International audienceThis paper describes the use of second order similarities for identifying similar texts inside a corpus of aviation incident reports written in both French and English. We use a second bilingual corpus to construct pairs of reference documents and map each target document to a vector so each coordinate represents a similarity score between this document and the part of the reference corpus written in the same language. We evaluate the system using a large corpus of translated incident reports. The results are promising and validate the approach.Cet article décrit l'utilisation de la technique de similarité de second ordre pour l'identification de textes semblables au sein d'une base de rapports d'incidents aéronautique...