wnja 2.0

Japanese WordNet 2.0

First major release of wnja 2.0, based on NTU-MC data with full WN-LMF 1.4 export.

What's new since v1.1

• Full WN-LMF 1.4 export with correct DOCTYPE
• Stub synsets for all missing PWN relation targets (E401 resolved)
• Improved handling of suru-verb lemmas (detect_sahen.py)
• Merges entries with the same (lemma, pos) across multiple hno homographs
• Passthrough entries for Japan-specific synsets not in wn+var
• 9 definition corrections: 装飾 → 観賞 for ornamental plant synsets
• Build is fully reproducible from source

Known limitations (planned for 2.1+)

• ~364 yojijukugo synsets lack Japanese entries
• W203/W202: ~280 overlapping entries from cross-pipeline duplicates
• W307: 464 cross-synset shared definition texts
• Sense-level antonyms not yet implemented
• Definition quality audit in progress (centaur LLM+human pipeline); corrections will appear in 2.1

Validation (wn validate)

Code	Count	Notes
E401	0	All PWN relation targets resolved
W203	282	Cross-pipeline overlapping entries
W301	8,451	PWN orphans + design stubs
W404	2,509	Stubs have no reverse relations (by design)

Contents

• wnja-2.0.xml — WN-LMF 1.4 file; 89,546 entries, 67,386 synsets