This paper is about the way language really works. I take examples from literary texts, i.e. fictitious, which nevertheless make sense when we read them. Linguistic features consist of operations and implications which, when all is said, are social and not linguistic. We could even say that literature reveals the gist of what language is about. As Stanley Cavell says, some texts reveal aspects of life often ignored by linguistics and mainstream philosophy of language, i.e. the “uncanniness” of reality. More specifically, I will try to show that literature offers: i) a critique of the way people use language — and more generally of the way society works; ii) new possibilities of life. Herman Melville’s short story “Bartleby” (1853) is a case in point. The new employee of a New York lawyer answers “I would prefer not to” when given some task to perform. There is very little to say about the employee, who can more or less be identified with his repetitive statement. The lawyer is certainly more interesting: he represents some of the most important values of 19th-century American society. My critique will bear on what is known as acts of language, more specifically performatives. My view is that they offer a very superficial description of language use and seem to be only able to account for a small number of interactions in society. Indeed the theory is too often nothing but a classification of what one can do with acts of language. When the employee of Melville’s story starts not behaving according to the unwritten rules prevailing in the lawyer’s world, he is ordered to be “reasonable”. The lawyer also says that what matters is our “assumptions” about the world in which we live. Melville’s text shows that these assumptions are only conventions claiming to possess a universal value. They give people places and roles to play in society. I believe that the questions we should ask when we study performatives is that of power, and possibly also of money. It should be obvious that performatives hide the reality that nothing justifies the fact that some people are more equal than others and that that situation must not change. I believe that performatives rely on a conception of language that denies the arbitrary of the sign. It tries to impose ready-made links (“assumptions”) between utterances and contexts. In the end, traditional linguistic classifications are political, whether they proclaim it or not. (They usually don’t.) They aim to give to some privileged individuals the control of language, and more generally of the fabric of society. The whole problem is of course that societies are divided and that consensus is a myth. I believe that language, if studied differently, could help us see the world as it is, accept the passing of time, chance and change, and not just repeat the same old artificial situations and hidden hierarchies.