You are on page 1of 23

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/233658154

Reconstruction in relative clauses and the copy theory of traces*

Article  in  Linguistic Variation Yearbook · January 2005


DOI: 10.1075/livy.5.02cec

CITATIONS READS

10 110

1 author:

Carlo Cecchetto
French National Centre for Scientific Research
87 PUBLICATIONS   787 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

SIGN-HUB - Task: Life stories of elderly Deaf signers View project

All content following this page was uploaded by Carlo Cecchetto on 19 May 2014.

The user has requested enhancement of the downloaded file.


A slightly revised version of this paper has been published as:

C. Cecchetto (2005), "Reconstruction in relative clauses and the copy theory


of traces" in P. Pica e J. Rooryck (a cura di) Linguistic Variation Yearbook
N. 5: 73-103, John Benjamin, Co., Amsterdam

1
Reconstruction in Relative Clauses and the Copy Theory of Traces
Carlo Cecchetto
Università degli Studi di Milano-Bicocca

Abstract
In this paper, after discussing the status of the copy theory of traces in the current formulation of the
minimalist program and the evidence for the “No-Tampering” Condition from which the copy theory
of traces follows, I focus on a specific case study, namely reconstruction effects concerning the head
of a relative clause. The common wisdom in the literature is that reconstruction of the relative clause
head can be observed by using variable binding as a diagnostic, while the diagnostic based on
Condition C gives opposite results. This split has led some researchers to propose that relative clauses
are structurally ambiguous, because they would receive both a raising analysis (which explains
variable binding reconstruction) and a non-raising analysis (which explains the absence of Condition
C reconstruction). One of the goals of this paper is showing that it is not necessary to postulate that
relative clauses are structurally ambiguous. In order to do that, I first show that the description in the
literature is partly inaccurate. If some methodological problems raised by the use of transitive nouns
are avoided, it can be shown that variable binding reconstruction occurs only when the relative clause
modifies the subject of an equative sentence. This suggests that variable binding reconstruction of the
relative clause head is not an ordinary case of reconstruction like the one found in canonical wh chains
but should be treated as a case of indirect binding, which is known to be sensitive to the
identificational (as opposed to predicational) character of the sentence. I then show that, if this
perspective is taken, the absence of Condition C effects can be explained without positing a structural
ambiguity. The final result of the investigation is that, despite the initially puzzling evidence, the copy
theory of traces can successfully explain the reconstruction pattern of relative clause head.

Keywords
Copy Theory of Traces, Reconstruction, Relative Clauses, Indirect Binding, Unaccusative Nouns

2
1. Introduction

Arguably, the most stable feature of the Minimalist Program, starting from the earliest formulations up
to the most recent version (for example Chomsky 2005), is the copy theory of traces. This aspect of
the program originated much empirical work and it is fair to say that our knowledge of the interface
between syntax and interpretation has significantly improved due to this work.
After briefly reviewing the conceptual motivation for assuming the copy theory of traces in the current
version of the minimalist program, I will discuss the type of empirical expectations that arise when the
copy theory is coupled with standard assumptions on binding. These expectations are met in
uncontroversial cases of movement, as extensively shown in the literature. There is one particularly
difficult case that arises in relative clauses constructions, though. In this paper, I will focus on this case
by trying to show that it stops being a puzzle for the copy theory of traces when some interfering
factors are taken into considerations. This paper is organized as follows. Section 2, 3 and 4 set the
scenario in which the debate on relative clauses is located. Section 2 contains a general discussion on
the status of the copy theory of traces. In section 3, I ask what specific forms the binding principles
should have once the presence of multiple copies in the LF configuration is taken into consideration.
Section 4 concludes the initial part of the paper by presenting the two main analyses for relative
clauses that have been proposed, namely the raising and the non-raising analysis. The remaining part
of the paper is entirely devoted to reconstruction effects in relative clauses. In section 5, I argue that,
in order to avoid some disturbing factors, reconstruction effects should be tested in relative clauses
which have an unaccusative noun as a head. In section 6, I claim that, if the methodology argued for in
section 5 is adopted, it is possible to show that variable binding reconstruction effects within a
complex NP (or DP) arise only if this complex NP (DP) is the subject of an identity sentence. I also
claim that the raising analysis cannot explain the limitation of variable binding reconstruction effects
to identity sentences in any obvious way. In section 7, I suggest that variable binding reconstruction
effects in identity sentences is a semantic phenomenon similar (and reducible) to the functional
interpretation of questions and relative clauses. Section 8 is devoted to discussing the total lack of
Condition C reconstruction effects in relative clauses. Section 9 concludes the paper.

2. The Copy Theory of Traces in the Minimalist Program


One of the leading idea of the minimalist program is that the computational system should be kept as
simple as possible. This requirement meets obvious methodological desiderata but is also inspired by
the hypothesis that the language organ might embody some abstract idea of perfection.
This simplicity requirement leads to the formulation of the so-called “No-Tampering Condition”,
which dictates that syntactic operations that apply to two arbitrary objects X and Y should leave X and
Y unchanged. Since the copy theory of traces can be seen as a corollary of the No-Tampering
Condition, let us spend a few words on it. Take the operation Merge, which combines smaller units in
bigger ones. The units to which Merge apply can either be lexical items (assembly of features) or
complex objects (objects that have been formed by a previous occurrence of Merge). Suppose that
Merge combines two complex objects X and Y and that Y is merged to X. Let us ask which kind of
object Y can be merged to X. There are two logical possibilities: Y can or cannot be part of X. The
former case is the operation that is commonly called Movement (Head Movement if Y is a lexical
item, Phrasal Movement if Y is the result of a previous occurrence of Merge). Movement is re-labeled
Internal Merge by Chomsky (2005), to stress its nature. The latter case is an extension of the structure
through insertion of material that was not already present in X. The new material can either come from
the lexical array from which the derivation stems (assuming that the access to the lexicon takes place
through a one-time selection of a lexical array that feeds all the derivation) or from a parallel
derivation. This type of extension is labeled External Merge.

3
The No-Tampering Condition dictates that Merge cannot break up X or Y. Therefore, the only
possibility is for Y to be merged to the edge of X. So, both Internal Merge (Movement) and External
Merge must extend the structure. This derives that syntactic operations cannot be counter-cyclical1.
The copy theory of traces follows straightforwardly from the No-Tampering Condition. To appreciate
this, think how movement was conceived in the Government and Binding Theory. In that framework,
when Movement applies, the original instance of the moved category is replaced by a trace which is
coindexed with the category in the new position. In modernized terms, this trace-insertion operation
would be an example of tampering. In fact, when Y is externally merged to X, X should not be
changed by this operation. So, the only legitimate option is the creation of a new copy of Y to the edge
of X and the conservation of the original copy of Y. In a nutshell this is the copy theory of traces:
movement creates multiple copies in the positions which the moved category goes through.
The persistence of a copy of a moved element in the original position is thus motivated on conceptual
grounds. It remains to be seen if this conclusion is empirically supported or not. There are two obvious
place in which this should be checked: the phonetic side and the interpretative component.
As for phonetic side, it is fair to say that the evidence for multiple copies of the moved element is slim.
As a matter of fact, in the overwhelming majority of cases we observe deletion of all but one copy.
Chomsky (2005) briefly speculates on this. He observes that there might be conflicting factors
operating. One factor that favors maintaining all the copies in the PF component is ease of processing.
It is not difficult to come up with cases in which maintaining all the copies would disambiguate a
sentence. One obvious case is (1), in which spelling out the lower trace would resolve the ambiguity
between the matrix reading (cf. 2) and the embedded reading (cf. 3).

(1) To whom did you say he had mailed a postcard ?


(2) To whom did you say to whom he had mailed a postcard?
(3) To whom did you say he had mailed a postcard to whom?

However, in the typical case only one copy is pronounced, even if this makes processing harder.
Chomsky conjectures that this might be due to a stronger requirement that imposes minimizing the
computation2.
I don't have much to add to this but for commenting that, if it is true that conflicting factors operate,
there should be cases in which specific features of the linguistic context determine a different outcome
and multiple copies survive at PF. I think that one such case might be spelling out of multiple copies
in sign languages. For example, wh chains are commonly characterized by the presence of multiple
spelled out copies in many sign languages (see Neidle et al. 2000, Nunes and de Quadros 2005 and
Petronio and Lillo-Martin 1997 for data presentation). A possible suggestion is that the way
articulation works in the visual-gestural modality makes reduplication less costly3. More specifically,
it is tempting to relate a bigger tolerance for reduplication to the fact that sign languages are
multidimensional in a degree which is not possible in spoken languages. For example, non manual
marking of grammatical features (for instance, lowered eyebrows which obligatorily marks wh signs in
many sign languages) can (often must) co-occur with signs that are manually articulated. There are
also cases in which one hand maintains a certain shape (a classifier shape or a wh shape) while the
other hand articulates other signs. So, reduplication might be less costly in sign languages because the

1
Incidentally, the formulation of the extension condition based on the No-Tampering Condition does not introduce any
new perspective on the problem introduced by Head Movement configurations, in which strictly speaking Internal Merge
does not apply at the edge. In this paper, I do not discuss Head Movement, though.
2
Note that this line of reasoning does not explain why it is the higher copy to be pronounced. Se Moro (2000) and Nunes
(1999) for specific proposal to explain this.
3
Reduplication is often associated to perseveration (see Neidle et at. 2000 for a review of the data). By perseveration one
refers to the maintenance of a particular articulation that will recur later in the sentence). For example, wh phrases co-occur
with a specific facial expression in many sign languages (roughly, lowered eyebrows). If two copies of the wh phrase are
pronounced in two distinct positions in the sentence, the facial expression associated to the wh phrase remains stable over
the manual material that intervene between the two copies.

4
articulation of some grammatical information in many cases occupies only one channel, leaving others
free for simultaneous articulations of other grammatical information.
Since in this paper we are concerned with the LF side of the copy theory of traces, I will leave the
suggestion that multidimensionality makes repetition less costly in sign languages as a speculation.
Even our cursory discussion suffices to make clear that strong evidence in favor of the existence of
multiple copies in the syntactic component cannot come from the PF side of the computation, at least
if one puts sign languages aside. Let us then switch to the LF side.

3. Binding Principles and the Copy Theory of Traces


A priori one might expect that the same copy cancellation mechanism that routinely apples at the PF
side (again, putting sign languages aside) applies at the LF side as well. If things were like that, there
would be no direct manifestation of the existence of multiple copies and the copy theory of traces
could not be easily evaluated on empirical grounds. The situation seems to be different, though. In a
moment we will review some clear pieces of evidence that allows us to test (and confirm) the
hypothesis that a LF chain is made of multiple copies. Assuming for the sake of the argument that
multiple copies do stay at LF, a natural question that arises is what set the semantic and the
phonological component apart in this respect. Why doesn't minimization of computation lead to copy
cancellation at LF, if minimization of computation does that at PF? Chomsky (2005) answers that "in
the semantic component minimization of computation is not a factor, because semantic computation is
universal, so automatic and costless. The phonetic interface, on the other hand, tolerates substantial
variety, and involves language specific and complicated computation".
Assuming that something like this is on the right track, let us switch to the data that shows the
existence of multiple copies, namely reconstruction effects. Reconstruction is intended here as an
informal label for cases in which a certain element which has undergone movement (or Internal Merge
in modernized terms) behaves as if it still were in the position occupied before movement. One
example is (4)4:

(4) *Which rumors about Billi did hei resent most which rumor about Billi?

He and Bill must be disjoint in interpretation in (4) and this follows from Condition C of Binding
Theory if which rumors about Bill for interpretative purposes behaves as if it occupied its base
position. Reconstruction effects like the one in (4) are often taken as strong evidence for the existence
of multiple copies, because it is possible to say that what triggers a Condition C violation is the lowest
copy (in the Government and Binding framework, in which it was assumed that movement leaves
traces instead of a copies, cases like (4) required an actual operation of reconstruction, which literally
undoes the effect of movement).
The syntactic approach based on the copy theory of traces is not the only possible treatment for
reconstruction effects. An alternative is the semantic approach that makes use of flexible types for
traces (cf. Sternefeld 1997, among others). The syntactic and the semantic approaches can be hard to
distinguish in term of predictions, at least in core cases of reconstruction (but see below for a case that
allows us to distinguish them). A conceptual argument in favour of the copy theory of traces is that it
simplifies the syntactic computation by allowing us to obey the No-Tampering-Condition (as
explained above), while the flexible type approach complicates the semantic computation, since it
must be assumed that the semantic type of traces is not uniform (traces in the semantic approach, in
addition of being of type e, can be of higher type). Admittedly, this type of conceptual argument is
hard to evaluate, because it involves a global evaluation. However, in this paper I will assume that
this argument is sound enough to consider the copy theory of traces a privileged approach to
reconstruction.

4
In this paper, when it is convenient, I will explicitly indicate all the copies in a chain and I will strike those that go
unpronounced. I use indexes just for convenience.

5
Reconstruction effects can be taken to be evidence that full copies of the moved constituent survive at
the level of computation in which binding principles are computed. In this sense, the existence of
copies is fully consistent with the No-Tampering Condition. A possible objection is that copies must
be interpreted as variables or elements that contain variables (see Fox 2002 for a possible
implementation). So, ultimately the No-Tampering Condition will be violated in the semantic
component, no matter what happens in core syntax. However, although something like that must be
certainly true, I do not see this as a real objection against the No-Tampering Condition but rather as an
observation that supports the modular organization of the different linguistic components. In the spirit
of much work on the syntax-semantics interface, one can say that there is a syntactic level of
representation (call it LF) at which the binding principles are computed and at which the No-
Tampering Condition still holds. This level is the input to the semantic component. In the semantic
component much of the syntactic structure is radically simplified and tampered, but this does not mean
that the No-Tampering Condition is not valid in core syntax.
Let us try now to be more precise and study in detail the interaction between the copy theory of traces
and the various grammatical conditions that are relevant for reconstruction effects. The question is
interesting because the definitions of binding principles that have been proposed typically are not
formulated by keeping in mind the presence of multiple copies (this is in part historical accident,
because much work in this area was done when the copy theory of traces was not developed yet 5). The
specific question that we should address is what happens when one of the copies in the chain obeys
some binding principle, while another copy in the same chain does not. Is the sentence in which this
happens grammatical or not?
Let us start from Condition C. (4) above suffices to show that the presence of a copy in the illicit
configuration makes the entire structure in which the violation occurs ungrammatical, no matter if
another copy of the same element sits in a licit configuration in another part of the structure. As matter
of fact, (4) is out because the lowest copy of the R-expression Bill is incorrectly bound by the pronoun
he, while the highest copy of Bill, which is in a licit position, does not help.
Assuming a very simplistic formulation of Condition C, which is nonetheless sufficient at this very
general level of discussion, we might propose the following formulation that explicitly takes into
account the existence of multiple copies6:

(5) CONDITION C
Each copy of an R-expression is prevented from being c-commanded by a coindexed pronoun.

The formulation in (5) is by no means a matter of conceptual necessity. It would not be difficult to
imagine an alternative formulation that rules in a structure in which at least one copy of the R-
expression is not c-commanded by a coindexed pronoun. The alternative formulation, would be
empirically inadequate given examples like (4), though. In addition, the formulation in (5) is also the
most natural, given the fact that Condition C is a negative condition (a sort of prohibition) and a
prohibition cannot be satisfied if it is selectively disobeyed.
There are other grammatical constraints that are positive in nature, because they require that a certain
configuration hold, instead of dictating that it cannot hold. Two obvious examples are Condition A
and the constraint that a bound variable must be c-commanded by its binder by LF. The interesting
case of interaction with the copy theory of traces is again one in which one copy of the bound pronoun
(or of the anaphor) in correctly c-commanded while a second copy is not. Two examples are (6) and
(7):

(6) Which rumors about himselfi did Billi resent most which rumors about himselfi?
(7) Which of hisi professor does [every student]i like most which of hisi professor?

5
However, specific devices to deal with reconstruction effects were proposed in the Government & Binding framework.
One sophisticated theory in which binding principles are defined in terms of chains was developed by Barss (1986).
6
In section 8, we will be more precise on the conditions that prevents a R-expression from having the same semantic value
of a c-commanding pronoun.

6
The grammaticality of (6) and (7) shows that nothing goes wrong with having one copy in the illicit
configuration (in the case at hand, the "bad" copy is the highest one) as long as there is another one
that sits in the right position (in the case at hand, the "good" copy is the lowest one). Here we witness
a difference between Condition C, on the one side, and Condition A and the constraint for bound
variables, on the other side. Clearly, the basis to explain this difference lies in the positive/negative
character of these conditions. Condition A and the constraint for bound variables are positive
conditions. We can think of a positive condition as a request and a request can be satisfied by a single
act of accomplishment. Stated differently, a positive condition has an existential (rather than a
universal) character. Therefore, we can propose the following formulation7:

(8) CONDITION A
At least one copy of an anaphor must be locally c-commanded by a suitable antecedent.
(9) CONSTRAINTS ON BOUND VARIABLES
At least one copy of a bound variable must be c-commanded by its antecedent by LF8.

After this discussion, we can conclude that the typical situation that arises in case of movement is the
following. Whenever we observe Condition C reconstruction effects, we also observe Condition A and
variable binding reconstruction effects.
The copy theory of traces proved useful also with more fine-grained cases. The typical example is
anti-reconstruction effects, exemplified in (10):

(10) Which rumors that Billi spread out did hei regret having trusted?

Although the data is delicate and the relevant generalization has been occasionally challenged (as a
reviewer points out), following the literature on this topic I will assume that (10) forms a minimal pair
with (4). (10) is considerably better than (4). Lebeuax (1989) and Chomsky (1995) propose that the
absence of Condition C reconstruction effects is easily explained if the copy theory of traces is
coupled with the assumptions that adjuncts can be inserted late in the derivation, since they do not
obey the Projection Principle. In particular, the relative clause that contains the R-expression in (10)
can be inserted in the derivation when wh movement has already taken place:

(10') Which rumors that Billi spread out did hei regret having trusted which rumors?

7
As previously observed for Condition C, it would not be difficult to imagine alternative formulations of Condition A that
require that all copies of an anaphor be locally c-commanded by a suitable antecedent. The alternative formulation would
be empirically inadequate, given examples like (6), and would also be at odds with the positive character of Condition A.
The formulation of Condition A given in the text is adequate also for a complex configuration like (i), which is routinely
used to argue in favor of the existence of intermediate copies (traces):
(i) Which picture of himselfi/j does Johni think that Billj likes?
The fact that the anaphor can be bound either by John or by Bill confirms that only one copy must be properly c-
commanded in order for Condition A to be obeyed. The copy in the base position licenses the reading in which Bill is the
binder, while the copy in the intermediate position licenses the reading in which John is the binder.
8
One reviewer objects to my reasoning in the following way. (S)he observes that if the upper copy of which of his
professors in (7) doesn’t get deleted, this means that his has to get bound (without being c-commanded by a “binder”). If
such binding is possible, then there is no need of the downstairs copy at all. Alternatively, one could say that this is not
possible and that the upper copy gets deleted. But if it got deleted, the whole discussion of the difference between the way
Condition C is formulated (as a universal condition) is formulated and Condition A is formulated (as an existential
condition) would become irrelevant. The reasoning of the reviewer presupposes that copies can be deleted, in violation of
the No-Tampering Condition. This tampering operation is surely necessary in the semantic component, in which traces
must be converted in variables, as I have mentioned in the text. However, my attempt is to give a syntactic account of the
binding principles under the assumptions that all copies survive in compliance to the No-Tampering Condition, which, by
assumption, holds in core syntax.

7
In the configuration (10'), no copy of the R-expression Bill is c-commanded by the coindexed pronoun
he. A similar escape hatch is not possible for sentences like (4), because the R-expression is within the
complement of the noun rumor. Anti-reconstruction cases like (10) are interesting because, while they
admit a straightforward treatment under the copy theory of traces, they do not admit any obvious
treatment under the semantic approach to reconstruction.
So far, so good. The copy theory of traces has been successful enough that the presence/absence of
reconstruction effects is considered today a reliable diagnostic for the presence of copies in
constructions in which it is not obvious if movement has occurred or not. From now on, we
concentrate on one such case, namely the relation between the head of the relative clause and the
position of the gap.

4. Two Possible Types of Analyses for Relative Clauses


A long standing debate in the literature on relative clauses concerns the choice between what is often
called raising analysis and what I will call non-raising analysis. Following a terminology which is
pretty standard, I will call head of the relative clause (or simply, head9) the restriction of the
determiner which introduces the complex NP containing the relative clause (for example, the head in
sentence 11 is the noun boy but the head in sentence "the Italian nephew of John that I cannot stand is
visiting him" is the category Italian nephew of John):

(11) A boy [I met e yesterday]

Clearly, the head and the gap inside the relative clause must be related to each other. The choice
between the raising and the non-raising analysis concerns how they are related10. According to the
raising analysis, the head and the gap are transformationally related. The gap is literally a trace (copy)
left by the relative clause head:

(12) A boy [I met boy yesterday]

According to the non-raising analysis, the link between the relative clause head and the gap is
mediated by a null operator. In particular, the gap is the trace of a null operator which turns the
relative clause into an open predicate which intersects with the predicate the head expresses:

(13) A boy [OPi I met ti yesterday]

One traditional argument in favor of the raising analysis is the presence of reconstruction effects in
sentences like (14):

(14) The picture of himself [that John likes e most] (was never on display)

The absence of Condition A effects in (14) can be easily explained if one assumes the raising analysis
together with the copy theory of traces. In this approach the empty category in (14) contains a copy of

9
This means that in this paper the term "head" does not refer to the concept of head of phrase structure theory. As should
be clear, the head of a relative clause does not need to be the head of the NP in the sense of phrase structure theory.
Although this terminology can be misleading, I will adopt it, because it is pretty standard.
10
The literature on relative clauses is extensive and I cannot summarize it. In particular, there are various variants of the
raising and of the non-raising analysis and it is also possible to find proposals that do not fit either of the two approaches as
I have sketched them. Kayne (1994) and Bianchi (1995) develops variants of the raising analysis while Borsley (1997)
defends the non-raising analysis. These papers contain a more complete discussion of the reasons supporting the two
competing analyses. Cresti (2000) is an example of a paper that does not fit either the raising or the non-raising analysis. I
will also avoid discussing non restrictive relative clauses, which have peculiar properties that set them apart (cf. Bianchi
1995).

8
the anaphor himself, which is commanded by its antecedent John. (15) illustrates a case of variable
binding reconstruction effect, which can be taken as further evidence in favor of the raising analysis.

(15) The relative of hisi that [every boy]i likes lives far away

Although the presence of reconstruction effects in sentences like (14) and (15) is often taken to be a
strong argument in favor of the raising analysis, I will argue that despite its strong initial plausibility,
this argument is spurious.
A first problem for this argument has already been discussed in the literature (see Munn 1994 for an
early discussion) and involves the absence of Condition C reconstruction effect in sentences like (16):

(16) The relative of Johni that hei likes lives far away

In the case at hand, we observe a dissociation between Condition C reconstruction effects, which are
missing, and Condition A and variable binding reconstruction effects, which are attested. This fact is
surprising. Empirically such a dissociation is unattested in uncontroversial cases of movement, as we
have shown by using wh movement in section 3. Technically, the formulations of binding principles
that we have seen to be adequate in section 3 do not explain the facts in (14) to (16).
Sauerland (2000) uses this dissociation as evidence that relative clauses are structurally ambiguous
between the raising and the non-raising analysis (see Carlson 1977 and Heim 1987 for earlier claims
that relative clauses are structurally ambiguous and Bhatt 2002 for independent evidence for this
claim). When relative clauses are given a raising analysis, Condition A and variable binding
reconstruction effects result. When relative clauses are given a non-raising analysis, no Condition C
reconstruction effect arises, for the R-expression in the relative clause head is not c-commanded by the
material inside the relative clause. Sauerland’s approach straightforwardly explains the dissociation
between the different types of reconstruction effects but is theoretically very costly. It introduces an
undesirable redundancy in the theory, because relative clauses are treated on a pair with structurally
ambiguous sentences like (17):

(17) Flying planes can be dangerous

However, there is a fundamental difference between classical cases like (17) and relative clauses. The
ambiguity in (17) can be eliminated and the two readings can be set apart simply by making a different
lexical choice (cf. 18) or by making agreement overt (cf. 19 and 20):

(18) Building airplanes can be dangerous


(19) Flying airplanes is dangerous
(20) Flying airplanes are dangerous

Disambiguating between the raising and the non-raising analysis is much less straightforward. Surely
the dissociation in reconstruction effects might be taken as evidence that points to this direction, but I
think that positing a structural ambiguity is such a radical move that it should be done only when all
the alternative possibilities have been unsuccessfully explored. In section 8, I will propose that there is
an alternative approach to the dissociations in reconstruction effects that is worth pursuing, so I won't
assume that relative clauses are structurally ambiguous.
The second observation that leads me to think that reconstruction effects in relative clauses are not
strong evidence for the raising analysis is the main empirical contribution of this paper: despite what is
commonly assumed in the literature (and what I myself have reported in the introductory part of this
paper), I will show that reliable cases of reconstruction effects are not present in subject-predicate
sentences but only occur in identity sentences. This limited distribution of reconstruction effects
cannot be easily accommodated by the raising analysis, but I will argue that the non-raising analysis
can do the job if it is coupled with some independently motivated auxiliary hypotheses.

9
The general conclusion that I will draw from my discussion of the interpretative properties of relative
clauses is that the presence of reconstruction effects is a reliable diagnostic for the presence of copies
(namely, for the fact that movement has occurred in a certain construction) only if:
i) there is no dissociation between Condition A, Condition C and variable binding reconstruction
effects and
ii) reconstruction effects are not limited to identity sentences but also arise in subject-predicate
sentences.

5. A Methodological Caveat: Transitive Nouns are not Reliable

Let us start discussing reconstruction effects involving the relative clause head. Data concerning
reconstruction effects are notoriously delicate and the data we are going to consider are not an
exception. There is a specific complication with measuring reconstruction of the relative clause head
that I would like to comment on. The complication is that NPs can have an implicit subject PRO (cf.
Giorgi and Longobardi 1991, especially chapter 4). So, one cannot exclude that the anaphor in cases
like (14), repeated as (21) below, is actually bound by PRO, which sits in the subject position of the
NP (the details of the DP internal structure are irrelevant for our purposes):

(21) [DP The [NP PROi picture of himselfi] [that Johni likes e most]] (was never on display)

If the structure in (21) is a viable one, the absence of the Condition A violation is not a case of
reconstruction, because the position in which himself overtly sits is c-commanded by a suitable
antecedent for himself (PRO). In turn, PRO is controlled by John, but this control configuration is a
case of backward pronominalization that does not require c-command. So, in order to explain the
grammaticality of (21), it is not necessary to assume that a copy of the picture NP is found in the
complement position of the verb like.
I believe that many alleged cases of reconstruction of the relative clause head can, and in fact should,
be treated as cases of binding by an implicit PRO. One case that clearly shows this is the Italian
sentence in (22):

(22) La descrizione di se stesso [che e aiuterebbe Gianni a passare l'esame] (non è stata presa in
considerazione dalla commissione)
The description of himself that would help Gianni to pass the exam (was not considered by the
committee)

The acceptability of (22) cannot be due to the fact that the head of the relative clause is "interpreted"
in the position of the gap, which I have indicated with e (that is, 22 cannot be a case of reconstruction).
This is so because the gap e (the subject position of the relative clause) is not c-commanded by the
alleged antecedent of the anaphor (Gianni). So, the only way to explain the acceptability of (22) is
assuming that it is PRO in the subject position of the complex NP which acts as a binder of the
anaphor:

(22') [DP La [NP PROi descrizione di se stessoi] [che e aiuterebbe Giannii a passare l'esame]]

This point is rather important because most alleged cases of reconstruction in relative clauses
discussed in the literature involve transitive nouns like picture that in principle can take an NP internal
subject PRO. If I am right, a large part of the literature on this topic rests on a shaky foundation.
So, in this paper, following Bianchi (1995), I make a radical methodological choice and I will test
reconstruction only in structures in which the head of the relative clause contains an unaccusative noun
(for convenience, I will call unaccusative head such a structure)11. The reason is that, as I am going to
11
I won't discuss reconstruction involving unergative heads (cf. i), because according to some theories unergative
structures are concealed transitive structures (cf. Hale and Keiser 1993).

10
show, unaccusative heads do not have a slot for an NP internal subject PRO, so the complication
illustrated in (22) can be avoided. Since in Italian there is a clear diagnostic for unaccusativity, I will
base my discussion of reconstruction on Italian data. However, for convenience of the reader I will
occasionally use English sentences when they duplicate the Italian pattern.
Unaccusative nouns are deverbal nouns which are derived from unaccusative verbs. The reason why
unaccusative nouns are more reliable than transitive ones can be made clear by using an example.
Take a noun like naufragio, which is derived from the verb naufragare. Naufragare is unaccusative
according to the standard tests of unaccusativity (cf. Burzio 1986). It takes the auxiliary essere (cf.
23), occurs in reduced relative clauses (cf. 24) and in absolute small clauses (cf. 25) and allows ne-
cliticization (cf. 26),:

(23) La nave è naufragata


the ship was wrecked
(24) La nave naufragata ieri (è stata ritrovata oggi)
the ship wrecked yesterday (was found today)
(25) Naufragata la nave (furono chiamati i soccorsi)
wrecked the ship was called help
(26) Ne sono naufragate tre
ne (=of them) were wrecked three

I assume that the noun naufragio inherits its thematic grid from the verb naufragare. So, it takes only
one argument (the internal one). Now consider (27):

(27) Il [naufragio della propriai nave] [che [ogni capitano]i/Giannii teme e] (è quello che può avvenire
durante la regata principale)
The shipwreck of self's ship that every captain/Gianni fears is the one that can happen during the main
regatta

If naufragio takes only one argument, there is simply no slot for a potential binder of the anaphor
proprio inside the relative clause head. Therefore, the acceptability of (27) shows that the head of the
relative clause is somehow "interpreted" in the position of the empty category e, in which proprio is c-
commanded by its binder. So, we can be sure that (27), unlike cases involving transitive nouns like
picture, is a genuine case of reconstruction. There is independent evidence for this conclusion. In (27)
the gap inside the relative clause is c-commanded and binding of proprio obtains. If one modifies the
structure in a way such that the gap inside the relative clause is not c-commanded, the prediction is
that binding of proprio should become impossible. This is what happens:

(28) ?? Il naufragio della propriai nave [che e umilia [ogni capitano]i/Giannii] (è quello che può
avvenire durante la regata principale)
The shipwreck of self's ship that humiliates every captain/Gianni is the one that can happen during the
main regatta

(28) has the typical status of Weak Crossover (WCO) violations, in accordance with the fact that the
empty category e is c-commanded by its binder only if the latter crosses over the former. All in all, we
can conclude that, when an unaccusative head is used, it is interpreted as if it occupied the position of
the gap inside the relative clause. Stated differently, reconstruction effects arise in this structure.
Before proceeding, I must make clear that the main empirical observation that I am going to argue for
(namely, the different behavior of relative clauses in identity and subject-predicate sentences) surfaces

i) La telefonata a sua madre che ogni italiano ha fatto (è costata molto)


The phone call to hisi mother that [every Italian]i made was very expensive
If these theories are right, reconstruction involving unergative heads might be as spurious as reconstruction involving
transitive heads.

11
in a clear way only if we consider unaccusative heads. If we extend our considerations to other cases
of relative clauses, the judgments become murky. My claim is that this is so because of the possible
presence of an NP internal PRO.

6. Reconstruction in Identity and Canonical Subject-Predicate Sentences

In this section, I claim that, if unaccusative heads are used, it is possible to show that reconstruction
effects with the material inside the relative clause head is only found in identity sentences. In each of
the following minimal pairs, there is a contrast between the a sentence, which is an identity sentence,
and the b sentence, which is a canonical subject-predicate structure12. In the a sentence, variable
binding reconstruction is observed and the structure is judged either perfect or fairly acceptable by the
native speakers I have consulted (including myself). On the contrary, the b sentence is significantly
degraded. Every speaker that I have consulted judges the b sentence markedly worse than the
corresponding a sentence. In all the examples, the relative head is unaccusative, for the reasons just
mentioned13.

(29a) The one accident of hisi that everyonei remembers is the one that affected himi first
(29b) *The one accident of hisi that everyonei remembers affected himi first

(30a) Il proprioi fallimento che nessunoi dimentica è quello che è avvenuto per primo
The self's failure that nobody forgets is the one that happened first
(30b) * Il proprioi fallimento che nessunoi dimentica è avvenuto per primo
The self's failure that nobody forgets happened first

(31a) ? Il suoi fallimento che nessunoi dimentica è quello che è avvenuto per primo
His failure that nobody forgets is the one that happened first
(31b) * Il suoi fallimento che nessunoi dimentica è avvenuto per primo
His failure that nobody forgets happened first

(32a) Il proprioi articolo che nessunoi apprezza è quello che sta in fondo alla lista delle pubblicazioni
The self's paper that nobody appreciates is the one that stays last in the publications list
(32b) * Il proprioi articolo che nessunoi apprezza sta in fondo alla lista delle pubblicazioni
The self's paper that nobody appreciates stays last in the publications list

(33a) ? Il suoi articolo che nessunoi apprezza è quello che sta in fondo alla lista delle pubblicazioni
His paper that nobody appreciates is the one that stays last in the publications list
(33b) * Il suoi articolo che nessunoi apprezza sta in fondo alla lista delle pubblicazioni
His paper that nobody appreciates stays last in the publications list

(34a) Il ribaltamento della propriai barca che nessunoi ricorda è quello che è avvenuto durante la
scuola di vela
The upsetting of self's boat that nobody remembers is the one that happened during the sailing school

12
I use the term "canonical subject-predicate structure" to refer to simple cases in which a property is predicated of an
individual (an expression of type e).
13
Given the limitation to unaccusative heads, I will be able to test reconstruction effects by using pronouns and the long
distance anaphor proprio. Interestingly, the anaphor se stesso/se stessa, which is closest counterpart of himself/herself,
does not naturally sit in the complement position of an unaccusative noun:
(i) ?* Ogni ragazzo ricorda il naufragio/fallimento di se stesso
Every boy remembers the shipwreck/failure of himself
Since it is not possible to reliably test reconstruction by using stesso/se stessa, I won't use these expressions in the
remaining part of this paper. So, I remain agnostic on the occurrence of Condition A reconstruction with the material in the
head of the relative clause.

12
(34b) *Il ribaltamento della propriai barca che nessunoi ricorda è avvenuto durante la scuola di vela
The upsetting of self's boat that nobody remembers happened during the sailing school

(35a) Il ribaltamento della suai barca che nessunoi ricorda è quello che è avvenuto durante la scuola di
vela
The upsetting of his boat that nobody remembers is the one that happened during the sailing school
(35b) *Il ribaltamento della suai barca che nessunoi ricorda è avvenuto durante la scuola di vela
The upsetting of his boat that nobody remembers happened during the sailing school

Each pair comes close to being a minimal pair in which the only relevant difference is that the a
sentence has an identity character, while the b sentence is a canonical subject-predicate structure. A
similar contrast between identity and canonical subject-predicate sentences can be illustrated by
reflecting upon sentences like (36):

(36) I dieci aerei che ogni tecnico ha controllato per ultimi sono quelli che hanno rischiato di cadere
per primi
The ten planes that every technician has checked at the end are those that have risked to crash first

(36) is ambiguous14. The source of the ambiguity is that the group of planes can, but need not, co-vary
with the different technicians. The reading which is more relevant for us is the one in which for each
technician there is a possibly different set of ten planes that risk to crash. Intuitively, this reading is a
case of reconstruction, because the scope properties of the numeral inside the relative clause head are
those associated to the position of the gap in the relative clause (in that position the numeral is c-
commanded by the universal quantifier). A very similar case of reconstruction is observed in (37). (37)
has a distributive reading in which the plan which risks to crash co-varies with the technician that has
checked it.

(37) L’aereo che ogni tecnico ha controllato per ultimo è quello che ha rischiato di cadere per primo
The plane that every technician has checked at the end is the one that has risked to crash first

(36) and (37) are identity sentences. It is possible to form a minimal pair with sentences that are
predicational. Interestingly, the reading that illustrates reconstruction becomes very difficult to grasp
in the predicational sentence of the minimal pair. In fact, in (38) and (39) the distributive reading is
very marginal (if possible at all).

(38) I dieci aerei che ogni tecnico ha controllato per ultimi hanno rischiato di cadere per primi
The ten planes that every technician has checked at the end have risked to crash first

(39) L’aereo che ogni tecnico ha controllato per ultimo ha rischiato di cadere per primo
The plane that every technician has checked at the end has risked to crash first

So, reconstruction is made possible, or at least strongly favored, by identity structures in a variety of
cases. This suggests two general considerations. First, a satisfactory account should explain the special
role of identity structures in licensing reconstruction.
The second observation is the pars destruens of this paper. I think that the contrast between identity
and canonical subject-predicate sentences is clear evidence against the idea that reconstruction can be
explained by combining the copy theory of traces and the raising analysis of relative clauses.

14
A sentence like i) is unambiguous. The only reading is the one in which the numeral takes wide scope.
(i) Dieci aerei che ogni tecnico ha controllato per ultimi sono quelli che hanno rischiato di cadere per primi
The absence of scope reconstruction in (i) is expected, because the numeral dieci is not a modifier in the head of the
relative clause. In (i) the head of the relative clause is the bare noun aerei and the numeral dieci is the determiner of the
entire complex DP.

13
According to this type of account, a copy of the head always sits in the position of the gap inside the
relative clause, so reconstruction effects should arise across the board. In this account, the fact that the
complex DP containing the relative clause is the subject of a predicative structure or the subject of an
identity sentence should play no role, since the internal structure of this DP does not change in the two
type of sentences. However, the identity character of the sentence seems to be a necessary trigger for
reconstruction. This fact remains totally mysterious under the account that combines the copy theory
of traces and the raising analysis of relative clauses. I conclude that such account should be rejected
and a more adequate one is needed. A more promising area to look at is the semantics literature on
relative clauses.

7. Reconstruction as a By-Product of the Functional Interpretation

The contrast in reconstruction properties between identity and canonical subject-predicate sentences is
directly reminiscent of another contrast that arises in English relative constructions and that was first
pointed out by Geach (1964). Geach observed that a quantified expression can bind a pronoun outside
its syntactic scope in an identity sentence like (40), while the same reading is much harder in a
subject-predicate sentence like (41). For convenience, from now on I will call sentences like (40)
Geach's sentences.

(40) The woman [every man]i loves is hisi mother


(41) ?? The woman [every man]i invited to the party came without himi

There is a striking analogy between Geach's sentences and the a. sentences in (29) to (35), since in
both cases a variable is bound even if it is not c-commanded by its binder. It is worth pointing out that
the bound variable reading in these sentences cannot be attributed to the fact that the quantified
expression escapes the relative clause and covertly moves to a position in which it c-commands the
pronoun. This analysis is excluded by two considerations. First, the quantified expression cannot
scope out of the relative clause because QR is a local operation15. Second, if QR could apply in (40)
and in the a. sentences in (29) to (35), it would not be clear why it cannot apply in (41) and in the b.
sentences in (29) to (35). So, the systematic difference between the two type of sentences would be
effaced.
I will now summarize Jacobson's (1994) approach to Geach's sentences, which attributes the
availability of variable binding in absence of c-command to the semantics of identity. She proposes
that equation can force a bound interpretation in absence of any actual syntactic binding. The informal
idea is the following. In (40), the variable is not c-commanded, given a non-raising analysis of relative
clauses, like the one sketched in (42):

(42) [ [The woman]j [OPj every mani loves tj ] ] is hisi mother

So, as far as syntax is concerned, the variable his can be freely assigned a value. This is when
semantics comes into play. Clearly, (40) is an identity sentence, so the pre-copular and the post-
copular phrases must denote the same object. What is this object? There are two possibilities. In one
interpretation, (40) is an equation between two individuals (the woman x such that every man loves x
is equated with the mother of some other individual y). In this interpretation, the value of the variable
his in the post-copular phrase does not depend on the quantified expression in the pre-copular phrase.
However, (40) has an additional reading that can be paraphrased by saying that it is a fact that every
man loves his own mother. This reading requires that the value of the variable his depend on the
quantified expression every man, although the latter does not c-command the former. Jacobson

15
Hulsey and Sauerland (2002) claim that a quantifier can move out of a relative clause, but their claim is inconsistent with
the present knowledge about the locality conditions of QR. See Cecchetto (2004) for extensive discussion.

14
observes that this reading strictly resembles the functional reading in questions, exemplified by (43).
In (43), the relevant reading is made explicit by the answer.

(43) Which woman does every man love? His mother

A question like the one in (43) is called "functional" because it is a question about a function, namely
what is asked is which relation holds between every man and the woman he loves. The answer part in
(43) points out the relevant function, namely the one that maps every man to his mother.
Jacobson proposes that Geach's sentences are also functional. In particular, the pre-copular phrase in
(40) denotes the function that maps every man to the woman he loves and the post-copular phrase
denotes the "mother of" function (the one maps every man to his mother). The sentence states that
these two functions are extensionally the same function (namely, they denote the same set of pairs of
individuals). Crucially, in order to have identity between these two functions, the pronoun his in the
post-copular phrase must co-vary with the quantified expression in the pre-copular phrase. So, it is the
semantics of identity that forces a bound interpretation of the pronoun. Following a standard
terminology, I will call the cases of variable binding without c-command of this type "indirect
binding".
Let us switch to the absence of the bound reading interpretation in (41). Note that the functional
reading is not available in this sentence. The pre-copular phrase in (41) cannot denote a function
because we would have a predicational structure in which the subject of predication is a function
rather than an individual. Banning special machinery that would require independent motivation, this
higher type of predication is not allowed. So, the only available reading in (41) is the one in which the
pre-copular phrase denotes an individual. But this reading does not allow variable binding without c-
command. In a nutshell this is the type of analysis for Geach's sentences which is due to Jacobson
(1994) and that has been further elaborated by Sharvit (1999).
Let us know switch to variable binding reconstruction. Although I cannot give an explicit
compositional semantic analysis of the a. sentences in (29) to (35), it seems pretty clear to me that they
can have a functional interpretation. Take (30a) as a representative. (30a) can be taken as a statement
that two functions are extensionally the same. The first function maps everyone to the failure that he or
she never forgets while the second function maps everyone to the accident that happened first to him
or to her. So, I propose that variable binding reconstruction in (30a) and in other a. sentences must be
analyzed as an example of indirect binding, parallel to the indirect binding which occurs in Geach's
sentences. Given this approach, the absence of variable binding reconstruction in the b. sentences in
(29) to (35) is expected. Since canonical subject-predicate clauses do not allow a functional reading,
indirect binding is impossible in this structure.
There is independent evidence that supports that hypothesis that reconstruction effects in identity
sentences are cases of indirect binding. As observed by Jacobson (1994) and systematically discussed
by Sharvit (1999), the functional reading of relative clauses displays a characteristic lack of
uniqueness16. For example, despite the presence of the definite article, (40) is true also in a situation in
which some man or other loves another woman in addition to his mother. If I am right in analyzing
(30)a and the a. sentences in general as cases of indirect binding due to the possibility of the functional
reading, this lack of uniqueness ought to arise in the a. sentences as well. This prediction is borne out.
(30)a, for example, is true in a situation in which someone remembers other failures, in addition to the
one that happened first to him or to her.
Admittedly, my approach to the reconstruction cases should be completed by a compositional
semantic analysis that I am not in the position to offer in this paper. In absence of that, some important
questions remain open. For example, it is not clear how the functional analysis can explain the scope
reconstruction pattern in (36) to (39), which is distinct, although related, to the reconstruction cases in
the a. sentences. However, the analogies between Geach's sentences and the variable binding
reconstruction cases are remarkable and clearly require a unified analysis. Crucially, the analysis for
16
In fact, it is well known that the lack of uniqueness is also a property of functional questions. "His mother" is a felicitous
answer to the question in (43) even if some man or other loves another woman in addition to his mother.

15
reconstruction cases that combines the raising analysis and the copy theory of traces cannot be
exported to Geach's sentences, since the pronoun that gets indirectly bound in Geach's sentences is not
even located in the complex DP containing the relative clause. So, assuming the raising analysis and
the copy theory of traces as an explanation for the reconstruction cases would imply giving up a
unified analysis for reconstruction cases and Geach's sentences. On the other hand, exporting the
analysis developed for Geach's sentences to the reconstruction cases is a promising move, because the
identity sentences in which reconstruction occurs seem to be amenable to a functional interpretation.
So, I conclude that bound variable reconstruction cannot be taken as strong evidence in favor of the
raising analysis of relative clauses.
Before concluding this section, I have to mention that there is another respect in which my analysis
should be completed by further research. According to Doron (1982) and to Sharvit (1999) non-
identity sentences also allow indirect binding, at least in special circumstances, that is to say, in
Hebrew and only when the quantifier is every. This introduces an interesting difference between
languages like Hebrew and languages like English and Italian. The limitation of the functional
interpretation to identity sentences makes sense because, as I said before, the functional reading in a
predicational structure would require an higher type of predication, which can be seen as a undesirable
complication of the semantic machinery. However, functional readings do exist in non-identity
sentences in Hebrew and this raises the question of why they don’t in English and Italian in the same
contexts. We same to witness a parametric variation in an area in which not much crosslinguistic
variation is expected (for example, it is hard to think of a morpho-phonological cue that children can
use to set the relevant semantic parameters). Unfortunately, I don’t have anything interesting to say
about this crosslinguistic variation and I have to leave this question without an answer.
Let me summarize. In this section, I discussed a case of variable binding without c-command that
arises only in identity sentences, namely Geach's sentences. The variable binding reconstruction cases
that we have discussed in section 6 also arise in identity sentences only, so they are directly
reminiscent of Geach's sentences. Therefore, I have proposed that the explanation that has been
proposed in the literature for Geach's sentences, namely that binding without c-command is a by-
product of the functional interpretation of relative clauses, should be extended to the reconstruction
cases. If I am right, the reason why canonical subject-predicate sentences do not display reconstruction
effects is that subject-predicate sentences are not amenable to a functional interpretation.

8. The Absence of Condition C Reconstruction Effects

In this section I go back to the dissociation between variable binding reconstruction effects on the one
side and Condition C reconstruction effects on the other side. We already discussed some data from
the literature in section 4. In this section, I double check the reliability of the pattern described in the
literature by contrasting variable binding and Condition C in identity sentences that contains relatives
clauses with an unaccusative head (this control is necessary for the methodological reasons that have
been already introduced). The pattern described in the literature is confirmed: variable binding
reconstruction obtains (cf. the a. sentences in 44 and 45) but Condition C reconstruction does not (cf.
the b. sentences in 44 and 45):

(44a) The accident of hisi that nobodyi forgets is the one that affected himi first
(44b) The accident of Johni’s that hei will never forget is the one that affected himi first

(45a) Il proprioi fallimento che nessunoi dimenticherà mai è quello che è avvenuto per primo
The self's failure that no one will ever forget is the one that happened first
(45b) Il fallimento di Bushi che proi ha dimenticato subito è quello che è avvenuto per primo
The failure of Bush that (he) has forgotten immediately is the one that happened first

16
Since the judgments in this area can be difficult, let us double check the data. I am claiming that (45)b
is acceptable17. In order to confirm its status, I compare it with two control sentences. The first one is
(46), which is clearly out in the intended interpretation. In (46) the R-expression is part of the
constituent which is pied-piped by the relative clause pronoun. Whatever mechanism accounts for
reconstruction with clear cases of wh movement (arguably, the copy theory of traces) accounts for the
Condition C effect in (46) as well.

(46) *Lui è il giornalista le cui critiche a Bushi proi ha dimenticato subito le cui critiche a Bushi
He is the journalist whose critics to Bush (he) has forgotten immediately

The contrast in acceptability between (45b) and (46) is very sharp and confirms that (45b) is not a
Condition C violation. The second control sentence is (47), which is also clearly unacceptable in the
intended interpretation. (47) shows that Condition C reconstruction does hold in Clitic Left
Dislocation (CLLD), a construction that differs minimally from relative clauses (see Cecchetto 2000
for a discussion of reconstruction in CLLD). The contrast between (45b) and (47) is strong and it
comes close to being a minimal pair that illustrates the non occurrence of Condition C reconstruction
for material inside the relative clause head.

(47) *[Il fallimento principale di Bushi]j proi loj ha dimenticato subito


The main failure of Bush (he) it has forgotten immediately

So, in agreement with what is generally assumed, I conclude that Condition C reconstruction does not
hold with the material in the head of the relative clause.
As we have already discussed, the raising analysis, coupled with and the copy theory of traces,
predicts that Condition C reconstruction effects should arise. The raising analysis might be weakened
by assuming that relative clauses are indeed structurally ambiguous between the raising analysis and
the non-raising analysis (this is Sauerland's 2000 position, for example) . However, for the reasons that
I already mentioned in section 4, I believe that positing an ambiguity should be avoided, if possible.
Therefore, let us consider the lack of Condition C effects in sentences like (44b) and (45b) in the light
of the discussion in section 7, in which bound variable reconstruction is not explained by assuming the
raising analysis but is seen as a case of indirect binding.
There are two devices that can make the semantic value of a certain category A co-vary with the
semantic value of another category B. The first is semantic binding; A semantically binds a variable B
if A removes the assignment dependency of B (typical examples of variable binders are quantifiers
and wh expressions). The second device that can make the semantic value of A co-vary with the
semantic value of B is coreference, namely the case in which A and B have different indexes (so there
is no binding relation between them) but A and B happen to be given the same individual by the
function that assigns a value to free variables.
We have to check if one of these two devices can explain why the R-expression John and the pronoun
he can co-vary in (44b), taken as a representative of the sentences that show the absence of Condition
C reconstruction effects. As a matter of fact, it is not difficult to show that co-variation arises because
John can undergo QR and raise to a position in which it c-commands the pronoun he18. There is solid

17
Bianchi (1995) observes that the Condition C effect is favored if the binder of the R-expression is “pro” and is partially
or completely avoided if the binder of the R-expression is a tonic pronoun. I won’t discuss the pattern with tonic pronouns
because they require some sort of emphasis that makes them the marked case.
18
In this footnote, I show that John and he cannot corefer in (44)b, at least if Reinhart's (1983) Rule-I is adopted. (i) is a
simplified formulation of Rule-I:
(i) Rule-I
NP “A” cannot corefer with NP “B” if replacing “A” with “C”, “C” a variable bound by “B”, yields an indistinguishable
interpretation.
Rule-I states that coreference is blocked in a certain structural configuration whenever variable binding is possible in the
same configuration. Let us apply Rule-I to (44b). Assume that “A” corresponds to the NP John and “B” corresponds to the
pronoun he. Rule-I states that coreference should be blocked if John can be replaced by a variable bound by a binder that

17
evidence that supports this analysis. If we replace the proper name with a quantified expression, we
can clearly see that variable binding obtains in this configuration. The examples in (48) to (50) show
that a quantified expression that sits inside the complement position of the head of the relative clause
can bind a pronoun that sits inside the relative clause:

(48) Il delegato da [ogni città]i che lai rappresenterà meglio riceverà un'alta onorificenza
The delegate from every city that it will-represent better will-receive an important reward
" The delegate from every city that represents it better will receive an important reward"
(49) L'incubo di [ogni bambino]i che loi perseguita tutta la vita è che gli rubino i giocattoli
The nightmare of every child that him persecutes all the life is that (they) from-him steal the toys
" Every child's nightmare, that continues for all his life, is that someone could steal his toys"
(50) Non conosco il relatore di [nessun tesista]i che non loi abbia maltrattato almeno una volta
(I) NEG know the advisor of no student who NEG him has mistreated at least once
"I do not know any student who has not been mistreated at least once by his advisor"

Although it is not crucial for us to decide the details of how binding is established (what is crucial is
that binding can obtain), I assume for concreteness that the quantified expression undergoes a short
occurrence of QR that moves it to some functional position FP that is still internal to the complex DP
modified by the relative clause, but is high enough to c-command the pronoun inside the relative
clause:

(51) [DP il [FP da ogni città]i [FP [ [delegato da ti] [che pro lai rappresenterà meglio]]]]

If a quantified expression inside the relative clause head can bind a pronoun inside the relative clause,
it is safe to conclude that John, which is also inside the relative clause head in (44)b, can bind he19. So,
we can conclude that the absence of Condition C effect in (44)b and similar sentences is the result of
the fact that the QR-ed R-expression can bind the pronoun.
Crucially, the explanation that I have just proposed for the obviation of Condition C effects is
perfectly compatible with the occurrence of variable binding reconstruction in the same configuration.
So, the dissociation between variable binding reconstruction (which holds) and Condition C
reconstruction (which does not) stops being puzzling. It is not necessary to postulate that relative
clauses are structurally ambiguous or, at minimum, the dissociation cannot be used as an argument to
posit this structural ambiguity20.

sits in the position occupied by he, with no change in meaning. But we know that a pronoun that sits in the position which
is occupied by John can be bound by a quantifier that sits in the position which is occupied by he. That variable binding
holds in this configuration is shown by the variable binding reconstruction case in (44a). So, Rule-I excludes that John and
he corefer in (44b). More generally, we can say that the occurrence of variable binding reconstruction blocks the possibility
of coreference between an R-expression in the head of the relative clause and a pronoun inside the relative clause.
However, and crucially for our problem, binding of the pronoun by the R-expression is allowed, as I explain in the text.
19
The assumption that proper names can undergo QR is not standard in the syntactic literature but is normally accepted in
the semantic literature. The reason is that, once QR is granted for quantificational phrases, blocking it for proper names
would require a special condition which is not independently motivated (cf. Heim and Kratzer 1998, chapter 8 for
discussion about this).
20
Note that the identificational or predicative character of the sentence does not play any role in my explanation of the
absence of Condition C reconstruction effects. Accordingly, Condition C violations shouldn’t be detected in either type of
sentences. This prediction is borne out. For example, sentence (i), which can be seen as the predicative counterpart of the
identity sentence (45)b, is not a Condition C violation either.
(i) Il fallimento di Bushi che proi ha dimenticato subito ha danneggiato solo gli altri
The failure of Bush that (he) has forgotten immediately has damaged only the others

18
9. Conclusion

In this paper, after discussing the status of the copy theory of traces in the current version of the
minimalist program and the way in which this theory interacts with the binding principles, I have
analyzed a case of reconstruction which seems to be recalcitrant to any simple treatment. This case is
reconstruction effects of the material contained inside the head of a relative clause. This case is a
challenge to the copy theory of traces because this theory explains reconstruction by the presence of a
copy in the pre-movement position. So, if a copy is present it should trigger all sort of reconstruction
effects (Condition C, Condition A, variable binding and scope reconstruction). In fact, this is what
happens with ordinary cases of movement. However, with the material contained inside the head of a
relative clause variable binding reconstruction effects are observed but Condition C reconstruction
effects are not. After rejecting, mainly for conceptual reasons, the hypothesis that relative clauses are
structurally ambiguous, I have suggested a different strategy. First, I have argued that some of the
generalizations that have been reported in the literature do not hold under a close scrutiny, because
relative clauses containing transitive nouns are not truly reliable. When some interfering factors are
factored out, the picture that emerges is that reconstruction effects arise only in identity structures.
This limited distribution, as I have suggested, can be made sense of, if variable binding reconstruction
is a by-product of the semantics of identity sentences, more particularly of the functional reading. The
approach that I advocate does not require that a copy be present in the reconstructed position (in other
terms, this approach does not assume a raising analysis of relative clauses). Finally, I have turned to
the fact that even the relative clauses which display variable binding reconstruction effects do not
display Condition C reconstruction effects. I have explained this dissociation by showing how the R-
expression can bind the pronoun it co-varies with.
I would like to conclude this paper with a general observation. The copy theory of traces has been
successful enough that the presence of reconstruction effects is often taken to be uncontroversial
evidence for a presence of a copy in the relevant position. In other terms, the occurrence of
reconstruction effects is commonly considered a clear diagnostic for movement. If what I have argued
for in this paper is correct, this type of reasoning requires a qualification. Movement, when it occurs,
triggers a uniform pattern of reconstruction effects, namely reconstruction is equally observed with the
different types of diagnostics used to detect it (Condition C, Condition A, variable binding and scope).
So, only a uniform pattern of reconstruction effects is a reliable diagnostic for movement. When
dissociation in reconstruction effects is observed, as it is the case with relative clause heads, this
suggests that there is no copy in the relevant position and that reconstruction effects should be dealt
with by advocating a semantic mechanism21.

21
Interestingly, the dissociation between different types of reconstruction effects is not restricted to relative clauses. There
is at least another syntactic context that illustrates a similar, although non identical, dissociation. This context, which is
discussed by Cecchetto and Chierchia (1999), is CLLD sentences in which a PP is left dislocated. In this context,
Condition C reconstruction occurs while scope reconstruction does not. Cecchetto and Chierchia, based on various
arguments that I cannot go into, argue that the dependency between the dislocated PP and its canonical (=non dislocated)
position is not created by movement and offer a semantic treatment for reconstruction effects with dislocated PPs. I am
mentioning this data because it gains further supports to the generalization that reconstruction is a unitary phenomenon
only when the dependency which is involved is created by movement, that is, when the lowest link of the dependency is
occupied by a copy of the moved category.

19
ADDRESS FOR CORRESPONDENCE
Carlo Cecchetto
Facoltà di Psicologia
Università degli Studi di Milano-Bicocca
Piazza dell'Ateneo Nuovo 1
20126 Milano (Italy)
e-mail: carlo.cecchetto@unimib.it

ACKNOWLEDGEMENTS
I thank the audience of XXIIX Incontro di Grammatica Generativa at the University of Lecce for
useful remarks and Gennaro Chierchia, Orin Percus and an anonymous reviewer for pointing out
problems and suggesting ways to overcome them. However, I take full responsibility for remaining
errors and shortcomings.

20
REFERENCES

Barss, Andrew. 1986. Chains and Anaphoric Dependencies, Doctoral dissertation, MIT.

Bhatt, Rajesh. 2002. The raising analysis of relative clauses: Evidence from adjectival modification.
Natural Language Semantics 10:43-90.

Bianchi, Valentina. 1995. Consequences of Antisymmetry for the Syntax of Headed Relative Clauses,
Ph.D. Dissertation, Scuola Normale, Pisa.

Borsley, Robert. 1997. 'Relative Clauses in the Theory of Phrase Structure', Linguistic Inquiry 28, pp.
629-647.

Burzio, Luigi. 1986. Italian Syntax, Reidel, Dordrecht.

Carlson, Gregory. 1977. ‘Amount Relatives’. Language 53:520–542.

Cecchetto, Carlo and Gennaro Chierchia. 1999. 'Reconstruction in Dislocation Constructions and the
Syntax/Semantics Interface', S. Blake, E. Kim e K. Shahin (eds.) WCCFL XVII Proceedings, CSLI
publications, Stanford University, pp. 132-146.

Cecchetto, Carlo. 2000. 'Doubling Structures and Reconstruction' Probus 12, pp. 93-126.

Cecchetto, Carlo. 2004. 'Explaining the locality conditions of QR. Consequences for the Theory of
Phases', Natural Language Semantics 12:4, 345-397.

Chomsky, Noam. 1995. The Minimalist Program, MIT Press, Cambridge, MA.

Chomsky, Noam. 2005. 'On Phases', manuscript, MIT.

Cresti, Diana. 2000. 'Ellipsis and Reconstruction in Relative Clauses' M. Hirotani, A. Coetzee, N. Hall
and J-Y. Kim (eds.), Proceedings of NELS 30, GLSA, University of Massachusetts, Amherst, 153-
163.

Doron, Edit. 1982. ‘On the syntax and semantics of resumptive pronouns’, Texas Linguistic Forum
19:1-48.

Fox, Danny 2002. ‘Antecedent-Contained Deletion and the Copy Theory of Movement’, Linguistic
Inquiry 33:63-96.

Geach, P. 1964. 'Referring Expressions Again', Analysis 24/5. Reprinted in P. Geach, Logic Matters,
1972, University of California Press.

Giorgi, Alessandra and Giuseppe Longobardi. 1991. The Syntax of Noun Phrases, Cambridge
University Press, Cambridge.

Ken Hale and S. J. Keiser. 1993. 'On argument structure and the lexical expression of syntactic
relations' in K. Hale and S. J. Keiser (eds.) A View from Building 20, MIT Press, Cambridge,
Massachusetts, pp. 53-109.

21
Sarah Hulsey and Uli Sauerland (2002) ‘Sorting out Relative Clauses: a reply to Bhatt’, manuscript,
MIT and University of Tübingen.

Jacobson, Pauline. 1994. ' Binding Connectivity in Copular Sentences' in M. Harvey and L.
Santelmann (eds.) Proceedings of SALT IV.

Kayne, Richard. 1994. The Antisymmetry of the Syntax, MIT Press, Cambridge, MA.

Lebeaux, David. 1989. Language Acquisition and the Form of Grammar, Ph.D. Dissertation,
University of Massachusetts, Amherst.

Moltmann, Friederike. 1992. Coordination and Comparatives. Ph.D dissertation, MIT, Cambridge,
Mass

Moro Andrea. 2000 Dynamic Antisymmetry. Linguistic Inquiry Monograph 38. Cambridge,
Mass.:MIT Press.

Munn, A. 1994. 'A minimalist account of Reconstruction Asymmetries' In Mercè Gonzàlez (ed.)
Proceedings of NELS 24, GLSA, University of Massachusetts, Amherst, 397-410.

Neidle, C., J. Kegl, D. MacLaughlin, B. Bahan & R.G. Lee 2000. The syntax of American Sign
Language. Functional categories and hierarchical structure. Cambridge, MA: MIT Press.

Nunes, Jairo. 1999. Linearization of chains and phonetic realization of chain links. In Working
minimalism, 217-249, ed. Samuel David Epstein and Norbert Hornstein. MIT Press, Cambridge,
Mass.

Nunes, Jairo and Ronice Müller de Quadros (2005) 'Duplication of Wh-elements in Brazilian Sign
Language', forthcoming in Proceedings of NELS 35.

Petronio, K. & D. Lillo-Martin (1997), 'WH-movement and the position of Spec-CP: Evidence from
American Sign Language'. Language 73 (1), 18-57.

Reinhart, Tanya. 1983. Anaphora and Semantic Interpretation. Croom Helm, London.

Sharvit, Yael. 1999. 'Functional Relative Clauses', Linguistics and Philosophy, 22, 447-478.

Sauerland, Uli. 2000. 'Two structures for English restrictive relative clauses'. In Mamoru Saito et al.
(ed.), Proceedings of the Nanzan GLOW, 351–366. Nanzan.

Sternefeld, Wolfgang. 1997: The Semantics of Reconstruction and Connectivity, Arbeitspapiere des
SFB 340, University of Tuebingen.

Heim, Irene. 1987. ‘Where does the definiteness restriction apply? Evidence from the definiteness of
variables’. In Eric Reuland and Alistair ter Meulen (eds.) The Representation of (In)definiteness, 21–
42, MIT-Press, Cambridge, Mass.

22

View publication stats

You might also like