PRO and subject of NP

Conclusion

In the preceding sections, we have examined a number of roles that PRO in the specifier of an NP might play in relating the interior of the NP to its exterior, with respect to the principles of the binding theory, bound pronouns, and control. In each instance, positing PRO in NP helped with only an arbitrary subset of the cases. The implicit arguments view outlined here encompasses the full range of cases.

If these conclusions are correct, then PRO in NP is unnecessary. However, these conclusions in no way undermine the case for PRO in S; if anything they strengthen it.

There are some ways in which the association of implicit arguments, mentioned in (9) and illustrated in subsequent examples, is similar to control of PRO. For example, the antecedent for both association and control is often determined by the verb. But association and control are different in one striking and important way: association assigns antecedents to argument slots of a specified kind (Actor, Patient, etc.) no matter how or if the argument is realized in a syntactic position, whereas control assigns antecedents to syntactic positions (the subject of infinitives, gerunds, etc...) no matter what kind of arguments are realized there.

A further difference between association and control follows from the one just mentioned. Since control is a relation between an antecedent and a specified position (the subject position of infinitives, gerunds, etc.), there can be only one instance of control per embedded clause, namely control of that specified position. But since association is a relation between an antecedent and an argument of an embedded complement, there may be as many associations per complement as there are arguments. And in fact, the following exhibits double associations:

(43) John gave Mary a kick.

John is associated with the Actor role of kick, and Mary is associated with the Patient role of kick. There is no analogue of this double association with control structures. Furthermore, the existence of double association cases is further evidence that PRO in NP can give only a partial account of the facts.

Implicit arguments are not the mysterious shadowy presences they are sometimes made out to be. They are really nothing more than the argument slots in the argument structure of predicates, and they must be visible to syntax for the purpose of theta-role assignment in any case, since theta role assignment is nothing more than the linking of an argument slot in the argument structure with a syntactic position. A ‘weak’ theta criterion, which does not require that every argument slot be so linked, is all that is needed to give implicit arguments, since these are nothing more than unlinked argument slots.

Because the theta roles must be available for syntax anyway for purposes of theta role assignment, no substantial part of the lexicalist hypothesis is given up by the implicit arguments theory outlined in this paper. The argument structure of the lexical item is a complex property of a lexical item that sentence grammar rules have access to. We have given a narrow characterization of the rules that have access to the argument structure: the rules whose domain of application is the first projection of the predicate (see section 7); this includes theta role assignment, and applications of the binding theory that have implicit arguments as antecedents.

The visibility of argument slots for the binding theory raises the possibility that the binding theory is strictly a relation among argument slots, and not among syntactic positions, even when the argument slots are linked to syntactic positions. This cannot be right, though, for a number of reasons. First, control of PRO in S is strictly a matter of position, not type of argument, as just mentioned. Second, when an argument is linked to a reflexive, or a pronoun, it behaves differently with respect to the binding theory than when it is filled with something else; a binding theory that took account only of argument-slot-to-argument-slot relations would not take account ot this. Also, the binding theory in general is not restricted to argument-to-argument relations. A syntactically realized NP can serve as an antecedent for adjuncts as well as co-arguments.

(44) John left in order to win.

So it appears that the binding theory (including control and the rule for bound pronouns) holds equally for implicit and explicit arguments, but with the visibility restrictions on implicit antecedents that follow from the fact that the argument structure does not X-bar project when the arguments are implicit.