Cancelability is one of the main tests to identify conversational implicature in general, and scalar implicatures in particular. Despite this fact, cancelability itself is a phenomenon rarely looked at. This paper presents an account of when the cancellation of a scalar implicature is an acceptable discourse move and provides experimental evidence to support our proposal. Our main claim is that the felicity of a scalar implicature cancellation depends on the discourse structure. More speci cally, cancellation is acceptable only if it addresses a Question Under Discussion that di ers from the previous one. As will be shown, this proposal has the additional bene t of permitting us to tease apart cancellations from self-repairs.

We are very grateful for the comments of two anonymous reviewers and to the audiences at the 21st Generative Grammar Colloquium, the ESSLLI 2011 Workshop on Projective Meaning and the Seminar of the Glif at Universitat Pompeu Fabra, where previous versions of this paper were presented. This research has been funded by research projects FFI2011-23356, FFI2011-23046, JDC-2009-3922, RYC-2010-06070 and FFI2009-07114.