linguistlist.org>
Editor for this issue: Terence Langendoen <terry linguistlist.org>
linguistlist.org
or Terry Langendoen at terry linguistlist.org.
rochester.rr.com>
Merchant, Jason (2001) The Syntax of Silence: Sluicing, Islands and the
Theory of Ellipsis. Oxford University Press, xv+262pp, hardback ISBN
0-19-924373-5, GBP 47.50; paperback ISBN 0-19-924372-7, GBP 18.99.
Oxford Studies in Theoretical Linguistics 1.
Jeffrey T. Runner, University of Rochester
[Note: The publisher had not yet announced this book on LINGUIST List
at the time that this review was posted. --Eds.]
SYNOPSIS
The book contains five chapters as well as a short introduction and
conclusion. Additionally, there are 17 pages of references, ten and a
half pages of indexes, including a short language index and name and
subject indexes. The first four chapters lay down the data, theoretical
groundwork and an interesting puzzle that set the stage for the fifth
and longest chapter, which is the heart of the book.
Sluicing, which Merchant argues is a type of IP ellipsis, is the topic
of the book. An example is the following (I will use '<x>' to indicate
an elided 'x'):
1. Gary visited somebody but he didn't say who <Gary visited>
His analysis makes the following claims, discussed in more detail below:
(1) Sluicing is a deletion process at the level of phonetic form (PF);
this builds on early work on the topic by Ross (1969). (2) No
morphosyntactic identity condition holds between the "sluiced" IP and
the antecedent IP which licenses it (contra e.g., Fiengo & May 1994):
the only identity condition is one of a particular kind of
semantic/pragmatic mutual entailment between the two IPs. (3) Certain
"islands" for syntactic movement involve features at PF which can be
deleted by Sluicing; certain other islands are true syntactic islands
and cannot be by-passed by Sluicing at PF.
The first chapter, Identity In Ellipsis: Focus and Isomorphism, is 28
pages long. Though the book focuses on Sluicing, Merchant begins by
presenting facts about a better-studied phenomenon, VP-ellipsis,
illustrated in a sentence like (2):
2. Gary visited Bill, and then the next day Mary did <visit Bill>
Following work by Rooth (1992), Swarzschild (1999) and Romero (1998),
Merchant argues that VP-ellipsis obeys a Focus condition, which
essentially states that a VP can be elided if it is GIVEN in a context.
GIVEN means that it is not in focus and that it has a particular type of
salient antecedent in the context. Merchant argues that the relevant
notion of GIVEN, which he calls e-GIVEN, is a mutual entailment
condition holding between the antecedent VP and the elided VP. That is,
it's not just enough for the elided VP to be entailed by the antecedent
VP, but the antecedent VP must be entailed by the elided VP.
This mutual entailment idea is the heart of Merchant's contribution to
the question of the relationship between the antecedent and the elided
material in ellipsis. Many approaches to ellipsis assume that a
morphosyntactic isomorphism condition holds between the antecedent VP
and the elided VP, in place of or in addition to a semantic/pragmatic
condition. Such a condition seems to be at play in failed VP-ellipsis
examples like (3):
3. *Abby was reading the book while Ben was <reading>
If all that were needed was that the antecedent VP entailed the elided
one, (3) should be acceptable, since reading a book entails reading.
The problem is that there is a direct object NP in the antecedent that
is not present in the elided VP. The standard solution to this problem
has been to propose a structural isomorphism condition, requiring that
the antecedent VP and the elided VP have the same syntactic structure.
Merchant discusses the approach along these lines presented in Fiengo &
May (1994), highlighting a challenge for that approach, illustrated in a
VP-ellipsis example like (4):
4. They arrested Alex-i, though he-i thought they wouldn't <arrest
Alex-i>
Without ellipsis this sentence would be ungrammatical since it contains
a Binding Theory Condition C effect: 'Alex-i' in the elided VP is
c-commanded by a coindexed pronoun, 'he-i'. However, the ellipsis seems
to make the problem disappear. Fiengo & May propose what they call
"vehicle change", which essentially allows under particular
circumstances (e.g., in VP-ellipsis) that the pronominal features of a
nominal may vary; this boils down to saying that 'Alex-i' can actually
be a pronoun, 'he-i', for the purposes of ellipsis, thus avoiding the
Condition C effect. Merchant points out that his mutual entailment
condition predicts this without a special vehicle change device. In
Merchant's view the sentence is actually (5):
5. They arrested Alex-i, though he-i thought they wouldn't <arrest
him-i>
Since all that matters is that the two VPs entail each other it is okay
for the elided VP to contain a coindexed pronoun rather than a full
NP--no morphosyntactic isomorphism condition forces the elided VP to
contain the full NP 'Alex-i' (I am glossing over certain details about
what it means for one constituent to entail another--see the book for
these details).
Chapter 2, The Syntax of Sluicing, is 46 pages long and argues that the
"Sluice" is a CP with an IP missing. In particular this chapter argues
that the missing IP contains the same syntactic structure that its overt
counterpart would contain. Merchant suggests that the null IP is
licensed by a particular combination of features in the head C position
of the CP sluice: [+wh, +Q]; this accounts for why sluicing is possible
in embedded questions ([+wh, +Q]) but not in relative clauses ([+wh,
-Q]). His particular proposal is that a special feature in the head I
of IP is what licenses the ellipsis itself. This feature, E, is argued
to move from I to C, to issue deletion instructions at PF (essentially
instructing the parser to "skip the complement of I"), and to impose the
Focus condition, mentioned in chapter 1. Merchant sees it as a positive
step to link the structural, deletion and focus conditions all together
in one feature of the morphosyntax.
One issue that arises is that the C head in Sluicing must always be
empty, which the following English example illustrates:
6. A: Max has invited someone. B: Really, who (*has)?
The Sluice never contains material in C. This is true in English as
well as in languages which more regularly allow material in C, such as
German, Dutch and Danish. It is also true for languages which place
special clitics in the C area, like the South Slavic languages.
Merchant provides two possible explanations for this. One is that the
movement to C "follows" IP-deletion; thus if the IP is deleted there
will be no verb or clitic to move to C. A second possibility is that
the "strong" feature (in the sense of the minimalist program, see
Chomsky 1995) which triggers the movement to C can be deleted at PF
(with the IP-deletion) obviating the need for movement to C (or perhaps
allowing the bare feature to move to C at LF but not at PF). This just
leaves one additional problem, which a movement/feature account cannot
solve: base-generated material, such as an actual complementizer, is
also banned from C in Sluicing. Merchant offers two suggestions. One
is a version of the doubly-filled Comp filter, specially stated to apply
in Sluicing contexts for languages that do not obey such a filter more
generally; and the other is to claim that the complementizers in
question perhaps normally cliticize to their right, so problems would
arise if their complement IP were deleted.
The third chapter, Islands and Form-Identity, contains 21 pages and sets
the stage for subsequent chapters by presenting a puzzle. On the one
hand, the wh-phrase that appears in the Sluice does not appear to have
arrived there by movement, since certain "islands" for movement do not
seem to be respected; this could suggest a base-generation account. On
the other hand, certain "form-identity" generalizations would be
difficult to state if Sluicing is not simply regular wh-movement; in
particular the morphological (e.g., case) form of the wh-phrase is
always what it would be if it were in its un-moved base position and
whether a language allows preposition-stranding under normal wh-movement
exactly predicts whether it allows preposition-stranding under
Sluicing. This would be difficult to ensure on a non-movement approach.
Merchant shows that Sluicing is fine even if the Sluice contains any of
the following types of islands: relative clause, adjunct,
noun-complement, sentential subject, embedded question, coordinate
structure, complementizer-trace, left branch and "derived" position
(=topicalization and subject) islands. An example of a Sluice
containing e.g., a relative clause is (7):
7. They wanted to hire someone who speaks a Balkan language, but I don't
remember which <Balkan language they want to hire someone who speaks>
If lack of sensitivity to islands indicates lack of movement, as is
often assumed, then these facts all point to the claim that Sluicing
does not involve wh-movement.
However, if Sluicing does not involve wh-movement then it becomes
difficult to explain what he calls "form-identity" generalizations. The
first is that the sluiced wh-phrase must bear the case that its
correlate bears; this shows up in case-marking languages like German,
where a verb like 'praise' assigns accusative case to its object but a
verb like 'flatter' assigns dative case; this distinction is maintained
in Sluicing:
8. Er will jemanden loben, aber sie wissen nicht, {*wer/wen/*wem}
'He wants to praise someone, but they don't know {who-*NOM/ACC/*DAT}'
9. Er will jemanden schmeicheln, aber sie wissen nicht, {*wer/*wen/wem}
'He wants to flatter someone, but they don't know {who-*NOM/*ACC/DAT}'
The second form-identity generalization is that a language will allow
preposition-stranding under Sluicing just in case it allows
preposition-stranding under regular wh-movement. He establishes this
generalization by looking at six preposition-stranding languages and 18
non-preposition-stranding languages. Again, this points to Sluicing
being a type of wh-movement. Thus, this chapter sets up a puzzle to be
solved later.
Chapter 4, Deletio nata atque mortua, is 50 pages long and reviews all
extant accounts of Sluicing. Merchant discusses five types of account,
including one of his own from previous work (Merchant 2000c), and shows
where each fails. He begins by discussing Ross's (1969) deletion
account. In favor of such an account is a straightforward explanation
of the identity-form generalizations discussed in Chapter 3. Ross
recognized that (some) islands were not respected by Sluicing and
suggested that grammaticality was calculated across a derivation, so
that if an early violation were later fixed grammaticality would
result. Since the island was deleted via Sluicing the violation was
repaired. However, Merchant points out that the fact that VP-ellipsis
does not repair island violations casts doubt on Ross's type of
analysis:
10. [Everyone wants to hire someone who speaks a different Balkan
language]
*Abby wants to hire someone who speaks Greek, but I don't remember which
(language) Ben does <want to hire someone [who speaks]>
A second analysis of Sluicing is that of "pseudosluicing". The claim
(made by e.g. Erteschik-Shir 1977 and Pollmann 1975) is that Sluicing
examples are produced by a kind of cleft construction:
11. Someone just left--guess who <it was>
Such an analysis would explain the island effects:
12. That he'll hire someone is possible, but I won't divulge who ?(it
will be).
Merchant provides nine arguments from various languages suggesting there
is a difference between Sluicing and this cleft-like "pseudosluicing"
construction. I will not outline them here for reasons of space.
The third type of analysis of Sluicing treats it as involving a
wh-operator that binds a resumptive pronoun. This would explain the
island effects straightfowardly since no movement would be required.
Merchant points out several problems with this idea. For one, Sluices
can occur with wh-words that do not have resumptive counterparts. A
second argument comes from languages like Irish which have a resumptive
pronoun strategy. These languages have a restriction that the
resumptive pronoun cannot be the "highest subject". No such restriction
holds of Sluicing.
The fourth type of analysis Merchant discusses is the best-known recent
one, that of Chung, Ladusaw & McCloskey (1995). Their analysis
base-generates the wh-phrase in CP and assumes that it binds an
indefinite NP, which they analyze as variable (following Heim 1982 and
Kamp 1981). Merchant points out several problems with this approach,
the most significant of which is its inability to account for "contrast"
Sluices like the following. (11) would have a logical form (LF)
representation like (12):
11. She has five CATS, but I don't know how many DOGS.
12. She has five CATS, but I don't know how many DOGS(x) [she has
CATS(x)]
Merchant's account can handle these cases because the mutual entailment
discussed in Chapter 1 and mentioned above replaces focused material
with variables so that the antecedent IP and the deleted IP both
contain: 'she has x'. Thus, they satisfy his e-GIVENness requirement,
which states that the two IPs must entail each other.
While each of the above approaches can account for the lack of island
effects in Sluicing none of them except Ross's can account for the
form-identity generalizations pointed out in Chapter 3. Merchant
(2000c) proposed a version of the Chung et al. (1995) approach that
attempts to capture the form-identity facts. On this view, which is an
LF-copying approach, the indefinite NPs in Sluicing undergo Quantifier
Raising (QR) leaving an IP containing a gap for copying into the
position of the Sluice. This alternative can explain the form-identity
generalizations since it is a movement account; however, it faces the
same problems the structural isomomorphism approach of Fiengo & May,
discussed in Chapter 1. In addition it relies on the dubious assumption
that indefinite NPs can QR out of islands at LF. For these reasons he
ultimately rejects this alternative as well.
The fifth and final chapter, Deletio Redux, contains 70 pages and is
intended to put together an analysis that can deal with everything
brought up in the preceding four chapters. To recap the issues,
Merchant has pointed out that the form-identity generalizations suggest
a movement approach to sluicing, along with PF-deletion; but the
apparent insensitivity to (some) syntactic islands suggests a
non-movement LF-copying approach. Merchant supports the movement plus
PF-deletion approach; thus, much of this chapter is devoted to
explaining the island (partial) insensitivity facts.
The fact that sluicing seems to be insensitive to some islands and not
to others suggests that islands come in different varieties sensitive to
different factors. Merchant develops this view by showing that what
have been called islands indeed fall into three distinct classes. The
first class, selective (or "weak") islands, Merchant claims are not
syntactic at all, but rather are semantic/pragmatic in nature. The
second class, which includes left-branch extraction, COMP-trace effects,
"derived" positions (topicalizations/subjects), and extraction of a
coordinate structure conjunct, Merchant argues are "undone" by PF
deletion. The third class, which includes extraction out of coordinate
structure conjuncts, complex NPs and adjuncts, all involve extraction
out of a propositional domain, which Merchant argues allow for an
analysis exploiting e-type anaphora. He sets aside the first class,
since their account does not interact with the question of whether
sluicing is PF-deletion or LF-copying (though addresses them at the very
end of the chapter). If Sluicing is PF-deletion and the second class of
islands are all due to some PF-feature conflict then the fact that
Sluicing can "undo" this type of island is explained. The third class
of islands are real non-PF islands but the use of e-type anaphora
combined with his pragmatic mutual entailment identity condition
accounts for the apparent violation of these islands. In fact, they are
not violated at all.
Space does not permit a detailed discussion of each island so just a few
examples will be outlined as illustration. Merchant devotes some time
to left-branch extraction, which is a member of the second class of
island types. He follows Kennedy & Merchant (2000a) which claims that
languages vary on whether their NPs can support a [+wh] feature in the
highest nominal projection. Those that can will allow left-branch
violations; those that cannot, like English, can only extract via this
highest nominal specifier if the unsupported [+wh] feature is deleted at
PF, e.g., via Sluicing (or VP-ellipsis). Left-branch extraction is
illustrated in (13). Sluicing seems to "fix" the problem (14)-(15).
13. *I don't know [how detailed]-i he wants [t-i' F[+wh][a t-i [list]].
14. He wants a detailed list, but I don't know how detailed.
15. He wants a detailed list, but I don't know [how detailed]-i <he
wants [t-i' F[+wh] [a t-i list]]>
A potential problem arises with respect to examples like (16):
16. *He wants a list, but I don't know how detailed.
This looks like a case where Sluicing hasn't fixed the left-branch
extraction problem. However, Merchant claims that (16) is not bad
because of a left-branch problem but rather because it violates his
mutual entailment identity condition. Essentially the issue is that
wanting a list does not entail wanting a detailed list.
Merchant points out that left-branch subextractions are not fixed by
Sluicing. Compare (17) and (18):
17. *[How badly]-i did you meet [a guy [t-i short of funds]]?
18. *She met a guy (badly) short of funds, but I don't know how badly.
Thus, whatever is causing the subextraction violation in (17) is not
being undone by PF-deletion. Merchant argues that the degree phrase
itself does not project the relevant FP through whose specifier the
measure phrase would be extracted, thus these remain islands
independently of whether an offending feature is deleted or not.
Another type of island that Merchant discusses in detail is what he
calls derived position islands. This refers to subject islands and
topicalization islands. An example of the latter is (19), with a
passive and with an unaccusative subject; and (20) illustrates how the
island effect is ameliorated by Sluicing:
19. *Guess [which Marx Brother] [a biography of t] {is going to be
published/will appear} this year.
20. A biography of one of the Marx Brothers {is going to be
published/will appear} this year--guess which!
Merchant's proposal is that (19) is out because extraction has taken
place out of a phrase that is not L-marked (following Chomsky 1986a).
Why is (20) good? Merchant proposes that within the Sluice the movement
to Spec,IP has not taken place and thus the extraction is from the
subject's base position, not its derived position. Since its base
position is L-marked no violation occurs. The question is why does the
subject in the Sluice not need to move to Spec,IP as it would in (19)?
The answer to that is that the feature that would normally drive the
subject-movement to Spec,IP, the EPP feature in I, is "strong" and thus
is uninterpretable at the PF interface. In (20), Sluicing has deleted
the IP, removing the offending feature.
Merchant explores the question of whether having the subject stay within
the VP at Spell-Out has any consequences for the interpretation. He
concludes that it does not, showing that indeed the subject can interact
with modals and negation as usual.
Before turning to the third class of islands that Sluicing seems to be
insensitive to, Merchant proposes an analysis of sentences like (21),
which forms the foundation for the analysis of that last class of
islands. The type of Sluice one might expect for these is in (22); the
problem is that the Sluiced IP contains an unbound trace:
21. The report details what IBM did and why.
22. The report details what-i [IBM did t-i] and why <IBM did t-i>
Merchant argues that (21) is not related to something like (22), but
rather to something like (23), which contains a pronoun in the second
IP:
23. The report details what IBM did and why <IBM did it>
The pronoun in (23) is an e-type pronoun licensed in the usual way (for
such pronouns) by a non-c-commanding quantifier, in this case the
wh-phrase in the antecedent IP.
This sets the stage for an analysis of how Sluicing appears to violate
the third class of islands, what Merchant calls propositional islands.
These include relative clauses, adjuncts and sentential subjects, and
coordinate structure conjuncts. A relative clause example is in (24).
In (25) is an example showing that usually extraction is not possible:
24. They hired someone who speaks a Balkan language--guess which!
25. *Guess which (Balkan language) they hired someone who speaks!
As in the previous examples, instead of relating (24) to something like
(25), Merchant argues that it is in fact related to something like (26),
which contains an e-type pronoun licensed by the quantifier in the
antecedent IP:
26. They hired someone who speaks a Balkan language--guess which-i <she
speaks t-i>
A similar analysis is given to other cases in which Sluicing apparently
violates propositional islands.
CRITICAL EVALUATION
Overall, I think this is an excellent book. It is very carefully
argued. It contains detailed discussion of alternative proposals,
examining their flaws. It brings in a wealth of data from other
languages with Sluicing. I am sure it will make a long-lasting
contribution to the study of both sluicing and ellipsis in general in
natural language.
There are two additional issues I would like to mention--one positive,
one possibly negative. First, part of Merchant's proposal is that the
only parallelism condition on ellipsis is his pragmatic/semantic
e-GIVENness condition and not a morphosyntactic condition. One positive
aspect of this proposal is that a certain amount of speaker variability
might be expected in ellipsis, which I believe is correct. A particular
example comes from Chapter 5 (judgments are Merchant's):
27. *He wants a list, but I don't know how detailed.
(27) and sentences like it seemed perfectly fine to me so I asked a few
other speakers and found that a few liked them and a few did not.
Recall that Merchant's analysis of (27) depends on whether one believes
that wanting a list entails wanting a detailed list. This seems like
something that could vary from speaker to speaker and from context to
context. Note that a morphosyntactic isomorphism requirement would rule
(27) out for everyone, including me and those speakers like me.
The other issue I want to bring up involves extraction out of DPs of
various sorts. It has been noted by a number of people that semantic
properties of the verb as well as referential/quantificational
properties of the DP itself both influence whether extraction out of a
direct object is possible (e.g., Erteschik-Shir 1977, Kuno 1987). For
example, verbs of destruction are worse than verbs of creation; and DPs
containing referential or strong quantifier determiners are worse than
those containing existential determiners:
28. Who did John write/??destroy a/??every/??the book about t?
Sluicing does not seem to care about this:
29. John wrote/destroyed a/every/the book about someone but I don't know
who.
An extension of Merchant's analysis might be to suggest that whatever
blocks (28) involves some strong PF feature, which Sluicing deletes in
(29). This seems somewhat counterintuitive, though, since the problem
in (28) seems likely to be an LF problem, not a PF problem, since the
relevant properties are semantic.
One could argue that (29) is good because the DP in the Sluice is in an
L-related position (cf., the discussion in the last chapter on "derived"
positions), but that would it seems commit one to the claim that the
object of the verb of destruction and the quantificational/referential
object in (28) are in non-L-marked positions at PF, while the object of
the verb of creation and the existentially quantified object are in
L-marked positions at PF. I know of no evidence for PF-differences in
position for these different DP-types. There have been proposals that
the relevant DPs in (28) are indeed in different positions at LF, due to
their interpretations, and that this distinction is what causes the
reduction in grammaticality (e.g., Diesing 1992, Runner 1995, 1998, who
both argue that this is an ECP violation at LF). But if the problem in
(28) is due to an LF violation then it would seem that the LF of the
Sluiced (29) is somehow different. This different type of island has
not been documented here. In addition, it seems to me that an
appropriate analysis of the amelioration caused by Sluicing in (29)
might actually extend naturally to the analysis of other derived
position islands, without having to posit such different LFs for the
otherwise interpretationally identical Sluiced material.
Overall, though, this book is worth reading and I would highly recommend
it.
REFERENCES
Chomsky, N. (1986a) Barriers. Cambridge, MA: MIT Press.
Chomsky, N. (1995) The Minimalist Program. Cambridge, MA: MIT Press.
Chung, S., W. Ladusaw & J. McCloskey (1995) "Sluicing and Logical Form."
Natural Language Semantics 3:239-282.
Diesing, M. (1992) Indefinites. Cambridge, MA: MIT Press.
Erteschik-Shir, N. (1977) On the Nature of Island Constraints.
Bloomington, IN: Indiana University Linguistics Club.
Fiengo, R. & R. May (1994) Indices and Identity. Cambridge, MA: MIT
Press.
Heim, I. (1982) The Semantics of Definite and Indefinite NPs. Ph.D.
dissertation, UMass-Amherst.
Kamp, H. (1981) "A Theory of Truth and Discourse Representation," in J.
Groendijk et al. (eds.), Formal Methods in the Study of Language.
Amsterdam: Mathematisch Centrum, 277-322.
Kennedy, C. & J. Merchant (2000a) "Attributive Comparative Deletion."
Natural Language and Linguistic Theory 18:89-146.
Kuno, S. (1987) Functional Syntax. Chicago: University of Chicago Press.
Merchant, J. (2000c) "LF Movement and Islands in Greek Sluicing."
Journal of Greek Linguistics 1:39-62.
Pollmann, T. (1975) "Een regel die subject en copula deleert?" Spektator
5:282-292.
Romero, M. (1998) Focus and Reconstruction Effects in Wh-Phrases. Ph.D.
dissertation, UMass-Amherst.
Rooth, M. (1992) "A Theory of Focus Interpretation." Natural Language
Semantics 1:75-116.
Ross, J.R. (1969) "Guess Who?", in R. Binnick et al. (eds.) Papers from
the 5th Regional Meeting of the Chicago Linguistic Society. Chicago:
Chicago Linguistic Society, 282-286.
Runner, J. T. (1995) Noun Phrase Licensing and Interpretation. Ph.D.
dissertation, UMass-Amherst.
Runner, J. T. (1998) Noun Phrase Licensing. New York: Garland
Publications.
Swarzschild, R. (1999) "GIVENness, AVOIDF, and Other Constraints on the
Placement of Accent." Natural Language Semantics 7:141-177.
ABOUT THE REVIEWER
Jeffrey T. Runner is an Assistant Professor in the Department of
Linguistics at the University of Rochester and has been teaching there
since 1994. He received his Ph.D. from the University of Massachusetts,
Amherst in 1995. His dissertation focused on direct objects in English,
exploring the relationship between syntactic position and interpretation
at various levels of representation. More recently, in addition to his
continued research on constructions involving objects, he has been
exploring the roles of syntactic structure and context in the domain of
Binding Theory, studying reflexives and pronouns from data collected
experimentally using a head-mounted eye-tracker.
Mail to author|Respond to list|Read more issues|LINGUIST home page|Top of issue