How etymfun works

A graph, not a list

Every word or character is a node; every arrow is a claim from a source: inherited from, borrowed from, made of these parts, or for Chinese characters meaning component and sound component. Click any node to zoom in on it and its neighbours; double-click to open its own page.

Evidence first

The graph is built from Wiktionary’s structured etymologies (via Wiktextract), the 說文解字, historical glyph forms and dictionary data. A local language model (gemma4, running on our own hardware around the clock) then reviews each parsed relation against the source text and writes a readable explanation. Each statement in an explanation must quote its evidence; explanations that fail the check are rewritten or discarded. Attested forms, reconstructions (marked *) and uncertain links are kept distinct.

Sources and licences

• etymagent (local gemma4:26b) — Generated
• Structural decomposition (chaizi) — See upstream
• Historical glyph forms (zdic.net 字形演变) — Rights unverified
• HSK / TOCFL word lists — Facts (levels)
• 教育部《國語辭典簡編本》 (Taiwan MOE) — CC BY-ND 3.0 TW
• 說文解字 Shuowen Jiezi (Xu Shen, c. 100 CE) — Public domain
• Written form — n/a
• Wiktionary (via Wiktextract / kaikki.org) — CC BY-SA 4.0
• wordfreq — CC BY-SA 4.0 (data)

Data snapshot: 2026-10-29 14:35