From conditions to Move α
In discussing the early work on the system of conditions—in particular, the Tensed-S, Specified Subject and Subjacency conditions—Chomsky makes the following important observations:
If these principles can be substantiated or improved, the class of potential grammars is vastly reduced. Many potential rules are eliminated outright. Furthermore, by limiting the possible application of rules, principles of the sort discussed make it unnecessary to make available in the theory of transformations as rich an apparatus as would otherwise be needed to delimit the application of particular rules. Thus, the principles constrain the variety of grammars by reducing the ‘expressive power’ of grammatical rules.
We might even set ourselves the goal, still distant but perhaps attainable, of so restricting the apparatus of the theory of transformations that rules can only be given in the form ‘move NP’, with other conditions on their application expressed either as general conditions on rules or as properties of initial phrase markers, or as properties of surface structures. (1975c:111–12)
Work on the system of grammatical principles progressed so rapidly, that in Chomsky 1976 the rule Move NP is taken as an established result.27 In particular, work on the syntax of Romance languages (e.g., French and Portuguese) demonstrated that not only did Chomsky’s Tensed-S and Specified Subject conditions apply crosslinguistically but that they applied to constructions not found in English—e.g., clitic constructions (see Kayne 1975; Quicoli 1976a,b).
To see how the Tensed-S Condition (henceforth TSC) and the Specified Subject Condition (henceforth SSC) mediate the syntactic distance between the movement and landing site with respect to Move NP, consider the following derived structures.
(1)
In essence, the TSC prohibits the application of a rule that links two positions across a finite clause boundary. Thus the application of Move NP in (1a.i—ii) violates this condition. The SSC prohibits the application of a rule that links two positions across a syntactic subject—hence the application of Move NP in (1a.ii) and (1b.ii) is excluded by the condition. In contrast, the application of Move NP in (1b.i) violates neither condition (or any other principle of grammar) and therefore is allowed.
Under the trace theory of movement rules, where ti constitutes an empty category left behind by the movement operation (e.g. substitution), it became possible to relate NP movement phenomena to anaphor binding by treating trace-binding as another instance of bound anaphora. Under this analysis it was possible to explain why NP-preposing in passive sentences was obligatory. As noted by Fiengo (1974, 1977), the trace binding relation in structures like (2) violates an independently motivated constraint on anaphor binding which accounts for (3).
(2)
(3)
Given that this constraint on anaphor binding is part of UG, we have derived a piece of the behaviour of NP-preposing from a general principle of grammar and therefore need not stipulate it in the formulation of the transformation. Moreover, as demonstrated in Chomsky 1976, the TSC and SSC apply to anaphor binding as well as NP movement.
(4)
In (12a.i–ii) anaphor binding violates the TSC; whereas in (12a.ii) and (12b.ii) the binding relation violates the SSC. In this way the paradigms in (9) and (12) are analogues. As a consequence, it appears that much, if not all, of the behaviour of NP movement can be derived from general principles of UG, and therefore that the movement transformation may be formulated simply as Move NP.
In Chomsky 1976 the TSC and SSC are treated as conditions on the application of rules. Thus in (9) they block the application of Move NP, whereas in (12) they block a rule of interpretation that links the anaphor himself to the antecedent John. Another way to interpret these conditions is as conditions on representations—or, more specifically, as conditions on binding, assuming that an NP trace is the empty category analogue of a lexical anaphor.
Restricting conditions (4) and (5) [the Prepositional Island Condition (PIC) and the SSC respectively], now, to rules of construal, we interpret them as applying to transformational rules as filters, in effect; the result of applying a transformational movement rule may or may not yield an appropriate case of ‘bound anaphora’. (Chomsky 1977b:77)
Under this interpretation Move NP (or the rule that links an anaphor to an antecedent) applies freely. If the result is deviant, then the derived representation will violate some principle of grammar. The latter interpretation captures a significant generalization and is in addition motivated by a redundancy argument, as demonstrated in Freidin 1978. If we interpret these principles as conditions on the application of rules, then a further condition on the application of rules, the Strict Cycle Condition (Chomsky 1973), is required to rule out one derivation of such deviant sentences as (5).
(5)
The Strict Cycle Condition (henceforth SCC) prohibits the application of a rule within a cyclic subdomain of a current cycle. The derivation of (5) which violates the SCC is given in (6), where np is construed as a base-generated empty NP.
(6)
On the first transformational cycle, α′, John is preposed to the complement subject position. Then on the second transformational cycle, α, John is raised to matrix subject position. If the derivation stops here, we have a well-formed sentence John was reported to have been given that book. However, if Move NP applies again within α′, now a cyclic sub-domain of a, and substitutes that book for the trace of John in complement subject position, the resulting derivation violates the SCC. If, on the other hand, the SSC is interpreted as a condition on representations, then (6d) violates the SSC because John now binds its only trace across a subject even though no application of Move NP violated the SSG. Under this interpretation we derive the empirical effect of the SCC. If all the empirical effects of the SCC can be derived in this way, as argued in Freidin 1978, then the SCC is redundant with respect to those other principles of grammar from which its effects are derived. This result, as well as the binding theoretic interpretation of the TSC and SSC, suggests a shift in focus from conditions on rule application to conditions on syntactic representations, thereby reinforcing the shift from rules to principles.
If this reduction of the ‘passive transformation’ to the rule Move NP is to succeed fully, then some account of the ‘special’ behaviour of the passive auxiliary be+en is required. Recall that in Chomsky 1976 the analysis in which the passive auxiliary is introduced via the rule of NP-preposing is explicitly rejected. Instead it is assumed that this auxiliary ‘derives from an independent source’ (p. 173). If this independent source is a phrase structure rule, then the argument in SS against base generating the passive auxiliary must be addressed. As Chomsky noted, the occurrence of the passive auxiliary is, unlike other auxiliary verbs, subject to a number of ad hoc restrictions:
(7)
Examples (7a–c) rule out sentential constructions like (8a–c) respectively.
(8)
Under the earlier transformational analysis of passives, these restrictions were accounted for by inserting the auxiliary as part of the passive transformation. However, if the passive auxiliary is base generated, then these restrictions must be handled by other ways—optimally, as the effects of UG principles.
Such an account can be derived from a theory of predicate/argument relations developed during the 1970s. The theory concerns the assignment of semantic functions (or thematic relations (henceforth θ-roles)) to arguments of predicates. In active sentences the assignment of θ-roles to arguments is straightforward.
(9)
In (9) the predicate praised assigns two θ-roles, one to the subject (θS) and one to the object (θO). In (10), the passive counterpart to (9), θ-role assignment works somewhat differently.
(10)
The syntactic subject Adam is assigned θO by virtue of the trace it binds. The object of the by-phrase, Bernie, is assigned θS by the preposition by.35 It is standardly assumed that the passive morphology on the verb praised prevents it from assigning θS to the syntactic subject position as in the active construction. Furthermore it is generally assumed that any licit application of NP-preposing moves an NP to a nonthematic position—i.e., one that is not directly assigned a θ-role. The possibility of NP movement between two thematic positions, as in (11), is prohibited by a general principle (12) governing the distribution of semantic functions.
(11)
(12)
Given the analysis of the active/passive pair (9–10), Adam would be assigned θS by virtue of occupying the subject position and would also be assigned θO via its trace in object position. Because the sentence Adam praised has no interpretation Adam praised himself, the structure in (11) is not licit—as predicted by the Functional Uniqueness principle. A further condition (13) rules out the possibility of a construction in which an argument bears no θ-role, as would occur if every syntactic position in a passive construction were filled with a lexical NP, as in (14).
(13)
(14)
In (14) θS is assigned to Bernie and θO is assigned to Barbie. Adam would receive no θ-role assignment in violation of (13). Thus Functional Relatedness accounts for restriction (7b) on the distribution of the passive auxiliary. Notice that it would also account for restriction (7a) under the assumption that passive morphology blocks θ-role assignment to the syntactic subject position.
The final restriction, (7c), falls under another principle of predicate/argument structure—namely that each semantic function of a predicate must be uniquely assigned (that is, assigned only once)—henceforth the principle of Unique Assignment. Given that θS is assigned to the syntactic subject position by a verb lacking passive morphology or by the preposition by, it would seem that in (8c) θS is assigned twice to two distinct arguments in violation of the Unique Assignment principle.37 Thus, given the principles of Functional Relatedness and Unique Assignment, the apparently ad hoc restrictions on the distribution of the passive auxiliary follow from the theory of UG. These two principles plus Functional Uniqueness form a subcomponent of UG referred to as θ-theory.
Given the principle of Unique Assignment, it would possible to base-generate a wh phrase containing a lexical object rather than one with an empty object position which is filled by a rule of Agent postposing—that is, an instance of Move NP. One argument for base-generating the object of the wh-phrase in place concerns the derivation of nominals like (15).
(15)
Under trace theory, the Agent-postposing analysis leaves behind an NP trace in prenominal position. The derivation of (15) would require the substitution of that trace with the determiner the, in violation of the Nondistinctness Condition on substitutions. Note that this analysis presupposes that there is only one prenominal position to be occupied by either a determiner or a possessional NP, a standard assumption during the 1970s. Granting basic assumptions about the analysis of categories, a noun (and hence the NP it projects) is distinct from a determiner. If we eliminate the Agent-postposing analysis in nominals, then there is little reason to retain it for sentential constructions.
In eliminating the application of Move NP that postposes the underlying subject from the derivation of sentential passive constructions, we can no longer rely on binding theory to explain the obligatory nature of NP-preposing as discussed above. Instead, the obligatory character of NP-preposing is subsumed under Case theory, another subcomponent of UG which was developed during the late 1970s.
The basic idea behind Case theory, due to Vergnaud (1977), is that there are two kinds of NP positions, those that are marked for Case and those that are not, and only Case marked NPs can be lexically realized. Chomsky (1980) formulates this idea in terms of a filter (16), generally known as the Case Filter:
(16)
Given the Case Filter, a lexical NP which occurs in an underlying Caseless position must move to a Case-marked position. This accounts for the syntax of predicate adjectives which take sentential complements.
(17)
In (17a) the complement subject the situation occurs in a Caseless position, in contrast to the finite clause complement subject in (17c). However, in (17b) the complement subject has been moved to the subject position of a finite clause (cf. (17c)) where it can receive Case. Thus (17a) violates the Case Filter, whereas (17b,c) do not. The NP-preposing that occurs in the derivation of (17b) is therefore required to satisfy the Case Filter.
This Case-theoretic analysis generalizes to NP-preposing in sentential constructions containing predicates with passive morphology, as illustrated by the raising paradigm given in (18).
Example (18a) shows that the subject of the infinitival complement of the verb expect is in a Case-marked position—presumably because of its syntactic relationship to the matrix verb since the subject of an infinitival is not itself a Case-marked position, as illustrated in (17a). However, the addition of passive morphology to the matrix verb renders the complement subject position Caseless, hence (18c) violates the Case Filter. Thus the complement subject more women must move to a Case-marked position to satisfy the Case Filter. As in (17b), movement to the subject position of a finite clause places the complement subject in a Case-marked position, thereby satisfying the Case Filter. The comparison of the two paradigms (17) and (18) suggests that passive predicates, in contrast to their nonpassive counterparts, share the same property as predicate adjectives—namely an inability to Case-mark an NP.41 ‘Suppose that the unique property of passive morphology is that it in effect “absorbs” Case: one NP in the VP with the passive verb as head is not assigned Case under government by this verb’ (Chomsky 1981:124). The analysis generalizes without qualification to simplex passive constructions. Therefore, the obligatory character of NP-preposing follows from a principle of UG, the Case Filter.
Given the Case Filter, the Functional Uniqueness principle of the θ-Criterion, the TSC and the SSC, UG severely limits the syntactic behaviour of the rule Move NP. NP movement is obligatory when an NP occurs in an underlying Caseless position. It must wind up in a Case-marked position at S-structure. NP movement cannot occur between two θ-marked positions, nor can it cross a finite clause boundary or a c-commanding subject.
The framework of principles developed over the 1970s demonstrated that the behaviour of transformations which had been stipulated in the structural descriptions and structural changes of more elaborate rules could be derived from general principles of UG. As a result, it was possible to dispense with structural descriptions for transformations altogether. Under this framework, a transformation could be given as a structural change which indicated the category (or categories—e.g., in the case of adjunction) affected and the elementary operation involved. In later work, circa 1980, Chomsky generalizes transformations still further by replacing specific reference to categories (e.g., NP or wh-phrase) with a variable α and characterizing the two distinct elementary operations which can perform movements (i.e., substitution and adjunction) under the designation ‘move’—hence the rule ‘Move α’.
This change of view concerning transformations is discussed in Chomsky’s Lectures on Government and Binding (1981; henceforth LGB).
In early work in generative grammar it was assumed, as in traditional grammar, that there are rules such as ‘passive’, ‘relativization’, ‘question formation’, etc. These rules were considered to be decomposable into more fundamental elements: elementary transformations that can compound in various ways, and structural conditions (in the technical sense of transformational grammar) that are themselves formed from more elementary constituents. In subsequent work, in accordance with the sound methodological principle of reducing the range and variety of grammars to the minimum, these possibilities of compounding were gradually reduced, approaching the rule Move-α as a limit. But the idea of decomposing rules such as ‘passive’, etc., remained, though now interpreted in a rather different way. These ‘rules’ are decomposed into the more fundamental elements of the subsystems of rules and principles (1) and (2). (p. 7)
In LGB the system of rules has the following subcomponents:
(19)
and the system of principles includes the following:
(20)
Given this view, descriptive categories like ‘passive’ have no status under the theory of grammar. Virtually all the properties of the ‘passive transformation’ have been derived from general principles of UG, including the movement operation itself, which is neither a language-particular nor construction-specific rule.
