Title: Quantifying the changing role of past publications

Abstract: Our current societies increasingly rely on electronic repositories of collective knowledge. An archetype of these databases is the Web of Science (WoS) that stores scientific publications. In contrast to several other forms of knowledge -- e.g., Wikipedia articles -- a scientific paper does not change after its "birth". Nonetheless, from the moment a paper is published it exists within the evolving web of other papers, thus, its actual meaning to the reader changes. To track how scientific ideas (represented by groups of scientific papers) appear and evolve, we apply a novel combination of algorithms explicitly allowing for papers to change their groups. We (i) identify the overlapping clusters of the undirected yearly co-citation networks of the WoS (1975-2008) and (ii) match these yearly clusters (groups) to form group timelines. After visualizing the longest lived groups of the entire data set we assign topic labels to the groups. We find that in the entire Web of Science multidisciplinarity is clearly over-represented among cutting edge ideas. In addition, we provide detailed examples for papers that (i) change their topic labels and (ii) move between groups.