In my paper, I would like to examine these two Firthian contributions, situational categories and restricted languages, which, although conceived separatly at two different moments of Firth's work and addressing distinct theoretical issues, evolved and merged into a single notion within Firth's work itself. We will see why restricted languages had been finally abandoned in favour of registers by the Neo-Firthians. In particular, the use of probabilistic methods which entail great amounts of data, especially large computer-based corpora, has given rise to a major turn in the treatment of linguistic events.