Extrinsic plagiarism detection quickly becomes ineffective when you do not have access to potentially sources documents of plagiarism or when the search space is large like the Web, which is often the case with current anti-plagiarism software. Therefore the intrinsic detection becomes much more effective. In this paper, the automatic authorship detection is exactly presented. It allows to know if a text's part does not belong to the same author as the rest of the text and so in theory to identify plagiarized passages of a document. We explain our contribution to the existing procedures and assess the limitations of our approach. The goal is to enable the detection and clustering of passages in a document by author.