Since the mid-seventies, supervaluationism has been supported because of its capacity to give a conservative semantics for vague predicates and to avoid the so-called sorites paradox. As many opponents point out, however, this framework has something elusive and leaves several things unexplained. Dynamic semantics, which I present here, is introduced to interpret the elusiveness positively from an information-theoretic point of view. Thus the aim of the paper is to present a dynamic semantics for vague predicates as a revised version of supervaluationism.