sudacode
5feed360ca
* fix: harden preload argv parsing for popup windows * fix: align youtube playback with shared overlay startup * fix: unwrap mpv youtube streams for anki media mining * docs: update docs for youtube subtitle and mining flow * refactor: unify cli and runtime wiring for startup and youtube flow * feat: update subtitle sidebar overlay behavior * chore: add shared log-file source for diagnostics * fix(ci): add changelog fragment for immersion changes * fix: address CodeRabbit review feedback * fix: persist canonical title from youtube metadata * style: format stats library tab * fix: address latest review feedback * style: format stats library files * test: stub launcher youtube deps in CI * test: isolate launcher youtube flow deps * test: stub launcher youtube deps in failing case * test: force x11 backend in launcher ci harness * test: address latest review feedback * fix(launcher): preserve user YouTube ytdl raw options * docs(backlog): update task tracking notes * fix(immersion): special-case youtube media paths in runtime and tracking * feat(stats): improve YouTube media metadata and picker key handling * fix(ci): format stats media library hook * fix: address latest CodeRabbit review items * docs: update youtube release notes and docs * feat: auto-load youtube subtitles before manual picker * fix: restore app-owned youtube subtitle flow * docs: update youtube playback docs and config copy * refactor: remove legacy youtube launcher mode plumbing * fix: refine youtube subtitle startup binding * docs: clarify youtube subtitle startup behavior * fix: address PR #31 latest review follow-ups * fix: address PR #31 follow-up review comments * test: harden youtube picker test harness * udpate backlog * fix: add timeout to youtube metadata probe * docs: refresh youtube and stats docs * update backlog * update backlog * chore: release v0.9.0
3.8 KiB
id, title, status, assignee, created_date, updated_date, labels, dependencies, references, priority, ordinal
| id | title | status | assignee | created_date | updated_date | labels | dependencies | references | priority | ordinal | ||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| TASK-209 | Exclude grammar-tail そうだ from subtitle annotations | Done |
|
2026-03-20 04:06 | 2026-03-23 03:22 |
|
|
high | 120500 |
Description
Sentence-final grammar-tail そうだ tokens can still receive subtitle annotation styling, including frequency highlighting, when Yomitan returns a standalone そうだ token and MeCab enriches it as an auxiliary-stem/coupla pattern (名詞|助動詞, 助動詞語幹). Keep the subtitle text visible, but treat this grammar tail like other grammar-only endings so it renders without annotation metadata.
Acceptance Criteria
- #1 Sentence-final grammar-tail
そうだtokens enriched as auxiliary-stem/copula patterns do not receive frequency highlighting or other subtitle annotation metadata. - #2 The preceding lexical token in cases like
与えるそうだkeeps its existing annotation behavior. - #3 Regression tests cover the annotation-stage exclusion and end-to-end subtitle tokenization for the
そうだgrammar-tail case.
Implementation Plan
- Add focused regression coverage for the reported
与えるそうだcase at both annotation-stage and tokenizeSubtitle levels. - Reproduce failure by modeling the MeCab-enriched grammar-tail shape (
名詞|助動詞,特殊,助動詞語幹) that currently keeps frequency metadata. - Update subtitle-annotation exclusion logic to recognize auxiliary-stem/copula grammar tails via POS metadata plus normalized tail text, not a raw sentence-specific string match.
- Re-run targeted tokenizer and annotation-stage tests, then record the verification commands and outcome in the task notes.
Implementation Notes
Investigated reported 与えるそうだ case. MeCab tags そう as 名詞,特殊,助動詞語幹 and だ as 助動詞; after overlap enrichment the Yomitan token becomes pos1=名詞|助動詞, pos2=特殊, pos3=助動詞語幹, which currently escapes subtitle-annotation exclusion and can keep a frequency rank.
Implemented a POS-shape subtitle-annotation exclusion for MeCab-enriched auxiliary-stem grammar tails. The new predicate keys off merged tokens whose POS tags stay within 名詞/助動詞/助詞 and whose POS3 includes 助動詞語幹, which clears annotation metadata for そうだ-style tails without hard-coding the full subtitle text.
Verification: bun test src/core/services/tokenizer/annotation-stage.test.ts, bun test src/core/services/tokenizer.test.ts --test-name-pattern 'explanatory ending|interjection|single-kana merged tokens from frequency highlighting|auxiliary-stem そうだ grammar tails|composite function/content token from frequency highlighting|keeps frequency for content-led merged token with trailing colloquial suffixes'
Final Summary
Added regression coverage for 与えるそうだ and updated subtitle annotation exclusion logic to drop annotation metadata for MeCab-enriched auxiliary-stem grammar tails. The fix is POS-driven rather than sentence-specific, so そうだ-style grammar endings stay visible/hoverable as plain text while neighboring lexical tokens keep their existing frequency/JLPT behavior.