What is encouraging is that many publishers are not only systematically tackling plagiarism, but have agreed to do so by sharing the full text of their articles in a common database.
Given the stereotypes of some companies, it was not finalized, yet it was a necessary step for the service to function – the iThenticate software used by CrossCheck works by comparing submitted articles against a database of existing articles. . CrossCheck’s 83 members have already provided the full text of more than 25 million articles.
What is disappointing is that plagiarism is widespread enough to warrant such precautions. In a notable pilot of the system on three journals, their publisher had to reject 6%, 10% and 23% of accepted papers, respectively.
Admittedly, there are reasons to believe that such levels of plagiarism are extraordinary. Very low rates were found in previous studies of samples on the Physics arXiv preprint server (see Nature 444, 524-525; 2006) and PubMed Abstracts (see Nature DOI: 10.1038/news.2008.0520; 2008).
But the reality is that there is a lack of data on the true extent of plagiarism, whether its prevalence is increasing significantly and what differences can occur between subjects. The hope is that the roll-out of CrossCheck will eventually yield reliable data on such questions over broad areas of the literature – while also acting as a powerful deterrent to plagiarism.
In the process, editors and publishers must remember that plagiarism comes in many varieties and degrees of seriousness, and that response should be proportionate. For example, previous studies suggest that self-plagiarism, in which a researcher copies his own words from a published paper, is much more common than plagiarism of the work of others.
Arguably, self-plagiarism can sometimes be justified, such as when a researcher is bringing up similar ideas to readers of journals in a different field. All plagiarism can also include honest errors or extenuating circumstances, such as a scientist with a poor command of English abbreviating a few sentences of an introduction from a similar work.
Such examples underscore that plagiarism detection software is not a substitute for human judgment, but an aid. A rule of thumb for considering an article’s degree of similarity to previous articles by Nature Journals and others—in particular, for small amounts of self-plagiarism in review articles—is whether the paper is otherwise of sufficient originality and is of interest.
Nature Publishing is a member of the group CrossCheck and is testing the service on submission of its own journals. It has detected levels of plagiarism only in research articles that are spot-checked, and often only in complementary methods. Plagiarism has been more common in submitted reviews, all of which are tested.
This is especially true in clinical reviews, although rates are still well below the 1% mark, and are related to some level of self-plagiarism in most cases.
Although the ability to detect plagiarism is a welcome advance, addressing the problem at its source remains a major issue. In recent years more and more scholarly societies, research institutions and journals have adopted broad ethical guidelines on plagiarism, many of which carefully distinguish between different levels of severity.
It is important that research organizations in all countries, and in particular the patrons of young researchers, establish their scientists to the international scientific community’s accepted norms in terms of plagiarism and publication ethics.