Detecting and handling suspected plagiarism in submitted manuscripts

Plagiarism is a great masquerade. Detecting plagiarism is daunting even for experienced reviewers and editors.1,2 Plagiarism constitutes scientific misconduct by violating the intellectual property rights of the creators of such content.3 Literature suggests that up to a sixth of manuscripts submitted to journals might be affected by plagiarism.4 Through this editorial, we share our perspectives on how to identify and manage instances of suspected plagiarism in manuscripts submitted to journals for publication.

Plagiarism – quite different from mere similarity!

It is imperative for all stakeholders (i.e. editors, reviewers, and authors) to understand that plagiarism is distinct from similarity.5 Plagiarism essentially refers to the reproduction of information without duly attributing the source of such information, i.e. passing it off as one’s own. This information could be text (as in manuscripts submitted to journals), previously published figures and tables in manuscripts, or ideas/hypotheses for research. On the other hand, similarity of text simply refers to how similar or dissimilar the text in the manuscript in question is, when compared with the available literature, either in previous manuscripts, web pages or grey literature.5 Software like iThenticate and Turnitin are commonly used to identify the extent of similarity in submitted manuscripts by journals.5 Furthermore, the extent of similarity detected by such software depends upon filters employed by the user, such as the number of words that are required to be similar in succession, or whether reference lists are included. The latter inevitably makes a manuscript appear more similar than it actually is, since references are generally listed in databases or in previous manuscripts. Generally, a limit of 8-10 words in succession is considered as an indicator of significant degree of similarity. However, even this might have limitations. There might be long phrases such as names of certain research tools or organisations similar with the published literature, however this does not amount to plagiarism.5 Organisations across the world responsible for regulating teaching and research, such as the University Grants Commission (UGC) in India, have set out limits for similarity in different parts of theses and manuscripts, dividing this further into areas like methods where similarity to some extent might be tolerable, and results and discussion where even minimal similarity might not be acceptable. Such guidelines appear to imply that thresholds of similarity are synonymous with plagiarism.6

However, plagiarism is not as simple as reproducing words. First, the mere reproduction of words need not indicate plagiarism at all, if this is done appropriately by placing the reproduced text within double quotes and referencing the source, however, such a practice should be minimised while writing manuscripts. Second, authors in the present world are aware of the ability of similarity checking software to identify such reproduced text, hence, might be able to partially bypass similarity detection by changing a few words here and there from the reproduced source.5 Third, such similarity checking software are unable to detect plagiarism of figures and tables. Fourth, the plagiarism of ideas or research hypotheses cannot be detected by checking similarity. Instead, this might be identifiable by cross-checking lists of references and their order in the source manuscript compared with the present manuscript, or based on evidence provided to the journal by the scientist who has claimed priority over the given idea. It is evident that the identification of most of these forms of plagiarism requires considerable manual input from editors and reviewers in the present day. Therefore, detecting plagiarism should not be simply thought of as an output of similarity checking software, which can only serve as a tool to support editors and reviewers in identifying instances of plagiarism. Ultimately, the identification of plagiarism (whether in the presence of any significant similarity or not) remains the purview of editors and reviewers. Table 1 summarises differences between plagiarism and similarity.

Practical considerations for journal editors and reviewers to identify plagiarism

It is preferable to use the output from similarity checking software as a starting point. Appropriate filters should have been used to exclude references as well as to set a word limit of 8-10 words before tagging a phrase as similar. Thereafter, such similarity reports should be manually reviewed in detail to identify portions that appear similar and span more than two or three lines. Editors should be careful that such stretches of words might have a few dissimilar words here and there which have been intentionally placed to avoid flagging of the text as unoriginal by similarity checking software.5 The source of such similar content as indicated by the software should be reviewed. Sometimes, the automated links to such sources do not function. In such a situation, those selections of text should be copied and pasted manually on to a search engine such as Google to identify the source, which should then be reviewed to confirm similarity. A common source of similarity is prior conference presentation abstracts, and this hardly indicates plagiarism. These can be easily identified by asking authors to declare prior conference presentations in the instructions to authors or during the process of peer review. Plagiarism of figures and tables is considerably difficult to identify. A search conducted on databases of figures such as Google images with the keywords related to the figure in question and manual checking of figures that appear similar with the submitted one is one way for editors and reviewers to identify plagiarised figures. Similarly, a manual scrutiny of tables in previous articles, including review articles, remains the only way to identify plagiarised tables. Even in the absence of output from similarity checking software, experienced editors and reviewers might suspect plagiarism in manuscripts if there are abrupt changes in flow and tone between sections. This might suggest that such sections have been reproduced from written work by other authors.

After completing such a review of the similarity report, authors might be asked to revise minor areas of similarity, or a decision might be taken to reject the manuscript if there are extensive areas of similarity across different sections, irrespective of the percentage of similarity. Furthermore, the instructions to authors (for new manuscripts) or revision comments (for revised manuscripts) might explicitly mention the need to seek permission from copyright holders along with appropriate referencing in instances where figures or tables have been reproduced with or without modifications from elsewhere. The authors might also be asked to add a disclaimer stating that the content and ideas expressed in the manuscript are original and have not been copied from elsewhere. Box 1 summarises such points for consideration by the journal editors and reviewers to detect plagiarism.

Box 1 How can editors and reviewers detect plagiarism in manuscripts?

Future perspectives about plagiarism detection and prevention

The emergence of artificial intelligence and machine learning as a tool to automate diverse processes might also be exploited to assist editors and reviewers to detect plagiarism, particularly of figures and tables. However, instances such as the difficulty of machine learning in distinguishing the image of a duck from a rabbit when rotated indicate that human oversight shall always be required no matter how sophisticated the automation might be.7 Since a significant proportion of retracted articles are due to plagiarism,8,9 databases of retractions such as the one provided by Retraction Watch might be linked to manuscript submission systems to flag authors whose manuscripts have been previously retracted due to research misconduct (including plagiarism). Such manuscripts might have a higher-than-average pre-test probability of misconduct, therefore, might be scrutinised more carefully by editors. Lack of education about what constitutes plagiarism remains an important determinant of such behaviour, especially in regions of the world such as Asia.6 Therefore, continuing attempts at educating authors worldwide regarding publication ethics and scientific writing should also include illustrative sessions
on plagiarism.10