kennethreitz 94a4594293 Exhaustive offline thesaurus + server-side describes/associations ...

Thesaurus: Wiktionary links symmetrized (A lists B => B gets A;
hypernym/hyponym invert), Moby Thesaurus II (public domain) merged as
a bulk layer ranked behind the curated links, headwords widened to
zipf>=2 plus anything Wiktionary tags as slang (yeet, rizz, finna).

Describes and associations move off Datamuse onto the server:
scripts/build_associations.py distills the Tatoeba corpus (CC BY,
2M sentences) into adjective-noun salience and windowed co-occurrence
PMI tables. New lookup modes desc/trig; the frontend makes no
third-party calls at all now.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

2026-06-12 00:09:52 -04:00

3.2 MiB

Raw Permalink History

View Raw