LECTURETTE #6 THE STRUCTURE-PRESERVATION CONSTRAINT, TRACES, AND DERIVATIONAL HISTORY One important development in the early '70's was 'trace theory'. This is the hypothesis that, whenever anything moves during the course of a deri- vation, it leaves a 'trace' behind. This trace (typically represented by a lower-case 't' for 'trace' or 'e' for 'empty', though in my own work i tend to favour a schwa) consists of an abstract copy of the moved consti- tuent, resembling it in all respects but lacking any phonetic realization. We will have a lot to say about traces later, especially when we focus on Binding Theory; at this point it is important to understand that a trace is identified through *co-indexation* with the constituent whose movement created it; this co-indexation is typically represented by a small sub- script or, here, by a parenthesized index-letter. (The letter most com- monly used is 'i', for 'index'; if more then one is needed the letters following 'i' in the alphabet are drafted.) I mention trace theory now in order to lead into the discussion of a ra- ther complex constraint on Move Alpha which i, at least, regard as pretty basic, in the implementation of which traces are significant. I'm refer- ring to what is commonly called the 'Structure-Preservation Constraint' or SPC. Joseph Emonds' 1970 MIT dissertation Root and Structure-Preserving Trans- formations was published in 1976 by Academic Press under the title A Transformational Approach to English Syntax: Root, Structure-Preserving, and Local Transformations. In this work, Emonds argued for a typology of transformations based on their domains or fields of operation and on a powerful constraint based on this typology. This typology divides all transformations into the three classes listed in Emonds' title: (I) Root Transformations. To oversimplify a little, these are trans- formations applying only to 'root' clauses, i.e., clauses that are not dominated by anything (other than S) -- roughly, the traditional notion of an 'independent' clause. (II) Structure-Preserving Transformations. Transformations that move constituents from one part of the constituent structure to ano- ther but do not create or destroy or rearrange structure. A transformation that leaves the hierarchical organization of con- stituent structure exactly as it found it is 'structure-preserving'. (III) Local Transformation. A local transformation affects precisely two constituents that are adjacent to each other and is not subject to any conditions outside of those two constituents. The affected con- stituents need not be sisters in constituent structure, merely adja- cent in linear order; several other constraints apply in the defini- tion of a local transformation (e.g., while the two affected consti- tuents need not be sisters there must be a c-command relationship between them, and at least one of them must not be a maximal projec- tion), but we needn't go into those details now. In its full form the SPC says that any movement-transformation, any in- stantiation of Move Alpha, must fall into one of these three classes. What this boils down to is that, unless a movement-transformation can be legitimately defined as belonging to either of two narrowly restricted classes, Root Transformations and Local Transformations, it may not alter the basic constituent structure in any way. (Back in Lecturette #3 i made an allusion to the possibility of 'node- deletion' as an example of 'Affect Alpha' as opposed to 'Move Alpha'. Node-deletion would be an example of a Local Transformation. Cf. below, discussion of Local Transformations and PF-Movement.) To give you an example of what this means in terms of practical theory, let's take a look at the history of the Standard Theory's Passive Trans- formation. In early Transformational Grammar (i.e., ca. 1957), a passive clause like (1) was assumed to be derived from the same deep structure as its active counterpart (2), a deep structure corresponding to the surface form of (2). The passive (1) was derived from it by an optional trans- formation that interchanged the subject and object NPs, the former sub- ject surfacing as complement of a PP, and provided passive morphology and auxiliary for the verb. (1) The door was opened by Hilary. (2) Hilary opened the door. In Aspects (1965), the main difference was the claim that the PP already exists at deep structure -- but without complement; the deep structure of (1) would be something along the lines of (3). (3) Hilary opened the door [by {}] (In the Aspects version, the complementless 'by'-phrase is introduced by the same procedure that introduces 'manner' adverbials, an analysis that predicts that manner adverbials and passive agent phrases should be in complementary distribution. When i was taking the introductory course in syntax in 1983, we were given as a homework assignment the task of refu- ting this analysis.) In the Aspects analysis, the PP in the passive clause (1) is already pre- sent at deep structure, not created by the Passive Transformation. In this respect, the Aspects version of the Passive Transformation is 'structure-preserving'. But in conjunction with trace theory it is still incompatible with the SPC. (Note that, since any clause, including em- bedded clauses, may be passive, the Passive Transformation cannot be classified as a Root Transformation. Since it involves two full NPs that are not adjacent to each other, it can't be a Local Transformation. Therefore according to the SPC it must be a Structure-Preserving Trans- formation.) Let's assume that the subject NP 'Hilary' has moved out of subject posi- tion into the PP in (3), giving a structure such as (4). This leaves a trace in subject position, coindexed with the former subject NP 'Hilary'. The SPC, in conjunction with trace theory, requires that this trace sur- vive throughout the derivation, so it is impossible for the direct object NP 'the door' to be moved into this position. In order for it to be li- censed as a 'new' subject, alterations would have to be made in the con- stituent structure, which would violate the SPC. (4) t(i) opened the door [by Hilary(i)] The upshot of this is that in REST the DS for a passive clause like (1) is necessarily something like (5) (i'm leaving out for the moment the question of the passive auxiliary). The Agent 'Hilary' is base-generated as the complement of a 'by'-phrase and there is an empty subject. (Typi- cally, a base-generated empty category is represented by a capital letter delta, which of course i can't represent here. I'm using a pair of curly brackets instead.) The direct object 'door' can move into this slot without creating any new structure. In REST, as we shall see in the near future, this analysis is additionally justified by concerns of Theta- and Case-Theory. (5) {} opened the door [by Hilary] Before wrapping up this lecturette i should mention the one flagrant ex- ception to the definition of a 'Structure-Preserving' Transformation: Adjunction. What is called 'Adjunction' in REST is equivalent to what in the '60's was often called 'Chomsky-Adjunction', though Chomsky himself disliked the label and avoided it as much as possible. In REST there isn't any other kind of adjunction, so we don't need the qualifier. 'Adjoin C to X' means to move C into a position adjacent to X and then create a new node X immediately dominating both X and C, such that C and X become sisters, as in (6). (6) X X / \ / \ -------------> C(i) X /_______\ adjunction / \ C /_______\ t(i) On the face of it, this transformation seems to be creating new structure -- it creates a new node X -- and therefore cannot be called 'structure- preserving'. However, the framework stipulates a loophole here; since the 'new' node X immediately dominates the 'old' X, for the purposes of the definition of a Structure-Preserving Transformation they are viewed as collapsible into each other. This understanding is made explicit in some of the work written and published in the mid-'80's, e.g., Chomsky's Barriers, in which for instance it is asserted that for certain purposes a constituent X cannot be said to dominate a constituent C unless 'every segment', i.e. every instantiation of X in (6), dominates C (which in fact in (6) it doesn't). (This understanding is also consistent with Emonds' definition of a 'root' clause. Strictly speaking, according to that definition in a structure like that in (7) every S qualifies as a root clause because it isn't dominated by anything *other than another S node*. In other words, for certain purposes any node N immediately dominated by another node of exactly the same category type is collapsible into it. I have no inten- tion at this time of addressing the question of whether this is a copout, or how consistently the methodological principle is observed in REST.) (7) S / \ S / \ S / \ S I should say a little something about 'Local Transformations'. It has been suggested by some (including myself, in my dissertation) that Local Transformations ought to be theoretically redefined as relevant to PF but not to syntax per se, meaning that they cannot take place between DS and SS but only between SS and PF. This notion is due partly to a general theoretical desirability of restraining the power of Move Alpha as much as possible and partly to the fact that many of the 'stylistic rules' discussed by Michael Rochemont in his 1978 Univ. of Massachusetts disser- tation A Theory of Stylistic Rules in English (which in REST terms is ba- sically an extended argument for a PF level and a set of movement-trans- formations deriving it from SS) are exactly equivalent to the examples Emonds gives of 'Local Transformations' (e.g., Subject-Aux Inversion). Whether this restriction on Local Transformations is tenable or not is an empirical question; in fact, in my dissertation research i found that it was impossible to describe 'scattering' (i.e., gratuitous discontinuity) in Sanskrit as a Local Transformation (specifically involving node-dele- tion) without doing serious violence to the relevant definitions. To sum up, it is important to note that trace theory and the SPC together provide at least a partial guarantee that the 'history' of a derivation is always recoverable. As we shall see in subsequent discussion, there are complications in the framework that can obscure derivational history, but they are relatively minor. The general consequence is that even at LF the base form (DS) of a string can be reconstructed. This is of some importance to the operation of the framework, besides being attractive, at least to many theorists. Best, Steven --------------------- Dr. Steven Schaufele 712 West Washington Urbana, IL 61801 217-344-8240 fcosws@prairienet.org **** O syntagmata linguarum liberemini humanarum! *** *** Nihil vestris privari nisi obicibus potestis! ***