In this paper, we discuss how different types of automatic annotation of digitised newspaper articles can be integrated into the iterative questioning of the source material and the creation of research corpora out of a collection of unstructured texts (kept in a structured collection). We annotate a sizeable collection of Swiss press articles (183,270), extracted via the impresso interface1 using topic modelling (MALLET)2 as well as a naïve Bayes classifier (script by Milan van Lange).
The methodological discussion we propose is to explore how text mining can help identify historical discourses that are difficult to query with keywords because of their inherent ambiguity and how to grasp them in a large corpus. We argue that the automated annotations can provide a body of corroborating evidence of the searched discourse, to be used as an intermediary and heuristic analysis step.
Afficher cette publication dans notre dépôt institutionnel (orbi.lu).