Automatic Tracking of Obsolescent Segments with Linguistic Cues

This paper deals with the description and the automatic tracking of text segments containing obsolescence in encyclopedia texts. We assume that despite the non-linguistic nature of this phenomenon, discursive cues are relevant to track those segments. For that purpose, we have worked on a corpus which has been manually annotated by experts and on which we have projected automatically tracked cues. We use machine learning techniques to evaluate the predictive power of our cues. We show, using supervised classification, that our hypotheses enable us to build an automatic procedure to assist human experts. MOTS-CLÉS : repérage automatique de l’obsolescence, indices discursifs, textes encyclopédiques, classification supervisée, aire sous la courbe ROC.