derived from the same operation, namely PF deletion.
2
Lasnik (2001) further observes that even short wh-movement of a direct object
resists VPE, as shown below:
(i) They studied a Balkan language, but I don’t know which Balkan language (??they
did).
(Lasnik 2001:318)
Lasnik speculates that the slightly improved status of (i), in comparison with (2b) and
(3b), might be due to the alternative possibility of analyzing the relevant part as an
instance of pseudogapping.
3
This still leaves us with the possibility that successive-cyclic A-movement is
forced by the shortest steps condition, since no phase intervenes in such movement.
See Bošković 2002 and Abe 2009 for relevant discussions.
4
For just expository purposes, it is assumed here that wh-movement takes place
through intermediate Spec-CPs.
5
Strictly, Chomsky (1993) holds that such an assumption applies only to
wh-arguments and that for wh-adjunct chains, the intermediate traces must not be
deleted since they constitute part of uniform chains. I argue in Section 4 that this in
fact holds for wh-adjunct remnants in sluicing and VPE and hence that Fox and
Lasnik’s line of explanation is maintained in such cases.
6
Merchant (2008) tries to defend the island repair approach by accounting for
the ungrammaticality of such VPE examples as (2b) and (3b) in terms of “a ban on
41 eliding less than possible under wh-extraction”, (p. 140) which he names MaxElide:
(i) Let XP be an elided constituent containing an A’-trace. Let YP be a possible target
for deletion. YP must not properly contain XP.
That such an economical constraint is operative might be correct as a descriptive
generalization, but the nature of this constraint is far from clear, as Merchant himself
notes. Especially, it “is crucial to note that MaxElide applies only to XPs that contain
a wh-trace, since no similar blocking effect is found in the absence of wh-movement”
(p. 142), as shown below:
(ii) a. Ben knows that she invited Klaus, but her father doesn’t.
b. Ben knows that she invited Klaus, but her father doesn’t know that she did.
Furthermore, it is crucial that MaxElide does not apply to an elided constituent
containing an A-trace even though it is ultimately bound by a wh-phrase, as noted by
Lasnik (2001) with such an example as the following:
(iii) a. Someone solved the problem.
b. i. Who?
ii: Who did?
I believe that unless such rather unexpected restrictions are properly dealt with,
MaxElide remains to be a simple generalization that wants explanation. For this
reason, I will not discuss Merchant’s (2008) approach in terms of this constraint any
further in the text.
7
Fox and Lasnik seem to take (10b) as basically grammatical, though they do
not give any comment on the apparently degraded status of (10b) indicated by ??. By
42 the way, Lasnik (2008) judges (10b) as ?. I follow Fox and Lasnik in regarding (10b)
as grammatical, ascribing its degradation to any other independent factor(s).
8
Strictly, Schuyler (2001) reaches a more specific conclusion with respect to
licensing wh-movement out of VP ellipsis:
(i) Contrast-locality condition for VPE-extraction:
For wh-movement out of the site of VPE to be licensed, there must be a
contrastively focused expression in the c-command domain of the VPE-extracted
wh-phrase.
According to this generalization, (10b) should be bad since the contrastively focused
expression YOU is not located in the c-command domain of the VPE-extracted
wh-phrase which one. To support this formulation, Schuyler provides the following
examples:
(ii) a. *PETE knows which puppy you should adopt, but JAN doesn’t know which
one you should.
b. *SOME guests wondered what Jan would eat, but OTHER guests already knew
what she would.
I must leave this factual issue for future research, and continue to adopt Fox and
Lasnik’s (2003) data as a target for explanation. I thank the two anonymous reviewers
for bringing Schuyler’s (2001) work to my attention.
9
Chomsky (1986) provides the following pair of sentences to give support to
what he calls Vacuous Movement Hypothesis (VMH):
(i) a. This is a paper that we need to find someone [CP who understands t].
43 b. This is a paper that we need to find someone [CP who we can intimidate with t].
In (ia), the null relative operator can undergo movement through the Spec-CP of the
embedded relative clause under the VMH, which yields “the very weak CNPC effect
typical with someone as the head of the NP.” (p. 51) In (ib), on the other hand, “the
corresponding derivation is impossible since the specifier position of CP is occupied
by the fronted object of intimidate, and the expression is less acceptable.” (ibid.) See
Agbayani (2006) for the claim that the impossibility of local topicalization of subjects
follows under the VMH.
10
This raises the question whether the wh-phrase in question cannot undergo
Pied-Pipe in such a case. I argue in Section 3.2 that in the contrast type of sluicing,
the remnant wh-phrase undergoes movement to Spec-CP to satisfy Parallelism.
11
Note that even though the remnant wh-phrase what in (19c) stays in situ, the
fact that the whole embedded clause is interpreted as interrogative as well as the fact
that the remnant wh-phrase takes scope over the whole embedded clause is captured
by the wh-feature of that phrase having moved to the C head. These facts will follow
more straightforwardly under Abe and Hornstein’s (to appear) modified version of the
in-situ analysis, according to which sluicing involves movement of the remnant
wh-phrase, just like regular wh-questions, but it is the bottom copy of the resulting
chain that is pronounced; in other words, “covert” movement is involved in this
construction, hence free from island effects. This version of the in-situ analysis may
be superior to Kimura’s (2010) original version in that it can take care of more
intricate scope facts such as the fact, pointed out by Agüero-Bautista (2007), that
44 sluicing allows intermediate scope in the interaction of the remnant wh-phrase with a
universal quantifier, just like regular wh-questions. Agüero-Bautista takes this scope
fact as indicating that the remnant wh-phrase undergoes successive-cyclic movement
in sluicing. Under Abe and Hornstein’s (to appear) approach, this will mean that the
remnant wh-phrase undergoes successive-cyclic covert movement. I am indebted to
one of the reviewers for pointing out the importance of the points maid in this
footnote.
12
This licensing condition excludes such elliptic sentences as given below:
(i) a. *Robin saw someone, but I don’t believe that [TP e].
b. *Ralph knows that I went, but his wife doesn’t know whether [TP e].
13
The comparable English examples do not really show the relevant contrast,
since wh-adjuncts such as why and how cannot sit in situ even when no island is
involved, as shown below:
(i) *What did you buy how/why?
I assume that the ungrammaticality of this sentence has to do with the licensing of
in-situ wh-adjuncts in multiple wh-questions, hence not affecting the reasoning given
in the text.
14
The version of (28a) with (in) which (way) as a wh-remnant originally comes
from CLM. There is a twist of facts about the locality of wh-adjuncts as remnants of
sluicing. Lasnik (2008) notes that they exhibit more severe locality effects than island
effects, i.e., clause-boundedness, an observation he attributes to Benjamin Bruening.
See Section 4 for an account of this locality in terms of Parallelism.
45 15
One reviewer brought to my attention Sauerland’s (1996) work, which
demonstrates that the following generalization holds true:
(i) The sluicing construction cancels only the effect of strong islands on extraction.
Weak islands remain in the sluicing construction.
As an example of this generalization with a negative island, Sauerland provides the
following example:
(ii) a. As a child, I behaved well, and my parents can tell you how well.
b. *As a child, I didn’t behave well, and I don’t remember how well.
(Sauerland 1996:301)
Based upon this generalization, Sauerland claims that “strong island effects occur due
to the misapplication of syntactic operations” and that “weak islands reflect
interpretive violations as proposed by Szabolcsi and Zwarts (1993).” (p. 297) The
present in-situ analysis to the sluicing construction may endorse such an interpretive
approach to weak islands, since if the effects of weak islands were syntactic in nature,
it would expect that those effects should also go away.
16
It is beyond the scope of this paper to discuss other types of sluicing that
behave differently from the merger type in locality effects. See CLM for the claim
that what they call the sprouting type of sluicing is island-sensitive. See also Lasnik
(2006) for the observation that in English multiple sluicing, the second remnant
wh-phrase shows signs of rightward movement, and see Abe (2010) for a possible
account of why this is the case, contrary to the remnant wh-phrase in standard single
sluicing.
46 17
See Kimura (2010) for further consequences of her in-situ analysis of sluicing.
Among them, she discusses what Merchant (2001) calls swiping, in which a sluiced
wh-phrase and its governing head P are inverted, as illustrated below:
(i) He was shouting to someone, but it was impossible to tell who to.
Kimura analyzes this construction as being derived by moving the sluiced wh-phrase
who to the Spec of the P to in order to satisfy the PF adjacency condition given in
(15).
As one of the reviewers correctly points out, the in-situ analysis appears to
have great difficulty dealing with the P(reposition)-stranding generalization
established by Merchant (2001), according to which P-stranding under sluicing is
possible only in those languages that allow P-stranding in regular wh-questions. How
serious this problem is to the in-situ analysis, however, really depends on how to
capture the (im)possibility of P-stranding. If “P-stranding is a purely movement
effect,” as the reviewer claims, then the P-stranding generalization might be fatal to
the in-situ analysis. Suppose, however, that a relevant condition is operative in the PF
component and that it is characterized in such a way that in the sequence P^DP, it is
impossible to pronounce only P or DP. Then, in those languages that do not allow
P-stranding, it is prohibited in regular wh-questions because in the sequence
P^wh-phrase, only P is pronounced, and it is also prohibited in sluicing because in the
same sequence, only the wh-phrase is pronounced.
18
Compare the sluices in (38) with the corresponding full-fledged wh-questions,
given below:
47 (i) a. I know that Meg’s attracted to Harry, but they don’t know who Meg’s attracted
to.
b. Since Jill said Joe had invited Sue, we didn’t have to ask who Joe had invited.
CLM claim that the well-formedness of (ia, b) “reveals that there is nothing wrong
with the interpretation ultimately intended for [38a, b].” (p. 253) The requirement that
inner antecedents must be indefinites is somewhat insecure when a most-phrase is
involved as an inner antecedent and the corresponding wh-remnant takes the form of
which-phrase, as one of the reviewer points out that (39b) is acceptable with which as
the wh-remnant and becomes even better with which ones.
19
The above account crucially assumes the trace theory of movement rather
than the copy theory standardly assumed in the current minimalist theory. This might
suggest that, contrary to what is widely believed, the trace theory is the right one for
characterizing a chain produced by movement. Alternatively, it might suggest that the
above account is not incompatible with the copy theory, but rather that the way the
relevant restriction on merger is formulated should be changed. Instead of (41), it
might be proposed that an inner antecedent cannot undergo merger with the
corresponding wh-phrase if the latter constitutes a non-trivial chain. Either option will
do for the present purposes. Since I have not found any obvious reason for choosing
one over the other, I will leave this matter undecided here and simply assume (41) as
well as the trace theory of movement in what follows. As will be seen directly, it is
easier and more straightforward to characterize the identity condition relevant for the
contrast type of sluicing and VPE under the trace theory of movement.
48 20
Given that who else in (48) does not have to move to Spec-CP to satisfy the
PF adjacency condition given in (15), it might be the case that the wh-phrase
undergoes covert focus movement, exactly like the corresponding focused phrase in
the antecedent clause. I will leave a further examination of this possibility for future
research.
21
See Stepanov and Tsai (2008) for the similar argument, based upon the
following contrast, that the wh-adjunct why is base-generated in Spec-CP:
(i) a. *John talked about a Balkan language, but I don’t remember which Balkan
language he did.
b. John talked about a Balkan language, but I don’t remember why he did.
However, they provide the following example, claiming that how is a VP-adverb,
hence behaving like an object wh-phrase in that it causes a violation of either
Parallelism or crossing barriers such as TP and AspP under Fox and Lasnik’s (2003)
framework:
(ii) *John talked about a Balkan language, but I don’t remember how he did. So far I have not been able to trace any good reason for such a discrepancy of data
point. Obviously, further research of relevant data is necessary.
22
This section is added to a previous version of this paper, following one
reviewer’s suggestion.
23
The other strategy adopted by Merchant (2004) to test island sensitivity of
fragment answers is concerned with multiple fragment answers. But this strategy also
seems ill-advised, since it makes the unmotivated presupposition that single and
49 multiple fragment answers should behave the same with respect to locality effects.
Such a presupposition is clearly false in the case of sluicing, as noted in fn. 16.
50 
Download