SEMANTIC AND SYNTACTIC FACTORS IN CONTROL 191where arbitrary control predictions can be overridden by speech act or
pragmatic considerations.
3.4 Passivization: Visser's Generalization
An interesting consequence of the analysis presented here is that it
accounts for the passivization facts of subject control complements. Visser's
Generalization, the observation that matrix clauses with subject controlled
complements cannot be passivized, is frequently discussed in the context of
the control problem.
(53) a. John was persuaded to leave early.
b. John was promised to leave early.
In sentence (53a), persuade is causative and undergoer control is predicted.
John is still the undergoer, despite appearing in subject position. The con
trol facts remain the same, whether or not the sentence is passivized. We
may account for this difference by the fact that promise is a commissive and
actor control is predicted; however, the actor is no longer present. Thus,
the sentence is ungrammatical. However, this does not provide a full
account of passivization with subject controlled complements, as certain
subject control matrix clauses may passivize. For example:
(54) a. John bought the book to give to the children.
b. John built a shed to store the tools in.
Here the infinitival subject gap is controlled by the actor John, yet these
sentences may passivize.
(55) a. The book was bought to give to the children.
b. A shed was built to store the tools in.
The difference in passivization facts, between actor controlled verbs such as
promise and purpose constructions such as (54) or (55), indicates that there
are syntactic differences between the two types of subject gaps. Indeed, the
gap with promise type constructions is an obligatory one, while the subject
gap in constructions such as (54) is not. Only actor control clauses with non-
obligatory control relations, such as purpose constructions, may passivize.
(56) a. John was promised to leave early.
b. *It was tried to go.