In my paper, I will discuss the articulation between the horizon of projection (the anticipation of the evolution of a field of knowledge) and the horizon of retrospection (its background) in the special case of early machine translation. Because its horizon of retrospection (World War II sciences and technologies, information theory and mathematical logic) did not include linguistics, the horizon of projection of machine translation could not be high quality machine translation, but formal and computational linguistics. The horizon of computational linguistics thus projected was piloted and installed by institutions, funded and evaluated by state agencies, with conclusions and recommendations stated by expert reports, illustrating a new form of relationship between institution and sciences.