Generalized cyclicity
Although it seems that all the empirical consequences of SCC violations identified so far can be handled by other mechanisms that have been proposed, the question remains whether derivations do in fact adhere to cyclic operations as a result of general constraints on the computational system. In the earliest account that attempted to derive the empirical effects of the SCC from principles of UG (Freidin 1978), these effects are a fortuitous consequence of the combined effects of several principles of UG. In the more recent analyses discussed above, countercyclic derivations are blocked by various stipulations—for example, that substitutions must create positions external to the targeted P-marker (the Extension Condition) or that an unchecked strong feature in a maximal projection cancels a derivation. These seem like retreats from earlier deductive accounts because they are based on a descriptive statement rather than on general principles that have some conceptual plausibility. The analysis based on FI and the MLC is an attempt to reconstruct a more deductive account under minimalist assumptions.
In this section, I review some more recent attempts, also based on minimalist assumptions, to deduce strict cyclicity from more fundamental principles of grammar. I begin with a brief discussion of the nature and role of transformation under a theory of bare phrase structure and end with a somewhat different perspective on cyclicity that follows naturally from the fundamental concepts of that theory.
For concreteness, let us assume the following version of the bare phrase structure theory. Phrase structure is constructed by classical adjunction, a structure-building operation that concatenates two categories and projects the label of one of them. This operation can apply to two items in a numeration (or perhaps directly in the lexicon), or to a lexical item and a P-marker that has been constructed as part of the derivation, or to two such P-markers. All three possibilities can generate a specifier-head configuration, whereas perhaps only the first two can generate a head-complement configuration. When the elements adjoined are not from the same P-marker, the operation acts as a generalized transformation. In contrast, the adjunction elementary can apply to two elements of a single P-marker, in which case the operation acts as a singulary transformation. Even though the adjunction elementary is involved in lexical insertion, phrase structure generation, and movement (both overt and covert), the conditions on its application differ depending on whether it operates as a generalized or singulary transformation. In the former case, adjunction operates on categories; in the latter, it is presumably restricted to the adjunction of a feature to a category. In standard practice, the former case is called Merge and the latter, Move (or Attract/Move). In addition, there must be some further operation that accounts for the pied-piping of a phrasal category in the case of overt category movement.
From this perspective, the issue of cyclicity concerns the application of the adjunction elementary to a nonroot category during a derivation, since presumably this application could have occurred at the point in the derivation where the adjunction site was the root node. Thus, adjunction to a nonroot, either by Merge or by Move, constitutes a countercyclic application. In other words, every category in a P-marker marks a cyclic domain in some sense, not just those categories that form domains for movement rules as in earlier analyses.
Kitahara (1997) attempts to account for generalized cyclicity in terms of a new analysis of the operations Move and Merge. He claims that there is a distinction to be made between the cyclic and countercyclic (noncyclic in his terminology) operations of Move and Merge. Cyclic Move, for example, consists of a single elementary operation: concatenate α and Σ, forming Σ′. It should be obvious that this is just classical adjunction. This operation maps (1a) onto (1b).
(1)
In contrast, countercyclic Move is more complicated. The targets of the operation, α and K, are contained in Σ, so concatenation of α and K forms L={γ, {α, K}}. The insertion of L into the P-marker now requires an extra operation, which Kitahara calls “replacement” mapping (2 a) onto (2b).
(2)
Thus, countercyclic Move on this analysis requires two elementary operations instead of one. Kitahara applies the same analysis to Merge. The preference for cyclic operations over countercyclic operations results from a general economy condition that prefers shorter derivations.
(3)
Under this analysis, cyclicity of rule application follows from general economy considerations.
Even though Kitahara’s analysis is more general and therefore may seem more appealing than one based on feature strength, the claim that there is an elementary operation of replacement seems problematic. For one thing, it seems doubtful that this elementary operation plays a role in legitimate derivations. If it does not, then there is no motivation for claiming that it exists at all. Its sole purpose would be to act as an extra operation so the economy analysis succeeds. We could just as easily say that the function postulated for this operation is illegitimate, thereby eliminating the need to appeal to economy—the line of inquiry I will pursue below. Kitahara claims that the erasure operation is an instance of replacement in which a feature F is “replaced” by the empty element Ø. However, there is no clear reason why this replacement is required if by erasure F becomes inaccessible to any further operation of CHL.16 The minimal solution should be simply to eliminate the feature.
Yet even if we grant that there could be an elementary operation of replacement that is responsible for erasure of features, it alone could not account for the work Kitahara attributes to it in the case of countercyclic Move. In erasure, the replacement operation has the effect of simply substituting a feature for the empty element. In Move, as shown in (2), replacement results in the deletion of Σ and the projection of Σ′ from L as well as the substitution of L for K.17 The deletion of Σ here cannot be the result of replacement so some other elementary operation would have to be involved.
Under earlier theories of elementary operations (see Freidin 1992), there were a maximum of three such operations, substitution, adjunction, and deletion, each of which performed one of the three basic operations on structure: structure preservation, structure creation, and structure destruction, respectively. Although a structure-preserving operation has been dropped in recent analyses, the other two kinds of operation are surely conceptually necessary. On minimalist assumptions, we would expect that elementary operations perform only these basic and simple structural tasks. From this perspective, Kitahara’s replacement operation, when it applies countercyclically, performs several truly elementary operations, including the structure-destroying erasure of Σ and the structure-building replacement of K with L.18 Hence, it is clear that the operation of replacement as characterized for countercyclic movement operations in Kitahara 1997 is simply not an elementary operation; therefore, it seems unlikely that it would be part of CHL.
Kawashima and Kitahara (1996), Collins (1995, 1997), and Epstein et al. (1998) make proposals to block countercyclic derivations generally as violations of Kayne’s (1994) Linear Correspondence Axiom (LCA), a general constraint on linear ordering in P markers. Rather than discuss the different proposals here, I will restrict my comments to the analysis presented in Collins (1997).
Collins claims that countercyclic operations generally yield structures that violate the LCA. Consider, for example, his discussion of countercyclic Merge. He assumes that given a set of syntactic objects Σ (i.e., a numeration) containing three elements (SOs), a countercyclic derivation like (4) can occur.
(4)
The step (4c), which is allowed by Collins’s definition of Merge (which I will not discuss here), creates a structure in which the two distinct nonterminals dominating β and γ are not in a c-command relation; hence, neither asymmetrically c-commands the other, in violation of the LCA. Collins gives (5) as an illustration of the structure generated by (4).
(5)
As a concrete example, Collins gives a derivation for the countercyclic insertion of a subject into [Spec, VP] after VP has been merged with T. The relevant steps are given in (6).
(6)
In this case, John and T are the terminals that fail to be ordered with respect to each other. Collins notes that this analysis for countercyclic merger applies as well to countercyclic movement.
Though intriguing, the LCA analysis of countercyclic Merge seems seriously flawed. First, note that Collins’s treatment of Merge would not be allowed under Chomsky’s characterization.
Clearly, then, CHL must include a second procedure that combines syntactic objects already formed. A derivation converges only if this operation has applied often enough to leave us with just a single object, also exhausting the initial numeration. The simplest such operation takes a pair of syntactic objects (SOi, SOj) and replaces then by a new combined syntactic object SOij. Call this operation Merge. (1995c, 226)
Under Chomsky’s formulation, Merge creates a new syntactic object, which surely must be treated as a single object by subsequent legitimate applications of Merge. Presumably by “replacing” the pair of syntactic objects with a new single object, Merge loses the ability to access the parts that have been replaced. Furthermore, there is a crucial difference between the unordered set of elements in the initial numeration and the set of elements created by Merge—namely, the latter constitute ordered sets (assuming of course that Merge gives a linear order to the objects it concatenates). So instead of (6a) as input to Merge, we have (7).
(7)
Under Chomsky’s characterization of Merge, (32) contains two syntactic objects (John and the ordered set>), not three (John, T, ) or four (John, T, saw, me).
Turning now to countercyclic Move, it may be that some version of the LCA will prohibit the operation from applying in legitimate derivations—presumably, such derivations will crash at PF because they contain elements that cannot be assigned a linear order.
The analysis based on the LCA assumes that countercyclic operations are in some sense possible, though they have consequences that are later filtered out. Alternatively, the bad consequences of a countercyclic operation might be immediate, as I will suggest in what follows. What I would like to propose is that the issue of countercyclic movement crucially concerns the question of derived constituent structure, which is central to transformational analysis.
Recall that from the beginning of modern generative grammar (e.g., Chomsky 1975a), transformations are defined as operations that apply to analyzed strings. As Chomsky notes (1975a, 320), “We must provide a derived constituent structure of transforms, for one thing, so that transformations can be compounded.” Thus, if we cannot specify the derived constituent structure for a transform, then it cannot serve as the input to another transformation. In earlier theories, a significant amount of information about derived constituent structure came from phrase structure rules. For example, Emonds’s structure preserving hypothesis (see Emonds 1970, 1976) proposed that the derived constituent structure from the application of structure-preserving rules was determined by the phrase structure rules of the base (cf. Chomsky’s (1975a) discussion of the derived structure of the passive by phrase). However, with the demise of phrase structure rules and the advent of bare phrase structure theory, we no longer have phrase structure rules or schemas to provide derived constituent structures. Under the bare phrase structure theory, the transformational operations that create constituent structure also specify derived constituent structure.
On minimalist assumptions, we might expect that the processes that create constituent structure are the only ones that provide derived constituent structure. In other words, there are no separate processes for determining derived constituent structures.
From this perspective, consider the problem of assigning derived constituent structure to an operation that moves a phrase to concatenate with a nonroot category. For concreteness, suppose that the object of a passive predicate is moved to form [Spec, IP] after the IP has been merged with a complementizer as in (8).
(8)
Notice that in (8), the target of the operation is IP, the maximal projection of I and the complement of that. When the NP Adam concatenates with IP, IP must project a category so that the NP will now by construction c-command the IP. The syntactic relations between the moved phrase and its target are no different from those that would have arisen if the target had been the root category at the point where the movement applied. However, it is not clear that the moved constituent or the newly projected IP bear any relation to the complementizer that c-commands the target IP. To see how this works, consider the structure derived from the cyclic movement of the object to [Spec, IP].
(9)
In (33), CP immediately dominates the IP was elected Adam, whereas in (34), CP immediately dominates the IP Adam was elected Adam. To construct (9) from the countercyclic derivation, the syntactic relation between the complementizer that and the moved NP Adam would have to be redefined; and, as a result, the relation between that and the IP was elected would also have to be redefined. These redefinitions would require some additional grammatical mechanism. Therefore, unless such a mechanism can be motivated on independent grounds, there is no reason to assume it is available.
Given that there is no redefinition mechanism, what happens when we attempt to move a phrase to concatenate with a nonroot category? One possibility is that when the concatenation occurs, the nonroot category projects to create a three-dimensional P marker, which can then be ruled out by the LCA. However, if there is little motivation for three-dimensional P-markers in the first place, then we want to avoid postulating such entities. However, if P-markers are only two-dimensional, there is no way for a category to project once it has become a constituent of another category. If the concatenation via movement cannot be carried out because the target cannot project in the normal way, then it seems reasonable to assume that such operations are not possible.
This analysis raises questions about the status of head movement and feature movement to heads more generally, as well as the status of covert movement. With respect to covert movement, there are two primary cases to consider: Quantifier Raising (QR) and object movement (e.g., in English). There are proposals for the analysis of quantifiers that eliminate QR (see Kitahara 1996; Hornstein 1995). If object movement is actually overt in English, as argued by Lasnik (1995c) and others, then perhaps there is no covert movement of categories at all. Turning to head movement, notice that this operation causes a problem only if we insist that the output is a string consisting of the moved head followed by the functional head it is adjoined to. The adjunction analysis appears to be redundant given that the head itself enters the derivation with whatever features the functional head represents (e.g., tense or agreement). Given that there is no need for such redundant constructions, a less problematic way to effect head movement would be to substitute the lexical head for the functional head. Thus, there may be a limited role for substitution in grammar after all. The substitution analysis for head movement would eliminate the problem posed by the absence of redefinition mechanisms.
If this analysis is on the right track, then the reason neither countercyclic Merge nor countercyclic Move can apply is that the elementary operation itself does not permit this. On minimalist assumptions, this should be the optimal explanation. It makes no appeal to external factors like economy, constraints on linear ordering, or other UG principles. Certain kinds of derivations are excluded simply on the grounds that the elementary operations of CHL are so constructed that they cannot occur.
