Text Spinners and the Problem of Paraphrase Plagiarism

In an episode of the sitcom Friends, Joey, a character known for his confidence in the dating world, finds himself feeling rather unconfident about his writing abilities. Joey has been asked by his friends Chandler and Monica to write a reference letter to their adoption agency, but he struggles to find the right words. He says, “I want it to sound smart, but I don’t know any big words,” so his friend Ross suggests that he use his computer’s thesaurus function to enhance his letter. As a result, Joey’s letter is turned into a kind of synonym soup: his assertion that “They are warm, nice people with big hearts” becomes “They are humid, prepossessing Homo sapiens with full-sized aortic pumps.”

I think about this scene sometimes when I read student-submitted writing that has been run through a “text spinner” such as SpinBot, QuillBot or other online “paraphrasing” tools that replace words in a text with synonyms. Since I first became aware of these tools in 2018, I have noticed an ever-increasing percentage of student submissions that show evidence of their use. Inside Higher Ed has previously covered students’ use of text spinners, resulting in “patchwriting” or “disguised plagiarism,” when this technology appears to have been relatively new. The prevalence of the use of such tools among students points to problems with the way we approach the issue of plagiarism.

Students’ usage of text spinners appears to be a direct result of instructors’ use of plagiarism detectors such as SafeAssign, Ouriginal or Turnitin. Many instructors and institutions use these plagiarism-detection services, which scan submitted text for similarities to other texts in their databases and generate an “unoriginality score” indicating what percentage of the submitted text is flagged as identical. In addition, numerous websites and apps claim to offer plagiarism-detection services, where students can submit their own text to see what they flag as “unoriginal.”

However, plagiarism detectors cannot reliably determine whether direct quotes integrated into a submitted paper have been accurately cited, and many assignments require the use of well-chosen quotes from external sources. Such quotes will usually be highlighted as “unoriginal,” and their words will be included in the overall “unoriginality” score. That can cause confusion or anxiety for students who fear that any text deemed “unoriginal” will automatically result in a charge of plagiarism. William Duffy has described how students would come to his writing center, anxious about the “similarity” or “unoriginality” score their paper received after being run through a so-called plagiarism detector.

Similarly, in one of my classes, a student submitted a short story analysis paper in which all of the direct quotes from the short story had been reworded, although the rest of the paper’s text sounded typical. The student explained that they had installed the premium version of the Grammarly app, which not only reviewed their paper for spelling and grammar but also offered “plagiarism detection.” The app had flagged the quotes from the short story as “unoriginal text,” and the student had made the app’s suggested synonym substitutions so their paper would not appear to contain plagiarized material.

While sometimes it may be readily evident that a text spinner has been used to alter a submitted paper, the substitutions are rarely as obvious as they were in Joey’s near-incomprehensible letter. For example, when I entered the sentence “They are warm, nice people with big hearts” into SpinBot, it rendered the sentence as “They are warm, pleasant individuals with huge hearts.” QuillBot generated the suggestion “They are folks with enormous hearts who are loving and friendly.” While both of these read awkwardly, it would be difficult to determine with certainty that a text spinner had been used.

When I recently talked with colleagues about this issue, it became clear that while many instructors had noticed the phenomenon of strange synonyms appearing in student papers, not all were aware of the tools that were used to produce these papers. A colleague recounted first learning about text spinners when one of his students described watching a classmate paste text into a text spinner, then copy the altered text it generated into an online plagiarism detector to see what was flagged and then repeat the process until the “unoriginality” score hit an acceptably low percentage. Based on conversations I have had with students and colleagues since, that seems to be a fairly common practice.

As a result, when students submit papers based on copied text that have been run through a text spinner, often a plagiarism detector will pick up only on a pattern of conjunctions, prepositions, proper nouns, pronouns and articles that form a similar pattern to texts in its database. (An example is included in Ann M. Rogerson and Grace McCarthy’s 2017 article.) It will highlight words scattered widely throughout the paper, so the overall “unoriginality” percentage may be low; however, if the original source text is found and compared with the student’s submission, it may be determined that the paper is much more unoriginal than its automatically produced “unoriginality” percentage would suggest. And while in the early days of text spinners, the structure of the original sentences remained largely intact, now many will alter the sentence structure as well, making it more onerous to find the text’s original source.

Ultimately, we must remember that, despite the advertised claims of any plagiarism-detection software, it has limitations and primarily assesses perceived “originality” versus “unoriginality,” as opposed to determining actual plagiarism, which requires more nuanced, human thinking to diagnose. (Efforts to better detect paraphrase plagiarism electronically are ongoing.) It is also true that paper mills also have access to plagiarism-detecting software and can use it to construct papers that are not flagged by these programs.

Steps to Take

Given these realities, we should consider how we can better design assignments to help students avoid the temptation to plagiarize. As John Warner suggests, having students respond to an assignment that allows them to personalize it from the perspective of their own lived experience or asking them to clearly demonstrate the steps they took as part of the process of the assignment can help curb the tendency to plagiarize. David Rettinger and Kate McConnell note the usefulness of assignments that promote higher-level thinking and metacognition. In my own experience, I have found it helpful to use very recent or lesser known texts as the focus of analysis, to construct distinct prompts that ask students to approach a writing assignment in a creative, original way unlikely to have been duplicated elsewhere, or to focus less on having a student produce a traditional essay and instead ask them to create a more meaningful, personal project that culminates in a video or presentation.

In some instances, you might have little say in the design of a course or its assignments; a department or institution may require a degree of uniformity across all sections of a particular course. If similar or identical sets of instructions for an assignment are mandated for multiple sections, the risk of plagiarism may be even greater, since a larger pool of previous responses will be available online. But you can still take some other positive steps.

• Educate students about plagiarism. Ensure that students have a clear understanding of what plagiarism actually means—for instance, that plagiarism is not just about words reused verbatim but applies equally to borrowed ideas—so they do not feel they must be dependent on the potentially bad advice of an app, bot or online tool. If students understand what constitutes academic dishonesty, they may be less likely to turn to online plagiarism-detection tools to assess their papers’ purported acceptability.
• Teach students how to cite correctly. Students need to know that it is OK—and sometimes even required—to use ideas and words from external sources. However, they need to understand how to use them appropriately to give clear credit. That can also help alleviate any potential anxiety about using external source material.
• Be transparent about what plagiarism detectors can and cannot do. Explain that a plagiarism detector can alert a user to patterns of text that are duplicated elsewhere, but it cannot accurately determine the presence or absence of actual plagiarism. Do not promote an overreliance on plagiarism detectors for your students: note that such tools usually do not provide users any resources for how to appropriately cite words or ideas from other sources.
• Don’t assume that a low unoriginality score means that submitted work is free from plagiarism. Open each submission with your plagiarism software enabled to see what text is flagged and determine if there is a pattern. Remember that text spinners replace most verbs, adverbs, common nouns and adjectives so those will not be flagged as unoriginal. In addition, let students know that a low unoriginality score assigned by a plagiarism detector is not an adequate defense against a charge of plagiarism.
• Don’t set a maximum unoriginality percentage score for student work. Consider that a paper’s header, bibliography, correctly cited quotes and any included instructions are likely to be flagged as “unoriginal” and included in the overall “unoriginality” score. There is no single numeric threshold that can accurately determine plagiarism.
• Be specific in your academic honesty policy. Explicitly discuss the unsanctioned use of text spinners, paraphrasing tools, text rewriters or paraphrase generators as part of your course or departmental plagiarism policy.
• Don’t jump to conclusions. Some of the most noticeable results of text spinners, such as distorted idioms or imprecise word use, can also be hallmarks of writing by English-language learners. As you become more attuned to the markers that may indicate the use of text spinners, take care not to unfairly target nonnative speakers. It is important to reach out to the student in cases of suspected paraphrase plagiarism to hear a student explain their writing in their own words.
• Initiate a dialogue and always be open to questions. At the beginning of each term, I ask students to discuss what they know about plagiarism, how they feel about it and what they would like to know. Students often have questions they are embarrassed to ask, but when asking questions is part of the discussion, they may feel more empowered to seek clarification. Like Joey, many students may lack confidence in their writing; an open dialogue with their instructor can help assure them of their own abilities.