Sluicing(:) between structure and inference

Full text

SLUICING(:) BETWEEN STRUCTURE AND INFERENCE*
Sandra Chung, William Ladusaw, James McCloskey
University of California, Santa Cruz
schung@ucsc.edu, ladusaw@ucsc.edu, mcclosk@ucsc.edu
Investigations of sluicing since Chung, Ladusaw, and McCloskey 1995 have profitably ex-
plored two approaches to this ellipsis process that differ significantly from ours. In one, the
ellipsis site is created by deletion of a fully articulated TP in which
Wh
-movement has applied.
In the other, the ellipsis site contains no internal structure at all, and its reference is resolved via
pragmatic inference. Here we reconsider some of the theoretical issues, focusing on sprouting,
the subtype of sluicing in which the remnant of ellipsis has no overt correlate in the antecedent
clause. We discuss evidence, some of it new, which suggests that sprouting involves the re-use
of existing material, much as we originally proposed.
1. Goals
In this paper, we reconsider some of the theoretical issues raised by sluicing, taking as a starting
point our 1995 article in
Natural Language Semantics
(henceforth CLM). Our aims are to (i) in-
corporate some of the insights and empirical discoveries that have emerged since (especially in the
work of Jason Merchant), (ii) refocus attention on the subtype of sluicing that we earlier called
sprouting, and (iii) pursue an analysis driven by the core intuition that at least this species of el-
lipsis involves the re-use of existing linguistic material. Our goal in this will be to illuminate the
interaction between formal linguistic structure and discourse interpretation in ellipsis processing.
Sluicing is the ellipsis of all but the interrogative phrase of a constituent question. In CLM,
we distinguished two subtypes of sluicing, which we called merger and sprouting. In merger, the
interrogative phrase that is the remnant of ellipsis has an overt correlate in the antecedent clause,
as shown in (1) (with the correlate italicized):
(1) a. They’ve made an offer to a phonologist, but I’m not sure which one.
b. She insulted somebody but she won’t tell me who.
In sprouting, the interrogative phrase that is the remnant of ellipsis has no overt correlate within
the antecedent clause, as seen in (2):
(2) a. They were firing, but at what was unclear.
b. She applied for the position but nobody could figure out why.
c. He finished on time, but with whose help?
*We are very happy to dedicate this paper to Judith Aissen, our friend and colleague of many years. The research
reported on here grows out of a Symposium on Ellipsis which was held at the 2006 Meeting of the Linguistic Society
of America in Albuquerque, New Mexico. We are grateful to all who took part for their help. We are especially grateful
to Jason Merchant for the excellence of his ongoing work in this area and for the many insights and challenges that he
continues to provide.

In such cases, the remnant
Wh
-phrase can correspond to an implicit argument of the predicate of
the antecedent (as in (2a)), or an adjunct (as in (2b) and (2c)).
We frame our discussion of the issues raised by sluicing in terms of the approaches to
ellipsis currently being explored. There are three:
APPROACH 1: The ellipsis site is an anaphoric element without internal structure, whose reference
must be resolved in the same way as the reference of any anaphoric element is
resolved, by way of pragmatic inference (Ginzburg and Sag 2000, Culicover and
Jackendoff 2005).
APPROACH 2: The ellipsis site is empty and unstructured at surface structure, but its content is
supplied by re-using (recycling, copying) an already built syntactic structure, with
its interpretation, from some accessible point elsewhere in the discourse (Williams
1977, Fiengo and May 1994, Lappin 1999, CLM).
APPROACH 3: The ellipsis site has internal structure, which is constructed in exactly the same
way as any audible piece of syntactic structure. The ellipsis site may, however, go
unpronounced—be rendered silent—if it is sufficiently similar to some antecedent
XP in some accessible position elsewhere in the discourse (Ross 1969, Sag 1976,
Hankamer 1979, Lasnik 2001, Merchant 2001).
Of these, the second and third approaches are much closer to each other than either is to the first,
since both assume detailed syntactic structure within the ellipsis site; they differ, however, in what
they assume about how that structure comes to be there.
Following CLM, Merchant 2001—and, ultimately, Ross 1969—we hold that at some point
in the derivation of examples like (1) and (2) the ellipsis site contains a fully fleshed-out syntactic
object; there is ‘syntax in the silence’, to use Merchant’s term. The phenomena discussed below
show sensitivity to syntactic properties that we take to be difficult to integrate into a general infer-
ential approach to sluicing like that found in Ginzburg and Sag 2000 or Culicover and Jackendoff
2005. We assume that the remnant
Wh
-phrase is contained within a CP (see especially Merchant
2001:Chap. 2). We further assume that what is missing in sluicing is the complement of whatever
head it is in a given languagethat attracts interrogative
Wh
-phrases to its specifier—Manetta 2005,
2006, Grebenyova 2006. That is, what is missing in sluicing is all but the edge of a phase defined
by a head which drives
Wh
-Movement to its specifier.
This general characterization yields for English the conclusion that the missing material in
a sluicing construction is the TP complement of interrogative C. Therefore, (2a) has the skeletal
structure shown in (3).
(3) They were firing, but [CP at what C[TP ]] was unclear
Sluicing, then, involves either the reduction to silence of the TP complement of C(as in APPROACH
3, e.g. Romero 1998, Merchant 2001), or else the recovery of a suitable TP from the discourse
context, supplying the content for the empty TP in (3), as in APPROACH 2 (for instance, CLM).
APPROACH 3 (deletion under identity or givenness) is the standard view in current research
in the Principles and Parameters framework and in the Minimalist Program. Jason Merchant’s
(2001) book, along with important work done around the same time by Maribel Romero (1998)

and Howard Lasnik (1999), were particularly important in establishing that view. In these works,
the core properties of sluicing are taken to derive from semantic conditions—such as givenness and
focal parallelism—which govern deaccenting and elision. For Merchant, for example, the crucial
elements are those in (4):
(4) a. Sluicing is derived by PF deletion of a fully articulated TP in which
Wh
-movement
has applied.
b. This deletion (like deletion in general) is subject to a semantic licensing condition, in
that TP can be deleted only if it is E-GIVEN.
E-givenness is in turn defined as in (5):
(5) An expression Ecounts as E-GIVEN iff it has a salient antecedent Aand,
1.Aentails the focus-closure of E
2.Eentails the focus-closure of A
What these requirements amount to in essence is the requirement that the non-focused portions of
the antecedent TP and the elided TP must entail each other.
Such theories give an admirably successful account of merger, but they arguably do not
generalize well to sprouting (see Chung 2005 and below). Here we take a different tack: taking the
sprouting cases as our starting point, we explore the idea that the interpretation of this subtype of
sluicing is best understood as involving the re-use of existing linguistic material (APPROACH 2).
2. Use and Re-Use
The central notion of use that we appeal to is, as might be expected, fundamentally pragmatic.
To use linguistic material is to introduce it into the collaborative game of constructing shared
contexts. Accepting this, to re-use linguistic material is to take an already-constructed syntactic
object with an interpretation, one which has already been deployed in discourse processing, and to
re-deploy it, with its interpretation, in a new anddifferent context. We assume a model of discourse
structure along the lines of that explored in Ginzburg 1996, Büring 2003, and Farkas and Bruce
2010, in which questions under discussion (Ginzburg’s (1996) QUD) are recorded and in which the
items so recorded are syntactic objects paired with their denotations. These syntactic objects are
presumably LF representations in the sense familiar from Government and Binding Theory and its
derivatives, and so may differ in important ways from the representations relevant for determining
phonological form (in the framework of Chomsky 2001, for instance, all uninterpretable features
will have been removed). We assume a framework for the interpretation of questions and sluices
along lines developed by AnderBois 2010a,b, which has the great advantage of letting us better
understand why disjoined terms pattern similarly to wide-scope indefinites in their ability to license
sluicing (CLM:268–269). We will have more to say later about some of the interpretive issues.
The re-use of linguistic material must be carefully distinguished from independent, and
distinct, introductions of an expression. In (6), we clearly want to say that there are two independent
token expressions of the DP a lawyer.
(6) A lawyer who sues a lawyer is crazy.

This determination has pragmatic consequences: each token of the expression a lawyer gives rise
to different discourse referents.
The situation is revealingly different in (7a), which in minimalist syntax is the pronounced
form of the structure sketched in (7b). Here, we clearly want to say that there is just a single use of
the DP a lawyer.
(7) a. A lawyer was sued yesterday.
b. [TP [ A lawyer ] was [VP sued [ a lawyer ] yesterday ]]
Current minimalist thinking holds that (7b) involves two syntactic occurrences of a single syntactic
token of the DP a lawyer, only the highest of which is pronounced. The basic idea is that when
movement (in its minimalist guise as Internal Merge) occurs, the DP a lawyer comes to serve both
as the sister of the Vsued and the specifier of T; in the terminology of Relational Grammar, it is
multi-attached. The distinction between multiple syntactic occurrences of a phrase (which amount
to a single pragmatic use of the phrase) and multiple syntactic tokens of a phrase (which lead to
distinct pragmatic uses) will be crucial in what follows.
2.1. Sluicing in the Absence of an Overt Correlate (Sprouting)
Consider, then, the examples of sprouting in (8):
(8) a. They were firing, but at what was unclear.
b. She applied for the position but nobody could figure out why.
c. He put in a bid, but on whose behalf?
d. A: I went to the movies last night. B: Who with?
e. Exchanges of gunfire took place, but it was not clear where from.
Let us suppose that the interrogative Cin (8a) has an empty complement whose content is supplied
by a TP already deployed in the discourse, so that the CP in (3) becomes what is shown in (9). What
sort of operation supplies the content of this TP is an issue to which we return; for the moment,
suppose it to be copying, and in this (metaphorical) sense to represent a re-deployment of available
content.
(9) [CP at what C[TP ]]
⇓
[CP at what C[TP they were firing ]]
Such a structure is uninterpretable as it stands (there is no way to integrate the
Wh
-phrase into
the composition of the meaning of the question), so another operation is needed—the creation of a
lower syntactic occurrence of the
Wh
-phrase within VP, an operation which will permit the needed
integration. That is, we add to the phrase marker a statement like (10):
(10)
at what
is immediately dominated by VP.
providing for (11):

(11) [CP at what C[TP they were firing ]]
⇓
[CP
at what
C[TP they were firing
at what
]]
Importantly, this operation is not specific to sluicing, but is an instance of the more generally
available operation that gives rise to multiple syntactic occurrences of a phrase. That is, it is (the
inverse of) Chomsky’s (2001)
Internal Merge
. This is the natural updating of our 1995 proposal in
a changed theoretical context.
The featural interactions in (10) and (11) are routine. If, for instance, interrogative
Wh
-
movement is driven by the combination of features [Q],[WH], and [EPP] on a C-head, then inser-
tion of the interrogative phrase into the specifier of Cin (10) satisfies only the third—the [EPP]
property. On the assumption that command, rather than the specifier-head relation, is the crucial
relation underpinning syntactic agreement relations (Chomsky 2001, 2008), the interrogative and
Wh
-features on C(and the corresponding features on the
Wh
-phrase) will be checked only when
the lower occurrence of the phrase at what in (11) is created—an occurrence within the com-
mand domain of interrogative C. Within the framework of Phillips (2003)—left to right, top-down
structure building—the necessary operation has a particularly natural home and is probably indis-
tinguishable from routine applications of
Wh
-movement.1
This updating of our 1995 proposal has a number of interesting consequences. First, it
eliminates the need for some stipulations required under the earlier proposal: for example, that only
traces can be added. Second, it preserves the empirical range of our earlier account of sprouting.
Third, it deals naturally with some more recent empirical discoveries in a way that CLM did not.
We note as an aside that the syntactic objects which are copied or re-used will have to
be abstract enough to permit certain morphological ‘mismatches’ between the antecedent and the
apparent requirements of the ellipsis site. This is to allow such cases as (12) (Merchant 2001,
2005):
(12) a. Decorating for the holidays is easy if you know how.
b. I’ll fix the car if you tell me how.
c. I can’t play quarterback. I don’t even know how.
d. I remember meeting him, but I don’t remember when.
e. John seems to be happy and I can guess why.
It seems reasonable to hope that these mismatches will reflect the kinds of differences between
surface syntax and LF syntax that we alluded to earlier.
1Our general approach to sluicing is very much in harmony with the research program laid out in Phillips and
Lewis 2009, in the sense that the grammatical computation for sluicing structures that we develop seems to mirror
what the processor must do when faced with the task of comprehending a sluice. For the production task, matters seem
a little less clear.

3. Consequences—Old and New
3.1. Albert’s Generalization
In the cases for which this mechanism must be appealed to, there can be no amnestying of island
and ECP effects. We thus understand an important property of sluicing. As noticed originally by
Chris Albert, reported by CLM, and confirmed recently in experimental work by Yoshida et al.
(2010), island violations are not repaired in sprouting (although they are famously repaired under
merger; see Ross 1969, CLM, Merchant 2001).
Consider, for instance, the examples in (13):
(13) a. *Sandy was trying to work out which students would speak, but she refused to say
who to.
b. *Agnes wondered how John could eat, but it’s not clear what.
c. *That Tom will win is likely, but it’s not clear which race.
(14) a. *Sandy is very curious to see which students will be able to solve the homework
problem, but she won’t say how.
b. *Clinton is anxious to find out which budget dilemmas Panetta would be willing to
tackle, but he won’t say how.
We will call this observation
Albert’s Generalization
, for its discoverer.
If the operation responsible for creating multiple syntactic occurrences in Internal Merge is
governed by the standard array of island and ECP effects, then we expect those effects to appear in
the subtype of sluicing for which this operation is crucial—namely,in the sprouting cases.
3.2. Fixed Diathesis Effects
We also understand another set of properties of sluicing. As observed first by Lori Levin (1982),
the interpretation of the elided TP in sluicing is limited by lexical choices made in the antecedent
TP. Compare (15a) with (15b), for instance.
(15) a. He served the soup, but I don’t know to who(m).
b. He served some of the guests, but I don’t know what.
The examples in (15) contrast sharply with the impossible (16):
(16) *He served the soup, but I don’t know who(m).
The problem here is that there are, essentially, two distinct but related verbs serve, which can be
illustrated crudely as in (17):
(17) a. serve1: <server>
SUBJ <meal>
OBJ <diner>
DATIVE
b. serve2: <server>
SUBJ <diner>
OBJ1<meal>
OBJ2

What goes wrong in a case like (16) is that the antecedent clause contains serve1while the elided
clause contains serve2. This is an impossibility under our proposal, one which follows, on our
view, from the fact that the missing material in a sluice is supplied by the re-use of a TP already
constructed from an array of lexical choices. There can be no subsequent return to the lexicon in
constructing the missing TP of the ellipsis site.2
The effect seems to be quite general. The examples in (18) show the same effect for the
verb send.
(18) a. He sent a package, but I can’t find out who to.
b. *He sent a package, but I can’t find out who.
c. He sent a package, but I can’t find out who he sent it to.
d. ?He sent a package, but I can’t find out who he sent it.
What goes wrong in (18b) is that the antecedent TP and the elided TP employ different argument
structures for the verb send: the antecedent TP employs the argument structure illustrated in (18c),
whereas the elided TP employs that illustrated in (18d).
Observations made by Jason Merchant (Merchant (2005)) suggest the same conclusion.
Beth Levin (2003) observed that the examples in (19a) and (19b) are close to synonymous, but
involve different versions of the verb embroider.
(19) a. They embroidered a table-cloth with peace signs.
b. They embroidered peace signs on a table-cloth.
Despite the semantic equivalence of (19a) and (19b), it is impossible, as Merchant points out, to
mix and match different versions of the verb under sluicing. That is, one cannot have a remnant
Wh
-phrase which implies one version of the verb embroider while the antecedent TP is built around
a different one. This is seen in the dual impossibility of (20):
(20) a. *They embroidered something with peace signs, but I don’t know what on.
b. *They embroidered something on the table-cloth, but I don’t know what with.3
Observations such as these pose severe challenges for purely inference-based approaches to ellipsis
resolution.
Merchant (2005) has observed a similar effect in cases such as (21), involving the causative-
inchoative alternation. In English, an example such as (21):
(21) They plan to close one of the schools, but they won’t tell us which one.
cannot be interpreted as in (22):
2The lexical entries in (17) are meant to be illustrative only. The central conclusion is unaffected if the different
argument structures for serve are realized syntactically via different arrays of functional heads (‘light verbs’) within
the vP. Such differences still reflect different lexical choices.
3(20b) is well-formed on a different and irrelevant reading—according to which the with-PP is an instrument rather
than a third argument of embroider.

(22) They plan to close one of the schools, but they won’t tell us which one will close.
Once again, this falls under our larger observation, since causative and inchoative close must reflect
distinct lexical choices—the first used in the antecedent, the second (impossible) in the ellipsis site.
Merchant (2005) observes that the point can be made more clearly in a language where the case
system lets one identify the grammatical function of the remnant interrogative phrase. Greek is
such a language, and once more (as can be seen in (23b)), the effect is as we now expect it to be:
(23) a. Eklisan
close-PL3ena
a-ACC dhromo,
road-ACC alla
but dhen
not ksero
know-S1pjon.
which-ACC
‘They closed a road, but I don’t know which.’
b. *Eklisan
close-PL3ena
a-ACC dhromo,
road-ACC alla
but dhen
not ksero
know-S1pjos.
which-NOM
‘They closed a road, but I don’t know which.’
(23b) must reflect the inchoative form of close in the ellipsis site, but the transitive form in the
antecedent clause—an impossible situation, given our general proposal.
The same pattern can be seen at work in the opposite direction in a case such as (24):
(24) *The window suddenly closed, but I don’t know who.
In a case such as this, we have the inchoative form in the antecedent and the transitive form in the
ellipsis site—an impossibility given our proposal. This is a case where it is particularly clear that a
treatment of sluicing based solely on pragmatic inference would not be adequate to the facts. For
(24), it is hard to see why the antecedent clause would not make salient a proposition like Someone
suddenly closed the window.
Finally, this set of observations further extends to the impossibility of voice mismatches
under sluicing (see Merchant 2001, Chung 2005, AnderBois 2010b):
(25) a. The candidate was abducted but we don’t know who by/by who.
b. *Somebody abducted the candidate, but we don’t know by who.
c. Somebody abducted the candidate, but we don’t know by who he was abducted.
As long as active and passive structures involve different lexical selections (in one sense or an-
other), we can understand the ill-formedness of (25b) in the same terms as (24) and earlier ex-
amples: the lexical resources used in the ellipsis site must necessarily be the same as those out
of which the antecedent TP is constructed. This we see as one of the consequences of re-use of
existing linguistic material.4
Verb phrase ellipsis, as is well known, behaves differently (Kehler 2002:53):
(26) a. This problem was to have been looked into, but obviously nobody did.
4Eric Potsdam (2007) observes that voice mismatches seem to be possible under sluicing in Malagasy. We must take
the position that such observations provide evidence for Pearson’s (2005) reanalysis of ‘voice’ in Malagasy in terms
of something like
Wh
-Agreement—see also Chung 2005 and Potsdam (2007:fn.11) for discussion of alternatives.

b. In March, four fireworks manufacturers asked that the decision be reversed, and on
Monday the ICC did.
c. Actually I have implemented it with a manager, but it doesn’t have to be.
d. The janitor should removethe trash whenever it is apparent that it needs to be.
Following Merchant 2007, 2008, we take cases like (26) to involve ellipsis of the complement of
the voice-determining head—a level of structure at which active and passive verbal phrases are
indistinguishable, both in terms of the lexical resources used in their construction and in terms of
the structures projected.
3.3. Chung’s Generalization
We are also now in a position to understand a more recent discovery. Merchant (2001) demon-
strated that exactly those languages which permit preposition stranding under
Wh
-movement also
permit prepositions to be stranded in the elided TP of sluicing. Chung 2005 has observed that even
in preposition-stranding languages, prepositions cannot be stranded in the elided TP in sprouting
cases—when the interrogative phrase that is the remnant of ellipsis has no overt correlate in the
antecedent clause. Compare (27), in which the interrogativephrase is a PP, with (28), in which the
interrogative phrase is the object of a stranded preposition.
(27) a. They’re jealous but it’s unclear of who/who of.
b. Last night he was very afraid, but he couldn’t tell us of what/what of.
c. Mary was flirting, but they couldn’t say with who/who with.
d. We’re donating our car, but it’s unclear to which organization.
e. The UN is transforming itself, but into what is unclear.
(28) a. *They’re jealous but it’s unclear who.
b. *Last night he was very afraid, but he couldn’t tell us what.
c. *Mary was flirting, but they couldn’t say who.
d. *We’re donating our car, but it’s unclear which organization.
e. *The UN is transforming itself, but what is unclear.
Of course, preposition stranding in the absence of ellipsis is unproblematic:
(29) a. They’re jealous but it’s unclear who they’re jealous of.
b. Last night he was very afraid, but he couldn’t tell us what he was very afraid of.
c. Mary was flirting, but they couldn’t say who she was flirting with.
d. We’re donating our car, but it’s unclear which organization we’re donating it to.
e. The UN is transforming itself, but what it is transforming itself into is unclear.
The puzzle here is why (28a–e) cannot be derived from (29a–e). We call this
Chung’s Generaliza-
tion
, also for its discoverer.
These observations are deeply puzzling for APPROACH 3—specifically, for the view that
ellipsis is the reduction to silence of a syntactic object whose content is ‘given’ in some sense
(among many others, see Romero 1998, Merchant 2001). On that view, it is hard to see how we
might distinguish the derivation in (30) from that in (31). Note the even more severe difficulty posed

by these observations for APPROACH 1, which involves only mechanisms of pragmatic inference.
Such theories too easily locate suitably salient content with which to fill out the interpretation of
the
Wh
-phrase. There is no challenge whatever in computing in context what the interpretations of
(28) ought to be.
(30) a. She is jealous, but we don’t know [ of who [ she is jealous of who ]].
b. She is jealous, but we don’t know [ of who [ ]].
(31) a. She is jealous, but we don’t know [ who [she is jealous of who ]]
b. *She is jealous, but we don’t know [ who [ ]].
But these observations already follow inevitably from our proposals. (28a), for example, would
begin with the fragment in (32):
(32) [ unclear [CP who C[TP ]]]
Re-using the antecedent TP will produce (33):
(33) [ unclear [CP who C[TP they’re jealous ]]]
But from (33), the only structure that can be created by way of the Internal Merge operation is that
in (34), which subsumes a violation of the lexical requirements of the adjective
jealous
.
(34) [ unclear [CP
who
C[TP they’re jealous
who
]]]
So as long as those requirements must be respected—either at the point at which the DP
who
is
(re)merged, or else at LF (if there is such a level), then the impossibility of (28a) is expected rather
than puzzling. In fact, on this view, (28a) is impossible for exactly the same reason that (35) is
impossible—a unification which seems entirely natural:
(35) *Who are they jealous?
As far as we are aware, there is no comparably natural treatment of these observations available at
present under other approaches to sluicing.
4. A Complementary Difficulty
The problem posed by the observations of (28) for versions of APPROACH 3 under a condition of
givenness is that the requirement of givenness appears to be met but sluicing fails. But there is also
a range of cases in which the requirement of givenness clearly is not met, but in which sluicing
nevertheless succeeds (Chung 2005).
(36) a. He put in a bid but I couldn’t tell on whose behalf.
b. She went to the movies but we don’t know who with.
c. She finished the project but we don’t know with whose help.
d. He’s on the no-fly list but it’s totally unclear for how long.

(37) a. She was babbling away, but about what, I have no idea. (RTE radio, December 31,
2005)
b. .. . with Argentina and Brazil increasingly worried about where they would get their
oil and at what price. (New York Times, May 5, 2006)
c. I agree with the NYT Executive Editor that the public did benefit from the Times’
disclosures about NSA and Treasury surveillance, though it’s impossible to know at
what cost. (David Ignatius,Washingon Post, July 5, 2006)
d. Batwoman is set for a comicbook return. But what as? (http://news.bbc.co.uk/,
Friday, June 2, 2006)
Cases such as (37) are handled without elaboration by the proposal sketched earlier. It is at best
unclear how they can be understood in a world in which sluicing is deletion under semantic ‘iden-
tity’ or givenness. Such a view would require that the pairs of propositions in (38)–(43) be in the
required relation (equivalence, mutual entailment, or whatever):
(38) a. [ he put in a bid ]
b. [ he put in a bid on someone’s behalf ]
(39) a. [ she went to the movies ]
b. [ she went to the movies with someone ]
(40) a. [ she finished the project ]
b. [ she finished the project with someone’s help ]
(41) a. [ he’s on the no-fly list ]
b. [ he’s on the no-fly list for some length of time ]
(42) a. [ she’s babbling away ]
b. [ she’s babbling away about something ]
(43) a. [ where they would get their oil ]
b. [ where they would get their oil at some price ]
But in none of these cases does the proposition expressed by the (a) example entail the proposition
expressed by the (b) example. In the case of (43), for instance, getting oil does not entail that the
oil be obtained for a price (there are many ways of obtaining oil other than buying it). Similarly for
(42)—one can babble without babbling about anything. And in (38), the proposition that he put in
a bid does not entail that he put in a bid on someone’s behalf. We believe that the observation is
quite general.
Given that the (b) examples entail the (a) examples, accommodation is sometimes sug-
gested as a means of upgrading the interpretation of the antecedent clause in such cases so that
equivalence or mutual entailment could be achieved (see Fox 1999 for one such proposal). The
challenge, it seems to us, would be to constrain accommodation so that it would permit sluicing
in (37), for instance, but not in many of the ill-formed cases that we considered earlier—in (24),
(25b), or in (28), for example.

In contrast, our proposal handles all of these cases without elaboration, because it requires
only that the elided TP be a monotonic extension, both syntactically and semantically, of the an-
tecedent TP. Internal merge can add new material to an antecedent clause in the ellipsis site, as
long as lexical and morphosyntactic requirements are satisfied. From this it follows that there will
be no general requirement that the interpretation of the antecedent clause be equivalent to, or even
entail, the interpretation of the elided clause.
Nominal-internal cases (Chung 2005) make the same point:
(44) a. She’s reading something, but I don’t know from which textbook.
b. She’s eating a pizza, but I don’t know from which restaurant.
c. She’s editing a manuscript, but I don’t know from what period.
Such cases are perfectly natural, but there is no entailment here from the interpretation of the
antecedent to the interpretation of the missing TP.
5. Semantic Consequences of Re-Use
In the view presented here, the empirical patterns surveyed above are seen as consequences of un-
derstanding sluicing as the re-use of existing linguistic material. So far we have been concerned
with the lexical and syntactic consequences of re-use—with the phenomena that support the as-
sumption that ‘the syntax in the silence’ is simply an interpreted syntactic object, which has already
been used in the discourse and which now serves as a resource in interpreting the sluice.
We now consider the semantic consequences of this re-use. In particular, we investigate
whether the syntactic re-occurrences of TP’s in sluices are understood as uses of the TP’s in the
strongest pragmatic sense—that the syntactic object in the ellipsis site counts as being introduced
into the collaborative game of constructing shared contexts.
We conclude that in the case of sluicing, it does not—in contrast to (some) other types of
ellipsis, notably VP ellipsis. We will see that sluices are understood as if the re-use of a familiar
linguistic expression constitutes re-use of its interpretation as well.
Here we will make the case by examining the interpretation of referential indefinites which
are subject to a novelty condition on their discourse reference (Heim 1982). As a result of the
novelty condition, each syntactic token of an indefinite introduces a new discourse referent. That
is, (45a) is interpreted as involving two perpetrators, in contrast to (45b).
(45) a. Someone committed a crime on Monday and someone committed a crime on Tues-
day.
b. Someone committed a crime on Monday and he committed a crime on Tuesday.
We will follow common terminology and say that each token of
someone
above introduces a
distinct discourse referent. Similarly,in (46) Jill and Jack know similar things, but weassume that
their knowledge involves distinct perpetrators.
(46) Jill knows that someone committed a crime, and Jack knows that someone committed a
crime.

The embedded questions in (47) behave similarly, in that the two syntactic tokens of the indefinite
a crime are associated with distinct discourse referents.
(47) Jill asked where someone had committed a crime, and Jack asked when someone had
committed a crime.
The association of an indefinite with a discourse referent can be used as a probe for the act of
using the indefinite. Each use, in this strongly pragmatic sense, is expected to involvecreation of a
new discourse referent. If we want to know whether a distinct syntactic occurrence of an indefinite
counts as a distinct pragmatic use of the indefinite, we can use this probe. Distinct discourse ref-
erents indicate distinct pragmatic uses; same discourse referent indicates that the second syntactic
occurrence doesn’t count pragmatically as a use—but rather, in our terms, as a re-use.
If ellipsis in general involves the re-use of a familiar linguistic expression, we should ask
whether that re-occurrence counts as a new pragmatic use of the expression. If re-occurrence con-
stitutes a new use, then indefinites occurring in an ellipsis ought to trigger new discourse referents
despite their silence. If re-occurrence does not constitute a new use, but merely re-use, then the
indefinite will not be associated with the creation of a new discourse referent: the interpretation
of the second occurrence would involve the discourse marker associated with its first (and only)
independent use.
Our current work suggests that ellipses are not uniform in this regard. Sluicing, at least,
involves re-occurrences that are not interpreted pragmatically as new uses. In most cases, the ma-
terial in the elided TP seems to be unable to introduce new discourse referents. Compare (47) with
(48):
(48) Jill asked where someone had committed a crime, and Jack asked when.
The only natural interpretation of this example, it seems, is that Jack’s question is about the same
perpetrator, and the same crime, that Jill’s question is about; in other words, (48) can be para-
phrased ‘Jill asked where person xhad committed crime y, and Jack asked when xhad committed
y’. This is the interpretation that would be expected if the discourse markers employed in the
antecedent TP are carried over into the interpretation of the elided TP. Similarly, in (49):
(49) Where someone commits a crime doesn’t determine how.
the only natural interpretation is that it is false that where the random person xcommits crime y
determines how xcommits y.
The non-synonymy of (47) and (48) is replicated in the example pairs in (50)–(54). In each
pair, the indefinite that putatively occurs in the sluice in the (b) example cannot be understood as
introducing a new discourse referent:
(50) a. We know what someone was reading, but we don’t know to who someone was read-
ing.
b. We know what someone was reading, but we don’t know to who.
(51) a. Although we know who someone spoke to, we don’t know what someone spoke (to

someone) about.
b. Although we know who someone spoke to, we don’t know what about.
(52) a. Jill wondered why Tracy dated a student, and Fred wondered for how long Tracy
dated a student.
b. Jill wondered why Tracy dated a student, and Fred wondered for how long.
(53) a. A high government official was critical of the
New York Times
, but it’s not clear
what other newspapers a high government official was critical of.
b. A high government official was critical of the
New York Times
, but it’s not clear
what other newspapers.
(54) a. Someone from Santa Cruz talked to SAM, but we’re not sure who else someone from
Santa Cruz talked to.
b. Someone from Santa Cruz talked to SAM, but we’re not sure who else.
At one level, the observation that the same discourse markers are employed in the elided TP as in
the antecedent TP seems expected. It ought to follow immediately from our proposal that sluicing
involves re-use of existing linguistic material; specifically, from the claim that the content of the
elided TP is supplied by copying of the antecedent TP, including its interepretation and associated
discourse markers. (We appeal to copying rather than the Internal Merge operation here, because
the relation must be able to operate across sentences uttered by different participants in discourse.)
However, the patterns illustrated in (50)–(54) are profoundly surprising when viewed from
the perspective of a general theory of ellipsis.
Since Hankamer and Sag 1976, it has been recognized that one of the hallmarks of ellipsis is
precisely the ability of elided material to introduce new discourse referents. Consider, for instance,
the VP ellipsis in (55a), which has an interpretation synonymous with (55b)—one in which each
syntactic token of a book introduces a new discourse marker.
(55) a. Kate is reading a book, and I am too.
b. Kate is reading a book, and I am reading a book too.
The ability of elided material to introduce new discourse referents lies behind the missing an-
tecedent phenomenon, which is used by Hankamer and Sag as a diagnostic of ellipsis as opposed
to deep anaphora (Grinder and Postal 1971, Hankamer and Sag 1976).
(56) a. *I’venever ridden a camel, and it was of the two-humped variety.
b. I’ve never ridden a camel, but Ivan has, and it was of the two-humped variety.
If the missing antecedent phenomenon is truly characteristic of ellipsis, then we need to ask
whether it is sluicing that is atypical in its interpretation and, if so, why it should be.
We conjecture here that the contrast is correlated with the size of the ellipsis site. Sluicing
and VP ellipsis differ in whether the content that must be supplied by copying of an antecedent XP is
larger or smaller than the domain of existential closure, which we take to be the smallest constituent
in which all the predicate’s arguments have had a chance to be introduced (see Chung and Ladusaw
2004). In sluicing, the missing content is larger than the domain of existential closure, so that the

re-used expression has a complete interpretation in terms of a discourse model. The re-occurrence
of the expression in the sluice simply provides that model to the interpretation.
In VP ellipsis, on the other hand, the missing content is smaller than the domain of existen-
tial closure, so any indefinites that are copied over from the antecedent VP can become existentially
closed ‘again’ in the new domain, with the result that new discourse markers will be introduced.
From this follow examples like (56b) (on the relevant interpretation), and the missing antecedent
phenomenon.
The observation that there are cases of sluicing in which the elided material cannot intro-
duce new discourse referents is both novel and—we believe—undeniable for examples of the type
(50)–(54).
It remains to be seen whether the observation is fully general; notice, to begin with, that
the examples cited above all involve sprouting. Even without a definitive answer, however, we can
bring the preceding discussion to bear on another sluicing pattern that CLM could not account for,
namely, the phenomenon of vehicle change.
5.1. Sluicing, E-type Anaphora, and Vehicle Change
As Romero (1998:67–69) and Merchant (2001:201–204) observe, and as Kyle Rawlins has also
pointed out to us, theories of sluicing that impose a syntactic identity condition on the elided TP
and the antecedent TP encounter a challenge in examples like (57).
(57) a. The Deans know who resigned, but they’re not sure for what reasons.
b. He told us which kids were eating, but he couldn’t tell us how much.
c. That’s a gazebo. But I don’t know who built it or why. (Merchant 2001:201)
d. What interveners are able to ‘get out of the way’, and how? (Merchant 2001:202)
e. Always, when a female physicist has been nominated, she wants to know for which
award.
f. Every female physicist who has been nominated wants to know for which award.
In cases like this, we cannot express what the elliptical sentence means without the ellipsis by
simply pronouncing the supposed antecedent in place of the ellipsis. We must change the indefinite
DP or the interrogative phrase to a pronoun. Following Fiengo and May 1994, we will refer to
this phenomenon as vehicle change. The elided TP’s in (57) have interpretations equivalent to the
interpretations of the non-elided questions in (58), which contain E-type pronouns.
(58) a. The Deans know who resigned, but they’re not sure for what reasons he resigned.
b. He told us which kids were eating, but he couldn’t tell us how much they were eating.
c. That’s a gazebo. But I don’t know who built it or why s/he built it.
d. What interveners are able to ‘get out of the way’, and how are they able to get out of
the way?
e. Always, when a female physicist has been nominated, she wants to know for which
award she has been nominated.
f. Every female physicist who has been nominated wants to know for which award she
has been nominated.

But if the non-elided questions in (58) are the source of the ellipses in (57), then sluicing cannot
require syntactic identity, because the E-type pronouns in the elided TP aren’t identical to anything
in the antecedent TP.
Such a syntactic mismatch, if real, could not be easily handled by CLM (or by Chung 2005).
But in the context of our discussion of the interpretive consequences of the re-use of linguistic
material in sluicing, it is natural to ask whether the syntactic mismatch in (57) is real or apparent.
If we are right that sluicing involves the re-use of a fully interpreted TP from previous discourse,
but that this re-use does not allow new discourse markers to be introduced, then a way of rising to
the challenge posed by (57) is at hand.
The elided TP’s in these examples do not, as a matter of morphosyntactic substance, contain
pronouns at all. The E-type pronoun effect in interpretation is the natural result of the assumption
that the antecedent TP is copied with its closed interpretation, including discourse markers.
Specifically, after copying, (57a) has the structure shown in (59).
(59) The Deans know [CP who1[TP who2resigned]], but they’re not sure [CP for what reasons
[TP who3resigned]].
In (59),
who
1and
who
2are different syntactic occurrences of a single token of
who
, related by
Internal Merge. In the ellipsis,
who
3is a further syntactic occurrence of this token of
who
, related
to the other two by the larger re-use of TP that sluicing involves. The E-type pronoun interpretation
of
who
3is, on this view, a natural consequence of the assumptions needed to interpret structures
like the antecedent, in which multiple syntactic occurrences of a phrase correspond to a single
pragmatic use.
This, we think, is a satisfying resolution to one of the most difficult issues faced by the
account of sluicing we advanced some fifteen years ago and return to here.
6. The Broader Picture and Some Open Issues
Part of the excitement of research on ellipsis is that every new investigation seems to raise as many
questions as it resolves. In this spirit, we would like to conclude by pointing to some issues raised
by the line of thought pursued here.
First, how far-reaching is the generalization that new discourse referents cannot be intro-
duced by material inside the ellipsis in sluicing? The judgements we reported for examples (50)–
(54) strike us as very clear. But other types of examples suggest that material inside the ellipsis site
might be able to introduce new discourse markers after all. Consider (60).
(60) MARY was swindled by a lawyer,and it’s not clear WHO ELSE.
In (60), the elided question seems to be about who other than Mary was swindled, with no require-
ment that the lawyer be the same one one who swindled Mary. What distinguishes this example
from those discussed earlier is that else is associated with a
Wh
-phrase which, in its origin site,
c-commands the indefinite in the ellipsis. When this relation does not hold, as in (61), the more
general pattern re-emerges: the ellipsis introduces no new discourse referents.
(61) A lawyer swindled MARY, and it’s not clear WHO ELSE.

Clearly there is some systematicity to the contrast between (60), on the one hand, and (50)–(54)
and (61), on the other. Consider also (62):
(62) Joe was swindled by a lawyer—Mary doesn’t know how many times.
The elided question in (62) allows an interpretation in which every time Joe was swindled, he
was swindled by a different lawyer. In this interpretation, the indefinite in the re-used TP must
have narrow scope. What we are tempted to propose for interpretations of this type is that the
meaning of the indefinite is composed by Restrict (in the sense of Chung and Ladusaw 2004)—a
mode of composition that would lead to the indefinite being associated with no discourse marker
at all. The antecedent clause in (62), in other words, would be roughly synonymous with Joe was
lawyer-swindled.
Some initial evidence appears to support this view. When distinct discourse markers are
introduced by distinct syntactic tokens of an indefinite, they can collectively serve as antecedents
for a plural pronoun.
(63) A woman committed a crime on Monday and a woman committed a crime on Tueday.
They were the same woman.
The same is true even when some of the markers are introduced within VP ellipsis.
(64) Kate has ridden a camel and Ivan has, too. They were the same camel.
But in contrast to the naturalness of these examples, there is something distinctly odd about the
plural pronoun in (65):
(65) #Joe was swindled by a lawyer—Mary doesn’t know how many times. They were the
same lawyer.
This oddness we take as an indication that the narrow-scope reading of a lawyer in (62) involves
Restrict, and no discourse marker is introduced. If so, there is no counterevidence here to our
claim that no new discourse referents are introduced inside the ellipsis in sluicing. Clearly, though,
further probing is required.
Another large issue is how the observations and arguments developed here are to be inte-
grated with the body of evidence (developed by Merchant 2001 especially) which argues in favor
of PF deletion approaches to sluicing. We fully recognize the force of these arguments. What is
striking is that the evidence in question seems to come entirely from the merger cases. Put differ-
ently, PF deletion accounts, like Merchant’s, offer admirably successful accounts of merger, but
deal less well with sprouting. Our approach does a good job of handling sprouting, but is less
successful when faced with the merger cases (and especially the connectivity effects they exhibit).
What remains elusive is a successful unified account of sluicing. Presumably, such a unification
will involvea reconceiving of the apparent choice between deletion and re-use, a reconceiving that
we cannot yet imagine.5
5Recent work by Howard Lasnik (2009), responding in part to an earlier version of the material developed here,

Finally, our discussion adds to the list of known contrasts between sluicing and VP ellip-
sis: tolerance of voice mismatches, island repair, cross-linguistic generality, and now the missing
antecedent phenomenon. The contrast we discuss here raises a question about the pragmatic conse-
quences of the ‘re-use’ of processed linguistic material as a resource for interpreting ellipsis sites.
If VP ellipsis and sluicing are to be treated uniformly, then it must be the case that re-use of a
coherently interpreted TP-sized unit differs in a principled way from re-use of a VP-sized unit, in
terms of the pragmatic consequences. This strikes us as a speculation well worth exploring.
References
AnderBois, Scott. 2010a. Focus and uninformativity in (Yukatek) questions. Manuscript, available
at http://anderbois.googlepages.com.
AnderBois, Scott. 2010b. Sluicing as anaphora to issues. Paper presented at SALT 20, University
of British Columbia and Simon Fraser University, April 29–May 1, 2010.
Büring, Daniel. 2003. On d-trees, beans, and accents. Linguistics and Philosophy 26:511–545.
Chomsky, Noam. 2001. Derivation by phase. in Ken Hale: A life in language, ed. Michael Ken-
stowicz, 1–52. Cambridge, Mass.: MIT Press.
Chomsky, Noam. 2008. On phases. in Foundational issues in linguistic theory: Essays in honor
of Jean-Roger Vergnaud, eds. Robert Freidin, Carlos Otero, and Maria Luisa Zubizaretta,
133–166. Cambridge, Mass.: MIT Press.
Chung, Sandra. 2005. Sluicing and the lexicon: The point of no return. in BLS 31, Proceedings
of the Thirty-First Annual Meeting of the Berkeley Linguistics Society, eds. Rebecca Cover
and Yuni Kim, 73–91. Berkeley, Calif.: Department of Linguistics, UC Berkeley.
Chung, Sandra, and William Ladusaw. 2004. Restriction and saturation. Cambridge, Mass.: MIT
Press.
Culicover, Peter, and Ray Jackendoff. 2005. Simpler syntax. Oxford and New York: Oxford
University Press.
Farkas, Donka, and Kim Bruce. 2010. On reacting to assertions and polar questions. Journal of
Semantics 27:81–118.
Fiengo, Robert, and Robert May. 1994. Indices and identity. Cambridge, Mass.: MIT Press.
Fox, Danny. 1999. Focus, parallelism and accommodation. in SALT 9, eds. Tanya Mathews and
Devon Strolovich.Department of Linguistics, Cornell University: CLC Publications.
Ginzburg, Jonathan. 1996. Dynamics and the semantics of dialogue. in Language, logic, and
computation, ed. J. Seligman. Stanford, Calif.: CSLI Publications.
Ginzburg, Jonathan, and Ivan Sag. 2000. Interrogative investigations: The form, meaning and use
of English interrogatives. Stanford, Calif.: CSLI Publications.
Grebenyova, Lydia. 2006. Sluicing puzzles in Russian. in Proceedings of the Annual Workshop
on Formal Approaches to Slavic linguistics 14 (FASL 14), 151–171. Ann Arbor, Michigan:
Michigan Slavic Publications.
Grinder, John, and Paul M. Postal. 1971. Missing antecedents. Linguistic Inquiry 2:269–312.
Hankamer, Jorge. 1979. Deletion in coordinate structures. New York: Garland Publishing.
takes a step in this direction in maintaining deletion for the merger cases, but analyzing the sprouting cases by way of
a combination of deletion and the kind of lowering that we appeal to in Section 2.

Hankamer, Jorge, and Ivan Sag. 1976. Deep and surface anaphora. Linguistic Inquiry 7:391–428.
Heim, Irene. 1982. The semantics of definite and indefinite noun phrases. Doctoral Dissertation,
University of Massachussets, Amherst.
Kehler, Andrew. 2002. Coherence in discourse. Stanford, Calif.: CSLI Publications.
Lappin, Shalom. 1999. An HPSG account of antecedent-contained ellipsis. in Fragments: Studies
in ellipsis and gapping, eds. Shalom Lappin and Elabbas Benmamoun, 68–97. Oxford and
New York: Oxford University Press.
Lasnik, Howard. 1999. On feature strength: Three minimalist approaches to overt movement.
Linguistic Inquiry 30:197–217.
Lasnik, Howard. 2001. When can you save a structure by destroying it? in Proceedings of the
North East Linguistic Society 31, eds. Minjoo Kim and Uri Strauss, 301–320. Amherst,
Mass.: GLSA.
Lasnik, Howard. 2009. Island repair, non-repair, and the organization of the grammar. in Inter-
Phases, ed. Kleanthes K. Grohmann, 339–353. Oxford and New York: Oxford University
Press.
Levin, Beth. 2003. Objecthood and object alternations. Handout from a talk given at the Depart-
ment of Linguistics, UCLA, May 2nd 2003. Available at http://www.stanford.edu/~belevin.
Levin, Lori. 1982. Sluicing: A lexical interpretation procedure. in The mental representation of
grammatical relations, ed. Joan Bresnan, 590–654. Cambridge, Mass.: MIT Press.
Manetta, Emily. 2005.
Wh
-expletives in Hindi-Urdu: The vP phase. Presented at the January 2006
meeting of the Linguistics Society of America, Albaquerque, New Mexico.
Manetta, Emily. 2006. Peripheries in Kashmiri and Hindi-Urdu. Doctoral Dissertation, University
of California, Santa Cruz.
Merchant, Jason. 2001. The syntax of silence: Sluicing, islands, and the theory of ellipsis. Oxford
and New York: Oxford University Press.
Merchant, Jason. 2005. Revisiting syntactic identity conditions. Paper presented at the Workshop
on Identity in Ellipsis, UC Berkeley, October 8, 2005.
Merchant, Jason. 2007. Voice and ellipsis. Manuscript, University of Chicago, available at http:
//home.uchicago.edu/~merchant/publications.html.
Merchant, Jason. 2008. An asymmetry in voice mismatches in VP-ellipsis and pseudogapping.
Linguistic Inquiry 39:169–179.
Pearson, Matthew. 2005. The Malagasy subject/topic as an A-bar element. Natural Language and
Linguistic Theory 24:381–457.
Phillips, Colin. 2003. Linear order and constituency. Linguistic Inquiry 34:37–90.
Phillips, Colin, and Shevaun Lewis. 2009. Derivational order in syntax: Evidence and architectural
consequences. To appear in Directions in Derivations, C. Chesi, ed., Elsevier.
Potsdam, Eric. 2007. Malagasy sluicing and its consequences for the identity condition on ellipsis.
Natural Language and Linguistic Theory 25:577–613.
Romero, Maribel. 1998. Focus and reconstruction effects in wh-phrases. Doctoral Dissertation,
University of Massachusetts, Amherst.
Ross, John R. 1969. Guess who? in CLS 5: Papers from the fifth regional meeting of the Chicago
Linguistic Society, eds. Robert Binnick, Alice Davison, Georgia Green, and Jerry Morgan,

252–286. Chicago, Illinois: Chicago Linguistic Society.
Sag, Ivan. 1976. Deletion and logical form. Doctoral Dissertation, MIT.
Williams, Edwin. 1977. Discourse and logical form. Linguistic Inquiry 8:101–139.
Yoshida, Masaya, Isaac Rottman, Jiyeon Lee, and Michael Walsh Dickey. 2010. Islands under the
predicted structure. Poster presentation at the 23rd Annual CUNY Conference on Human
Sentence Processing, March 18–21, 2010, New York University.