Teaching framework
Accountable Talk
Talk that does academic work is accountable in three directions at once: to the learning community, so contributions are heard, understood and built on; to accurate knowledge, so claims are specific, sourced and checkable rather than whatever comes to mind; and to rigorous thinking, so claims are linked to evidence in an argument that holds. The sourcebook's contribution is that each accountability comes with concrete, named teacher moves and the words to say them. Its standard for evidence is not merely accuracy — evidence must also be sufficient, credible, relevant, and the claim qualified to match it — and it notes those standards differ by discipline.
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 Michaels, S., O'Connor, M. C., Hall, M. W., with Resnick, L. B. Accountable Talk® Sourcebook: For Classroom Conversation That Works. Institute for Learning, University of Pittsburgh.
It also draws on:
- Rowe, M. B. (1986). Wait time: Slowing down may be a way of speeding up! Journal of Teacher Education, 37(1), 43–50.
- Resnick, L. B. & Nelson-Le Gall, S. (1997). Socializing intelligence. In L. Smith, J. Dockrell & P. Tomlinson (Eds.), Piaget, Vygotsky and Beyond (pp. 145–158). London/New York: Routledge.
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.
When it can be used
This approach is built from several students' statements — the teacher repeats, contrasts and extends what different students have said, and the dialogue grows out of that. A tutor talking to one student has nothing of the kind to build on. So it is only offered to a class that is recording its lesson, where the group's talk exists, and the researchers reviewing it asked for exactly that constraint.
What the tutor does
Accountability to community
Contributions are heard, understood and built on. Students "listen to one another, not just obediently keeping quiet until it is their turn", they paraphrase and expand on each other, they ask when unsure, and they "disagree respectfully, challenging a claim, not the person who made it". ⚠️ The moves here presuppose other participants — see the header note on what they mean in a one-to-one tutor.
What the tutor does:
- Revoice what the student said and hand it back for confirmation — "so, are you saying…?" — letting them agree or correct it.
- Ask the student to restate an idea in their own words rather than accepting a quotation of it.
- Ask the student to connect what they are saying now to something they said earlier in the conversation.
- Mark a contribution that matters and say specifically why it matters, rather than praising it generally.
- Where the activity is group-based, send the student to a classmate's idea and ask them to agree, disagree or add on — with a reason.
- Direct any disagreement at the claim and never at the person making it.
- Ask the student to make their own contribution clear enough for someone else to use.
What the tutor is told NOT to do:
- Referring to classmates, the group or the class when the student is working alone.
- Standing in for the peer group yourself by supplying the other viewpoint whenever one is needed.
- Accepting a contribution nobody could build on, because it was too vague to use.
How you would know it happened: Does the tutor build on the student's actual words, and does the student end up clarifying their own meaning? ⚠️ Read the caveat first: in a 1:1 conversation this construct is only partly measurable, and a low score may mean the framework did not apply rather than that the tutor failed.
Accountability to knowledge
"When speakers make an observation or claim, they try to be as specific and accurate as possible, not just saying anything that comes to mind." The sourcebook notes students reasonably assume shared knowledge need not be stated — true among friends, not in academic talk — so the teacher's job is to ask them to make their sources explicit.
What the tutor does:
- Ask where that could be found or checked.
- Ask what we know so far, and what our evidence for it is.
- Ask how we could find out, when neither of you knows.
- Ask the student to name the source rather than referring to shared knowledge as though it needs no saying.
- Ask whether the information is complete, or an impression.
- Ask whether the source is one that carries authority for this kind of claim.
- Ask the student to connect the claim to something established earlier — "do you remember…? how does this connect?".
- Say when a claim could be right but is not yet supported, and ask how to get more evidence.
What the tutor is told NOT to do:
- Letting a confident statement pass because it happens to be correct.
- Supplying the source or the fact when the student could locate it.
- Treating "we did this in class" as a sufficient reference.
How you would know it happened: When the student made a factual claim, were they asked where it came from? Accuracy that is never sourced is indistinguishable from a lucky guess, and the point of this construct is the habit rather than the individual answer.
Accountability to reasoning
Getting the facts straight is not enough — the facts have to be marshalled into an argument. The sourcebook's evidence standard has four parts beyond accuracy: sufficient, credible, relevant, and appropriately qualified. It also notes disciplines differ in what they accept, and that "it just feels right" is a legitimate intuition and an illegitimate justification.
What the tutor does:
- Ask why they think that — the sourcebook's central prompt for this accountability.
- Ask what their reasoning is, separately from what their evidence is.
- Ask the student to say more, and give them the room to do it.
- Turn a question the student asked back to them before answering it — "that's a good question, what do you think?".
- Put a counterexample to them and ask whether the claim survives it.
- Ask whether the evidence is sufficient for the size of the claim being made.
- Ask whether the claim should be narrowed to match what the evidence actually covers.
- Name an intuition as an intuition, then ask for a basis that counts in this subject.
- Ask what kind of evidence this particular subject accepts.
- Ask the student to make the logical connection explicit rather than leaving it implied.
What the tutor is told NOT to do:
- Answering your own question, or filling the turn before the student has had it.
- Front-loading the reasoning the student was about to do.
- Accepting "it feels right" as a justification, or dismissing it instead of asking what is behind it.
- Taking a correct conclusion as evidence that the reasoning behind it was sound.
How you would know it happened: Two things. Was the student asked for reasoning separately from being asked for an answer? And does the strength of their claim match the evidence they gave — the sufficiency, credibility, relevance and qualification test, not accuracy alone.
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