The bottleneck in marking Paper 2 writing

Marking an English writing paper spends most of its time judging each item against the scheme, flagging language errors, and writing specific feedback on content and organisation. The problems across a class tend to repeat: errors of tense and article, arguments with no support, stiff transitions between paragraphs. By the later scripts, the teacher is rewriting the same set of comments over and over. The result is that each piece usually gets one round of feedback, and students find it hard to polish the same one.

The AI for Empowering Learning and Teaching Funding Programme asks schools to bring AI-assisted teaching into at least three subjects. English writing is a common entry point, because the marking load is heavy and the results are easy to quantify and write into a report. The condition is that the tool fits the HKDSE marking requirements rather than a generic English score.

How English AI marking supports the writing paper

First marks on content, language and organisation

The reliable approach is a system with the HKDSE English Paper 2 marking scheme built in, giving a first grade for each of the three areas, Content, Language and Organisation, rather than putting out a single total. The teacher moves from marking from scratch to reviewing and correcting, looking at the system's item-by-item judgement first and then deciding whether to agree.

Every mark comes with the candidate's own words

Every grade should point back to a specific place in the essay, so the teacher can see at a glance what the judgement rests on and the student knows which sentence the problem is in. A mark with no supporting text cannot be reviewed. With item-by-item evidence, the review becomes a quick confirmation, and the teacher only handles the parts in doubt.

A shorter feedback cycle to support multiple drafts

Students revise on the feedback and resubmit, the system gives a fresh first mark on the revised draft, and the teacher only reviews what is still in doubt. The marking cost does not rise in proportion to the number of drafts, so one draft with one round of feedback can become several drafts with several rounds. Writing ability builds through repeated revision, and a faster feedback cycle is exactly what Paper 2 teaching most needs.

Criteria for choosing an English AI marking tool

  • Whether the marking scheme matches the HKDSE: what is built in should be the English Paper 2 itemised scheme, calibrated against past papers, rather than a generic English writing score.
  • Whether item-by-item marks come with the source text: run a few essays through and check that each grade points back to a specific sentence in the piece.
  • Whether teacher review is convenient: changing a mark should take a few steps, and it should keep a record of changes for the report and for professional accountability.
  • Whether the language analysis is reliable: test its identification of common errors such as tense, article and collocation on local student essay samples.
  • How student data is handled: where it is stored, how long it is kept and whether it is used to train models, all in line with the Personal Data (Privacy) Ordinance.

The real limits of AI marking in English

Some judgements cannot be given to AI. Whether an argument persuades, whether the tone suits the genre and the reader, whether the word choice is apt and has style: these still rest with the teacher. AI suits the high-volume, repetitive first pass and the flagging of language, while the teacher handles the judgement of quality and knows an individual student's progress across a whole year. The system sees this one script; the teacher sees this student's path.

High-stakes assessment holds the line all the more firmly. The final grade in internal exams and banding assessments must rest in the teacher's hands. A sensible division of labour is AI for the volume, teachers for the quality, and we lay out the evidence for every mark when we build our tool precisely so that the teacher's final review is both quick and confident.