A language is not a list of words — it is a machine for making words: a finite kit of affixes that recursively bolt onto roots, flipping part of speech at each click. And the result chains — nation → national → nationalize → nationalization. The shape of that derivation graph is a portrait of how a language builds meaning, and it differs measurably between languages.
Pick a signature root and watch its derivation tree grow outward — each edge a real, POS-annotated MorphyNet edge, labelled with the affix that built it and tinted by the part of speech it produces. Hover a word to lift its lineage back to the root; click it to open its Babel entry; click a dashed node to load its deeper children. This is what's attested in MorphyNet — not every word the language could make.
Count every derivation edge by the part of speech it goes from → to, and a signature appears. English does its heavy lifting noun→noun and noun→adjective; Russian's single largest move is verb→verb, because Slavic builds via verbal aspect and prefixation. Same affix-bolting machine — demonstrably different gears.
Productivity here is honest: the count of distinct words an affix has been attested to build (not raw edge tags). Each affix carries its grammatical job — its dominant POS shift and the real share, long tail and all. Click an affix to see real chains it sits in, then jump into the forest.
Start from a root and add one real next-affix at a time. Each rung is checked against an actual Babel /word/ entry (measured 99% hit rate) — built words that exist link out; the rare miss is shown but not faked.