Speakers use the adversative markerbutin different ways, for instanceto express parallels, contrasts, denial of expectations and corrections. Twotraditions aim to account for these uses ofbutin a unified manner: for-mal contrast and inferential approaches. Both make the case thatbut-sentences have an affirmation-denial (yes-no) polarity structure. Theyalso require pragmatic processing to cover all uses ofbut.I introduce new data to show that we need to account for the pragmaticprocessing at stake in assessingbut-sentences. If unrestricted, pragmaticprocessing could makeany but-sentence acceptable.I analyze the yes-no polarity inbut-sentences in terms of speech acts(assertion-denial). A speech acts account has two benefits. It clarifiesthe pragmatic dimension ofbut’s contribution (an indirect speech act). Itexplains how speakers pick claims that are salient in the context to enrichinferences. Moreover, the speech act picture covers the variety of uses ofbutand provides the needed restrictions on pragmatic processing. It alsoaccounts for the new problematic data.I conclude with a broader theoretical prospect: studyingbutin terms ofspeech acts provides a method to analyze parts of meaning as expressionsof speakers attitudes and commitments.