With language we can say much more than we can actually understand. It follows that the algorithm for creating representations is not the process of understanding, since it does not have the same limitations. On the other hand, people will not be content with a representation of the message: they want to understand it. In this talk, I propose a particular theory of understanding called "gnosis". The theory is based on a similar tension between the meaning of a logical formula and its proof. The theory has various consequences. (1) Representations are not unique and subject to permanent change. (2) Language possesses various means to direct the gnostic stream in the hearer. One such mechanism is the topic/focus articulation. (3) Truth conditional equivalence is not all that matters for semantics.