Plagiarism detection systems are unfair: detecting biases in automated systems
For more than a decade Professor Lucas Introna and Dr Niall Hayes of the Department of Organisation, Work and Technology have been identifying the biases inherent in automated systems of various kinds. One of their most far-reaching findings has been that the design and use of plagiarism detection systems such as Turnitin – used by many Higher Education institutions worldwide – may be unfair due to their embedded values and assumptions. This has had a major impact on policy and practice at institutions around the world.
The essence of this project, says Introna, was to demonstrate that “technologies are not neutral machines/tools but are relatively ‘frozen cultures’. They embody, in their design, certain beliefs, values and interests. The fundamental problem with plagiarism detection systems is that certain types of copying are detected and other types are not.”
Students who retain a sufficient number of consecutive characters will be detected by the system’s algorithm. Often these students will be those who are copying words from a source in the system’s database in order to construct their own ideas through a complex patchwork of copied phrases or sentences – referred to as ‘patchwriting’.
In these cases there is normally no intention to cheat as such, explains Introna:
“The practice often stems from a lack of confidence in English, or the technical language of the subject discipline, or from a lack of experience in academic writing – all of which is mostly true for international students. International students therefore often get detected as ‘plagiarists’ disproportionately to ‘home’ students whose experience and linguistic ability allows them to integrate the phrases and sentences they take from other sources into their own writing, rendering them undetectable. Thus, the assumption that copying is equal to plagiarism, which is embedded in the algorithm of these systems and in the way they are often used, can lead to very unfair outcomes for some students.”
The dissemination of these insights, through a series of reports, workshops and supporting resources, has transformed writing support and teaching practices at no less than 32 institutions across Britain, North America and other parts of the world. It has also resulted in at least ten institutions developing plagiarism policy frameworks that are less punitive and taking a more developmental approach towards those students accused of plagiarism, leading to much fairer outcomes for them.
Southern Illinois University, for example, actively drew upon the work to formulate its institutional policy for dealing with plagiarism. The University encourages staff to appreciate 'that students from non-Western cultures may have different concepts of authorship and little or no training in how to use sources and therefore may need extra help in avoiding plagiarism' and that they should 'expect some ‘patchwriting’ (developmental plagiarism) that is unintended, and allow time for revision of patchwritten texts'. These are recommendations that come directly from the Lancaster research.
The students’ union of Concordia University in Montreal used the research as part of an ‘Academic Fairness Campaign’ to provide a more nuanced way of dealing with students identified as plagiarists by detection systems. Drawing on Lancaster’s research, they suggested that university staff should try to turn around these apparently bad situations to use them as learning opportunities. Their recommendations state that “if an international student, especially a new student, is suspected of plagiarism, do not automatically assume intent to be dishonest… Adopting an educative approach to plagiarism is preferable to one based solely on punishment.”
This campaign is just one example of how this work has helped students to get fairer treatment when accused of plagiarism, and it continues to inform both individual and institutional practice around the world.