In this paper, we present LIMSI participation to one of the pilot tasks of QA4MRE at CLEF 2012: Machine Reading of Biomedical Texts about Alzheimer. For this exercise, we adapted an existing question answering (QA) system, QALC, by searching answers in the reading document. This basic version was used for the evaluation and obtains 0.2, which was increased to 0.325 after basic corrections. We developed then different methods for choosing an answer, based on the expected answer type and the question plus answer rewritten to form hypothesis compared with candidates sentences. We also conducted studies on relation extraction by using an existing system. The last version of our system obtains 0.375.