The formal analysis in SS
The SS analysis of the English verbal system begins with various assumptions about the underlying phrase structure for English sentences, embodied in the phrase structure rules of SS. Thus consider the subset in (1) that directly involves verbal elements. (The numbering follows Appendix 2 of SS.)
The lexical category V, which in the SS analysis is rewritten as a lexical verb, is dominated by two phrasal categories, Verb and VP. The rule for Aux is based on the assumption that the aspectual auxiliaries are associated underlyingly with affixes that they never take in Phonetic Form (PF). Furthermore, they occur as bare formatives without any category designation. This turns out to be a crucial aspect of the SS analysis, as shall become clear—one that has a particular relevance with respect to Lasnik’s reformulation. And most important, C, which comes to represent the finite tense and agreement affix, is introduced into the derivation as an independent element.
Given these rules, the verbal affixes en and ing, and those represented by C start out as independent from and to the left of the verbal stems they will be affixed to in PF. In the case of the aspectual affixes, the analysis assumes an underlying dependency that becomes discontinuous at PF. Note that the exact nature of this underlying dependency is not clear (see n. 3).
To account for these discontinuous dependencies as well as the underlying independent tense affix that can not remain independent in PF, SS postulates a transformational rule named Auxiliary Transformation (but commonly referred to as affix hopping).
The structural change (SC) not only inverts the order of an affix and the adjacent verbal stem to the right but also inserts a word boundary symbol # to the right of the inverted affix.
Although the formulation in SS is not explicit about the nature of the elementary operation the transformation performs, Lasnik makes the natural assumption that affix hopping involves adjunction of the affix to the verbal stem rather than a permutation of the two elements, noting that this is explicit in Chomsky 1975a. About the inexplicitness of the formulations in SS, Lasnik makes an important point. This actually indicates a virtue of the SS analysis, given that Chomsky was assuming that the analytic details not specified in the rules would be provided by general principles of syntactic theory rather than by stipulation in the formulation of the rules.
One serious problem with this formulation of affix hopping, which Lasnik points out, is that it introduces symbols, Af and υ, which are not, strictly speaking, symbols of the grammar. They are neither terminal nor nonterminal symbols introduced via phrase structure rules, nor are they variables. Rather, they are abbreviations for a Boolean combination of symbols, one of the possibilities for specifying a term in a transformation. In this case, the terms are Boolean disjunctions: {s,Ø, past, en, ing} and {M, have, be, V}.
Lasnik adapts Ross’s critique of the use of Boolean disjunction in the formulation of transformations (Ross 1969).6 He points out that under the SS theory the elements grouped in the disjunctions do not constitute a natural class even though they behave similarly. Furthermore, the evaluation metric makes no distinction between the disjunction of these elements and another disjunction of random elements. Ross’s solution was to designate modals and the aspectual auxiliaries as the same category, [+V, +Aux], thereby making use of a feature theory which wasn’t at the time available to the SS theory To reformulate affix hopping in this way, we would have to substitute [+V, −N] for υ ([−N] to exclude adjectives, and no designation for the feature [Aux] to allow either + or −). Finding a non-ad hoc feature designation for the set of verbal affixes, however, might still be a problem. I won’t pursue this issue further because in current analyses the role of affix hopping is either eliminated or limited to moving only the tense affix, as in Lasnik’s revised theory.
The affix-hopping analysis in SS is crucial to an account of the distribution of auxiliary do. In yes/no-questions, for example, a verbal element must occur in front of the subject. Because the main verb in modern English cannot be so displaced, some verbal auxiliary must occur in the clause-initial position. If the derivation contains neither a modal nor an aspectual auxiliary, then periphrastic do occurs, carrying the tense and agreement. SS accounted for this by postulating a question transformation under which the tense and agreement affix C would front just in case it was adjacent to the main verb V
The question transformation is given in SS as (3), where the ellipsis symbol stands for a variable.
Using Boolean disjunction, the SA can be abbreviated as (4), thereby eliminating eight symbols.
The formulation in (4) highlights the fact that there seem to be two distinct cases of question formation: one that affects a single constituent C (now realized as some particular tense and agreement affix) and the other that affects a nonconstituent, C and the following verbal auxiliary (which have yet to be conjoined via affix hopping).
Lasnik criticizes this second case for violating the structure dependence of transformations. He equates the statement that transformations are structure dependent with “transformations always affect constituents” (p. 76). Under this interpretation of structure dependence, affix hopping would have to apply prior to Tq so that Tq would be moving a constituent. This would, of course, create a problem for the derivation of interrogatives with periphrastic do. If the tense and agreement affix concatenates with the main verb before Tq moves it in front of the subject, then do-support can never apply— never, because in SS affix hopping is obligatory One solution, essentially the analysis of Lasnik 1981, is to make affix hopping an optional rule and deal with the unwanted consequences of this formulation with a stranded affix filter. In effect, affix hopping must apply unless do-support applies, thereby eliminating the stranded affix.
Lasnik raises other theoretical problems for the transformations formulated in SS, almost all of them involving unnecessary descriptive power in formulating rules, which creates problems for learnability For example, given the formulation in (3) a term in the SA of a transformation can be any of the following:
Lasnik points out that the two sequences (5c) and (5d) are problematic because they do not constitute single constituents. As formulated, the rule violates the tacitly assumed principle that transformations can only affect (i.e., move or delete) constituents (i.e., not a string that does not form a single constituent)9 and therefore allows for a wider class of rules than seems necessary. Moreover, just being able to specify terms of a transformation in so many different ways allows for excessive descriptive power.
Along these lines, Lasnik also criticizes the general use of Boolean and quantificational conditions in the SAs of transformations in SS. He also discusses problems for learnability posed by postulating an obligatory/optional distinction for transformations, extrinsic rule ordering, and the placement of variables in SAs. This material is quite intricate and worth careful study.
However, instead of reviewing Lasnik’s account, I would like to approach the solution to these problems from a somewhat different perspective. One reason early transformational accounts were so complicated in their formulations of rules was that the emphasis was on generating all and only the grammatical sentences of the language under analysis using only phrase structure and transformational rules. If the rules generated a deviant expression, then this was a serious problem, construed as a direct refutation of that formulation of the rules. It was not until well after the advent of general principles like the A/A Condition (Chomsky 1964) that it became possible to allow rule mechanisms to misgenerate structures and then deal with the problem in terms of general conditions on rule application and/or representations. And it was not until the development of the conditions framework began to accelerate in the early 1970s that it started to become clear that the detailed articulation of a system of general principles/conditions had a salutary effect on the formulation of rules—namely that the formulations could become optimally simple. For transformations, this culminates in the hypothesis that a grammatical transformation cannot compound elementary operations. That is, each grammatical transformation consists of a single elementary operation.
From this perspective, two questions arise: what are the elementary transformational operations? and what further specifications must be added to formulate the grammatical transformations in various languages? From the perspective that has developed over the past three decades, the answer to the second question seems clear: ideally none.
To see how this works, consider the elementary operation of adjunction, which minimally and maximally affects two elements: the element that is to be adjoined and the element to which it is adjoined. If we assume that the nature of what can be adjoined to what is fully determined by general principles (e.g., that only single constituents are affected), and further that the syntactic distance between the two (in the case of movement by adjunction) is also determined by general principles, then there is no need to stipulate the behavior of the rule in terms of a complicated SA and/or SC that contains more constant terms than minimally required by the elementary operation itself. Furthermore, there is no opportunity to utilize Boolean or quantificational conditions to specify SAs of transformations. Given this, the placement problem for variables (see n. 10) disappears. If the elementary operation involves two constants, then there are going to be exactly three variables—two cosmetic end variables and the one intrinsic variable that separates them. Furthermore, the optional/obligatory distinction becomes superfluous. If the operation is “obligatory” and it fails to apply in the derivation, then the result will be ruled out by some general principle on representations. Thus, for example, the obligatory character of do-support in SS can be handled by Lasnik’s Stranded Affix Filter, a UG condition on representations. And finally, given the simplicity of elementary operations, it becomes virtually impossible to specify intrinsic or extrinsic ordering relations. Here again, if operations apply in an order that yields a deviant result, general conditions—virtually always on representations—will rule out the derivation.
Although this perspective on transformations rules out the specific formulations of transformations in SS, it is not obviously incompatible with a rule of affix hopping or do support. And of course Tq remains in essence, though now generalized as head movement (T-to-C and V-to-T). Tnot, another transformation intimately connected to the verbal morphology system, has been abandoned. As Lasnik points out, the SS system does not generate simple negative sentences with not. Tnot only inserts the negative affix n’t into the verbal string. The separation of C from the main V by n’t (or by the subject NP (with Tq) or by the emphasis marker A [with TA]) blocks affix hopping of C onto V and triggers do-Support. This account of the distribution of periphrastic do remains the major insight of the SS transformational analysis. It is preserved in Lasnik’s revision, to which I now turn.
