This article, based on a study of a prototypcal case (the works on textual linguistics written by Jean‑Michel Adam), analyses the tensions prevailing in scientific discourse which must respond to two partially different objectives, the production of high level knowledge addressed to peers, on the one hand, and their vulgarisation, on the other. These tensions are examined on the basis of reformulation between texts and diagrams, quotations and comments. Reformulation aims not only to explicit, it also legitimizes a scientific paradigm, in this case textual linguistics, justifies the evolution of its relationship with discourse analysis, while showing the advantages of a model capable of accounting for a large number of data in an orderly way. As a result, according to the strategies adopted, diagrams and comments are either a means of vulgarisation or of the development of a scientific approach.