This project is a part of nature language processing and its aims to develop a system of recognition inference text-appointed TIMINF. This type of system can detect, given two portions of text, if a text is semantically deducted from the other. We focused on making the inference time in this type of system. For that we have built and analyzed a body built from questions collected through the web. This study has enabled us to classify different types of times inferences and for designing the architecture of TIMINF which seeks to integrate a module inference time in a detection system inference text. We also assess the performance of sorties TIMINF system on a test corpus with the same strategy adopted in the challenge RTE.
