12 Subjacency and the ECP

• Subjacency
  • Two possible derivations for long-distance wh- movement
  • IP as a barrier to wh- movement
  • DP as a barrier to wh- movement
  • The coordinate structure constraint revisited
• The Empty Category Principle (ECP)
  • Antecedent government
  • Lexical government
• Further issues and refinements
  • Is subjacency an independent principle?
  • Movement out of ECM complements
  • Movement out of DP
• Exercises and problems
• Supplementary material
  • Node relations

Given then-current assumptions concerning phrase structure (X' theory hadn't yet been introduced), the A-over-A constraint is illustrated by the contrast in (2).

Against the backdrop of what was known about wh- movement in the early 1960s, the island constraints discussed in Chapter 11 represented significant empirical progress. But from a more theoretical point of view, the island constraints are simply a list of stipulations, and they sharply raise the question of whether it is possible to reduce them to fewer, deeper structural principles (ideally, a single such principle). In this chapter, we discuss two influential proposals with this aim: the subjacency condition and the Empty Category Principle (ECP). As we will see, the reduction of the constraints on wh- movement to more general principles has proven a considerable challenge to syntactic theory, and one that persists to the present day. In particular, no satisfactory overarching framework has yet been found that subsumes the entire range of island constraints. The island constraints themselves therefore remain as an empirical benchmark against which to measure any theoretical proposal concerning constraints on wh- movement.

Subjacency

Two possible derivations for long-distance wh- movement

When we consider examples of long-distance wh- movement like those in (3), two possible derivations come to mind.

(3)

What did he say that he was reading?

On the one hand, the wh- phrase might move from the position in which it is interpreted, however deeply embedded that is, to the sentence-initial Spec(CP) position in one fell swoop, yielding a wh- movement chain with two links, as in (4). On the other hand, wh- movement might take place in more than one step. The first step takes the moved constituent from its original position to the nearest Spec(CP), and each subsequent step takes it to the next higher Spec(CP). This derivation of (3), which involves two steps and yields a wh- movement chain with three links, is shown in (5).

(4)

(5)

The derivation in (5) is known as a cyclic derivation (the idea being that each successively higher clause (= CP) forms a separate cycle in the derivation of the entire sentence), and the derivation in (4) is accordingly known as noncyclic. Notice that the noncyclic and cyclic derivations in (4) and (5) differ in the presence of an intermediate trace, which is highlighted by a box in (5).

IP as a barrier to wh- movement

Parentheses indicate intermediate traces that are posited in a cyclic, but not in a noncyclic, derivation.

However, the existence of syntactic islands forces us to choose the cyclic alternative. For instance, consider the ungrammatical question in (7a) (= (23b) of Chapter 11 and its cyclic derivation in (7b).

Remember to interpret how in (7a) as modifying solve, not forgotten, as indicated by the lowest trace.

(7)	a.	*	Howi have they forgotten [ which problem they should solve ti ] ?
	b.

If wh- movement were able to occur in one fell swoop, then there would be nothing to stop long-distance wh- movement in (7a), and the question should be grammatical, contrary to fact. But the ungrammaticality of the question can be made to follow from the assumption that wh- movement is cyclic. Specifically, let's assume that wh- movement is subject to the condition in (8), and that IPs form barriers to movement, as indicated by the boxes in (7b).

If the only barriers in English were IPs, then wh- movement out of noun complements and left branch structures like those in (11) (= (20b) and (26b) of Chapter 11) would be expected to be grammatical, contrary to fact. The structures are given in (11); the green IP nodes are intended to indicate that the derivation is consistent with subjacency under the (incorrect) assumption that the only barriers are IPs.

However, the ungrammaticality of (11) can be derived straightforwardly if the set of barriers in English includes not only IPs, but also DPs. This is illustrated in (12), where red indicates barriers that cause subjacency to be violated. As before, green indicates barriers that are consistent with subjacency.

In determining whether subjacency is violated, we can consider the relevant movement chains in either top-down or bottom-up fashion. Here and in what follows, we have chosen bottom-up. This choice has the consequence that the barriers that cause subjacency to be violated in (12) are IPs, not DPs. This doesn't mean, though, that DPs aren't barriers! If they weren't, we'd be back to the representation in (11).

As we will see later on in our discussion of the ECP, the predictions made by subjacency in connection with movement out of DPs, while correct in the case of (13b) and (14), are too strict. In other words, subjacency (as it stands, with all instances of IP and DP as barriers) incorrectly rules out grammatical instances of wh- movement.

The coordinate structure constraint revisited

Although the internal structure of coordinate phrases is not well understood, it is clear that a coordinate noun phrase like tiramisu and espresso is a recursive structure consisting of a DP that dominates two further DPs, as in (16).

Given this structure, the ungrammaticality of the examples in (16) follows from subjacency, as shown in (17).

However, not all examples that violate the coordinate structure constraint also violate subjacency. For instance, in (18) and (19), where syntactic categories other than DP are coordinated, only a single IP barrier intervenes between the moved phrase in Spec(CP) and its trace.

Take care to read the (b) examples without an intonation break before the conjunction.

A further example of this sort is shown in (20b).

As (21) shows, (20b) violates a parallelism constraint known as the across-the-board (ATB) constraint, according to which movement of a constituent out of a coordinate structure must affect all co-conjuncts simultaneously.

The exceptional behavior of coordinate structures with regard to subjacency (expected to be grammatical, yet in fact not so) and the additional special restriction imposed upon them (the ATB constraint just mentioned) strongly suggest that the coordinate structure constraint is sui generis, and that it should not be grouped together with the other island constraints. Accordingly, most proposals to reduce the island constraints to more general principles make no attempt to include the coordinate structure constraint, and we, too, will make no further mention of it.

The Empty Category Principle (ECP)

The material from here to the end of "Is subjacency an independent principle?" will not be covered in the exam. Resume reading at Movement out of ECM complements.

The island constraints and the subjacency condition that subsumes them correctly account for much, but not all, of the spectrum of relevant facts concerning wh- movement. For instance, Ross himself observed that it is more acceptable to move complements out of indirect questions than it is to move adjuncts. Subjects (also non-complements) behave like adjuncts, giving the pattern in (22).

(22)	a.	?	[ Which problem ] ]i have they forgotten howj they should solve ti tj?
	b.	*	Howi have they forgotten [ which problem ]j they should solve tj ti?
	c.	*	[ Which problem ]i have they forgotten [ how ]j ti should be solved ti tj?

In order to account for the contrast between complements and non-complements with respect to long-distance wh- movement, it has been proposed that traces of movement must satisfy a condition distinct from subjacency, the so-called Empty Category Principle (ECP). Early formulations of the ECP (Aoun, Hornstein, and Sportiche 1982, Huang 1982, Lasnik and Saito 1984) were disjunctive; that is, they consisted of two mutually exclusive conditions. In the course of the 1980s, attempts were made to reformulate the ECP in conjunctive terms; that is, as two conditions that traces of movement must satisfy simultaneously. In our view, these attempts have not been successful, because they continue to impose different conditions on the movement of complements and of non-complements. That is, even if the ECP itself is no longer formulated in a disjunctive way, the disjunction it contained is not resolved, but simply appears elsewhere in the proposals in question. For instance, Rizzi 1990 distinguishes two ways of establishing a legitimate antecedent-trace relation, one for complements, and one for non-complements. We will therefore continue to maintain a disjunctive version of the ECP, specifically (23).

We discuss the two conditions on traces in turn. The notion of antecedent government is defined in (24) (see Node relations for a definition of binding).

As is evident, clause (ii) of the definition of antecedent government recapitulates the subjacency condition, and it is easy to see that the antecedent government clause of the ECP therefore enforces cyclic movement. This of course derives the ungrammaticality of (22b,c), but leaves the relative acceptability of (22a) unexplained. However, as we will see directly, (22a), though violating the antecedent government condition of the ECP, satisfies the alternative lexical government condition.

Lexical government

The notion of lexical government to be presented in what follows relies on the concept of lexical government domain (which is based on the concept of g(overnment)-projection proposed in Kayne 1984). The term 'governor of XP' in (25a) refers to the head that stands in a head-comp configuration with XP.¹

The term 'govern' is used here in a slightly different sense than it was in Chapter 8. There, 'govern' referred to a morphological requirement imposed by a head on a noun phrase expressing one of the head's arguments. Here, the term 'govern' makes no reference to morphology, but refers instead to a purely structural relation, the head-comp relation.

As is evident, the definition in (25) is recursive. We begin by considering the nonrecursive case in (i), which is very simple. Consider the configuration in (26), where we take the DP complement of the preposition as XP and the PP as YP.

(26)

In (26), the governor of the DP is the preposition, and so the PP (the preposition's maximal projection) is a lexical government domain for the DP.

Now consider the more complex structure in (27), where we continue to take the DP complement of the preposition as XP, but it is now VP that is the YP.

(27)

Is VP a lexical government domain for the lower DP? (25.i) is not met, since the head of VP, V, does not govern the DP in question. But V does govern PP, which we determined to be a lexical government domain for DP in (26). Therefore, VP is a lexical government domain for DP in (27) by (25.ii).

Given (26) and (27), it is now easy to see that any lexical government domain can be extended simply by substituting it as a complement of a higher head; that head's maximal projection is then in turn a lexical government domain. Thus, all the structures in (28) are lexical government domains for the lowest complement DP.

(28)	a.				b.
	c.				d.				e.

Lexical government itself can then be defined as in (29).

The notion of 'proper governor' in (29.i) is introduced in light of crosslinguistic contrasts like that between (30) and (31).

Vem har  du  talat  med ti ?
who have you talked with

Qui as-  tu  parlé  avec ti ?
who have you talked with

Wem hast du  mit  ti gesprochen?
who have you with    talked

The idea is that only a proper subset of governors (= heads) is 'strong' enough to license a trace by lexical government. Exactly which heads belong to the set of proper governors can vary by language. For instance, prepositions are proper governors in English and Swedish, but not in French or German. As a result, the traces in (30) are lexically governed, whereas those in (31) are not, despite the analogous configurations in both cases. It should be emphasized that proper government figures only in the definition of lexical government itself, not in the definition of the concept of lexical government domain. So although the French preposition à 'to' cannot itself license a trace, it can license the extension of a lexical government domain. This is shown by the contrast in (32) (Kayne 1984:167). As in (30) and (31), green and red indicate heads that are and are not proper governors, respectively.

Qu'i est-ce qu'  elle tient à ti ?
what is  it that she  holds to
'What is she keen on?'

Qu'i est-ce qu'  elle tient à  faire ti ?
what is  it that she  holds to do 
'What is she keen on doing?'

At first glance, the notions of lexical government and lexical government domain might seem to permit any complement to undergo wh- movement. But the definition is more restrictive than that. Consider, for instance, the contrast between the (b) examples in (33) and (34), where the constituent undergoing wh- movement is the complement of admit in both cases.

The structure of (33b) is given in (35); the successive lexical government domains for the DP dominating the trace are indicated by boxes.

(35)

Since the maximal lexical government domain for the trace, the matrix CP, contains the trace's antecedent (what), (33b) satisfies the lexical government clause of the ECP, and its grammaticality is expected. Notice, incidentally, that lexical government is satisfied regardless of the presence of an intermediate trace in the lower Spec(CP). That is, even if the trace were not antecedent-governed, (33b) would satisfy the disjunctive version of the ECP assumed here.

By contrast, the lexical government domain for the wh- trace in the structure for (34b), given in (36), extends only as far as the complement IP.

(36)

The reason that it extends no further is that the subject clause (the lower IP) is not governed (in other words, it is not a complement). Since the infinitival subject clause does not dominate the trace's antecedent (what), the trace of wh- movement, though governed by admit, is not lexically governed. Moreover, two IPs intervene on the path between the trace and its antecedent, and so the trace is not antecedent-governed either. (34b) is therefore correctly predicted to be ungrammatical.

When there is an independent reason for the complement clause to be a CP (when it is an indirect question, for instance), wh- movement within the bounds of the complement clause is possible, as shown in (37).

(37)	a.		Which of one's weaknesses to admit can be a tricky question.
	b.

We leave it open here why the complement clause in (37) is an IP rather than a CP.

Further issues and refinements

Given the role that the notion of barrier plays in both antecedent government and subjacency, it would be desirable to eliminate subjacency as a separate condition by subsuming it under the antecedent government clause of the ECP. Is this feasible? In light of (38a), the answer must unfortunately be 'no.' The matrix CP in (38a) is a lexical government domain for the trace (cf. the configuration in (28e)), and in the absence of subjacency, (38a) should therefore be completely acceptable. The degree to which it is not, then, provides evidence in favor of maintaining subjacency as a separate constraint on wh- movement. The acceptability contrast between (38a) and (38b,c) could then be attributed to the violation of only one principle in the former case, but of two in the latter.

Movement out of ECM complements

However, if (39b) violates both clauses of the ECP, this raises the question of how subjects of ECM complements are able to satisfy the ECP (or for that matter, subjacency), given that two barriers intervene between the antecedent and its trace in both (39b) and (40).

A standard proposal is to include the head-spec configuration as an instance of proper government, along with the head-comp configuration (Kayne 1984). This would allow the trace in (40) to satisfy the lexical government clause of the ECP. But this proposal neither explains why such examples are completely acceptable (suggesting that they satisfy both the ECP and subjacency), nor why it is possible to move not just subjects out of ECM complements, but adjuncts as well, as in (41).

An alternative that immediately comes to mind is simply that nonfinite IPs do not count as barriers, but this proposal fails to account for the contrast in (42), which essentially parallels that in (38).

However, a minor revision to the proposal that nonfinite IPs aren't barriers makes it empirically adequate (though we know of no independent motivation for the revision). We will say that IPs that are governed by V (or one of the other "contentful" categories Adj, N, or P) do not count as barriers. This has the result that the traces in the sentences in (40) and (41) satisfy the antecedent government clause of the ECP as well as subjacency. (43) shows our revised assumptions about these sentences; IPs that are not barriers are highlighted in blue.

Resume reading here. Ignore any references to the ECP.

Movement out of DP

As mentioned

(45) gives some further, naturally-occurring examples (the struck-out which in (45a) is included for clarity; it is silent in the original).

The most acceptable examples of this type involve movement out of DPs without a possessor, so it might be proposed that a further refinement of the notion of barrier is in order. In particular, let us assume that branching DPs barriers, whereas nonbranching ones are not. (What I mean by "branching DP" is a DP - not a D' - with more than one daughter; in other words, a possessive DP.) Then the examples in (44) and (45) all satisfy subjacency, as indicated by the revised representations in (46) and (47), whereas an example like (48) would continue to violate it.

Finally, given the contrast between (38a) and (38b,c), we would expect movement out of DP that violates both subjacency and the ECP to be strongly unacceptable. Movement out of left branches satisfies this expectation.

An important remaining puzzle is why some instances of movement out of nonbranching DPs are perfectly acceptable, whereas others (even ones that satisfy the ECP) are not. In our view, an explanation for the successive decrease in acceptability of examples like (50) should be sought not in syntactic, but in pragmatic factors. We indicate this explicitly by using the pound sign, the sign for pragmatic infelicity.

We begin our attempt to explain the pattern in (50) by noting that any question is associated with a so-called existential presupposition. The presupposition expresses the backdrop of knowledge against which the question is raised, and the question itself solicits information that is missing in the questioner's knowledge store. For instance, the question in (51a) is associated with the presupposition in (51b).

Now consider the existential presuppositions in (52), which range from ordinary to implausible.

If we make the plausible assumption that, all other things being equal, the acceptability of a question matches the plausibility of the presupposition with which it is associated, then questions that are perfectly well-formed from a grammatical point of view might nevertheless be judged as unacceptable if they are associated with a highly implausible presupposition. This, then, would account for the range of acceptability in the questions in (50), despite their structural parallelism.

1. The structural notion of government (to be distinguished from the morphological notion of case government is defined as in (i).

English has two sorts of though clauses: ordinary ones that do not involve movement, as in (1), and ones that do, as in (2). The construction in (2) is often referred to as the though preposing construction.

The term though preposing is potentially confusing. What is preposed is not though itself, but some constituent in the though clause. In other words, though preposing is XP preposing that is licensed by though.

In Chapter 5, we proposed treating "contentful" subordinating conjunctions like though as P. For this exercise, for simplicity, we suggest treating though as C (on a par with that). However, if you prefer the P analysis, you can solve the exercise by making the auxiliary assumption that PPs can "sprout" specifiers in the same way that CPs as a landing site for moved constituents.

A. Using the grammar tool in xbar ch12, build structures for the sentences in (1) and (2). For simplicity, you can build chunks and indicate how they go together.

B. Using the same grammar tool as in (A), build structures for just the though clauses in (3). (Omit the material in parentheses, given for context.)

C. Which, if any, of the though clauses in (3) are expected to be ungrammatical given the principles covered in this and previous chapters? Explain.

D. (4) is ungrammatical. Why?

A. Using the grammar tool in xbar ch12, build structures for both of the questions in (2).

B. Can the judgments in (2) be derived from the principles introduced in this chapter? Your explanation should be succinct, but specific. For instance, if a question violates subjacency or the antecedent government clause of the ECP, indicate which barrier causes the subjacency violation. If a question violates the lexical government clause of the ECP, indicate which clause in the definition of lexical government is violated.

C. Is the specifier position in the elementary tree that the grammar tool postulates for the complementizer for necessary?

Exercise 12.4

B. Using the grammar tool in xbar ch12, build a structure for the question in (1b) that is consistent with your answer to (A).

C. Is the acceptability of (1b) consistent with the principles introduced in this chapter? Explain.

Exercise 12.5

B. Using the grammar tool in xbar ch12, build trees for the following questions.

The grammar tool provides a choice of two elementary trees for review. Use the one that is consistent with your answer to (A).

C. Record your judgments concerning the questions (use "✓", "?," and "*" as your options).

D. For each of the questions, briefly explain whether it obeys subjacency and the ECP. Assume, as in Further issues and refinements, that neither nonfinite IPs governed by V nor nonbranching DPs are barriers. How do the predictions of the model of wh- movement developed in this chapter mesh with your judgments from (B)?

Exercise 12.6

A. Using the grammar tool in ***, build structures for the (simplified) subpart of (1) in (2).

In building the relative clause in (2), you may find it helpful to consider the declarative clause Only of few of them had been critically studied.

B. Why did you put had where you did in the structures you built in (A)?

C. The grammar tool forces you to attach of Europe as an adjunct of fishes rather than as a complement in (A). Provide evidence in favor of the attachment.

D. Is the trace of wh- movement in (2) antecedent-governed? In your answer, indicate clearly which nodes (if any) are barriers that intervene between the trace and its antecedent. Assume the definition of barriers in Further issues and refinements.

E. Is the trace of wh- movement lexically governed? Explain. In your answer, indicate clearly all (if any) lexical government domains for the trace.

F. Which (if any) clauses of the ECP does the trace of passive movement satisfy in the structure you built for (A)? As in (D) and (E), your answer should include all (if any) barriers that intervene between the trace of passive movement and its antecedent, and all (if any) lexical government domains for the trace of passive movement.

Exercise 12.7

B. Based on your answer to (A), use the xbar ch12 grammar tool to build structures for both of the sentences in (1).

C. Using the same grammar tool, build the structure for the variant of (1b) in (2).

Assume that nonfinite IP sisters to Adj are not barriers, on a par with nonfinite IP sisters to V.

D. For each trace of movement in (C), briefly explain whether it satisfies the ECP. For each case, your answer should include which clause of the ECP is satisfied, or whether both clauses are satisfied. (Don't consider verb movement.)

Exercise 12.8

A. Using the xbar ch12 grammar tool, build the structure for the phrase in (1). Treat every as a determiner.

B. Assuming that non-branching DPs are not barriers to wh-movement, is (1) consistent with the ECP? Explain. Once again, your answer should include which clauses of the ECP are violated or satisfied.

Problem 12.1

B. Using the xbar ch12 grammar tool, build a structure for (1) that reflects your judgment from (A) in light of the syntactic principles introduced in this chapter. You will need to decide what syntactic category to assign to the gerund clause.

C. Explain how your judgment from (A) can be made to follow from the structure you propose in (B).