This is not a "why" post. He did an N x N contingency table of how many people moved from language X to language Y, then rendered his findings as a set of directed graphs.