Literature is the one teachers doubt most, and they are half right.
Nobody worries much about a Physics derivation being machine-marked. The method is either right or it is not. Literature feels different, and the instinct is sound — but it is usually aimed at the wrong half of the paper. Most marks in a Class 9 or 10 literature answer are not awarded for insight. They are awarded for covering the points the question demanded, from the text. That part is checkable. The insight part is not.
Everything below is from one real paper. A Class 9 ICSE English test on Julius Caesar, Act 1 Scene 3, sat on 11 September 2026 — eighteen questions, forty marks, marked by ClassPulse. The whole thing is published, every page and every mark, so the examples here can be checked rather than taken on trust.
What it can judge
| The question asks for | Can it be marked? |
|---|---|
| Identifying a character, speaker or event from an extract | Yes. There is a right answer and it is in the text. |
| Explaining what a line or image means | Yes, against the points the scheme names. |
| Naming a set number of reasons, sights or devices | Yes — and see the failure below, because this is the shape that breaks. |
| Supporting a claim with evidence from the text | Yes. Whether the support is present is checkable. |
| Contrasting two characters' attitudes | Yes, where the contrast is stated in the play. |
| Whether an original interpretation is good | No. This is the part a teacher is for. |
| Whether the writing is elegant | No, and it should not try. It marks content, not style. |
What that looks like on a real answer
On question 16 — what Casca believes about the unnatural events, and how Cassius uses that belief — the student got 3 of 3, with this comment:
“You correctly identify Casca’s fear of the unnatural events and recognise them as signs from God. You also explain that Cassius uses these signs as a warning connected with Caesar and persuades Casca to join the conspiracy. The answer could be expressed more clearly and accurately, but all three required ideas are present.”
That last sentence is the whole design in one line. The writing was clumsy; the marks were for the ideas. An examiner marks the same way, and a tool that quietly docked marks for awkward phrasing would be marking a different paper from the one the board set.
Question 17 does the same thing from the other side. Full marks, plus a note: “Avoid calling Cicero’s view ‘just a kind of bad weather’, as this is a slightly imprecise expression, though the intended contrast is clear.” Style comment, no penalty. That is the right division.
Where it got it wrong on this very paper
Two questions on this paper have the same shape — name three things — and they were not marked consistently.
- Question 15 asked for three unnatural sights. The student gave two, and added one that was not Casca’s. It scored 2 of 3, and the feedback named both the missing sight and the wrong one. That is correct marking.
- Question 11 asked for three reasons. The feedback says, in its own words, “You also give two valid reasons… This fully answers the question” — and awarded 3 of 3. Two reasons is not three, and the mark should have been docked. It was not.
We publish that because you would find it anyway — the whole paper is up — and because it is the honest shape of the risk. The failure is not that a machine cannot read Shakespeare. It read it fine. The failure is that on a question requiring a count, it credited the content it found and did not check the arithmetic of the demand. That is precisely the sort of thing a teacher catches in two seconds while approving, which is why approval is not optional and why there is no setting that removes it.
So what is it actually for, in literature
Not for deciding who deserves the top band. For the forty scripts underneath, where the work is repetitive and a tired teacher marking at eleven at night is measurably less consistent than at four in the afternoon. Every student gets a written comment on every question instead of a tick and a total, a teacher reads and adjusts rather than composes, and the disagreements — there will be some — are visible on the screen rather than buried in a total.
Common questions
Can AI mark essay-type literature answers?
It can mark them against the points a scheme names, which is what most school and board literature schemes actually ask for. What it cannot do is rank two well-argued readings against each other, and automated essay scoring in the sense of predicting a grade is the weakest thing you could ask of it. Treat the output as a structured read, and keep the judgement.
Will it penalise a student for poor English?
Not for the content marks. The example above is the pattern: awkward expression noted in the feedback, full marks awarded because the required ideas were there. If your paper has separate marks for language, those are set by your question paper and marked on it — the scheme is derived from your paper, not from an assumption about how literature should be weighted.
What about a student who argues something unexpected but defensible?
This is the weakest case in any subject and the most acute in literature, and we will not pretend otherwise. A reading the scheme did not anticipate can be under-marked. It is the single best reason to read the flagged questions before releasing marks, and the reason the marks quote the words on the sheet that earned them — so a teacher can see what was credited without re-reading the whole script.
Does it work for Hindi, Marathi or Bengali literature?
Not something we claim. Our testing has been on English-medium answer sheets, so treat Hindi, Marathi, Gujarati and Bengali literature alike as untested here. ClassPulse is built for English-medium sheets today.
Send a literature paper you have already marked.
That is the subject to test it on, precisely because it is the one you doubt. Put our marks beside yours and see where we disagree. The first paper is free.
We reply the same day. The whole marked paper · ICSE checking