Reuse and plagiarism in Speech and Natural Language Processing publications

Abstract

The aim of this experiment is to present an easy way to compare fragments of texts in order to detect (supposed) results of copy and paste operations between articles in the domain of Natural Language Processing (NLP), including Speech Processing. The search space of the comparisons is a corpus labeled as NLP4NLP, which includes 34 different conferences and journals and gathers a large part of the NLP activity over the past 50 years. This study considers the similarity between the papers of each individual event and the complete set of papers in the whole corpus, according to four different types of relationship (self-reuse, self-plagiarism, reuse and plagiarism) and in both directions: a paper borrowing a fragment of text from another paper of the corpus (that we will call the source paper), or in the reverse direction, fragments of text from the source paper being borrowed and inserted in another paper of the corpus. The results show that self-reuse is rather a common practice, but that plagiarism seems to be very unusual, and that both stay within legal and ethical limits.