Writing Style Similarity Reflects Academic Genealogy
This paper demonstrates that academic genealogy significantly influences writing style, showing that advisors and their students (as well as academic siblings) exhibit substantially higher stylistic similarity than random peers, which leads to a high rate of misattribution errors in authorship detection systems that fail to account for these inherited stylistic traits.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
In the world of academic publishing, a writer's voice is often treated as a unique fingerprint. When a computer program is asked to determine who wrote a specific paper, it scans the text for subtle patterns in word choice, sentence structure, and rhythm, assuming these habits belong to a single, independent mind. This assumption is the foundation of authorship attribution, a technology increasingly used to catch ghostwriters or identify papers generated by artificial intelligence. However, this system rests on a fragile premise: that a researcher's style is entirely their own. In reality, scholars do not develop in isolation. They study under mentors, absorb the habits of their departments, and often share the same native language and specialized vocabulary. If a student's writing style is heavily shaped by their advisor, then a computer designed to spot the true author might mistakenly point to the mentor instead, creating a false accusation that is nearly impossible to disprove during the peer-review process.
A researcher set out to test whether this mentorship effect is strong enough to confuse these automated systems. They built a massive dataset using records from the Mathematics Genealogy Project, a curated history of who trained whom in the field of mathematics. By linking these verified family trees to the actual abstracts of thousands of papers, they created a ground-truth map of academic relationships. They focused specifically on authors who had written at least two papers alone, ensuring the text reflected an individual's voice rather than a group effort. The researcher then used computer models to measure how close the writing styles of different people were, comparing advisors to their students, and students to one another, while carefully controlling for other factors like shared universities or research fields.
The results revealed a clear and measurable echo of mentorship in the written word. When the researcher compared the writing of an advisor and their student, the two were found to be significantly more similar than two random people working in the same field would be. In fact, the advisor's style sat nearly forty percent closer to the student's than a random peer did. This similarity was not just a matter of them working in the same building; the effect persisted even when the students had moved to different institutions to work. The study also uncovered a phenomenon the researcher calls "academic siblings": two students who trained under the same advisor but may never have met each other. These pairs, who shared no direct connection other than their mentor, wrote with a striking similarity to one another, sitting about thirty percent closer in style than random pairs. This suggests that the influence of a mentor travels through the academic lineage, shaping the voices of students who never interacted directly.
Crucially, the researcher found that simply sharing a university or a department was not enough to create this effect. When they looked at people who earned their doctorates at the same school but had no advisor-student link, their writing styles showed almost no extra similarity compared to strangers. This ruled out the idea that the effect was caused merely by being in the same physical environment or using the same local jargon. The true driver appeared to be the direct line of mentorship. To see how this impacted real-world tools, the researcher ran a test where a computer system tried to identify the author of a paper from a list of candidates. The system made mistakes in about eight percent of the cases where the true author was not the top guess. In those specific errors, the computer was far more likely to pick the true author's advisor or an academic sibling than it would be by random chance. In fact, the system made these specific types of errors eleven times more often than chance would predict.
The study concludes that the assumption of total stylistic independence is flawed. While a researcher eventually develops their own voice, the imprint of their mentor remains detectable for years, creating a pattern that automated systems can mistake for identity. This does not mean the technology is broken, but rather that it must account for the reality that academic writing is a learned craft, passed down from one generation to the next. The findings suggest that when a system flags a paper as belonging to a mentor, it may not be a sign of fraud, but rather a reflection of a deep, historical connection in how that scholar learned to write.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.