polyglott

Learn a new language with ALL the languages you already know. Open-source multilingual language-learning platform.

---

The one-sentence pitch

Every language-learning app forces you to pick one source language and learn the target from it. Polyglott lets you learn from all the languages you already speak, using cognate networks, grammar bridges, and cross-linguistic transfer to move faster than any 1:1 app can.

Why it matters

• ~60% of the world is multilingual (~4.8 billion people).
• Every major app (Duolingo, Babbel, Busuu, Memrise) assumes you have one native language and one target. That's a constraint from 1990s edtech, not from how the brain actually learns.
• Decades of applied linguistics research — Cumulative Enhancement Model (Flynn 2004), Linguistic Proximity Model (Westergaard 2017) — show that all prior languages contribute, property-by-property, to learning a new one.
• Nobody has built this yet.

See docs/market-research.md and docs/user-research-multilingual-pain-points.md for the full case.

The simplest version (v0.1 MVP spec)

One page. No backend. No account. No blockchain. No analytics.

1. User picks their known languages from a dropdown (e.g. Spanish, Portuguese, French).
2. User picks their target (e.g. Italian).
3. User sees ONE word at a time from a hardcoded deck. For every word, the UI shows:
   • The Italian word.
   • The cognates in each known language that the user speaks.
   • A warning for false friends.
   • A short etymology line showing the shared root.
4. User clicks "I knew this" or "I didn't" (Leitner-style SRS).
5. Next word. Repeat.

That's it. The activation moment is step 3 — the "this app sees me" moment — where a Spanish + Portuguese + French speaker learning Italian sees a word like "acqua" radiating to agua / água / eau with the Proto-Italo-Celtic root akwā- underneath.

No step 3b where you sync with a server. No step 3c where you mint an NFT. Just the flashcard.

If v0.1 is compelling for one user on one language triple, v0.2 adds more words. v0.3 adds more languages. v1.0 adds a backend. In that order. Not before.

Status

Piece	State
Market research	Done (see docs/)
User research	Done (see docs/)
Mission + objectives	Done (see data/mission.json)
Cognate dataset seed	Planned — sourced from Wiktionary + WALS
v0.1 MVP (static flashcard)	Not started. Good first issue.
v0.2 more languages	Not started
v1.0 backend + accounts	Not started

This is pre-product. The research is deep; the code is zero. If you want to show up and build the simplest version, open an issue and claim it.

Data sources (planned)

• Wiktionary — cognate sets, etymologies, translations
• WALS — World Atlas of Language Structures, typological distances
• PanLex — cross-language translation graph
• IDS — Intercontinental Dictionary Series
• Academic SLA datasets (public) where available

All data is open-licensed or CC-BY and will be attributed in the repo.

Prior art

• LeitnerLang — a prior hackathon prototype by the same team. Concepts kept: 32-day Leitner circular queue, concept-based decks. Concepts killed: blockchain everything.

Contributing

See CONTRIBUTING.md.

Good first issues:

• Seed a JSON of ~100 Italian words with cognates in Spanish/Portuguese/French/Romanian.
• Build a single-file HTML prototype of the flashcard experience.
• Write a cognate-schema.json spec for the data shape.
• Document the Leitner SRS algorithm in ALGORITHM.md.

License

MIT. Fork it.