A good deal of attention has been given to the question whether quantification into referentially opaque contexts is possible or not. No one to my knowledge, except Jaakko Hintikka, has ever suggested that quantification into referentially opaque contexts is sometimes necessary. The suggestion occurs in 6.7 of his book, Knowledge and Belief. 1 My aim in this section is to ascertain, on the basis of that discussion, the conditions under which this necessity is thought to exist.
From all indications, the author takes the following conditions to be sufficient: the contexts are epistemic (the contexts are governed by a phrase like 'knows that'); and (trivially) we quantify into these contexts (a quantifier outside the context binds a variable inside the context). The line of reasoning is apparently this: It is necessary that, if we quantify into epistemic contexts, then these contexts are construed as referentially opaque, because the alternative -construing these contexts as referentially transparent -always allows the inference from (1) Tom knows that the dictator of Portugal = the dictator of Portugal and
(2) to (3) The dictator of Portugal = Salazar Tom knows that the dictator of Portugal = Salazar, and from (3) to (4) (Ex) (Tom knows that the dictator of Portugal = x); but this inference is invalid, from an intuitive point of view. "It follows that no one can help knowing who Portugal's dictator is as soon as one