Cyclicity and Minimalism
In the course of his minimalist critique of all of syntactic theory, Chom sky (1993) argues that deep structure—the level of representation resulting from recursion in the base—should be eliminated and generalized transformations reinstated. He notes that in Aspects, ‘‘Elimination of generalized transformations in favor of cyclic base generation [sic] is...justified in terms of explanatory adequacy. But the questions under discussion then do not arise in the far more restrictive current theories’’ (p. 46). But it seems that the questions do still arise, and the answer is the same as it was, as noted above. Chomsky continues to assume that we must eliminate the kinds of derivations that concerned him in Aspects. Recursion in the base (and the concomitant level of deep structure) was not the antidote. Rather, the (strict) cycle was, as pointed out by Freidin (1978).
The minimalist approach to structure building and cyclicity that Chom sky proposes is as follows. We have ‘‘bottom-up’’ creation of a phrase marker by generalized transformations. There is no deep structure. The derivation is constrained by an ‘‘extension condition.’’
(1)
Chomsky observes that ‘‘[(7)] yields a version of the strict cycle’’ (p. 23).
Actually, I regard the necessity of the principle of strict cyclicity (in any of its incarnations) as still an open question. Certainly, it is not unreason able to regard as best the theory that allows the smallest set of derivations (the classic motivation for strict cyclicity). But it is also not unreasonable to regard as best the theory that makes the fewest stipulations. In this instance, the two methodological desiderata appear to be at odds. Sub sequent to the formulation of the classic theoretical argument for strict cyclicity, empirical arguments were developed for cyclicity. The first type of argument centers on derivations where one transformation must both precede and follow another in order to produce a particular sentence. One standard example from the late 1960s involved Passive in both an embedded and a matrix sentence, as in (2).
(2) Mary is believed by John to have been unfairly criticized
At the time, it was taken for granted that Passive can only move an ‘‘object’’ into subject position. Thus, in (2), following Passive in the embedded clause that moves Mary into embedded subject position, Subject Raising makes Mary the object of the matrix clause. At this point, it is available for Passive in the matrix. Thus, Passive must be able to both precede and follow Subject Raising. Strict ordering of Subject Raising before Passive combined with cyclic application of rules makes this possible. However, Freidin (1978) shows that this entire class of arguments—potential undergeneration—carries no weight. Assuming free ordering (i.e., eliminating the stipulation of strict linear ordering), all of the grammatical sentences that could have been generated by cyclic derivations can be generated without it. As Freidin discusses, of greater potential import is overgeneration. The standard case involves illicit extraction out of an embedded question, as in (3).
(3) *Wherej did [you wonder [whati [John put ti tj]]]?
(3) is, of course, immediately excluded by Subjacency, but only if where is prevented from moving ‘‘from Comp to Comp.’’ If what undergoes wh movement first, the embedded Comp (Spec,CP in later developments of the theory) is unavailable for the first step of the movement of where. But if where moves first, the embedded Comp is available. Then, once where moves to the higher Comp, what can move into the now vacated lower one. It is this kind of derivation that strict cyclicity forbids. How ever, again as noted by Freidin, under plausible assumptions this kind of derivation is independently excluded. Given that movement leaves a trace, the position into which what must move is not empty, being occupied by the trace of where. This suffices to prevent the derivation.
Chomsky (2000, 136) proposes a new theoretical argument for (the minimalist version of) strict cyclicity. The idea is that derivations conform to a condition demanding as little tampering as possible with existing structure. As Chomsky puts it, ‘‘operations preserve existing structure.’’ I will use a schematic example to illustrate this point. Consider the phrase structure tree in (4).
(4)
Now suppose b is adjoined to X, in conformity with (1), the Extension Condition. The resulting tree is (5), which clearly includes (4) as a sub tree, thus preserving the existing tree structure.
(5)
Suppose, on the other hand, we were to adjoin b to C, the most flagrant possible violation of the Extension Condition. The result is shown in (6).
(6)
This time the original tree is not a subtree of the derived tree. Thus, if it really is a priori desirable to preserve existing structure, we have a strong argument for strict cyclicity as instantiated by the Extension Condition.
Interestingly, this strong argument might actually depend on a some what arbitrary notational choice. Given the graph-theoretic representation of phrase structure, the argument trivially went through. But Chomsky’s original conception of phrase structure, as in Chomsky 1955 and numerous subsequent works, was set-theoretic. Consider, then, the corresponding derivations on phrase markers construed as sets. The initial phrase marker corresponding to (4) is (7).
(7) {X, ZA, ZBC}
With b adjoined ‘‘high’’ (i.e., to the node that dominates all others), the derived phrase marker is (8), corresponding to (5).
(8)
With b adjoined ‘‘low,’’ the derived phrase marker is (9), corresponding to (6).
(9)
Surprisingly, with the ‘‘low’’ attachment, the original structure is a subset of the derived one, while with the ‘‘high’’ attachment, it is not. More generally, the higher the attachment site, the more radically the set is altered; the lower, the less radically. Thus, surprisingly, the relationship between the Extension Condition and preservation of structure appears to be notation dependent. Again, an argument for strict cyclicity is less conclusive than it originally seemed.
Chomsky (1995) gives what can be construed as an argument against the Extension Condition, though he doesn’t present it quite that way. Developing his notion of ‘‘feature strength,’’ which forces overt movement, he proposes that strong features are elements that are not tolerated by the derivation and therefore must be eliminated (almost) immediately upon insertion into the structure. More precisely, he defines strength as follows:
(10)
Chomsky observes that ‘‘cyclicity follows at once’’ (p. 233). This is because this conception of strength disallows acyclic checking of heads with strong features. Bosˇkovic´ and Lasnik (1999) explore this deduction in detail, particularly emphasizing what Chomsky implies: the fact that the Extension Condition and strength radically overlap in their effects. Typically, if overt insertion of a head a with a strong feature takes place acyclically, then the derivation will immediately cancel, since, by hypothesis, a is contained in a category not headed by a. Similarly, if overt movement takes place acyclically, then the strong feature F of a driving the movement must not have been checked until later structure has been created headed by something other than a. Consider, for example, an acyclic derivation of (3), repeated here as (11).
(11)
Suppose Subjacency, at least in wh-island configurations, reduces to Relativized Minimality. Then there is a potential derivation in which where moves directly to the matrix Spec,CP before what moves into the embedded Spec,CP. Relativized Minimality would not prevent this derivation, given the acceptability of (12).
(12) Where did you put what?
That is, (12) shows that prior to wh-movement, where can be higher than what. The Extension Condition rules out (11) on this derivation. How ever, (10) redundantly blocks the derivation, since the strong feature of the embedded interrogative C is not checked until too late, or, more precisely, the derivation is canceled as soon as the complement CP is embedded. As Bosˇkovic´ and Lasnik (1999) note, this redundancy is conceptually problematic. Bosˇkovic´ and Lasnik propose removing the redundancy by eliminating the Extension Condition, since they argue that the condition is problematic in several respects. In particular, head adjunction and LF movement appear to freely violate it. First, consider head adjunction. Suppose X has merged with YP, creating X’, as in (13).
(13)
Now suppose Y is to raise to adjoin to X, the typical case. The difficulty is that this adjunction does not extend the tree upward, as (8) shows.
(14)
Similarly, if there is covert wh-movement in a module following all of the overt operations, such movement in an embedded question like (15) will fail to extend the tree upward (with what moving to join who in the embedded Spec,CP).
(15) Mary wonders who bought what
If cyclicity is reduced to feature strength as in (10), then head adjunction and LF movement are no longer exceptional. For example, no additional stipulations are now needed to accommodate overt (strong-feature-driven) V-to-I raising in French, as long as the relevant IP projection is not embedded within another category before V-to-I adjunction takes place. Also, since LF movement is, by definition, not driven by strong feature checking, it is automatically consistent with the strength conception of the cycle.
An interesting property of this conception of the cycle is that, unlike the Extension Condition, it does not in principle ban elements that do not have any strong features from entering the structure acyclically. Bosˇ kovic´ and Lasnik (1999) show that permitting such derivations has several positive effects. One instance involves Empty Category Principle/ Subjacency asymmetries. Quite generally, traditional ECP violations with extraction of adjuncts go hand in hand with Subjacency violations with extraction of arguments. More precisely, in the contexts in which extraction of adjuncts leads to an ECP violation, extraction of arguments leads to a Subjacency violation. This is illustrated in (16) with an adjunct island.
(16)
Chomsky and Lasnik (1993) develop a system in which traditional ECP violations with extraction of adjuncts and Subjacency violations with ex traction of arguments reduce to the same economy condition, the only distinction between the two being that with argument Subjacency violations the offending trace is deleted in LF, whereas with adjunct ECP violations it remains present in the final LF representation. In this system, we would not expect to find a configuration in which extraction of adjuncts would lead to an ECP-type violation, but extraction of arguments would not lead to a Subjacency-type violation. It is well known, however, that such configurations exist. For example, as discussed by Rizzi (1990), pseudo-opacity effects and inner island effects obtain with adjunct extraction, but not with argument extraction. (17) illustrates this for pseudo-opacity.
(17)
The strong feature approach to cyclicity can explain this surprising asymmetry between the ECP and Subjacency, provided that we adopt Lasnik and Saito’s (1984, 1992) theory that adjunct traces are checked with respect to locality restrictions only in LF, whereas argument traces can be checked in overt syntax.4 Assuming that beaucoup does not have any strong features to check and is not required to be present in the structure to check strong features of another element, it could enter the structure acyclically. However, since beaucoup is phonologically realized, we know that it must have entered the structure in overt syntax. Given Lasnik and Saito’s proposal, beaucoup then does not have to be present in the structure when argument chains are checked with respect to locality restrictions, but it does have to be present when adjunct traces are checked. The surprising asymmetry between the ECP and Subjacency exhibited by pseudo-opacity is thus captured in a relatively straightforward way.
