EPI isn’t ‘sentence builders’. A unit runs modelling → intensive receptive processing (90–95% comprehensible) → forced output → intensive retrieval that weans learners off the builder → optional grammar → planned then unplanned fluency, with recycling across units.
The seven phases
- Model target chunks (sentence builders, parallel texts).
- Receptive processing: ~2 lessons of listening/reading flooded with the chunks until receptive mastery.
- Forced output: highly structured tasks recycling new + old chunks (read-aloud games for decoding only).
- Intensive retrieval (≥1 lesson): oral translation games (No snakes no ladders, Oral ping-pong), drills, structured tasks.
- Optional, highly controlled grammar (needed more for inflected languages).
- Fluency: planned tasks with task repetition (Bygate 2009, 2015), then unplanned tasks.
- Recycle via the “Conti matrix” in later sub-units.
- Warning: a few Quizlet/Kahoot tasks a day is not EPI.
Later framing (2026)
Conti maps each of Rosenshine’s ten principles onto MARS EARS stages; he notes this is conceptual alignment, not validation.
Critique (Claude)
No outcome studies of EPI as a package; Jordan calls the sentence-builder phase a synthetic syllabus in disguise. Its strongest components (input flooding, retrieval, task repetition) each have independent support.
Connections
MARS-EARS - the EPI instructional sequence · Sentence builders and patterns-first lexicogrammar · Check trainers’ SLA claims against primary sources