Fouille de textes hiérarchisée appliquée à la détection de fautes

This paper presents an original text mining approach based on text segmented units, mixing data mining techniques and text linguistics. First, we describe the text units and their descriptors designed for detecting mistakes in scientific papers in English. These include text organisation, emphasizing the relative positions and the context of stylistic markers. Secondly, the paper explains the text mining methods required for such tasks and provides the techniques to extract rules characterizing classes and emerging patterns techniques. Experiment results show the usefulness of text organisation set as an hypothesis. The paper concludes on the text mining benefits for linguistic tasks. MOTS-CLÉS : fouille de textes, linguistique textuelle, mise en forme matérielle, structure typographique, stylistique, règles d'association, motifs émergents, caractérisation de classes, détection de fautes d'anglais, correction, style.