This paper is a theoretical interrogation of the way language (actually) works. One of its conclusions is indeed that language cannot be studied apart from its use, by which we mean (real) use by (real) ‘subjects,’ possessing a body, an unconscious, and being part of social processes (which — it should be recalled — are always conflictual, whether the conflicts are apparent or not). The paper analyses examples taken from a novel by Laurence Sterne and a poem by Federico García Lorca. We consider that theories must be borne out by examples, and that these examples should be complex and extended (hence in our case the use of literary texts), and on no account like the skeletal examples a lot of linguists and philosophers seem to relish and which account for the naivety of a great many of their pronouncements on language. The paper contains a critique of the so-called concept of the performative, and more generally of the dismal superficiality and idealism of post-Wittgensteinian theories of language. Sterne’s Tristram Shandy enables us to reassess the revolutionary impact of true British empiricism (Locke, and especially Hume). We posit that language alternates at varying degrees between what can be called “power” [pouvoir] (that is power over other people, whether open or not, and whether efficient or not), and “potential” [puissance] (that is the discovery of new possibilities of life and of new kinds of relationships with others). It follows that utterances should not only be submitted to semantic interpretations, but above all analysed along pragmatic lines. What mental operations, what networks of relations do we construct when we make sense of a text? We propose, among other hypotheses, that co-ordination is probably more important than predication, which sadly still enjoys today an undue hegemony in the study of language