feat(seed): file bilingual senses under one headword #25

Merged

raisondetredev merged 1 commit from refs/pull/25/head into main

2026-08-03 21:01:27 +02:00

raisondetredev commented

2026-08-03 21:01:15 +02:00

(Migrated from codeberg.org)

PONS has no nested sense list — it prints one headword per sense and
tells them apart with a parenthetical: roll, roll (list),
roll (meat), roll (roller), eighteen of them for roll alone.
Seeded verbatim that is eighteen catalog entries, so a lookup returns a
wall of near-identical results and a bulk-add makes eighteen cards for
one word.

Splitting the qualifier off recovers the structure PONS is really
expressing: roll is the entry, (list) and (meat) label its senses.
The base form becomes the expression, the qualifier a per-sense label,
and the existing merge collapses the siblings into one entry with
numbered senses. 114,830 English rows become 102,381.

Whatever follows a parenthetical is kept, so encroachment (incursion) on files under encroachment on rather than encroachment: a
construction with a preposition is a different headword, and PONS lists
it separately too. Square brackets are PONS's notation for optional text
(to encroach [up]on sb), not sense disambiguation, and are left alone.

The merge now keys on expression alone rather than (expression,
reading). PONS prints the IPA on whichever sense-entry it pleases, so
siblings share a base form but not a reading, and keying on reading
would preserve the very fragmentation this exists to undo. The cost is
that true homographs (read /riːd/ vs /rɛd/) collapse into one entry —
which is how PONS files them anyway, one page per spelling. A merged
entry keeps the first non-empty reading and unions the search blobs, so
"boarding school" still reaches house.

Signed-off-by: Tizian 「ティツィアーン」 raisondetredev@pm.me

PONS has no nested sense list — it prints one headword per sense and tells them apart with a parenthetical: `roll`, `roll (list)`, `roll (meat)`, `roll (roller)`, eighteen of them for `roll` alone. Seeded verbatim that is eighteen catalog entries, so a lookup returns a wall of near-identical results and a bulk-add makes eighteen cards for one word. Splitting the qualifier off recovers the structure PONS is really expressing: `roll` is the entry, `(list)` and `(meat)` label its senses. The base form becomes the expression, the qualifier a per-sense `label`, and the existing merge collapses the siblings into one entry with numbered senses. 114,830 English rows become 102,381. Whatever follows a parenthetical is kept, so `encroachment (incursion) on` files under `encroachment on` rather than `encroachment`: a construction with a preposition is a different headword, and PONS lists it separately too. Square brackets are PONS's notation for optional text (`to encroach [up]on sb`), not sense disambiguation, and are left alone. The merge now keys on `expression` alone rather than (`expression`, `reading`). PONS prints the IPA on whichever sense-entry it pleases, so siblings share a base form but not a reading, and keying on reading would preserve the very fragmentation this exists to undo. The cost is that true homographs (`read` /riːd/ vs /rɛd/) collapse into one entry — which is how PONS files them anyway, one page per spelling. A merged entry keeps the first non-empty reading and unions the search blobs, so "boarding school" still reaches `house`. Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>