Editing AI, Building Translators: Machine Translation Post Editing, Employability, and Reflective Learning in AR3461
Dr Orhan Elmaz
School of Modern Languages
- What motivated you to use AI in this module, and what goals or challenges were you aiming to address?
I used AI because Machine Translation Post-Editing (MTPE) is increasingly common in translation services globally and is governed by the industry standard ISO 18587:2017. I therefore wanted an employability-focused assessment that mirrors contemporary practice. Within AR3461, the goal was not to use AI for speed, but to treat AI output as an unstable draft through which students could practise: (a) diagnosis grounded in Classical Arabic grammar and rhetoric; (b) ethically transparent correction, particularly through clear annotation; and (c) sound judgement under ambiguity – an essential skill when translating the Qur’an, where interpretive responsibility is high.
- How did you design or adapt the assessment, and how did you prepare students for using AI appropriately?
I designed the second assignment around machine translation post-editing of Sūrah 54 (al-Qamar, “The Moon”) of the Qur’an, specifically because of its stylistic features – refrain, rhyme and rhythm, and its sequence of prophetic narratives. The assessment required students to critique and revise an AI-generated translation rather than accept it. Students were prepared through the module’s content on Classical Arabic grammar and literary analysis based on passages from the Qur’an, alongside framing the Qur’an within everyday Muslim life and religious practice. They were also guided by academic translation conventions that students repeatedly referenced in their reflections: transparency about additions, careful handling of ellipsis and context, and avoiding unmarked interpretation – for example, several students explicitly stressed in their reflections that changes to the source text must be clearly indicated in the target text.
- What challenges did you encounter, and how did you address them?
Some students initially felt intimidated or pressured by AI, particularly in terms of translation speed and perceived quality, when I introduced the MTPE workflow for the remainder of the semester after Independent Learning Week. Others maintained that AI would never be fully successful because of its inconsistency; this was mitigated by treating prompt iteration and enhancement as an expected part of the workflow rather than a sign of failure. A few students concluded that AI cannot replicate the quality of human translation and learning, given the unreliability of MT – whether through mistranslating stylistic features, flattening rhetoric, or failing to provide annotations that transparently signal interpretive choices.
- What benefits did you see for students and for your own teaching practice?
Students demonstrated stronger editorial agency and confidence: they justified their choices and explicitly overrode MT where necessary. They also practised MTPE by applying grammar and rhetorical knowledge with ethical responsibility to a high-stakes, sacred text: the Qur’an. In doing so, they developed employability-aligned skills, including evaluating MT quality, documenting interventions, and managing risk in sensitive domains.
For my own teaching practice, the task made students’ thinking visible. Their reflections revealed how well they understood key grammar and rhetoric concepts, where MT output misled them, and which conventions they adopted. This, in turn, has been useful for refining my teaching emphasis.
- How did you evaluate the usefulness of this assessment to ensure that it reflected the desirable learning outcomes?
I evaluated the assessment through students’ reflective write-ups and the edits they described, focusing on the extent to which they applied Classical Arabic grammar and rhetorical analysis, adhered to annotation conventions, and articulated reasons grounded in Qur’anic context and reader responsibility. I also coded patterns across the full set of reflections and extracted quotations that indicated confidence, intimidation, ethical awareness, and process learning – evidence that the assessment elicited the intended outcomes: Classical Arabic grammar and style, critical judgement, AI resilience, and responsible practice.
- What would you do differently next time, and what advice would you give to colleagues?
Based on this experience, sensible refinements would be:
-
Make the post-editing protocol explicit from the outset: provide a short checklist (e.g., annotate all non-source additions, track omissions versus additions, note rhetorical losses, and maintain consistency across repeated phrases).
-
Require a brief “quality report” aligned with industry practice: students summarise which error types occurred, what they changed, and why to directly support MTPE employability aims.
-
Build in a comparison task: include one passage where students compare two AI outputs and justify which is safer and why, strengthening opportunity recognition and risk management.
- Normalise initial intimidation: explicitly discuss that MT can feel impressive at first, but the “quality costs” become visible during post-editing.
Advice to colleagues: use AI tasks where the discipline can genuinely go beyond the model – where students must apply domain knowledge, ethics, and judgement. Frame AI as a draft generator and the student as the accountable human-in-the-loop editor. Assess the quality of reasoning, not speed. Finally, choose texts where context and responsibility matter, so students practise transparency and risk-aware decision-making rather than surface-level correction.