QuickChat + Sentence Engine

Child tapped "want" on the bottom edge. Bubble shows the composed word + 4 noun predictions ranked by kitchen scene context. Dashed borders = predictions from the sentence engine. Child taps a prediction to add it, then taps ▶ to speak the 2-word phrase. Tier 2 max = 2 words.

Child tapped "I" then "want". Bubble shows both composed words + 6 predictions ranked by scene context + grammar transitions. More predictions at Tier 3 because children at this level handle more choices. Multi-word building with grammar-aware next-word prediction.

After speaking "Want pizza", 3 follow-up cards rise from center stage. Each is a complete 1-2 word sentence spanning a different pragmatic function: requesting, commenting, social. One tap speaks the whole thing. Scene dims 15% to draw focus. Tapping a card speaks it → tray refreshes with new follow-ups. 3 options matches clinical validation for choice-making research + Switch Control compatibility.

After speaking "I want pizza", 6 follow-up cards appear in a 3×2 grid + scrollable strip. Grid shows top-ranked engine-generated sentences spanning all pragmatic functions. Strip catches topic shifts and overflow ("More things to say..."). Each card is one tap to speak. Tier 3 children communicate rapidly — QuickChat becomes an accelerator, not just a scaffold.