Italian POS viewer

Offline Italian grammar trainer: Tatoeba sentences → spaCy POS/lemma + LLM word alignment → a single self-contained HTML viewer (hover for POS colors, lemma, alignment).

Install

• uv sync — pulls spaCy, both IT models (sm + md), polars

Try the demo

• open viewer.html in a browser, load sample_sentences.json

Build your own deck

• uv run fetch_data.py → it-en-pairs.tsv (downloads Tatoeba, ~2 min)
• cp .env.example .env, add your OpenRouter key
• uv run pipeline.py --input it-en-pairs.tsv --n 200 --output sentences.json
• load sentences.json in the viewer

Data sources

• Tatoeba IT-EN pairs — auto-fetched by fetch_data.py; manual fallback: Tatoeba export tool
• freq/ — itWaC frequency lists (MIT), drive the vocabulary filter

Notes

• Phase 0 rationale (freq filter, md model, dual-model gerund override): see pipeline.py docstring
• dev/disagreements.py — sm-vs-md POS diagnostic