Digitalization has led to massive changes in the sciences, especially in the humanities, as shown by the Digital Humanities. Using linguistics, we discuss these changes on theoretical, epistemological and methodological levels, which consist in particular in the fact that methods of machine language processing are increasingly alinguistic. These are methods that are not based on linguistic theories and do not follow linguistic modelling of language or language usage. It remains to ask how a scientific discipline such as linguistics must be structured so that it becomes antifragile rather than robust in the face of such revolutions. In other words, how it can get richer out of unforeseeable changes and does not primarily have to endure fights of demarcation and defence against such transformations.