Nature 456, 30 (6 November 2008)
Detectors could spot plagiarism in research proposals
Victor Maojo, Miguel García-Remesal & Jose Crespo

Sir
Your News story ‘Entire-paper plagiarism caught by software’ (Nature 455, 715; 2008) follows other reports of systems to detect plagiarism (see M. Errami and H. Garner Nature 451, 397–399; 2008, and S. L. Titus et al. Nature 453, 980–982; 2008). Having all been involved in proposal evaluation, we believe the studies indicate that a text-matching analysis of research proposals could reduce plagiarism in subsequent publications.

For instance, when European Commission evaluators have met in the past to evaluate research proposals, they received printed copies which had to be returned before the panel members left, and had no computer access during deliberations. A plagiarism-detector using text-mining methods could be used instead of the current security measures. Such a system could, in principle, detect similarities to previous submissions and uncited sources using advanced document segmentation.

Only official agencies have access to confidential proposals and the funds to experiment with automated plagiarism-detectors. It is important that they should investigate these approaches to reducing the possibility of scientific misconduct.

Garner estimates that among the 181 papers they have identified so far as duplicates, 85% of the text is similar on average, but one-quarter share close to 100%. For a full list of the most similar pairs of articles, click here. There are currently 22 ‘repeat offenders’ in the database. These are authors who have published at least two articles that do not share authors (and so are putative or known plagiarisms). On average these people have ‘authored’ four papers, ranging from two to ten, and spanning 12 countries.

Garner has begun to systematically contact editors and authors of the duplicates he has identified to assess how other cases have been followed up, and is submitting the results for publication. Many journal editors seem reluctant to pursue cases of plagiarism, and half of the articles that editors are alerted to remain uncorrected, Garner says. Few journals have communicated their retraction decision to PubMed, the most widely used abstracts database.

When confronted with their plagiarism, some researchers can be brazen. One offender, whose paper shared 99% of its text with an earlier report, wrote to Garner: “I seize the opportunity to congratulate [the authors of the original paper] for their previous and fundamental paper — in fact that article inspired our work.”

One correction on my part: I’d thought that the website contained only curated entries, which is not the case, although the “false positive” rates they claim are far lower than what I’m perceiving. OK, let’s look at some double-curated “true positives”. This is a routine bit of salami publication, with resulting similarity between abstracts. Not exactly the proudest moment in the history of science, but hardly misconduct. This, on the other hand, seems like a completely legitimate overlap between two reviews on the same subject.

Anyway, how is some grad student even supposed to know whether an abstract in an obscure, possibly defunct, foreign-language journal is for a poster or a paper? No, I’m going to have to vehemently differ with “the authors acknowledge the limitations of the data included”.

My take on the Errami and Garner article: utter garbage. Look at this. Or this. A lot of others look like poster abstracts of results published elsewhere as papers. Others are, as you say e-pub/print overlaps. And these entries are supposedly curated!

Frankly, the more I look, the more I think it rises to the level of misconduct in its own right, publicly accusing people by name over nonsense like this.

I’ll agree that the public presentation of duplicates in Deja Vu seems premature given the significant number of false positives. But there are true positives in there, and they do put a disclaimer up front about leaving it up to the visitor as to how the information is used. However, I’d rather the bulk of the introductory text be less zealous and accusatory given the lack of, um, peer review here, and the authors acknowledge the limitations of the data included. The tool clearly works and is needed but will only be practically useful at the level of a journal checking for prior publication or perhaps screening PubMedCentral, which won’t include meeting abstracts or e-pub ahead of print duplicates … but not all of PubMed or any other citation database in bulk. – writedit