ja-panel-corpus: 28 model-written Japanese passages and 8 Aozora Bunko Test sentences #50

Merged
weksa merged 3 commits from weksa-lowering/ja-panel-corpus into main 2026-11-12 13:52:37 +00:00

2026-11-12

Weksa
data/original: ja passages replaced, none a rendering of an en-US original or a Dao line
All 28 Train/Validation ja passages are new Japanese sentences on the inventory's
concepts with their own imagery (paddies, charcoal, kites, the bath, the market); none
renders an en-US original, and none restates a data/daodejing line (ch3, 5, 7, 8, 78) or
other canon (the old ones on 'true words', 'long-enduring', 'bellows', 'water vs hard
things' are gone). corpus.cc restored from 21eaa07 and rebuilt through weksa corpus add:
numbering 1..28 unbroken, holdout-ja:1..8 and original:1..28 byte-identical.

The concept test now matches Japanese by multi-character surface and English by word
stem, strips false friends (天気, 心配, 腹が立つ, 水曜 ...), and has a negative test.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
2026-11-12 13:51:06 +00:00
Weksa
data/original: ja passages rewritten onto the Dao r3 concepts; ja Test sentences pinned verbatim
The 28 Train/Validation ja passages are new Japanese sentences on the concepts the
pinned inventory holds (water, mind, belly, bone, sage, will, heaven, earth, fill, empty,
weaken, strengthen, compare, resemble ...), written in Japanese idiom, not translated.
The branch was unmerged, so corpus.cc was restored to 21eaa07 and rebuilt through
weksa corpus add; holdout-ja:1..8 and original:1..28 come out byte-identical.

Tests: the eight ja Test texts, groups, speakers, split and edition URLs are literal
expectations; every passage of both languages rests on a live inventory concept
(en 28/28, 18 concepts; ja 28/28, 21 concepts).

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
2026-11-12 13:51:06 +00:00
Weksa
data/original: ja-JP slice, 28 model-written Japanese passages and 8 Aozora Bunko Test sentences
Train and Validation (14 each) are Japanese passages written in Japanese, ModelWritten
Anthropic, generator claude-sonnet-5-5; none sits in a Test group. Test is eight
one-sentence excerpts (Soseki, Kyoka, Akutagawa, Ogai, Arishima, Kenji, Kajii, Kunikida)
taken from the card ruby zips (cp932, ruby and the pipe marker dropped, NFC) and checked
verbatim against each card's XHTML file; licence strings keep 底本, 入力 and 校正 within
the store's 200-byte cap. Entities appended (keeper gains 番人; eight pd-* authors).
Items 1..28 and holdout:1..8 unchanged (digest pinned). Tests per language: counts,
provenance, Test public-domain only, no markup, panel round reads both languages, speaker
forms, recall 56, no Test text in any prompt.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
2026-11-12 13:51:06 +00:00