Zoe is 3 years old. She uses dozens of words reliably and has started combining them on her own. Her SLP has moved her to Tier 3. The sentence engine is active.
Zoe builds a sentence
She doesn't just want milk. She wants cold milk, and she wants to say please.
1Zoe taps "want" from the bottom edge. The sentence engine predicts what comes next based on grammar rules.
2Speech bubble shows grammar-aware options — nouns from the kitchen scene, descriptors that pair with "want":
🥛 milk
🥣 cereal
🧊 cold
👍 more
3Zoe taps "cold" from the top edge, then "milk" from the scene. The speech bubble shows: want → cold → milk.
4Zoe adds "please" from the bottom edge. Taps the speak button.
5"Want cold milk please." — four words, natural prosody. Her mother says "Cold milk with please? Coming right up."
This is generative language — Zoe assembled a novel sentence that no one programmed. The sentence engine helped by offering predictions, but Zoe chose her own path.
At Tier 3, the engine shifts from templates to grammar-aware predictions. It understands word order (verb → adjective → noun) and suggests accordingly. The child constructs novel utterances with scaffolding — the hallmark of the 24-48 month language explosion (Bates & Goodman, 1997).
Zoe tells Mama
Communication isn't just requesting. It's directing.
1Zoe taps "mama" from the right edge, then "help" from the bottom.
2Sentence engine predicts extensions:
👆 open
🧃 juice
🙏 please
3Zoe taps "open" then "juice."
4"Mama help open juice." — four words, directed at a specific person, about a specific action and object. Sophisticated pragmatic language.
Person + action + object combinations (agent-action-object) are a key milestone at MLU 2.5+, emerging 24-36 months. The sentence engine scaffolds word order so the child can focus on meaning, not grammar.
