SPROUTING SUBDUED - Linguistics at Marylandling.umd.edu/~bradl/publications/Larson_sprouting.pdftle...
Transcript of SPROUTING SUBDUED - Linguistics at Marylandling.umd.edu/~bradl/publications/Larson_sprouting.pdftle...
SPROUTING SUBDUED
BRADLEY LARSON
[email protected] UNIVERSITY OF MARYLAND
1. Introduction
Sprouting instances of sluicing are about as under-studied as they are theoreti-
cally interesting. Aside from the forays by Chung XXXX, Chung, Ladusaw, and
McCloskey (XXXX), Lasnik and Park (XXXX) and Chung 2011 there has been lit-
tle full-length investigation of the phenomenon. Luckily, the previous analyses not
only have fascinating conceptual repercussions but also have developed a comfortable
empirical base that promises fecund future exploration.
In this paper I will resuscitate the sluicing analysis of Merchant, XXXX in light of
Chung, Ladusaw, and McCloskey’s criticisms of it when it comes to sprouting. I will
discuss both analyses and their shortcomings and eventually show that an alteration
of Merchant’s analysis can straightforwardly avoid the problems it faces as well as
those that Chung et. al.’s does.
Another consequence of this approach is that it proffers up yet another another
argument in favor of the decomposition of the Merge operation as proposed in Horn-
stein (2008).
In essence, I will argue for a Merchant-style analysis: that sprouting is derived
by phonological ellipsis of a fully structured TP and that this ellipsis is licensed by
1
mutual entailment of the relevant TPs, but with a slight modification. That is, the
sentence in (1) is derived like in (2).
The wh-phrase initially extraposes to the right of its VP because it is new infor-
mation with respect to the antecedent clause. This is to say, everything else in the
elided TP is entailed by the antecedent TP (and vice versa) except this wh-phrase.
The wh-phrase being the odd element out in not being entailed by the antecedent
TP, moves to adjoin to the VP.
I argue that this trace avoids being taken into consideration when the TP is elided
due to its only partially being introduced into the structure (Having undergone Horn-
stein’s Concatenate, but not his Label operation). This also avoids the potential
problem of deleting a non-constituent, which is what looks like is happening in (2).
(1) Ivan was eating, but I don’t know what.
(2) Ivan was eating, but I don’t know whati [TP Ivan was [V P eating ti]∧[ti]].
The major constraint on sprouting under this account is that elements can be
sprouted only if they can extrapose. This condition will turn out to have sufficient
predictive power to handle the problematic cases for both current analyses.
But first, let us discuss the current analyses.
2
2. Current Analyses
Chung, Ladusaw, and McCloskey (CLM) sluice away extraneous details and present
a clear summary of the main approaches to sprouting in the beginning of their article.
In (3) below I offer an amended version of their summary.1
(3) a. Approach 1 The ellipsis site is empty in overt syntax and is supplied with
material by covert movement of the relevant material into the ellipsis site
followed by covert downward movement of the sprouted element.
b. Approach 2 The ellipsis site has internally structured material throughout
the derivation and is deleted phonologically under a condition of semantic
mutual entailment with its relevant antecedent.
2.1. Approach 1. Take the example in (4). For the first approach this sentence
has, at what CLM call Surface Structure, the structure found in (5).
(4) Ivan was standing, but I don’t know where.
(5) Ivan was standing, but I don’t know [CP where [TP ] ]
In covert syntax, the antecedent TP is copied and moved into the empty TP like
in (6). The wh-phrase where is then moved into what is in common parlance its
base-generated position (7). These movements happen covertly. Thus (4) is derived
with the relevant interpretation without the second conjunct’s TP being pronounced.
1 I omit discussion of analyses in which the ellipsis site is without internal structure at any levelof representation and anaphoric to its antecedent. See Merchant 2001 and Chung, Ladusaw, Mc-Closkey 1995 for details
3
(6) [TP Ivan was standing], but I don’t know [CP where [TP Ivan was standing ] ]movement
(7) [TP Ivan was standing], but I don’t know [CP where [TP Ivan was standing
where] ]movement
2.2. Approach 2. The second approach derives the sprouting sentence in the fol-
lowing manner. The last step of the first approach is now the first step: the sprouted
wh-word wh-moves in the normal fashion from its base-generated position (8). The
TP from which it moved is later phonologically deleted (9).
(8) [CP where [TP Ivan was standing where] ]movement
(9) [CP where [TP Ivan was standing where] ]
This deletion is licensed semantically if there is a discourse TP antecedent which
entails the deleted TP and vice versa: something like (10).
(10) [TP Ivan was standing], but I don’t know [CP where [TP Ivan was standing
where] ]
Thus the sprouting sentence is derived by recourse to semantic entailment that the
licensing of the deletion operation depends on.
2.3. Discussion. One effect of the first approach is that the content of the ellipsis
will be syntactically identical to that found in the antecedent because it will be the
exact same material, just moved. This contrasts with the second approach in which
4
the syntactic material of the ellipsis site need not be identical to that the relevant
antecedent material.
This is the way the approaches stand in the literature, but the means by which the
ellipsis site derives its interpretation does not always entail a particular condition on
the ellipsis. That is, the fact that the second approach as currently stands requires
mutual entailment (e-GIVENness in Merchant’s terms) as a condition for the ellipsis
does not logically follow from the fact the phonological deletion of the material in
the ellipsis site has occurred.
It could be the case that the material is phonologically deleted as in the second
approach but the condition on that ellipsis is (mostly) syntactic and not semantic.
This will be the approach taken here.
Before going into the details of the proposed account, let us investigate the short-
comings of the current ones that urge us toward a new account.
3. Problems with Current Analyses
3.1. Approach 2. Chung, Ladesaw, and McCloskey provide strong arguments against
the second approach so far as the mutual entailment condition is concerned. I will
rehearse them summarily here and offer a few further problems with the approach
in general.
3.1.1. e-GIVENness Violations. First, there are sprouting cases that clearly violate
the mutual entailment condition suggested by Merchant. Take an example like (11)
from CLM. Although the elided TP entails the antecedent one, the antecedent TP
does not entail the elided one.
(11) He put in a bid, but I couldn’t tell one whose behalf.
5
The unelided TPs are given below and the entailments can be easily discerned:
(12) a. He put in the a bid
b. He put in a bid on someone’s behalf
3.1.2. Chung’s Generalization. Further there is what is considered “Chung’s General-
ization". Even in languages that generally allow preposition stranding, the following
sentence is unacceptable:
(13) *They’re jealous but it’s unclear who.
Under the PF deletion approach, the ungrammaticality of the above sentence goes
unexplained. First, the existence of the preposition of is arguable solely for case
reasons and has no semantic import (Chomsky, XXXX). CLM point out that the PF
deletion account cannot distinguish between the derivation in (14) from the derivation
in (15) on its own terms, especially considering the lack of semantic import in of.
(14) *They’re jealous but it’s unclear who [TP they’re jealous of who ]
(15) They’re jealous but it’s unclear of who/who of [TP they’re jealous of who ]
3.1.3. More e-GIVENness Violations. Even in cases where the mutual entailment of
the TPs seems relatively trivial, like in example (4) repeated here as (16), this might
not be the case. Given the background above, the two TPs that mutual entail each
other are in (17).
(16) Ivan was standing, but I don’t know where.
(17) a. Ivan was standing
6
b. Ivan was standing somewhere
Now it seems clear to us that if one is standing that one must be standing somewhere.
But this is beside the point. The question is whether standing entails standing
somewhere as far as the grammar is concerned. This certainly might be the case,
but as such it is a crucial, yet unstated assumption on Merchant’s part.
The split between human intuition and grammatical logic is made more clear in
the following example (18). The TPs are given in (19) and there is certainly no
mutual entailment here.
(18) I exist, but I don’t know why.
(19) a. I exist
b. I exist for a reason
So CLM’s criticism can be extended from a subset of sprouting types to the general
case.
3.1.4. Island Effects. Finally, though CLM mention what is know as Albert’s Gen-
eralization, they do not explicitly show that the PF deletion account cannot account
for it. The generalization states that sprouted wh-phrases are subject to islands as
seen in (20).
(20) *I saw the movie that showed Ivan eating, I just can’t remember what
This contrasts with non-sprouting examples (so-called “merged" examples) which
obviate islands (21).
7
(21) I saw the movie that showed Ivan eating something, I just don’t remember
what.
The PF deletion analysis holds that the deletion is what obviates the island. But
such accounts analyze the two sentences above in the same way (22) and as such do
not account for the differences above.
(22) a. I saw the movie that showed Ivan eating, I just can’t remember what
[TP I saw the movie that showed Ivan eating what]
b. I saw the movie that showed Ivan eating something, I just can’t remem-
ber what [TP I saw the movie that showed Ivan eating what]
If it is PF deletion that rescues sentences from island violations, sprouting should be
no exception.
3.1.5. Summary. It has been seen that Merchant’s PF deletion approach. The re-
liance on e-GIVENness to license the ellipsis cannot handle sprouting cases because
the TPs that are relevant seem to have asymmetrical entailments, even in the general
case. Also, some cases in which there seems to actually be mutual entailment fail
to license deletion (Chung’s Generalization). Second, such approach fails to capture
differential island obviation effects between regular sluicing and sprouting.
3.2. Approach 1. I take CLM’s criticisms of Merchant’s approach applied to sprout-
ing to be fairly damning. I will alter his analysis so as to maintain its general viability
in the following sections. First however shortcomings of CLM’s approach must be
discussed. Given that their approach involves an entirely novel means of structure
8
building, the criticisms here will only be relatively to my understanding of their
formulation. That said, I think there are clear enough cases here.
3.2.1. TP movement. First, it is clear that the movement of the antecedent TP into
the ellipsis site is a non-canonical type of movement. There is no obvious c-command
relation between the two copies of the antecedent TP and the movement is more or
less downward. This should be a violation of cyclicity.
For the sake of argument let us say that since this is covert movement, cyclicity
does not apply and movement does not need to move to a c-commanding position.
Even if this is granted, the result of the movement is not something that is well
formed. Recall the structure prior to the TP movement (23).
(23) Ivan was standing, but I don’t know where [TP ]
In the above structure, there exists already a TP position for the antecedent to
move to. What moves to this position must be a constituent so the result would
presumably be (24). In tree form the lower TP would have the structure like (25).
(24) Ivan was standing, but I don’t know where [TP [TP Ivan was standing] ]
(25) TP
T TP
9
It is unclear to me how the above structure is well-formed. What sectional aspect
of the T head is satisfied by the TP that it takes as a complement? Why is this
apparently the only instance of such a construction?2
Say we were to alter CLM’s approach and say that instead of an empty TP position
that the antecedent TP moves into, it was instead the case that the C-head had an
empty complement position (26). When the TP moves into that empty position we
have a well formed object:
(26) Ivan was standing, but I don’t know where [C′ C ]
(27) Ivan was standing, but I don’t know where [C′ C [TP Ivan was standing]]
(28) CP
C TP
This creates another problem. Certain C heads subcategorize for particular types
of TPs. That is, the complementizer that subcategorizes for a finite TP whereas
for subcategorizes for a non-finite TP for example. Sprouting is indeed possible in
instances where the antecedent TP is one tense and the elided TP is a different tense
(29).
(29) Ivan decorates cakes, but I don’t know how [TP to decorate cakes]
2 It could conceivably be the case that the moved TP substitutes for the the TP already there,but this would violate Full Interpretation (Chomsky, 1986). If does not violate it, it will surely fallafoul in the subcategorization mismatch cases discussed presently. It also presumably could notadjoin to the T head because then the subsequent movement of the sprouted wh-phrase into theTP should violate an adjunct island.
10
Were the TP to move into the configuration in (28), this would create a subcatego-
rization mismatch. CLM might suggest that this sort of mismatch can be obfuscated
sufficiently by working covertly, but how this works exactly is unclear to me given
that the subcategorization features presumably still exist in covert syntax. That is,
when the wh-element moves downward after this step, it does so for featural reasons.
So at this time in the derivation syntactic features must still be live.
3.2.2. Island effects. Following the movement of the TP from the antecedent into the
other clause, CLM propose that the sprouted wh-phrase then moves from its [Spec,
CP] position into its lower position (30).
(30) ...I don’t know [CP where [TP Ivan was standing where] ]movement
This movement is no different from any canonical wh-movement and is subject to
the same restrictions. Movement across an island is prohibited in regular upward
movement and the same is the case for this downward movement. Were the wh-
phrase to move into an island, we would expect the sentence to be bad. And this is
the case:
(31) a. *I saw the movie that showed Ivan eating, I just can’t remember what
[TP I saw the movie [Island that showed Ivan eating what]]movement
b. *I saw the movie that showed Ivan eating, I just can’t remember what
So, when this movement happens island still exist as things to be moved into to
deleterious effect. Given that, it should be impossible to move the TP into an island
11
in the step before the wh-phrase does. Say there were a structure like (32) prior to
movement of the antecedent TP.
(32) Ivan likes Ivy, but at first I didn’t believe [Island the claim that Iris knew
why [TP ]
Movement into that island should effect an island violation just as the movement of
the wh-phrase in (31) did. However, the resulting sentence is acceptable. There is
no apparent island violation
(33) a. [TP Ivan likes Ivy], but at first I didn’t believe [Island the claim that Iris
knew why [TP Ivan likes Ivy ]]
b. Ivan likes Ivy, but at first I didn’t believe [Island the claim that Iris knew
why.
3.2.3. Multiple Sluicing and Minimality. Given this approach the following sentence
should show minimality-induced unacceptability, but it does not:
(34) Ivan was talking, but I’m not sure who to about what.
The above sentence, an example of swiping (Merchant, XXXX), involves the sluicing
of two wh-phrases. In order to be pronounced, the wh-phrase about what must be
outside of the lower TP at spell-out to PF, before the movement of the antecedent
TP (35).3
(35) Ivan was talking but I’m not sure who to about what [TP ]
3 The fact that English does not normally allow [Spec, CP]s to be overtly doubly filled with wh-phrases should be cause for concern here as well
12
After the TP has moved, in order to satisfy its wh-feature or at least be interpreted
in the correct location, the wh-phrase who to must move into the VP. In doing so it
will cross-over the wh-phrase about what which I believe would be an instance of a
minimality violation under this approach.4
Another example is shown below that does not involve swiping:
(36) I had a list of who met the candidate when, but I lost track of who for how
long.
The above sentence (with both clauses having a pair-list reading) is acceptable to me,
but how would CLM’s approach handle this sentence. Again, deriving this sentence
under a CLM-style regime would induce a minimality violation. The wh-phrase who
must move across the wh-phrase for how long after the TP is copied and moved from
the antecedent clause.
3.2.4. Indefinite Interpretations. Given that under this approach it is the self-same
copy of the TP in the antecedent and second clause, CLM make the intuitive claim
that an indefinite within the TP should maintain a consistent meaning across its
positions. That is, in (37) the someone should be interpreted as the same person in
both clauses
(37) [Clause1 [TP Someone likes Iris]] ... [Clause1 [TP Someone likes Iris]]
This will contrast with traditional ellipsis cases like VP-ellipsis. In these cases an
elided indefinite can be interpreted as a different person than the indefinite in its
4 It should not matter that this happens in covert syntax given CLM’s discussion of wh-islandeffects.
13
antecedent. Example (38) can have the interpretation in which Ivy and Iris like two
different people.
(38) a. Ivy likes someone, and Iris does [V P like someone], too
b. Ivy likes someone, and Iris does, too
CLM offer examples which they claim show that this is borne out. They claim
that the sentence in (39) can only have the interpretation in which there is only one
suspect.
(39) Jill asked where someone had committed a crime, and Jack asked when.
I find the reading in which there is only one suspect possible, but unfortunately I
find it very easy to get a reading in which there are two suspects and two crimes.
Suppose there were a reported in Seattle named Jill and one in Washington DC
named Jack. Jill was curious as to where a crime had been committed and so asked
about it. Unbeknownst to here Jack did a similar thing in Washington DC about
the time at which a different crime had been committed. The sentence in (39) could
be truthfully and felicitously uttered to describe the situation just laid out.
So, in this case it seems that sprouting seems to function like a type of ellipsis.
They make a similar case about the following sentence:
(40) Where someone commits a crime doesn’t determine how.
In the above sentence, I agree that it is strikingly difficult to get a reading in which
there are two different people committing crimes. Unfortunately in the non-sluiced
version of the sentence (41) that same reading is equally unavailable.
14
(41) Where someone commits a crime doesn’t determine how someone commits
a crime.
3.2.5. Summary. In this subsection is was shown that the CLM approach to sprout-
ing has some substantial problems. One, it is unclear how exactly the lowering of
TP is to function exactly given current theoretical assumptions (even after granting
this odd downward movement). Second, TP movement is expect to be subject to
syntactic islands under this account, but the facts do not bear this out. Third, this
account cannot easily handle mutliply-sluicing wh-phrases given that they must all
move from [Spec, CP] positions into the lower phrase. This downward movement is
claimed to work identically to upward movement and as such we would predict min-
imality violations or violations concerning multiple multiple wh-phrase movement in
English. Finally, the predictions that CLM make with respect the constraints on
indefinite interpretation are not borne out empirically.
4. New Approach
In explicating this proposed approach I will address each of the problems that
Merchant’s analysis as it stands faces. The solution to the first problem will feed
solutions to the other ones. By the time all the problems have been address and
solved, the basis of the current proposal will be clear.
4.1. Mutual Entailment. The first problem with Merchant’s approach is that the
TPs under sprouting do not entail each other. Let us a take a clear case like (42) in
which the speaker cannot remember what sort of prize or award Ivan had won.
(42) I knew Ivan had won, I just couldn’t remember what.
15
It is clear that Ivan winning does not entail Ivan winning a prize or award. What it
is that Ivan had won is something new introduced into the discourse in the second
conjunct. This is a shortcoming of Merchant’s constraint of e-GIVENness. How
could the second TP have been deleted when it obviously fails to meet this criterion?
Following Schwarzchild (XXXX), an element cannot be GIVEN (entailed by prior
discourse) if it is F(or focus)-marked. Elements that are entirely new in the dis-
coursed will be F-marked and F-marked elements cannot be GIVEN. When we com-
pare the TPs in the above sentence, it is clear that the wh-phrase in the second one
is a discourse new element:
(43) a. Ivan had won
b. Ivan had won what
Given that the wh-phrase is not going to end up being entailed by the previous
discourse, it must be F-marked like in (44) prior to movement. Anything else would
cause a crash presumably.
(44) [TP Ivan had won [what]F ]
This framing of the situation allows us to reconsider Merchant’s licensing condition
of ellipsis. Instead of ellipsis being licensed only if the constituent is e-GIVEN, I posit
(45).
(45) Exhortation on ellipsis: If Ellipsis is going to occur, elide as little non-GIVEN
material as possible
We might want some way to cash this out in the syntax. It could be the case
that merely marking an element as F-marked in the syntax suffices to avoid the
16
entailment problem. F-marking may be a code for, roughly, “Ignore this element
when computing entailment". But were this the case it would be difficult to explain
the fact that F-marked elements consistently avoid elision. In (46) for VP-ellipsis
and (47) for sluicing, discourse new/contrastive adjuncts avoid deletion.
(46) Ivan ate quickly and Ivy did eat slowly.
(47) I know what Ivan ate quickly but I don’t know what Ivan ate slowly.
Note that in the above cases it is exactly what escapes ellipsis by adjunction that is
contrastively focused. In other words, what is not entailed by the antecedent to the
ellipsis is not elided but is instead focused.
In order to handle the case of Ivan winning something above, we should want the
complement of win to function like the adverbs above.
Say it were the case that the F-marked complement extraposed out of its base-
generated position and adjoined to VP, like in (48).5 Now, qua adjunct, the wh-
phrase is of a type with the adjuncts above.
(48) [TP Ivan had [V P won t]∧[what]F ]]
It is intuitively plausible that the TP can be targeted to the exclusion of the F-marked
element for both deletion deletion and entailment computations.6
There is no direct evidence of this movement as it is string vacuous, but we can
find evidence for it investigating island effects, the second problem with Merchant’s
analysis.7
5 Take the wedge to indicate adjunction.6 Though it is the case that a non-constituent is being elided here, in a following section I willpropose a way to avoid this apparent problem.7 Further evidence for this idea is provided in upcoming sections.
17
4.2. Conditions on Extraposition. In the above section it was suggested that
F-marked arguments extrapose. But certainly there are phrases that cannot ex-
trapose. For instance, subjects cannot extrapose. That is, something like (49) is
unaccepatable:
(49) *There were ti eating some cheeseburgers [big fat men]i
To give the idea that sprouted items have extraposed some teeth, I posit the ex-
hortation below. Given this, we expect that spouting will only occur with elements
that can otherwise extrapose and elements that cannot will not be able to serve as
sprouted elements.
(50) Exhortation on extraposition: F-marked elements must have extraposed if
they would have otherwise been in an ellipsis site.
The above exhortation has a distinct move α feel to it and as such will end up
mediating PF and LF requirements in a way that does not require communication
between them. That is, extraposed elements will be able to be excluded from deletion
on the PF side due to their being adjuncts. These same extraposed elements will
have done so because they were not entailed by the discourse. That new material
will be excluded from ellipsis falls out from this: As little non-GIVEN material as
possible will be deleted and this material will be in a position to evade deletion if it
has extraposed.
4.3. Island Effects. Recall that Merchant’s analysis of sluicing fails to predict the
differential island sensitivity between regular sluicing and sprouting. However, if the
18
sprouted element were to have move to a position where it would not be elided, the
island effects should still hold were it to move from that position.
After extraposing the F-marked wh-phrase we are left with (50) above. Ellipsis
can work on the TP to the exclusion of the newly minted adjunct as in (51).
(51) [TP Ivan had won t what]
How it can delete a ostensible non-constituent will be discussed later. For now
what is important is that any further movement of the wh-phrase will leave an un-
elided trace in its wake. If this movement takes place across an island, the sentence
should be bad. And this is the case we find in sprouting: it is subject to islands.
(52) *I saw the movie that showed Ivan eating, but I can’t remember what
The ellipsis site is ambiguous as to whether the trace of the wh-phrase is elided or
not, but the extraposition of F-marked elements precludes the structure in which all
island-internal traces of the wh-phrase are within the ellipsis site.
This is very similar in spirit to the CLM approach, just couched in a Merchant-style
derivation without recourse to novel theories of movement.
4.4. Chung’s Generalization. Recall that the mutual entailment license for dele-
tion led to problems concerning preposition stranding. That is the following deriva-
tion should be fine (there is relevant e-GIVENness) but the sentence is quite bad.
(53) They’re jealous, but we don’t know who [they’re jealous of].
Under the extrapostion account, the fact that the above sentence is bad follows
straightforwardly. The actual structure of the sentence is in (54). The wh-phrase
19
has extraposed and adjoined to the soon to be elided structure so as not to take part
in the deletion. In doing so it left behind its selecting preposition.
(54) They’re jealous, but we don’t know whoi [they’re jealous of ti ti].
There are two (probably related) reasons why such a derivation would be ruled out.
First, as Schwarzchild claims, “F-marking of an internal argument of a head licenses
the F-marking of the head" (Schwarzchild, XXXX:144).
Since the complement of the preposition of above is F-marked, so should the
preposition itself. Since the entirety of the prepositional phrase is F-marked, under
this analysis the whole phrase should extrapose. This has not happened in the above
structure and it should be thus ruled out.
Another way to rule out the above derivation is to note the following. Although it
is exceedingly easy to extrapose a prepositonal phrase (55), it is exceedingly difficult
to extrapose the complement of a preposition (56). But in the derivation in (54)
exactly this is done.
(55) Ivan talked yesterday to Ivy.
(56) *Ivan talked to yesterday Ivy
It seems that whatever might prompt the complement of a preposition to extrapose
requires that its prepositional head move as well.
4.5. Summary. In this section I have laid out the basics of the altered PF deletion
approach to sprouting. It address the problems with the Merchant-style analysis of
sluicing has when met with sprouting examples. In the following section I will extend
20
the analysis and show that it can account for the other aspects of sprouting in the
literature.
4.6. How to Exclude Adjuncts. As noted before, it is a simple empirical fact that
ellipsis can target XPs to the exclusion of adjuncts. But how can this happen?
Hornstein (2009) presents the following conundrum. According to Bare Phrase
Structure (BPS) (Chomsky, 1995), there can only be one maximal projection per
head. Prior to BPS, this was not the case and in particular adjunction extended the
tree but did not change the bar level information. As seen in (57), an adjunct could
adjoin to a VP and the label dominating that would in turn be another VP.
(57) Ivan [V P [V P saw Ivy] on Saturday]
This was advantageous. Certain operations only worked on maximal projections, say
VP-ellipsis. In the above structure VP-ellipsis could operate on the inner (22a) or
outer (22b) VP.
(58) Iris saw Ivy on Sunday...
(59) a. ...and Ivan did on Saturday.
b. ...and Ivan did, too.
But with BPS, we can no longer capture these facts. What is considered a maximal
projection is now relative and not inherent to any node. As such, the structure in
(20) only has one maximal projection, the outer VP. We no longer have a means of
operating on the VP to the exclusion of the adjunct.8
8 See Hornstein, 2009 for arguments against Chomsky’s reformulation of adjunction which could inprinciple avoid this problem.
21
(60) Ivan Ivan [V P [saw Ivy] on Saturday]
5. Decomposed Merge
To solve this dilemma, Hornstein proposes a decomposition of the Merge oper-
ation9. Merge, as construed in Chomsky, 1995 takes two syntactic elements and
combines them, projecting one of them as the label of said combination (61).
(61) Merge(x,y) –> {x,{x,y}}
Hornstein instead posits that the above operation should be broken down into two
operations: Concatenate (62) and Label (63). The Concatenate operation takes two
atomic syntactic units and combines them into a complex of atomic units. Label
makes said complex atomic itself by choosing one of the elements of the Concatena-
tion operation to serve as the label of complex.
(62) Concatenate(x,y) –> {x,y}
(63) Label(x,{x,y}) –> {x,{x,y}}
According to the theory, normally both of these operations are carried-out, but
with adjunction this is not the case. Adjuncts, not being necessary to the derivation,
do not necessarily have to undergo Label.
This decomposition allows for a elegant account of the differential behavior of
adverbial modification. When an adverb Concatenates with a verb and does not
project (64), the verb+adverb complex is, in Hornstein’s words, “invisible" to the
9 Precursors to this theory can be found in Chametzky (2000) and Uriagereka (2002) and thetheory is further discussed in Hornstein and Nunes (2008). See Larson (2010) for an extension tocoordination
22
rest of the structure.10 So when an operation like VP-deletion targets a VP with a
Concatenated adverb, the VP deletes leaving the adverb behind (65).
(64)
VPrun
Aquickly
(65) Ivan ran slowly and Ivy did quickly.
When an adverb is both Concatenated and Labeled into the structure (66), VP-
deletion applies to the adverb as well (67).
(66) XP
Xrun
Aquickly
(67) Ivan ran slowly and Ivy did too.
5.0.1. Application to Sprouting. We can straightforwardly apply the above discussion
to adjunct wh-phrases attaching to the VP. So long as they do not Label, the ellipsis
will leave them unscathed. Take ‘∧’ to signify Concatenation without Label, we can
analyze sprouting more accurately as (68) below.11
(68) Ivan was talking, I just can’t remember wheni [IP Ivan was talking]∧[ ti ]
10 Take the dashed line to indicate Concatenation with Labeling. Note that there is no c-commandrelation between the two atomic elements.11 It is not relevant here where the adjunct adjoins, but I assume it to be to the VP.
23
5.0.2. Sprouting Complements. It is of course not the case that only adjunct wh-
phrases can be sprouted. As seen in (69), complements can just as easily be sprouted.
The sprouted what is interpreted as the complement of an elided eat.
(69) Ivan was eating, but we couldn’t make out what.
In this account, the sprouted complement wh-phrase undergoes extrapostion. I take
this movement to result in an adjunction structure following Stowell XXXX and
others (Drummond, XXXX Hunter, XXXX, etc.).
Instance where sprouting occurs with noun complement wh-phrase works in a
similar fashion. For the example below, the noun complement wh-phrase need only
extrapose to adjoin to the phrase that it is a complement to for this analysis to work.
This allows us to avoid deleting the PP without resorting to long distance rightward
movement.
6. Extending the Analysis
In this approach I will go various other facts of sprouting and show that this
approach easily and accounts for them, and in an explanatory fashion.
6.1. Fixed Diathesis Effects. CLM rely on there being two different lexical entries
for verbs like serve in order to account for the fact that following sentence in (70) is
bad
(70) *Ivan served the food, but I don’t know who
The sentence above is to be interpreted as (71). For CLM the ellipsis is not licensed
because, since there are two lexically different serves, the TP copying and movement
24
could not have taken place. That would require there be two instances of the same
word.
(71) Ivan served the food, but I don’t know who Ivan served the food.
Under the extraposition analysis, the facts fall out without recourse two two ho-
mophonous verbs. Say that the serves in (71) are the same lexical item. The indirect
object is optional in the sketch of a lexical entry in (72).12
(72) serve: <server> (<diner>) <meal>
The tree structure for the second TP would be something like (73) prior to movement.
(73) TP
T vP
server v
v VP
diner V
V meal
Under this analysis the diner argument would need to extrapose to survive the
ellipsis. However, extraposition of this argument is generally not allowed (74).
(74) *Ivan served the meat many big hungry oafs.
12 See Merchant for argument that the indirect object of serve is optional while the direct objectis not.
25
Any other effects of fixed diathesis can be reduced the same ban on extraposition of
a complement to a preposition as discussed before.
In instances where extraposition is not a potential cause for ungrammaticality, we
expect swiping of indirect objects to be fine. In German, the indirect object analogue
of the one in above sentence can easily appear to the right of the direct object.
(75) Ivan
Ivanhat
hasdas
the.ACCFleisch
meatdem
the.DATdicken
fatMann
manserviert.
served.Ivan served the fat man meat.
Given that (75) is acceptable, we expect the analogue of example () to be as well:
(76) Ivan hat das Fleisch serviert, aber ich weiss nicht wem.
Ivan has the.ACC meat served, but I know not who.DAT
Ivan served the meat but I don’t know who to.
As it stands, this approach to fixed diathesis effects is superior to that in CLM
because it captures the same data without recourse to multiple homophonous nouns.
6.2. Multiple Sluiced Elements. Discussed earlier with respect to CLM’s ap-
proach, we noted that sentences like (77) were acceptable.
(77) Ivan was talking, but I’m not sure who to about what.
We now easily derive such sentences. For the above case, first the wh-phrase who
moves to some vP position. Following this, the lower VP extraposes and as such the
evades deletion like in (78).
(78) ... [vP whatj Ivan was talking ti]∧[V P to tj V about what]i
26
As the to who phrase c-commands the about what phrase, minimality allows move-
ment of the what and not movement of the who. This wh-phrase moves to the [Spec,
CP] position and the TP is elided (79) and the result is the sentence in (77).13
(79) ... [CP whok [TP was Ivan talking ti]]∧[VP to tk V about what]i
In sum, CLM’s approach seems to predict that the above sentence be unacceptable
whereas this approach predicts the opposite.
6.3. No sprouting subjects. CLM’s syntactic identity restriction straightforwardly
accounts for the fact that (80) is unacceptable. The antecedent is passive and the
sluicing sentence is active and there is no syntactic identity.
(80) *Ivy was hit, but we don’t know who
A Merchant-style analysis runs into difficulty here because there seems to be a fairly
obvious mutual entailment relation between the clauses. Why could the above sen-
tence not have be derived like (81).
(81) Ivy was hit, but we don’t know whoi [IP ti hit Ivy]
Under the analysis proposed here, the sentence in (80) is ungrammatical because
sprouted actor is in a subject position in the sluiced sentence. As such, it should
be required to extrapose before movement to [Spec,CP]. Since we have seen that
subjects cannot extrapose in English, the sentence is correctly ruled out.
This account can also easily handle otherwise problematic examples from Chung
2011. Here it is apparently the case that infinitival subjects have been sprouted. She
attributes this to the fact that the case assigner of the wh-phrase (a finite T-head)
13 See Larson, XXXX for a broader application of this idea into swiping instances.27
is not present in the antecedent clause. The Merchant-style analysis here can handle
this by resorting to the fact, again, that subjects cannot extrapose.
(82) *Having to compromise is inevitable, but they have no idea who has to
compromise.
7. Conclusion
In this paper I have argued against the covert TP- and wh-phrase-lowering anal-
ysis of Chung, Ladusaw, and McCloskey. Their analysis turns out to have serious
conceptual and empirical problems that can be avoided adopting a Merchant-style
analysis. It has also been shown that Merchant’s analysis as it stands is untenable
and I offer a revision of it here. This revision relies on extraposition of focused mate-
rial that escapes TP deletion in sluicing and the consequences of this capture a wide
swath of data and rescue the PF deletion account. Sprouting is fed by extraposition
and thus only those elements that can extrapose can be sprouted.
28