1 Specifiers and projection
with Margaret Speas
The overlap between the constraints imposed at the level of D-structure by the Projection Principle and those imposed by X′ theory has been noticed by various linguists, including Hale (1978, 1980b), Chomsky (1981a), Stowell (1981b), Marantz (1984), and Emonds (1985). As long as no empirical generalization is lost, we may wish to attribute this overlap to inherent redundancy in cognitive mechanisms. However, if there are linguistic generalizations which are obscured by this redundancy, then it is appropriate to investigate the properties of the two systems of constraints, in order to isolate the unique properties of each, so that the observed generalizations will be expressible.
In this paper, we will argue that there are certain important generalizations which cannot be expressed in terms of X′ theory and the Projection Principle as they are currently conceived. We will propose a theory of well-formed D-structures which captures these generalizations. Our theory is based on what we take to be a fundamental asymmetry between lexical categories and functional (non-lexical) categories: functional categories project to X″, and are limited to a single specifier position and a single complement position, while all projections of lexical categories are X′, which is indefinitely iterable (in the sense of Harris 1951), limited only by the Projection Principle and other independent principles of licensing.
After having introduced this “non-uniform” bar level hypothesis in section 1, we will argue in the following sections that various desirable consequences follow from this hypothesis: (1) The so-called “implicit argument” (Roeper 1983), whose syntactic characterization has been the subject of much recent discussion, can now be explicitly represented in the phrase structure under this hypothesis; (2) We can capture certain typological variations which are not expressible in other theories; (3) We can predict which positions will be iterable and which will not; (4) We can simplify the principles of θ-marking (especially θ-marking for “external arguments”); (5) We can eliminate the need for “VP-adjunction” in syntax, and can also simplify the definition of “barrier”; and finally (6) We can eliminate the undesirable overlap between constraints on possible D-structures imposed by the Projection Principle and those provided by X′ theory.
1 Projection of lexical and functional categories
Following Chomsky (1970), we will assume that the primitive vocabulary of the grammar includes the category features [±N] and [±V], and that these features allow a partition of lexical items into four categories. It is not clear to what extent the above features may be labels for some semantic or other property of the categories, but there is an important distinction between categories which bear these features and those which do not: the categories bearing these features are those which may take arguments. In the theory of Higginbotham (1985), these and only these are the categories which have a θ-grid as part of the lexical entry. Following the longstanding tradition, we will call these four categories the lexical categories.
Lexical Categories: | [+N −V] | (noun) |
| [−N +V] | (verb) |
| [+N +V] | (adjective) |
| [−N −V] | (preposition) |
In the framework of Chomsky (1986a, b), the relationship between the lexicon and the syntactic level of D-structure is one of projection; properties of lexical items, including θ-marking properties, are projected from the lexicon into syntax, constrained by the Projection Principle and the schematic “X′” well-formedness conditions on phrase markers.
The Projection Principle (informal statement): lexical properties are maintained at all syntactic levels.
The X′ Schema:
(i) X′= X X″* (order irrelevant)
(ii) X″ = X″* X′
where X″* stands for zero or more occurrences of some maximal projection.
(Chomsky 1986a: 2–3)
In English at least, the lexical categories do not exhaustively partition the set of items in the lexicon. In particular, the items such as Comp and Infl, which have been called non-lexical or minor categories, act as heads but do not appear to bear the N and V features.1 Since these categories are projected from the lexicon and have independent lexical entries, we will avoid the term non-lexical, and will refer to these categories as functional categories.
It has long been observed that the cross-category generalizations captured by the X′ schema were fuzzy with respect to the functional categories; even Jackendoff (1977) resorted to some extra features to get the generalization to work out right. (Specifically, to the features [±subject], [±object], he added [±comp] and [±det].) Until Chomsky (1986a), it was thought that the categories IP and CP were defective in some way; Chomsky suggests extending the X′ schema so that CP and IP would both have specifier positions.
We would like to propose a different view, in which functional categories have a unique specifier position, but lexical categories may iterate “specifiers”, as long as all “specifiers” are fully licensed and can be interpreted at LF. We maintain that only the specifiers of functional categories close off projections, therefore the node dominating the maximal projection of a functional category should be X″ (or XP), while all projections of a lexical category are X′, since there is no inherent limit to their iteration.
Before we proceed, let us be clear about exactly what we mean by “specifier”. Chomsky emphasizes that the notion “specifier” is strictly a relational one, used as a label for whichever maximal projections happen to appear in a given category as immediate daughters of X″. However, this version of the X′ schema does not give us an explanation for the contrast between (1) and (2).
(1) | a. | the very very old man |
| b. | Mary's big red book |
| c. | Susan never could have been eating cabbage. |
(2) | a. | *the the old man |
| b | *yesterday's Chomsky's book. |
| c. | *It Mary ate a bagel. |
| d. | *the John's cat |
| e. | *What who did buy? |
These data show that there are some types of “specifiers” which may iterate and others which may not. Of course, it is not a priori necessary that the ungrammatical cases be ruled out by X′ theory alone. For example, some of the ungrammatical examples might be ruled out by other principles, such as the θ-Criterion. However, these data are important because it is routinely assumed in current theory that cases like (2) are ruled out by the supposed fact that there is only one available specifier position, yet X′ theory as it is formulated in the most recent treatments provides no such restriction. Chomsky's formulation of X′ theory allows any number of specifiers for each category.
It should also be pointed out that the presence of apparent subjects across categories (cf. Stowell 1982) does not provide evidence that each category has some unique subject position given by X′ theory, since extraction data reveal an underlying difference in the status of the “subject” from category to category, as shown by the examples below:
(3) | a. | We saw Bill's book. |
| b. | We saw Bill drunk. |
| c. | *Whose did you see book? |
| d. | Who did you see drunk? |
| e. | Whose book did you see? |
| f. | *Who drunk did you see? |
The subject of the adjective can be extracted (d), while the subject of the noun cannot (c). The noun plus its subject can move as a constituent (e), while the adjective plus its subject cannot (f ). These examples indicate that the status of the “subject” of the adjective drunk in (b), (d) and (f) differs in some fundamental way from the status of the “subject” of the noun book in (a), (c) and (e).
In order to avoid terminological confus...