I argue that the implementation of the Dummettian program of an "anti-realist" semantics requires quite different conceptions of the technical meaning-theoretic terms used than those presupposed by Dummett. Starting from obvious incoherences in an attempt to conceive truth conditions as assertibility conditions, I argue that for anti-realist purposes non-epistemic semantic notions are more usefully kept apart from epistemic ones rather than being reduced to them. Embedding an anti-realist theory of meaning in Martin-Löf's Intuitionistic Type Theory (ITT) takes care, however, of many notorious problems that have arisen in trying to specify suitable intuitionistic notions of semantic value, truthconditions, and validity, taking into account the so-called "defeasibility of evidence" for assertions in empirical discourses.