Skip to content

feat(hanzi-structure): normalize D2 kana/bopomofo glyph aliases - #315

Merged
wangzhizhou merged 2 commits into
mainfrom
dev
Nov 14, 2026
Merged

wangzhizhou merged 2 commits into
mainfrom
dev

Conversation

@wangzhizhou

Copy link
Copy Markdown
Contributor

内容

标准 v1 下逐条复查 1,000 字语料的 73 条 positionMismatch,把「D2 用假名/注音符号代替汉字部件」的同形写法归一化:

  • 新增 modules/hanzi-structure/corpus/d2-glyph-aliases.tsv:リ→刂(j)、ㄗ→卩(b)、ク→夕(q);只做同形归一化,不新增字根、不改键位;
  • build_b0.py --glyph-aliases:归一化后仍走标准 v1 的字根清单与码位校验,别名本身不能把条目送进 full;
  • 彑(X 键)加入标准字根清单(彖 xeu/椽 sxey/喙 kxey/篆 txeu 等 7 例);ノ 作撇笔画参与识别码槽判定。

结果

指标 前 后
B0 1,000 full 916 928
B0 1,000 partial 80 68
B0 3,000 full 2,578 2,614
B0 3,000 partial 268 232
运行时 full 2,578 2,614

零降级:3,000 字运行时逐条比对,36 条全部 partial → full。

剩余 232 条 partial 多为 D2 视觉拆法(如 交 记作 亠八乂,86 取 六乂)与占位 PUA,需 canonical 根序数据,继续不猜。

验证

  • validate_b0.py PASS
  • make macos-offline-tests 67 文件 / 336 例 PASS
  • make verify PASS(538 产物零网络符号)
  • 新增 test_d2_glyph_aliases_are_normalized

ADR-359 / 报告 285。

- add corpus/d2-glyph-aliases.tsv: リ→刂(j), ㄗ→卩(b), ク→夕(q), same-shape only
- build_b0 --glyph-aliases; normalized glyphs still pass standard v1 checks
- add 彑 (X) to root inventory; ノ counted as 撇 stroke for recognition-slot drop
- B0 1k full 916→928/partial 80→68; 3k full 2578→2614/partial 268→232; zero downgrades
- update ADR-359/report 285/standard/schema/provenance/sources/status docs

Signed-off-by: wangzhizhou <824219521@qq.com>
…zation

Signed-off-by: wangzhizhou <824219521@qq.com>
@wangzhizhou
wangzhizhou merged commit 7b95ba6 into main Nov 14, 2026
4 checks passed
wangzhizhou added a commit that referenced this pull request Nov 14, 2026
Signed-off-by: wangzhizhou <824219521@qq.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant