Parallel areas detection in multi-documents for multilingual alignment (Détection de zones parallèles à l'intérieur de multi-documents pour l'alignement multilingue) [in French]

Parallel areas detection in multi-documents for multilingual alignment This article broaches a central issue of the automatic alignment : diagnosing the parallelism of documents. Previous research was concentrated on the analysis of documents which are parallel by nature such as corpus of regulations, technical documents or simple sentences. Inversions and deletions/additions phenomena that may exist between different versions of a document has often been overlooked. To the contrary, we propose a method to diagnose in context the parallel areas allowing the detection of deletions or inversions between documents to align. This original method is based on the freeing from word and sentence as well as the consideration of the text formatting. The implementation is based on the detection of repeated character strings and the identification of parallel segments by image processing. MOTS-CLÉS : détection et alignement de zones, appariement de N-grammes de caractères, corpus de multidocuments.