The first sentence in this abstract is true. What does the previous sentence—a contingent truth-teller—say? Well, truth is transparent: to say a sentence is true is just to reassert what that sentence said. Hence, to fully determine what a sentence says—what proposition it expresses—we must be able to eliminate any use of truth from it. We cannot do so with the first sentence in this abstract; it does not say anything. In this paper I will defend such a `no proposition' solution to the truth-teller paradox and show that it extends cleanly to the liar. Accounts of this kind have been defended before. My first novel contribution will be to show that whether a sentence expresses a proposition should be determined via the weak Kleene-3 logic, rather than the strong Kleene-3 logic as standardly assumed. This is motivated directly, but it also allows us to deal with a very broad class of paradoxes in the same way—including Curry's paradox. Despite not expressing a proposition we seem to be able to understand the first sentence in this abstract. We also want to be able to talk about it, and especially its liar friends, without reintroducing paradox. My second novel contribution will be to address how we can do both by appeal to independently motivated claims about ways sentences can fail to say what they seem to say. Finally, I will briefly discuss what all this means for attempts to reason with, and/or formalise the logic and semantics of, natural languages.