Tackling the PAN’09 external plagiarism detection corpus with a desktop plagiarism detector

Abstract: Ferret is a fast and effective tool for detecting similarities in a group of files. Applying it to the PAN’09 corpus required modifications to meet the require-ments of the competition, mainly to deal with the very large number of files, the large size of some of them, and to automate some of the decisions that would nor-mally be made by a human operator. Ferret was able to detect numerous files in the development corpus that contain substantial similarities not marked as plagiarism, but it also identified quite a lot of pairs where random similarities masked actual plagiarism. An improved metric is therefore indicated if the “plagiarised ” or “not plagiarised ” decision is to be automated