Why is the syntax of matrix and subordinate clauses different? The view I will present is that while the constraints which interact to determine the grammatical structure of optimal clauses are the same for both types of clause, the constraints are relativized to the matrix and subordinate domains. (This contrasts sharply with for example, Emonds' 1975 conclusion A Transformational Approach to English Syntax that matrix clauses are not subject to the same types of transformations as subordinate clauses.) The ranking of the constraints relativized in this way is the source of the selection of matrix and subordinate optima, and results in limited variation between the two.

The constraints (most important among them HeadLeft, ObligatoryHead, ObligatorySpecifier), have a general form, a form which assesses matrix projections and a form which assesses subordinate projections. (Cf "positional markedness" proposals in the phonological literature.) These lie at the heart of the system which determines the syntax of matrix and subordinate clauses, and allow us to derive their principle characteristics: presence or absence of a complementizer, presence or absence of inversion, and admissibility of adjunction.

The fundamental commitment of the theory is to the view that complexity in observed grammatical behavior must result from the interaction of simple, general constraints. The interaction explains variation in clause types, within and between languages, without positing complex stipulative constraints such as the "Projection Principle" (Rizzi and Roberts 1989, McCloskey 1992) or "Pure Extended Projection" Grimshaw 1997).

Complements to "bridge verbs" show mixed properties: I will argue that their syntax can be explained if they are subject to both the matrix-domain constraints and the subordinate-domain constraints. The syntax of such complements shows some, but not (necessarily) all, of the characteristics of matrix clauses. English dialectal variation wrt inversion in interrogative complements (McCloskey 1992) follows under this proposal from an alternative ranking of two of the constraints.

In sum, the syntax of matrix, subordinate and root clauses is systematically related, and the system is due to the interaction of general constraints on the shape of projections.