Language learning resources — an open dataset

Around 800 curated language-learning resources across 30+ languages, as JSON and CSV: apps, courses, podcasts, YouTube channels, books, dictionaries, tools and communities — each with honest notes and machine-readable facets (type, CEFR, level, skills, method, price).

A list of links is a bookmark folder. This is meant to answer "what should I use, at my level, for this language" — and to stay queryable while it does.

The data behind howtolearn.app, under CC BY 4.0.

Contents

• Browse it
• Get the data
• Examples
• Documentation
• License

Browse it

The whole dataset as pages you can read on GitHub, one per language — with the notes and the caveats, not just the links: catalog/ · Spanish · Japanese · French · all 33 languages

One table per language holding every field: type, level, CEFR, price, skills, method, platform links, pitch and the full notes. Facets first so they stay on screen; the prose columns are at the end, since those are the ones you read rather than scan. Generated by node scripts/catalog.mjs — corrections go to data/resources/, never to the Markdown.

Get the data

File	What it is
resources.json	every resource in one array
resources.csv	one row per resource, for spreadsheets
data/resources/*.json	the source of truth, one file per language

The root files are generated and committed, so nothing needs building. After editing data/, regenerate with node scripts/build.mjs — plain Node, no dependencies. Exact counts live in stats.json; the numbers above are approximate on purpose, so they cannot go stale here.

Examples

A record, trimmed:

{
  "language": "spanish",
  "id": "notes-in-spanish",
  "name": "Notes in Spanish",
  "url": "https://www.notesinspanish.com/",
  "types": ["podcast", "course"],
  "levels": ["beginner", "intermediate", "advanced"],
  "cefr": "A1–C1",
  "skills": ["listening", "vocabulary", "speaking"],
  "price": "freemium",
  "methods": ["comprehensible-input"],
  "pitch": "Unscripted conversations between Ben and Marina, graded across three clear levels.",
  "notes": "… the audio is free; worksheets and transcripts are sold as paid packs, which is the main catch …",
  "links": [{ "label": "Apple Podcasts", "url": "https://podcasts.apple.com/…" }]
}

Free podcasts a beginner can start with, in any language:

jq -r '.[] | select((.types | index("podcast")) and .price == "free"
                    and (.levels | index("beginner")))
       | [.language, .name, .url] | @tsv' resources.json

Every RSS feed, one per line — a ready-made podcast OPML source:

jq -r '.[] | .links[]? | select(.label == "RSS") | .url' resources.json

Everything teaching Japanese through comprehensible input:

import json
rows = json.load(open("resources.json"))
[r["name"] for r in rows
 if r["language"] == "japanese" and "comprehensible-input" in r.get("methods", [])]

More recipes, the CSV columns and the full record: docs/usage.md.

Documentation

• Schema — every field, type and allowed value
• What is in it — counts by type, method and language
• Using the data — recipes, CSV columns, repo layout
• How it was curated — selection, link verification, link check
• Limitations — what the data is not, and known issues
• Contributing — corrections are the main thing this repo wants

License

Data — CC BY 4.0. Use it commercially, remix it, build on it; attribution is the one requirement. Code (scripts/) — MIT.

Language learning resources dataset, curated by How to Learn, licensed CC BY 4.0.