(1) 	a.	[Which picture of himselfi] did Johni buy?
	b.	[These pictures of himselfi] seem to Johni to be excellent.

In the particular model of movement that we've adopted, although the structures in (1) appears to violate Principle A, there is at least a copy of the moved wh-phrase in a position where Principle A is not violated, as we can see in this tree for (1a):

Since we're modelling movement as an instance of Internal Merge, movement creates multiple copies of moved items (and we need some algorithm, not yet specified, telling us which of the copies to pronounce), as in the tree above. In this tree, we at least have one copy of himself that's in the right kind of relation with John. In other words, we apparently need to restrict Binding Theory so that it's satisfied as long as one of the copies of an object obeys the binding principles. Now, how do we achieve that?

One possibility would be to restate our principles of Binding Theory, or rethink our assumptions about the shapes of trees, so that syntactic elements could effectively be in more than one position at once. For example, our Binding Theory could have statements in it like 'At least one copy of each anaphor must have a binder within its binding domain'. The tree in (2) would satisfy a condition like that; one copy of himself isn't bound, but the other copy is. Or we could posit a different kind of model of movement, one in which phrases are allowed to be immediately dominated by more than one node (yielding what are called multidominance trees):

In a multidominance tree like the one in (3), rather than creating a new copy of the wh-phrase, we would model 'movement' as the creation of a new immediate dominance relation; the wh-phrase would become immediately dominated by CP, as well as by VP. Allowing trees like this would force us to rethink a lot of our assumptions; it's not exactly clear how to define c-command in a way that gets us the results we want, for example. But intuitively, it seems clear how a tree like (3) might help us with the reconstruction problem; after movement, it's still true that every node dominating John dominates himself, which is our classic definition of c-command.

In fact, we'll see that these solutions are on the wrong track. We really do want to have multiple copies, and to have a process that chooses one of the copies (not necessarily the pronounced one) as the one that binding theory applies to. In order to see this, we need to look at another type of reconstruction, having to do with scope.

Scope refers to the order in which certain elements are interpreted. Consider an example like (4):

(4) Someone from New York is likely to win the lottery.

The sentence in (4) can mean several different things. It could mean, for example, that there is a particular person from New York who is likely to be the winner of the lottery (maybe Donald Trump, who has been buying thousands of lottery tickets in an attempt to get into the news again). Alternatively, it could mean that the lottery in question is one in which most of the ticket-buyers are New Yorkers, and that therefore chances are good that the eventual winner will be from New York.

We can capture this difference in meanings as a difference in the order of interpretation of 'someone' and of 'likely'. Informally, we can paraphrase the two readings in this way:

(5) There is someone from New York who is likely to win the lottery.
(6) It is likely that there is someone from New York who will win the lottery.

The difference between (5) and (6) is described as one of scope; either someone takes scope over likely, as in (5), or the reverse, as in (6). If we were seriously doing semantics, we would try to make our understanding of scope more formal, but this level of formalism will do for what we're interested in here.

In (7) we see another instance of scope ambiguity: (7) A guard is standing in front of every building. This sentence has two readings, one of them much more likely-sounding than the other:

(8) Every building has the property that there is a guard who is standing in front of it.
(9) There is a guard who has the property that he is standing in front of every building.

On the first reading, in which every building takes scope over a guard, the sentence simply says that every building is guarded by a guard. On the second reading, the assertion is that there is a single (apparently very large) guard who is standing in front of all of the buildings.

Interestingly, the ambiguity of (7) is not present in (10):

(10) A guard said that I should stand in front of every building.

(10) only has a reading parallel to (9), in which a single guard said that I should stand in front of every building. It has no reading parallel to (8); in other words, it cannot mean that for each building, there was a (potentially different) guard saying that I should stand in front of it. The ambiguity of (7), then, is apparently crucially linked to the fact that the two quantifiers are in the same clause; when quantifiers are separated by a clause boundary, as in (10), then the first quantifier unambiguously takes scope over the second.

Now, having established that quantifier scope ambiguity only holds when the two quantifiers are in the same clause, we need to ask ourselves what counts as a 'clause' in the relevant sense. Consider, for example, the sentence in (11):

(11) A guard seems to be standing in front of every building.

(11) is ambiguous, again, with two readings parallel to the ones in (8-9). In particular, it has a reading that was missing in (10), in which every building seems to be guarded by a (different) guard. Our condition on quantifier scope ambiguity, then, will have to allow the two quantifiers in (11) to be scopally ambiguous. In principle, there are at least two ways we could achieve this.

We could distinguish between (10) and (11) by saying that quantifiers can be scopally ambiguous just if they are not separated by any tensed clause boundaries. On this view, (11) is ambiguous because the clause boundary separating the two quantifiers is an infinitival one.

Alternatively, we could appeal to the fact that (11) involves A-movement; a guard begins the derivation in the embedded clause and raises into the matrix clause. At the beginning of the derivation, the two quantifiers are not separated by any clause boundaries, and we could say that the condition on quantifier scope ambiguity can be satisfied at that earlier point in the derivation.

The sentence in (12) offers us evidence that the second of these approaches is the correct one:

(12) I seem to a guard to be standing in front of every building.

(12) is unambiguous; for (12) to be true, there must be a single guard who believes that I am standing in front of every building. The sentence cannot mean that for every building, there is a (possibly different) guard who thinks I am standing in front of it. The two quantifiers in (12) have a fixed scope. We can cope with this fact if we take the second of the two approaches above; quantifiers separated by clause boundaries, whether tensed or untensed, are scopally unambiguous, but we can compute scope in (11) as though A-movement had not happened, which allows us to treat the two quantifiers as though they were still in the same clause. In (12), the two quantifiers are never in the same clause at any point in the derivation, and we therefore expect, correctly, that the quantifiers should be scopally unambiguous.

We have arrived at the conclusion that movement can be 'undone' for computation of scope; the sentence in (11) is ambiguous because, prior to movement, the two quantifiers were in the same clause. Scope computation is therefore like binding, in that both can show the effects of reconstruction.

So now we know that reconstruction for scope and reconstruction for binding both exist. Let's move on to consider how these two types of reconstruction interact. We'll discover facts which will turn out to bear on the questions about the nature of reconstruction raised above. In particular, we'll see that the semantic computation has to pick out a particular copy for interpretation (that is, for computation both of scope and of binding); we'll see evidence that we shouldn't simply allow all the copies to contribute to interpretation, or posit trees, like multidominance trees, in which moved objects are effectively in multiple positions at once.

Consider (13), which closely resembles (11) above:

(13) Two guards seem to be standing in front of every building.

Like (11), (13) has two readings, one in which every building is guarded by two guards (so that there are twice as many guards, total, as there are buildings), and another, less reasonable-sounding reading in which there are two guards who are somehow standing in front of every building at once. In other words, (13), like (11), is scopally ambiguous, in a way we've decided to associate with reconstruction of A-movement. Now, compare (13) with (14):

(14) Two guards_i seem to each other_i to be standing in front of every building.

Unlike (13), (14) is not scopally ambiguous; it only has the odd reading, in which two guards are somehow standing in front of every building at once. (14) differs from (13) only in the addition of to each other in the matrix clause. The difference is not simply due to the presence of a PP after seem, as (15) shows:

(15) Two guards seem to me to be standing in front of every building.

(15) is like (13) and unlike (14) in that it is scopally ambiguous; it can have the 'normal' reading in which I think that every building has two guards in front of it. The lack of ambiguity in (14), then, is not simply the result of the presence of a PP after seem; rather, it must have to do with the anaphor each other, which is bound by the A-moved quantifier two guards. Apparently reconstruction is blocked here because, if it were to take place, each other would be deprived of its binder. In other words, when we perform reconstruction, we do not simply freely choose which copy to subject to each of the various processes of interpretation; rather, we must choose a single copy which will be the relevant copy both for scope computation and for binding. In (13) and (15), we have the option of choosing either copy of the A-moved phrase for scope computation, since binding is not an issue, and the sentences are therefore scopally ambiguous. In (14), by contrast, we apparently are forced to choose the highest copy of two guards in order to satisfy Principle A, and as a result, scope reconstruction is blocked.

We can construct a similar argument for the same result, using Principle C instead of Principle A. (16) is scopally unambiguous:

(16) [At least one of John_i's guards] seems to him_i to be standing in front of every building.

The example in (16) can only mean that John thinks that one of the guards he employs has the peculiar ability to stand in front of multiple buildings simultaneously; it cannot mean that John thinks that every building has one of his guards in front of it. In other words, scope reconstruction is apparently blocked in (16). We already have the means to understand why; if scope reconstruction were to take place in (16), then him would bind John, in violation of Principle C. Again, we see that scope reconstruction is blocked when it would create a configuration that violates binding theory; in other words, reconstruction must create a single representation that is interpreted both for scope and for binding.

This representation is clearly based on the tree created by the syntax; we have seen that it operates on the movement chains created by syntax. At the same time, it is clearly not the syntactic tree as that tree is pronounced; reconstruction of a moved DP can be to a position where that DP is not pronounced. Reconstruction is thus another case of a post-syntactic operation, of the type we have already seen in our discussion of covert wh-movement, which feeds the semantic component but not the phonological component. Operations of this sort are depicted in the model represented below, sometimes called the T-model or Y-model:

In this model, the syntactic derivation first constructs a tree. This tree is then sent both to the semantic and phonological components, a process which involves various post-syntactic operations. On the semantic side, operations like covert wh-movement and reconstruction create a new syntactic tree, called an LF (or Logical Form), which is a representation of the meaning of the sentence, and it is this LF which is interpreted by the semantics. Because the operations which create the LF are post-syntactic, they feed only the semantic representation and not the phonological representation, and hence have no effect on pronunciation. Conversely, on the phonological side, we posit post-syntactic operations which feed pronunciation but not meaning, eventually yielding a Phonetic Form (PF) which is interpreted by the phonological component.

The above discussion of reconstruction is one piece of evidence that we need something like this tree. We have seen that reconstruction operates on the result of the syntactic derivation, affecting the interpretation of moved objects. Moreover, it appears to be an operation which affects the semantic representation but not the phonological one.

Reconstruction is a complex topic with a rich literature, and we can't do it full justice here. Before we conclude, let's just note two more things about the nature of reconstruction.

All of the examples of reconstruction we've seen so far have to do with A-movement. We can show that the conclusions drawn here hold for A-bar movement as well, however--in particular, we can see that reconstruction involves a privileged status for a single copy of the movement chain, which the one interpreted for binding theory.

Consider (17):

(17) [Which of the papers that he_i gives to the teacher] does every student_i worry about the most?

Example (17) has a reading in which he is a variable bound by every student; in other words, it has a reading that we could more or less informally paraphrase as in (18):

(18) For every student (x), which of the papers that x gives to the teacher does x worry about the most?

The answer to a question like (17) could be something like "the first paper"--that is, every student worries the most about the first paper that he gives to the teacher, because he doesn't yet know how tough a grader the teacher will be.

We can independently show that when a pronoun is interpreted as a variable bound by a quantifier, the quantifier must c-command the pronoun. In (19), for example, the pronoun can be interpreted as a variable, while in (20) and (21), it cannot be:

(19) Every mother_i loves her_i child.
(20) *[The friends of every mother_i] love her_i child.
(21) *Her_i child loves every mother_i.

Returning to (17), then, we can now see that this is an instance of reconstruction; the pronoun he must be c-commanded by the quantifier every student in order to be interpreted as a variable, and while this c-command relation does not hold on the surface, it did hold prior to movement.

Now consider (22):

(22) [Which of the papers that he_i gave to the teacher_j] does every student_i hope that she_j will read carefully?

In (22), again, we must posit reconstruction, so that he can be c-commanded by every student. If we simply reconstruct the wh-phrase to its lowest position as the complement of read, however, we have a problem. The wh-phrase contains an R-expression, the teacher, which is subject to Principle C, and reconstruction to the lowest position will create a Principle C violation; the teacher will be bound by she.

Fortunately for us, this problem has a solution. We have already had reason to posit successive-cyclic movement; we have seen arguments that the wh-phrase in (22) must be stopping in the specifier of the embedded CP (the one headed by that) on the way to its eventual landing site. And if we reconstruct the wh-phrase to this intermediate position, then our problem is solved; the specifier of the embedded CP is c-commanded by every student (so that he can be a bound variable), but not by she (so that the teacher can satisfy Principle C).

This approach makes a prediction, which appears to be correct. If we change the sentence so that she c-commands every student, then there should be no position for the wh-phrase to reconstruct to that will satisfy all its requirements, and the result should be ill-formed on the relevant reading:

(23) *[Which of the papers that he_i gave to the teacher_j] did she_j say that every student_i should revise?

In (23), reconstruction to the intermediate Spec CP position is of no help; the wh-phrase will still be c-commanded by she, and hence the R-expression the teacher will violate Principle C. More generally, the pronominal variable he must be c-commanded by every student, yet any reconstruction position c-commanded by every student will also be c-commanded by she, which will trigger a Principle C violation. The ill-formedness of (23) thus confirms that reconstruction for A-bar movement obeys the same general principle as reconstruction for A-movement; reconstruction must be to a single position. In this case, we cannot simply reconstruct to one position for Principle C and to another for pronominal variable binding.

Let us end our discussion of reconstruction with a puzzle. Consider the ill-formed example (24):

(24) *He_i seems to John_i's mother to be expected to win.

(24) is a violation of Principle C: the R-expression John is bound by he. But why can this violation of Principle C not be saved by reconstruction? Suppose we were to reconstruct he to a lower position, something we have seen we need to be able to do. Would the resulting structure lack a binder for John?

Apparently the answer to this question is 'no'; even after reconstruction, the position which has been 'vacated' by reconstruction is not empty, but contains something capable of triggering Principle C. We will leave matters at this fairly mysterious point, for now.