A static site that puts 7,992 languages and 13,706 dialects on one map, with search in 15 languages, filters, a language card for each entry, and four ways to see how any two languages relate: family tree, grammar (WALS), vocabulary (ASJP) and distance. Built from Glottolog, Wikidata, WALS and ASJP; no backend, no accounts.
Try it: mrwd.github.io/langs-db (live app) · product page
- Search by Russian and English name, ISO 639-3 code, Glottocode and family
("Altaic",
ket,Turkic…). - Filters: macroarea, language family (242), country, number of speakers, level of documentation (is there a grammar/dictionary?), endangerment status (AES scale).
- Map (Leaflet + CARTO): every language is a dot coloured by endangerment status; when few results are left, the map flies to them automatically.
- Language card: family, countries, speakers, the best published description, and links — Glottolog, Wikipedia (ru/en), OLAC (speech recordings and materials), Ethnologue, Endangered Languages Project.
- Light and dark themes, mobile layout, permalinks of the form
#l=<glottocode>. - 15 interface languages: Русский, English, Español, Deutsch, Français, Português, Italiano, 中文, 日本語, العربية (RTL), Türkçe, Polski, Українська, हिन्दी, Indonesia — switcher in the header; statuses, countries and number formats are localised, and language names come from Wikidata labels (33k labels).
- Search in any language: the index includes all localised names plus 84k
alternative names from Glottolog (
altnames.json, loaded in the background) — "wenecki", "Idioma veneciano" and "Altaï méridional" all resolve to the same language. Dialects are always searched, even when the filter is set to "Languages only". - Visit statistics right on the site (the "Statistics" button): visits, visitors, a 30-day trend, and breakdowns by country, city and interface language.
- Geographic hint: if an idiom is not found, the site checks whether the query is a place name and shows the languages of that area within a 200 km radius. That way "sandonatese" leads to San Donà di Piave and "sandonese" to Sandono, and in both cases to Venetian — even though no catalogue lists those languoids. Sources (both free, no API keys): the primary one is Nominatim, which knows what actually is a populated place; the fallback is Wikipedia full-text search, which picks up derived forms of names but also returns unrelated objects with coordinates, so it only kicks in when Nominatim comes up empty. The hint is only shown when the place name and the query share a common stem — accidental street-name matches are filtered out.
- Dialects (13,706): the "Languages and dialects" toggle in the filters; a dialect inherits its parent's status and countries, a language card lists its dialects, and a dialect card links back to its parent language.
- Compare two languages (the "⚖ Compare" button on a card): basic parameters plus grammatical features from WALS (192 features, 2,501 languages) with matches highlighted; for dialects the parent language's data is used.
- Language relations on the card — four independent dimensions:
- branch and relatives — the full Glottolog tree (up to 10 levels: "Indo-European › … › East Slavic"), with relatives taken from the nearest branch;
- grammatically similar — the share of matching WALS features (970 languages with at least 30 features coded);
- lexically similar — ASJP Swadesh lists compared using LDND (5,544 languages, 870k pairs);
- neighbours on the map — the nearest languages by coordinates.
python3 -m http.server 8642 --directory web
# → http://localhost:8642(The server is only needed so the page can load data.json; any static host will
do.)
web/ the entire site (deployable as is)
index.html
app.js logic: filters, table, map, cards, comparison, relations
i18n.js interface translations (15 locales)
stats.js visit counter and statistics panel
style.css themes and layout (including RTL)
data.json the main database: 7,992 languages + 13,706 dialects (4.5 MB)
altnames.json 84k alternative names for search (1.4 MB, lazy-loaded)
wals.json 192 WALS grammatical features (0.8 MB, lazy-loaded)
related.json grammatical and lexical similarity per language (1.6 MB, lazy-loaded)
stats.json aggregated statistics, refreshed daily by CI
vendor/ Leaflet 1.9.4
data/
build_data.py builds data.json and altnames.json
build_wals.py builds wals.json
build_related.py builds related.json (WALS overlap + ASJP/LDND)
fetch_wikidata.py pulls speaker counts and names in 14 languages
collect_stats.py polls the Abacus counters → web/stats.json (for CI)
tz_countries.json time zone → country code (from the IANA zone.tab)
raw/ downloaded sources (glottolog-cldf, wals, asjp, wikidata), not in git
| Source | What we take | Licence |
|---|---|---|
| Glottolog (glottolog-cldf) | languages, dialects, families, coordinates, countries, AES status, MED | CC BY 4.0 |
| Wikidata | speaker counts (P1098), localised names, Wikipedia links | CC0 |
| WALS (cldf-datasets/wals) | 192 grammatical features for comparing languages | CC BY 4.0 |
| ASJP (lexibank/asjp) | Swadesh lists for lexical similarity | CC BY 4.0 |
- AES (Agglomerated Endangerment Status) — a combined endangerment status based on ElCat, Ethnologue and UNESCO data: 1 "not endangered" … 6 "extinct".
- MED (Most Extensive Description) — the most complete published description of a language, from a full grammar (300+ pages) down to a word list.
- Speaker counts exist for ~1,800 languages; for small languages they may be out of date.
cd data
# 1) fresh Glottolog data
curl -sL -o raw/languages.csv https://raw.githubusercontent.com/glottolog/glottolog-cldf/master/cldf/languages.csv
curl -sL -o raw/values.csv https://raw.githubusercontent.com/glottolog/glottolog-cldf/master/cldf/values.csv
# 2) a fresh Wikidata dump (~1 min, 26 SPARQL queries)
python3 fetch_wikidata.py
# 3) rebuild web/data.json
python3 build_data.py
# 4) (optional) refresh WALS for language comparison
for f in values parameters codes languages; do
curl -sL -o raw/wals_$f.csv https://raw.githubusercontent.com/cldf-datasets/wals/master/cldf/$f.csv
done
python3 build_wals.py
# 5) (optional) language similarity: grammatical and lexical
for f in forms languages; do
curl -sL -o raw/asjp_$f.csv https://raw.githubusercontent.com/lexibank/asjp/master/cldf/$f.csv
done
python3 -m venv ../.venv && ../.venv/bin/pip install rapidfuzz # needed once
../.venv/bin/python build_related.py # ~6 minutes: 870k pairsAnalytics is assembled from two parts, neither of which needs an account:
- The browser (stats.js) bumps several counters on
Abacus — a public, registration-free
counter API — on the first hit of a session:
total,uniq,d-<date>,tz-<time zone>,ui-<locale>. No cookies and no personal data; country and city are derived from the browser's time zone, and the IP is never stored. Counting is disabled on localhost so development does not skew the numbers. - A nightly GitHub Action (stats.yml) polls
every key once a day via collect_stats.py and commits
web/stats.json— Abacus has no "list keys" endpoint, so CI assembles the breakdown and the site serves the ready-made file instantly.
Worth knowing: the counter is public (the namespace is visible in the JS, so in theory anyone could inflate it), ad blockers cut off some visits, and the city is inferred from the time zone — that is a region, not a precise location. Swapping Abacus for another service only takes edits to stats.js and collect_stats.py.
The site is published to GitHub Pages automatically: every push to main
triggers the deploy.yml workflow, which
publishes the web/ folder. After a data refresh, committing the new
web/data.json and pushing is enough.
The web/ folder is self-contained, so it can be moved to any other static host
(Cloudflare Pages, Netlify, Vercel) unchanged.
The code is under the MIT licence (see LICENSE). The data keeps the licences of its sources, listed in the table above.