Control without PRO

In this thesis I develop a syntactic representation of control that is PRO-free. I implement a mechanism of theta-role assignment that uses theta-role percolation (Neeleman and Van de Koot (2002)), which enables the apparent subject properties of controlled infinitives to be reinterpreted as properties associated with an external theta-role, rather than a subject. I first analyse Obligatory Control using a decompositional analysis of theta-roles (independently motivated in Samek-Lodovici 2003), according to which theta-roles are divided into two selectional requirements, one that formally licenses arguments and another that regulates interpretation. It is the latter one that regulates Obligatory Control, by being copied to the controller in the super-ordinate clause. The resulting theory makes the same predictions as one based on PRO, yet avoids dependence on this ill-defined element inherited from the GB era. I then offer an account of the case-agreement properties of predicates in Icelandic infinitival clauses; this data shows not only that PRO is unnecessary but that it cannot actually fulfill the task for which it was proposed. An extra-syntactic analysis is developed for Non-Obligatory Control, since this relation is not syntactically constrained in the way that Obligatory Control is. I claim that in non-obligatory control structures the external theta-role is not saturated syntactically. This unassigned role is subject to an LF-interpretative rule that attributes it with a [+human] specification. This rule suffices for cases in which the interpreted subject is understood generically, but in long-distance control structures, this [+human] specification is further supplemented by a discourse rule that links the theta-role to a highly accessible antecedent, as developed in Ariel (1996). An attempt is made to generalise the rules operative in non-obligatory control structures, to the null-objects