Entanglement

The whole classification — every branch point, every language, every parent-to-child edge — thrown into one space and left to resolve itself. What you saw on load was it settling: 12,575 nodes and 12,154 edges finding their own arrangement.

The 421 families share no edges with each other at all. Nothing links them, nothing holds them apart, and nothing arranges them — so what keeps them separated is only the pressure they exert on one another, and they end up threaded through each other rather than sorted into neat regions.

Click any node and its whole family lights — every sibling and cousin in the component, not just the line it sits on — with its own path from the family root picked out in white over the top. Selecting one English language lights all 941 nodes of Indo-European, scattered right across the field, which is the clearest way to see how thoroughly the families are threaded through each other.

The radial tree draws the same relationships forced into a circle, which makes families comparable and makes them all look alike. This is the other extreme: no imposed geometry at all, so the only structure visible is the structure that is actually there.

Structure from Glottolog 5.3. Colour is family, the same colour it carries on the other three pages. The layout is computed in the browser each time — Barnes-Hut repulsion over a quadtree, rebuilt every tick — from a seeded start, so it resolves the same way every time without being precomputed. Non-genealogical buckets are left out, since their members are not relatives.