ABSTRACT
LetusconsiderVisser'sgeneralizationfirst.Take asubjectcontrolverbsuchaspromise.Therelevant instanceof(84)forpromise,irrelevantdetailsaside,is thefollowing:
(85) h(promise')(x)(~P)(y)~ MaP( So(<"h(promise' ),x;P,y>)
some x and some P, the following must be the case:
(86) PAS' (promise') (x) (·"P) Recall that, by definition, (84) is a schema for a finite set of meaning postulates of the form:
(87) Ma P(So(< PAS'(promise'), x,~P>) But (87) will be undefined, for neither x nor P is the source-argument of PAS'(promise). PAS' drops the arguments of verbs that correspond to the subject position, which in the case of promise happens to be the source
forpromise',wederiveacontradiction,namelythat
promise'doesanddoesnotobey(84).HencePAS'mustbe undefinedforpromise':promisecannotpassivize.The
reasonwhythisissoisthattherealwayshastobea
thatthepresenceofaby-PPwouldn'tmakeanydifference
Wehavejustarguedthatitfollowsfromthefactthat promiseobeys(84)thatPAS'(promise')mustbeundefined. Evidently,afunctorthattakesanundefinedfunctionas
argumentwillyieldanundefinedvalue.Externalargu-
mentsareVPmodifiers(i.e.functors).Hence,anyby-PP appliedtoPAS'(promise')willalsobeundefined.Thus thepresenceofaby-phrasedoesn'tmakeanydifferencein
Considernowthefollowing:
(88)a. b.
