Created: 24-09-2026 · Last updated: 24-09-2026

title: Etymology component census toc: false

Etymology component — European derivation vs indigenous vyutpatti

A full entry-level census (H5334) of where and how nine Cologne dictionaries state etymology: the share of entries that state one at all, the surface marker that announces it, the type of statement, and the position the statement occupies inside the article — then the European and the indigenous treatment compared on the shared-lemma set.

This is the measured successor of two narrower pieces of prior art, and replaces neither:

• Etymology style witness (/tools/etymology-style on the site) renders a frozen five-dictionary marker spike vendored from csl-observatory (one hand-read style label per dictionary, no position, no shared lemmas).
• data/etymology-oracle.json (npm run build-etymology-oracle) consumes csl-orig's *_etymology.tsv extraction layer and answers which root a dictionary attributed — on the subset an extractor could parse. It does not say how many entries carry etymology at all.

Generated by scripts/build-etymology-census.py over the csl-orig display text; npm run build-etymology-census regenerates it.

Trust block

• Scope. Every <L>…<LEND> entry of each dictionary's csl-orig display text, article body only (the repeated headword before ¦ is cut).
• Credit rule. A type is credited once per entry, at the offset of its first marker. Types are not mutually exclusive, so per-type counts can sum above entriesWithEtymology.
• Markers are surface evidence, not semantics. A marker census measures how the dictionary announces etymology, not whether the etymology is right. Root agreement between dictionaries is the separate, established result owned by csl-orig's etymology_stats and consumed in data/etymology-oracle.json.
• Known floors and ceilings. MW's rate is diluted by ~287 k entries, most of them compounds carrying no derivation; PWG's near-zero cognate count is genuine (Böhtlingk–Roth state Sanskrit sources, not Indo-European ones); AP90 has no <lang> markup at all, while the 1957 AP revision does.
• Not covered. ARMH and ABCH (verse thesauri — no etymological component to measure) and dictionaries without a derivation-bearing display text.

How often does each dictionary state an etymology?

Share of entries carrying at least one etymological marker, oldest first.

Which type of statement?

root-reference = an explicit verbal-root or base attribution · cognate = a comparative reference to a non-Sanskrit etymon · vyutpatti = an indigenous derivation formula (kāraka + kṛt affix, a dhātu citation) · affix = a morphological decomposition with no root attributed · nirukta-citation = a European dictionary citing the indigenous tradition rather than deriving.

Where in the entry does it sit?

initial = the first marker falls in the first 15 % of the article body (or its first 120 characters) — the etymology is stated before the gloss, the way Apte's bracket and the SKD formula do. final = the last 15 %, the way a comparative block is appended after the senses.

Markers actually found

The surface string each count rests on. A zero is kept when its absence is the finding (PWG's comparative vgl., AP90's missing <lang>).

Shared lemmas — European vs indigenous

The lemmas present in all four of : the only place where the two traditions can be compared on the same word rather than on different vocabularies.

Which type does each side reach for, on the same lemma?

Sample — lemmas etymologised by all four

Headwords in SLP1 as keyed by csl-orig; each cell lists the types that dictionary states for the lemma.

Dr. Mārcis Gasūns