Patterns that combine verbs, nouns, and modifiers into full clauses.
Building Kifuliiru sentences means combining pieces you already have — conjugated verbs, pronouns, nouns, adjectives, adverbs — according to patterns (SV, SVO, verb + place, verb + time, questions, negatives).
It does not invent new grammar; it instantiates templates.
When estimating how many distinct sentences can be built in principle:
S_gen ≈ |verbs| × |templates| × |tenses|Real output is smaller because:
Two kinds of output matter:
| Kind | Meaning |
|---|---|
| Result | A sentence built from the pattern draw even if semantics are shaky after retries |
| Product | Passed semantic compatibility — suitable as a finished learning line or publishable row |
Gate in words:
product ⇔ validateSemantic(verb, subject, object, adjective)Products are what you ship to learners; results are useful for QA and bulk sampling.
When importing corpus pairs to discover templates:
PES = (P_valid / P_total) × Q_matchHigh PES means automated sentence expansion is safe; low PES means humans should fix templates first.
Generation expects your word lists:
Without that vocabulary, open-ended generation cannot run — by design.