In this paper a new approach is shown for a very fast monolingual external plagiarism detection system based on an altered n-gram concept (contextual n-gram), a new high precision contextual Information Retrieval engine, and a new pruning strategy (Referential Monotony) for plagiarism detection and its limits. The assessment results can be compared with the carried out by the winner team at PAN'09, but achieved with remarkable speed (35 min) and low hardware requirements (single laptop).