This is a theoretical article hailing at classroom practice. It was originally intended to be a report on an ongoing study introducing the fundamental theorem of calculus in a STEM freshman course, but the necessary explanation of our epistemological position has taken up most of the available space. Based on Hegel and Lacan, we argue for the reality of language against the implicit realistic positions that dominate mathematics education research. We describe how Cauchy's "one says that" (on dit que) created a bubble of semantic conventions. From inside this bubble, mathematicians only recognize the students by their shadows on the wall, like in Plato's cave. We argue that learning occurs outside the bubble. Instead of dragging the students in, to teach them ready-made notations and methods, our study exercises the dialectics of language in interaction with the students. We stick faithful to our lemma: one teaches by listening and learns by talking.