Plagiarism software can be beaten by simple tech tricks
IT scholar says PDF tweaks allow students' copied work to evade detection. Hannah Fearn reports
Technological loopholes allow savvy students to beat academic plagiarism software, an IT expert has warned.
James Heather, senior lecturer in computing at the University of Surrey, has revealed that plagiarism detection systems such as Turnitin that are routinely used by universities are open to simple cheats allowing students to evade detection when submitting copied material.
The software works by extracting text from an essay or assignment and checking whether it matches text from other sources, such as documents available online.
But in a new paper, "Turnitoff: identifying and fixing a hole in current plagiarism detection software", Dr Heather reveals that beating the system is simple.
"In their current incarnation, one can easily create a document that passes the plagiarism check regardless of how much copied material it contains. When there are loopholes that can be exploited, they give the operator a false assurance that a submission is original."
The study, which appears in the journal Assessment & Evaluation in Higher Education, demonstrates ways in which students can modify plagiarised work to avoid detection.
"If we can stop the text from being properly extracted from the document, without affecting how the document looks and prints, then the software will not be able to identify any plagiarised material," Dr Heather writes.
Students aware of this loophole could get around the system by converting a plagiarised essay to PDF format, he says, and then altering the corresponding "character map" - a map of the sequence of characters used in the text. Although the text would remain visually unaltered, extracts tested by the plagiarism software would be garbled, and so matches would not be detected.
Or, he says, students could rearrange character codes, or "glyphs", in the PDF so they no longer correspond to the alphabet and "the link between the text and its printed representation will be broken".
In this scenario a tutor could print out and read the essay, but the computer running the detection software would scan nonsense.
Finally, students could convert text into a series of Bezier curves to represent the shape of letters rather than using the characters themselves.
"If there is no text, then the plagiarism detection cannot function," the paper notes.
Dr Heather argues that requiring students to submit work in Microsoft Word is not a solution to the problem; students could simply convert a doctored PDF into Word.
Instead, he says, universities should supplement detection systems with a secondary optical character recognition (OCR) program.
"The only reliable way to make certain that the extracted text matches what is represented on the printed page is to use OCR," the paper concludes.
Such a system attempts to "do the same thing as the human reader of the submission: take a rendered copy of the work and interpret the marks that appear on the page. This immediately counters all attempts to alter the internals of the document."
This method places a burden on a university's server, costing time and money. But free OCR software is available and universities should make use of it, Dr Heather says.
A spokesman for Turnitin said the cheating methods required a high level of technical skill but the company is working to detect when tricks have been used.