Commands
/linguistic:propose
Generate a phase plan for the current target language (or for a freshly-named one).
Syntax
/linguistic:propose [language]Description
Produce a structured plan covering all 5 phases (Scope → Acquire → Analyze → Evaluate → Release) for the target language, including:
- Recommended data sources (with ethics flags)
- Tokenizer and vocabulary strategy
- Adapter / fine-tune choice
- Eval suite and metrics
- Release gating
Output is a plan.md written to the current workspace, suitable for review and editing before execution.
Options
| Argument | Description |
|---|---|
[language] | Target language (uses workspace context if omitted) |
Example
/linguistic:propose Khmer
> Generating plan for Khmer (khm, Class 2)...
Writing plan.md to workspace...
Plan: Khmer (khm) — 5-phase summary
Phase 0 Scope: ISO khm, Glottolog khmr1253, Class 2, abugida script, Austroasiatic
Phase 1 Acquire: MADLAD-400 khm slice + Wikipedia + CommonCrawl; OFA vocab extension
Phase 2 Analyze: morph (isolating-lo, skip morph skill); syntax (no UD treebank; cross-lingual)
Phase 3 Eval: chrF++ + COMET; Belebele; FLORES+ (flag contamination)
Phase 4 Release: OPEN (CC-BY sources; standard model card)
Full details: workspace/plan.mdRelated Commands
/linguistic:lifecycle— execute the plan/linguistic:decide— force individual decisions within the plan
Was this page helpful?
Edit on GitHub
Last updated on