The paper outlines a picture of belief reports as speech acts that comprise or put on stage an embedded speech act whose virtual enunciator is distinct from the main enunciator of the belief report. The role assigned to context is to provide the background against which the putting on stage of a certain embedded speech act is, or fails to be, an admissible move. As to the problem raised by the changes in evaluation due to substitutions among co-referring expressions in the that-clauses ("Lois believes that Superman flies" vs "Lois believes that Clark Kent flies"), the paper attempts to outline the conditions under which such substitutions preserve or threaten the correctness of a belief report.