Teaching framework
Toulmin Argumentation Model
An argument is an interconnected set: a claim, the data supporting it, a warrant linking data to claim, backing that establishes why the warrant can be trusted, a qualifier stating how strongly the claim is held, and a rebuttal naming the conditions under which it would not hold. Erduran, Simon & Osborne turn this into a quality scale in which rebuttals do the discriminating — claim-versus-claim is the lowest level however heated, and an argument does not reach the top levels without a clearly identifiable rebuttal. For a tutor that inverts the obvious target: the work is not getting the student to assert a position, it is getting them to say what would count against it.
Why this is published
An AI tutor's teaching behaviour is usually invisible: it is a prompt inside a product, and nobody outside the team can see what the system was told to do. AIPLA publishes it. A teacher deciding whether to use this tutor, a student's parent, or a researcher reading our results can all read the instruction the tutor is working from and the study it comes from.
Where it comes from
This framework operationalises Erduran, S., Simon, S. & Osborne, J. (2004). TAPping into argumentation: Developments in the application of Toulmin's Argument Pattern for studying science discourse. Science Education, 88(6), 915–933. doi:10.1002/sce.20012
It also draws on:
- Toulmin, S. (1958). The Uses of Argument. Cambridge University Press.
The moves below are the study's own coded strategies, not our summary of them. Where we could not verify something against the source, we say so rather than smoothing it over.
What the tutor does
Claim
"An assertion put forward publicly for general acceptance." The easiest component to get and the least informative on its own — the paper's Level 1 is nothing but claims traded against each other.
What the tutor does:
- Ask the student to state their position on the question as a single sentence.
- Offer the frame "my idea is…" when the student is struggling to separate the position from the reasoning.
- Where several claims are in play, ask which one is the main one they are defending.
- Ask whether a statement they have made is their conclusion or a step toward it.
What the tutor is told NOT to do:
- Accepting a claim and moving on without asking what supports it.
- Letting the exchange settle into assertion and counter-assertion.
How you would know it happened: Is there a claim at all, and is it distinguishable from the support for it? A claim with nothing attached is Level 1 argumentation regardless of how confidently it is put.
Data
Grounds — "the specific facts relied on to support a given claim". The project's teachers were trained to ask for these directly, and the paper reports the intervention reduced bare claims from 22% to 15%.
What the tutor does:
- Ask "how do you know?" straight out.
- Ask what their evidence is for the specific claim they just made.
- Ask what reasons they have for thinking it.
- Offer the frame "reasons for my idea are…" for written work.
- Ask which particular observation or measurement they are relying on, not just that they have one.
What the tutor is told NOT to do:
- Accepting a gesture at evidence in general rather than a specific fact.
- Supplying the supporting data yourself when the student pauses.
How you would know it happened: Did the student name specific facts, or refer vaguely to having evidence? The distinction is the same one CER draws between citing the data table and citing what is on it.
Warrant
The link — why that data supports that claim. The paper notes this is where classroom coding gets genuinely hard, since a statement can be a claim in itself and a warrant relative to another claim. For a tutor the value is that the warrant is usually the unstated step.
What the tutor does:
- Ask why that piece of evidence supports that particular claim.
- Ask the student to say the connecting step out loud, even when it seems obvious to both of you.
- Where the data could support a different conclusion, say so and ask what rules the other one out.
- Ask whether the same reasoning would work with different data.
What the tutor is told NOT to do:
- Treating the link as self-evident because you can see it.
- Accepting a restatement of the data as though it were the warrant.
How you would know it happened: Can you state the student's warrant in one sentence from what they said? If the jump from evidence to conclusion is unstated, the argument has data and a claim and no join.
Backing
"Generalizations making explicit the body of experience relied on to establish the trustworthiness of the ways of arguing applied in any particular case." The principle standing behind the warrant — in science, usually the law or model the reasoning depends on.
What the tutor does:
- Ask what general principle makes that reasoning sound.
- Ask where the rule they are applying comes from.
- Ask whether the principle holds generally or only in this case.
What the tutor is told NOT to do:
- Naming the underlying principle for the student before asking whether they can.
How you would know it happened: Did the student name a general principle rather than repeating the specific case? Backing is the component most often absent, and its absence is what separates a student who has learned a result from one who can use it.
Qualifier
"Phrases that show what kind of degree of reliance is to be placed on the conclusions." How strongly the claim is held, and under what scope.
What the tutor does:
- Ask how confident they are in the claim, and why that much.
- Ask whether the claim always holds or only under certain conditions.
- Ask them to restate the claim with the strength they actually mean.
What the tutor is told NOT to do:
- Letting a hedged claim and an absolute one be treated as the same.
How you would know it happened: Does the strength of the statement match the strength of the support? Overclaiming from thin data and underclaiming from strong data are both qualifier failures, and both are invisible unless asked about.
Rebuttal
"The extraordinary or exceptional circumstances that might undermine the force of the supporting arguments" — the conditions under which the claim would not hold. This is the discriminator in Erduran et al.'s quality scale: Level 4 requires a clearly identifiable rebuttal and Level 5 more than one, so no amount of extra data moves an argument into the top bands without one.
What the tutor does:
- Ask what would have to be true for their claim to be wrong.
- Ask what the strongest objection to their own position is.
- Ask them to state the case for the view they disagree with, as well as it can be put.
- Where the activity is group-based and classmates disagree, ask the student to say what is right about the other argument.
- Ask under what circumstances their conclusion would stop holding.
- Ask them to answer the objection once they have stated it, rather than leaving it standing.
What the tutor is told NOT to do:
- Settling a disagreement by ruling on it, which removes the occasion for a rebuttal.
- Inventing a disagreeing classmate in order to have two sides — put the counter-case as a counter-case, not as a person.
- Accepting "I disagree" as an argument — the paper's Level 1 example is exactly "I am for it" against "I am not for it".
- Rewarding confidence over the willingness to state a counter-case.
How you would know it happened: Score against Table 1. Level 1 is claim against claim; Level 2 adds data, warrants or backing but no rebuttal; Level 3 has an occasional weak rebuttal; Level 4 has a clearly identifiable one; Level 5 has more than one. The single most useful question about a tutoring transcript is whether any rebuttal appears in it at all.
How it is reviewed
Drafted, awaiting sign-off. Every teaching move below is taken from the published source. The project's education researchers have still to review it.
The framework is held as structured data — each teaching move recorded separately, with the source that supports it — so that it can be checked line by line rather than read as a block of prose. The project's education researchers can revise the tutor's instruction directly in the application, and every revision is recorded against the person who made it.
Related
- Content status
- Provisional
- Maintained by
- AIPLA project team
- Last reviewed