I point out that Rhetorical Structure Theory, a widely known formalism for representing discourse relations, lacks a native semantics and that recent attempts to provide a semantics for it tend to construct it as an addition to the semantics for other aspects of language. To the contrary, I propose that proper semantic interpretation for rhetorical relations should at least in significant part be sought in the realm of various update and test operations on Context Sets, i.e. in the various aspects of information flow in communication. Several more concrete preliminary suggestions are outlined, and known linguistic evidence is adduced to corroborate the feasibility of the proposal.