LECTURETTE #13: DS AND THE PROJECTION PRINCIPLE In lecturette #12 i introduced the Theta-Criterion, which requires every NP in a clause to bear a unique theta-role and every theta-role assigned by any of the heads in a clause to be borne by a unique NP. The Theta-Crite- rion is crucially to be satisfied at DS. Indeed, that is in a sense the definition of DS: it is the formal representation of the clause's theta- structure, the heads and the structural representation of the theta-roles they assign. So at DS there are Bar-0 heads and positions to which they assign theta-roles. These latter can be called 'theta-positions', more commonly 'A-positions'. ('A' stands for 'argument' in this case; as far as i know, the two terms are at least approximately synonymous; i prefer to speak of 'theta-positions', in order to avoid confusion with 'A' as the abbreviation for 'adjective'.) Positions to which theta-roles are not assigned are referred to as 'theta-bar' or 'A-bar' positions. (This usage of 'bar' has nothing to do with X-Bar Theory. It's derived from mathematical set theory, where the *complement* of a set S (i.e., the set of all members of the domain of S that are not members of S) is designa- ted S-bar. Thus, if we define the set of all integers as the domain and define the set of even integers as E, then the set of all odd integers could be designated E-bar.) If DS is the represenation of the clause's theta-structure, then by definition there aren't any theta-bar comple- ment positions in it (theta-bar specifiers are a different story; many specifier positions are theta-bar, but are nevertheless required by the fundamental definitions of X-Bar Theory. More on this below.) In some variants of the framework, it is hypothesized that ordering con- straints can be imposed on DS by restricting the *direction* in which theta-roles can be assigned. For instance, if the grammar insists that heads can only assign theta-roles to the right then all subcategorized complements will automatically be base-generated to the right of their governing heads. As will be seen in subsequent discussion, particularly when we get to Case Theory, this base-order can be altered at SS by cri- teria independent of Theta-Theory. And not all versions of REST sub- scribe to the assumption of an 'ordered base', as it is sometimes put. It is possible to view DS as inherently unordered -- in the terms intro- duced in Lecturette #2, under this hypothesis DS would be characterized by dominance but not precedence relations. (Of course, under Kayne's Antisymmetry hypothesis this option of an 'unordered base' is not available.) However, as will be seen shortly, the possibility of reordering at SS is not unconstrained, as should be evident if we take the Structure-Preser- vation Constraint (SPC) seriously. So, by definition, DS is a direct representation of a clause's theta- structure. But as Chomsky points out in his 1980 Pisa Lectures, if LF is to be regarded as a semantically-interpretable level of structure then the clause's theta-structure must also be represented there; how else can we recognize, e.g., the direct object of a transitive verb as satisfying that verb's subcategorization requirement? And if the object is moved, as in (1), how can this detail of the clause's interpretation be achieved unless the theta-structure is somehow preserved at LF? (1) a. How many hot dogs did you eat at the stadium? b. Which book did you give Claude? c. How much did you buy that table for? As you will remember from Lecturette #6, part of the answer to this ques- tion lies in trace theory: the moved NPs in (1) leave behind traces in their base positions, and the coindexing between overt fronted NP and trace in base position is supposed to guarantee the interpretation of one in the position of the other. But this only works if the SPC is respec- ted: if the trace is in the NP's base position not only in terms of line- ar order but in terms of hierarchical structure as well, i.e. if it's in the NP's theta-position. Which means the clause's full theta-structure must be represented at LF as well as at DS. But if LF is also supposed to be (in part) a representation of the clause's theta-structure then SS must be too, given that REST's T-model presumes that nothing can be true of both DS and LF without being also true of SS. And this, as is clear from Chomsky's exposition in the Pisa Lectures, is the logical foundation of what is usually called the Projec- tion Principle: the clause's theta-structure, directly represented at DS, must be *projected* to SS and LF as well. (This doesn't apply to PF, because the derivation of PF from SS crucially involves the elimination of any structure that is not phonologically relevant -- including traces. Remember from Lecturette #6 that it has been plausibly argued that the SPC in its strictest form doesn't apply at PF -- i.e, PF-movement isn't structure-preserving.) Therefore, whatever transformations derive SS from DS and LF from SS must preserve the clause's theta-structure. This means, among other things, that nothing can be moved *into* a theta-po- sition. Movement can be to (1) a head-position (an option reserved for constituents that are themselves lexical heads, and which is sometimes called 'incorporation'), (2) a theta-bar Spec position, or (3) an adjunc- tion-site. (Remember that, according to the SPC, X-max constituents can only adjoin to maximal projections. The adjunction of X0 constituents to X0 sites is formally indistinguishable from incorporation.) This restriction of movement to theta-bar positions is a large part of the reason Kayne rejects the possibility of rightward movement. If, as Kayne maintains, at DS all complements are to the right of their gover- ning heads and all specifiers to the left, and complements can only move to theta-bar positions (typically specifiers), then they can only move to the left. One of our participants has perspicuously asked: What about adjunction? An adjunction site is always, by definition, theta-bar, be- cause adjunction-structures can't be base-generated. What does it matter if the adjunction is to the left or to the right? The answer, to the best of my knowledge, lies in something developed early in the 80's called the 'Extended Projection Principle'. Part of this 'extension' is the claim that SS, at least in its basic structural organization, should bear as close a resemblance as possible to DS. Which boils down to the- ta-positions in the places where theta-positions typically belong and theta-bar positions in the places where they typically belong. So if, as Kayne claims, at DS all theta-positions are to the right and all theta- bar positions to the left, then the 'Extended Projection Principle' would presumably if not require at least strongly favour all theta-bar posi- tions to be to the left also at SS. In which case movement can only be to the left. A more substantive part of the Extended Projection Principle has to do with specifiers, and essentially formalizes the implicit assumption in X-Bar Theory, alluded to in Lecturette #9, that 'functional heads' are entitled to the same projection-structure as lexical heads, including specifiers as well as complements. Within projections of lexical heads, the specifier position is at least potentially a theta-position; this is most obviously true of VPs, since, if we accept the Internal-Subject Hypothesis, the subject's theta-role is assigned to the spec-VP posi- tion. Likewise in NPs, in order to satisfy the Theta-Criterion it is necessary for genitive NP modifiers of NPs as in (2) to get theta-roles, which are presumably assigned to them by virtue of the structure they in- habit. So these specifier positions, insofar as they are theta-posi- tions, are licensed at DS. (2) a. [the man's] dog b. [the city's] destruction c. [the child's] birth d. [the letter's] receipt The Extended Projection Principle extends this licensing also to speci- fier positions of functional heads, which being non-lexical have no power to assign theta-roles. As a result, specifier positions can also be li- censed for functional heads such as Infl or Agr, i.e., the surface posi- tion of the subject is licensed at DS. This approach is intended to assimilate into the theory sentences like (3), which on the face of it would appear to violate the Theta-Criterion. As Nicolas Ruwet has made clear in his paper 'On Weather Expressions' (cf. his collection Syntax and Human Experience, University of Chicago Press, 1991, pp. 82-142), a verb like 'rain' has absolutely no theta- roles to assign; in which case it can't license even a subject position. The 'dummy' NP 'it' bears no theta-role; it and its structural position are licensed not by the Theta-Criterion but by the Extended Projection Principle. (3) a. It's raining. (English) b. Il pleut. (French) In this regard the Extended Projection Principle is more obvious in some languages than in others, and indeed in 1980 Chomsky allowed that its manifestations might be parameterized. Obviously, unlike the English and French sentences in (3) the Italian and Icelandic equivalents in (4) have no overt subjects, and indeed to the best of my knowledge typically do not allow overt subjects, a fact that could be attributed simply and straightforwardly to the fact that there is no theta-role for such an NP to bear (in distinction to other verbs in these 'pro-drop' languages, as they are called, which can take *optional* overt subjects). Presumably, the Extended Projection Principle does not apply with the same force in these languages as it does in English and French, to force a subject NP even when there is no theta-role for it to bear. (4) a. Piove. (Italian) b. Ragnar. (Icelandic) 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! ***