Browsing by Author "Mansouri Rayane Moussa"

Now showing 1 - 2 of 2
  • Thumbnail Image
    Item
    Algorithms as Language Managers: The Linguistic Penalty and the Rise of Anti-Bot Language, in Algerian EFL Academic Writing
    (2026) Mansouri Rayane Moussa
    Universities across the world are deploying Artificial Intelligence (AI) detection tools to manage academic integrity. In Algeria, where English is taught as a foreign language (EFL), this policy has created an unexamined problem: the same writing features that EFL instruction builds into students, formulaic phrasing, standard transitions, and structured argument, are the features that detection software flags as machine-generated. This study investigates that contradiction, examining what the anti-bot language register looks like in student writing, and how Algerian higher education students and educators respond to detection pressures. The research employed a mixed-methods design, collecting data through a 21-item Likert questionnaire completed by 23 EFL students (Cronbach's α = .772), semi-structured interviews with a group of four students and two educators, non-participant observations at three academic sites, and writing sample analysis using aidetector.com. Two groups of student writers were compared: one that produced natural, unmonitored academic writing, and one that wrote knowing their text would be scanned. Results indicate that naturally written EFL essays were classified as AI-generated , despite being fully human-authored. Conversely, texts written under detection pressure, full of deliberate errors and structural disruptions, students reported consciously changing their writing style when scanned, and that writing for a detector feels fundamentally different from writing for a reader. Educators confirmed these patterns from the classroom perspective. Ultimately, this study demonstrates that better EFL academic writing is more likely to be flagged as AI-generated than deliberately degraded writing. This is not a minor technical error. It represents a structural consequence of how detection tools work, disproportionately panelizing the students who have worked hardest to master academic English
  • Thumbnail Image
    Item
    Algorithms as Language Managers: The Linguistic Penalty and the Rise of Anti-Bot Language, in Algerian EFL Academic Writing
    (2026) Mansouri Rayane Moussa
    Universities across the world are deploying Artificial Intelligence (AI) detection tools to manage academic integrity. In Algeria, where English is taught as a foreign language (EFL), this policy has created an unexamined problem: the same writing features that EFL instruction builds into students, formulaic phrasing, standard transitions, and structured argument, are the features that detection software flags as machine-generated. This study investigates that contradiction, examining what the anti-bot language register looks like in student writing, and how Algerian higher education students and educators respond to detection pressures. The research employed a mixed-methods design, collecting data through a 21-item Likert questionnaire completed by 23 EFL students (Cronbach's α = .772), semi-structured interviews with a group of four students and two educators, non-participant observations at three academic sites, and writing sample analysis using aidetector.com. Two groups of student writers were compared: one that produced natural, unmonitored academic writing, and one that wrote knowing their text would be scanned. Results indicate that naturally written EFL essays were classified as AI-generated , despite being fully human-authored. Conversely, texts written under detection pressure, full of deliberate errors and structural disruptions, students reported consciously changing their writing style when scanned, and that writing for a detector feels fundamentally different from writing for a reader. Educators confirmed these patterns from the classroom perspective. Ultimately, this study demonstrates that better EFL academic writing is more likely to be flagged as AI-generated than deliberately degraded writing. This is not a minor technical error. It represents a structural consequence of how detection tools work, disproportionately panelizing the students who have worked hardest to master academic English.