Dialogue Factory

Two speakers, one scene, no script. Students rehearse an AI-generated dialogue, then progressively lose the lines — until they own it from memory.

Retrieval practice beats re-reading at every level

Roediger & Karpicke (2006) show that testing memory — even with partial cues — produces dramatically better long-term retention than re-reading the same material. The gap-fill mechanic is a retrieval practice task; each time students reconstruct a blanked word from context, they strengthen the memory trace more than reading the word passively would.

Dialogic rehearsal develops pragmatic competence

Kasper & Rose (2002) argue that pragmatic competence — knowing what to say, how to say it, and when — develops through exposure to and rehearsal of authentic conversational patterns. Dialogue Factory provides level-appropriate authentic patterns and requires students to internalise them through progressive memory load.

From controlled to automatic processing

DeKeyser (2007) models skill acquisition as a progression from controlled, effortful processing to automatic retrieval. The three-stage Dialogue Factory progression maps directly onto this model: full text (controlled reading) → partial gaps (controlled retrieval) → no text (automatic production).

Generates 4-turn dialogues in everyday situations (café, shop, school). Gaps target common phrases and basic past tense forms. One gap per turn keeps the memory demand manageable.

A2 learners need exposure to how everyday transactions sound in natural English. The gap-fill ensures they actively process each phrase rather than skimming — and the memory performance makes those phrases productive, not just familiar.

Generates 4–5 turn dialogues in professional or functional scenarios. Grammar focus is placed on modal verbs (could, would, should) — gaps are positioned on exactly these forms.

B1 learners use modals but default to 'can' in every situation. Grammar-targeted gaps force them to encounter and produce 'could' and 'would' in the exact functional contexts where they are natural — making the grammar upgrade feel communicatively necessary rather than arbitrary.

Generates 5–6 turn dialogues in nuanced situational roles. Gaps target idiomatic expressions, hedging devices, and register markers. Two gaps per turn require sustained attention across the dialogue.

C1 learners often speak fluently but with a limited pragmatic register — they say 'I think' when 'it might be worth considering' is what the context calls for. Gaps on hedging expressions make the pragmatic choice visible and retrievable.

Role Reversal

After completing the standard dialogue, pairs swap roles and perform it again. The language that came naturally as Role A often needs effort as Role B — the reversal exposes asymmetries in students' knowledge of each role's language.

Dialogue Extension

After the memory performance, pairs must add two extra turns to the dialogue that continue the scene naturally. The extension must use at least one word from the gap-fill. Tests whether the vocabulary from the dialogue is now productively available.

Cross-Dialogue Compare

After working on all three generated dialogues, the class identifies which phrases appeared in more than one dialogue. These are the most functionally productive phrases of the session — they deserve explicit attention and repetition.

Reconstruction from Memory

Without seeing the original dialogue again, students write the complete text from memory as a homework task. Compare their reconstruction to the original at the next lesson — gaps and variations reveal exactly what was retained and what was lost.