170 L. MICHELLE CUTRER
icate takes subject control or nonsubject control will ultimately turn out to
follow directly from the meaning of the predicate concerned.
However, semantic approaches such as those of Jackendoff (1972),
Rûzicka (1983), and Nishigauchi (1984) are not free from criticism. The
crucial deficiency of syntactic accounts is that syntactic predictions of con
trol always require lexically marked exceptions. However, the same criti
cism can be raised for certain semantic accounts. Rûzicka (1983) treats the
property of subject or object control as a semantic feature. Although the
control feature is semantically based, the assignment of such a feature
involves lexically marked exceptions. The prediction of this semantic con
trol feature is not based on any more general theory or principles, but
rather is arbitrary and ad-hoc. Jackendoff (1972) appeals to verb classes in
his analysis of control; however, the verbs are grouped arbitrarily according
to the control properties they exhibit.
The approaches of Jackendoff (1972) and Nishigauchi (1984), which
base the prediction of controller on thematic relations, are subject to the
criticism that the thematic roles are ill-defined. There is no small amount of
disagreement among different semantic analyses as to what exactly counts
as a certain thematic role. In such analyses, one cannot always distinguish
whether it is the thematic role which predicts the controller or the control
relation which predicts the thematic role. The thematic roles which these
accounts appeal to need to be more clearly defined. Furthermore, it is the
verb which assigns these thematic roles. Hence, an analysis should appeal
to the semantics of the verb where possible.
Nishigauchi (1984) includes meaning in his analysis of control and in
certain respects his analysis is similar to portions of the analysis presented
in this paper. However, Nishigauchi deals only with a limited subset of con
trol cases and his apparatus does not easily extend to examples of control
phenomena outside of his limited subset. His subset does not include most
of the examples of control phenomena which a control theory is typically
asked to account for. Moreover, he has not accounted for cases of object
control and doing so would require not only the construction of a different
apparatus for prediction of control relations in the case of object gaps, but
also a revision of the GB account of the distribution of controlled elements.