Successive Cyclicity and A-Movement
Chomsky(1973) introduced yet another, and independent, aspect of cyclicity: successive cyclicity. Successive cyclicity, following from Subjacency (and perhaps other locality constraints as well), demands that apparent long movement is actually the result of a successive series of short movements. There is a massive literature on this for A ¯-movement (as in interrogatives or relative clauses), and a wealth of evidence has been uncovered confirming the theoretical prediction of Subjacency. Torrego (1983, 1984) and McCloskey (1990), among many others, have shown how the successive short steps composing a long instance of A ¯-movement are overtly marked, for example in Spanish and Irish. In both instances, the marking shows up in or near the complementizer position (via obligatory fronting of a finite verb or by a change in the form of the complementizer), significant in light of Chomsky’s Subjacency proposal that long-movement proceeds via that position. A-movement has been much less discussed with respect to successive cyclicity. In the remainder of this chapter, I will consider potential arguments for and against successive cyclic A-movement, ultimately concluding that A-movement does indeed proceed in this way.
Chomsky (1977, 74) presents the following example, noting that ‘‘if the rule of NP-movement that yields [(1)] applies successive cyclically...then the rule will observe subjacency.’’
John seems [to be certain [to win]]
In Lectures on Government and Binding (LGB; 1981, 44), he discusses a similar example, (2), with the representation in (3).
(2) John was believed to have been killed
(3) John INFL be believed [S t’ INFL have been killed t]
This time, though, he offers a different theoretical motivation for the assumed successive cyclicity of the movement, saying, ‘‘In the case of sentence [(3)], we were led by the projection principle to assume that the rule Move-a applies twice, leaving the two traces t and t0, successively.’’ This is evidently incorrect. While the movement might be successive (in fact, I will argue that it is), it cannot be the Projection Principle that is forcing this, since that principle demands only that y-positions be present at all syntactic levels of representation; but crucially in the LGB frame work, positions to which, or through which, movement takes place are never y-positions. Instead, Subjacency might be the relevant principle, as proposed in Chomsky 1977. Or perhaps not. Certainly in the 1970s theory of Subjacency, with S (= IP) as a bounding node, successive movement is forced. But in a version of the Barriers framework (Chomsky 1986a), IP is never an ‘‘inherent’’ barrier, and in the typical structure of A-movement such as (1), there might not be anything for the embedded IPs to ‘‘inherit’’ barrierhood from.
With that in mind, it is worth considering whether A-movement is in fact successive cyclic, and, if it is, what forces this. Chomsky (1981, 44 45) presents an empirical argument for intermediate A-traces, hence, for successive-cyclic A-movement. He observes that (4) is acceptable, indicating that Condition A is satisfied.
(4)
On this basis, he argues for intermediate traces (hence, for successive cyclic movement). His line of reasoning is as follows: ‘‘The GF y [nonthematic grammatical function] filled by medial traces such as t’ in [(3)] may... berelevant to LF; for example in the sentence [(4)],...where the medial trace serves as the antecedent of each other, which requires an antecedent in the same clause in such cases in accordance with binding theory’’ (1981, 44–45). Interestingly, though, it actually isn’t completely clear that the antecedent must be in the same clause, given the binding theory developed in Chomsky 1981 (or those developed in Chomsky 1973 and Chomsky 1986b, for that matter). Consider the formulation in Chomsky 1981.
(5)
(6)
(7)
If there is a trace in the intermediate clause, then that clause is the governing category (GC) of each other. But that is no argument that there is such a trace. Suppose there were none. Then the intermediate clause, lacking a SUBJECT, would not be the GC, and Condition A would correctly be satisfied with they as the binder of each other. The result is the same under the formulation in Chomsky 1986b. Here, I will give only the portion of the theory relevant to Condition A:
(8)
(9)
As Chomsky puts it, ‘‘For...an anaphor...the licensing condition amounts to saying that the relevant governing category for a is the mini mal one in which the binding theory could have been satisfied by some indexing’’ (p. 172). Once again, if there is an intermediate trace, each other must be (and is) bound in that clause. But in the absence of that trace, the licensing condition for the anaphor could not, hence need not, be satisfied in the intermediate clause. Therefore, matrix binding is allowed. Once again, then, there is no argument for successive cyclicity.
I will return to this issue, but first, I will consider an argument that A movement is not successive cyclic.
Epstein and Seely (1999) note that successive-cyclic A-movement should create an A-chain, by a chain formation mechanism. According to Chomsky (1995), a chain is a set of ‘‘occurrences,’’ where each occurrence is defined in terms of sisterhood. The occurrences at issue here are then intermediate projections of I, I0s, since we are considering movement through Spec,IPs. But it is widely assumed that syntactic operations can’t target intermediate projections; thus, it is reasonable to assume that the chain formation mechanism can’t either. Therefore, the needed chain links can’t exist, so movement must take place in one fell swoop.
On the other hand, even if intermediate projections are not available for any purposes, thus possibly rendering Chomsky’s definition of ‘‘occurrence’’ unstatable, an alternative definition in terms of ‘‘motherhood’’ instead of ‘‘sisterhood’’ seems equally capable of characterizing what Chomsky set out to characterize. And ‘‘motherhood’’ raises no issues at all about intermediate projections. Further, even if chains as representational objects are ultimately eliminated (for Epstein and Seely’s reasons or others), the successive movement itself is still perfectly statable in derivational terms. That is, at the derivational point where a head—say, I—is merged into the structure, the resulting configuration is, by definition, a maximal projection. It should also be recalled that there is overwhelming evidence, alluded to above, that A ¯-movement is successive cyclic. But it is very hard to see how we could accept Epstein and Seely’s argument for A-movement but reject it for A ¯-movement, since type of movement is ir relevant to their line of reasoning.
With this argument against successive-cyclic A-movement called into question, I return to Chomsky’s Condition A argument for successive cyclicity. Above, I showed that Chomsky’s argument does not go through on standard formulations of the binding domain for anaphors. Chom sky’s argument relied on a clause-mate condition, but those formulations do not give that result in the crucial cases. However, in Lasnik 2002, 2003, I argue that clause-mate is, in fact, the correct characterization, as argued much earlier by Postal (1966, 1974). My argument is initially based on a very interesting verb-particle construction first discussed by Kayne (1985) and later analyzed by Johnson (1991) in terms relevant to the present discussion. Johnson provides an insightful account of examples like (10) involving overt raising of the ECM (exceptional case marking) subject John.
(10) Mary made John out to be a fool
Both Kayne and Johnson convincingly treat (10) as an infinitival counter part of (11).
(11) Mary made out that John is a fool
Very plausibly, John in (10) has overtly raised into the matrix clause. Now consider that for many speakers, the raising seen in (10) is optional. For those speakers, (12) is an acceptable alternative to (10).
(12) Mary made out John to be a fool
This time, plausibly John has remained in the embedded clause, and in Lasnik 2002, 2003, I provide evidence that this is so. Now note that an anaphor on the upstream side of out is much better than one on the downstream side, with antecedent in the matrix clause.
(13)
(14)
None of the standard formulations of the notion of governing category predict the disparity between the (a) and (b) versions of these examples, but, under the stated assumptions, a clause-mate formulation could. In each of the (b) examples, the anaphor is separated from its antecedent by the boundary of the infinitival clause, while in the (a) examples, no clause boundary of any sort intervenes. It must be acknowledged that there is an interfering factor: verb-particle-NP order with reflexives and reciprocals is somewhat degraded even with only one clause, as in (15)–(16).
(15)? Jack called up himself
(16)? They called up each other
However, the deviance is less severe than that seen in ECM instances like (13)–(14). Thus, the argument for the clause-mate condition stands.
Given the clause-mate requirement on anaphors, examples like the following, pointed out to me in this connection by Adolfo Ausı´n (also attributed to Danny Fox, via David Pesetsky, in Castillo, Drury, and Grohmann 1999), argue for successive-cyclic A-movement:
(17)
In the absence of successive-cyclic movement, himself would incorrectly be predicted to violate Condition A, and, on fairly standard assumptions, herself to be in conformity.
A similar though somewhat less direct argument can be constructed on the basis of Condition B. Examples (18)–(19) display standard Condition B effects.
(18)
(19)
These contrast with (20)–(21), where the antecedent is more distant in relevant respects.
(20)
(21)
As would be expected, (18) remains unacceptable under VP-ellipsis.
(22)
What is not expected is that (19) is substantially improved under VP ellipsis.
(23)
How is it that deletion, which I take to be a PF process, remediates a presumably semantic violation? Suppose that in addition to their proper ties with respect to anaphora, the pronouns in question also have to satisfy a morphosyntactic requirement: as weak pronouns, they must cliticize onto the verb, as suggested by Oehrle (1976), who cites data like these from Chomsky 1955:
(24) The detective brought John in
(25) The detective brought in John
(26) The detective brought him in
(27) *The detective brought in him
If the relevant structural configuration for Condition B is based on the notion of clause-mate, an account of the ellipsis paradigm presents itself. In particular, one might hypothesize that it is not actually the Condition B violation (presumably an LF effect) in (19) that is repaired by ellipsis in (23). Rather, it is failure to cliticize (a PF violation) that is repaired (by PF deletion). And without cliticization, the pronoun can remain in the lower clause (if, as I argue in Lasnik 1999, 2001, ‘‘subject raising’’ in these constructions is generally optional). On the other hand, in (45), the pro noun and its antecedent are clause-mates independent of cliticization, so there is no possibility of ‘‘repair’’ of the Condition B violation by ellipsis.
Given the clause-mate character of Condition B, (28) constitutes evidence for successive-cyclic movement, as John must have moved through the intermediate Spec,IP to yield the observed obviation. John in its sur face position is not a clause-mate of him.
(28)
This raises the question of what forces the movement to be successive cyclic. It is possible that it is a locality constraint on movement, as argued most recently by Bosˇkovic´ (2002). But it is not so clear that the most ‘‘minimal’’ theories of locality guarantee this. At the moment, it seems to me that the likeliest culprit is the Extended Projection Principle, as I argue in Lasnik 2003.
