Nusantara Languages

The problem

Indonesia has more than 700 languages, one of the densest concentrations on Earth. For most people, the shape of that diversity is invisible. Which languages are related? How did the families spread across the islands? Language maps exist, and family trees exist, but nothing links the two into one thing you can explore.

I built Nusantara Languages for students, linguists and anyone curious about where Indonesia's language families meet. The clearest example is the seam where the Austronesian and Papuan families touch, which runs through Halmahera. It is a personal, open-source project. It isn't Badan Bahasa's official "Peta Bahasa", so it doesn't use that name or look.

What I built

The map and the family tree are one object

The map and the tree read from the same data. Hover a branch of the tree and every territory in that group lights up on the map. Click a territory and the tree scrolls to it and opens its ancestry. Because both views read one structure, they can't fall out of sync. I wrote the tree and geometry logic as plain functions, so the trickiest parts can be tested without a browser.

I measured speed before styling anything

The site is static and makes no network requests at runtime. A build step fetches the worldwide sources, keeps Indonesia, simplifies the shapes and writes one data bundle with a coverage report. Before any styling, I measured whether hovering would feel instant with 726 languages drawn. It came in at about 0.01 milliseconds against a 2-millisecond budget. That let me keep plain vector drawing and hand-build both the map and the tree parser.

I dropped the planned atlas over its licence

The plan was to draw territories from a 1980s atlas, the source meant to make the map look good. At the licence check, it turned out to allow non-commercial use only. That conflicts with the open licence the site's data is published under. I refused the source, recorded why on the method page, and switched to two openly licensed datasets from 1990 to 2020. The map's era changed with it. Ethnologue is banned entirely, and a test proves none of its data ships. Where a figure exists only there, the page leaves it out and says so.

A language without a territory stays a point

Of the 726 languages, 421 have mapped speaker areas. The other 305 appear as labelled points, drawn differently. I didn't stretch them into shapes to make the map look complete. The build computes that 58% figure from the data, so it can't drift. The coastline comes from a geographic dataset and is never traced from language areas. I chose family colours to stay distinct for three kinds of colour blindness, and show endangerment as hatching over the family colour.

Result

Nusantara Languages is live in English and Indonesian, fully static, and works offline. Beyond the map and tree, it answers three questions from the same data. Which languages are a language's closest relatives? How far does a family spread? Austronesian spans about 5,010 km, Timor-Alor-Pantar about 350 km. Which languages from different families touch? There are 67 such pairs. Each answer prints its rule and its limit next to the number. You can also colour by subgroup, follow guided views of the seams, and export a PNG with attribution built in, as the licence requires. The site states that boundaries between languages are gradual, and how many languages are areas and how many are points. I built it alone between August and September 2026.