--- id: jp-learning-and-anki-tooling/05-mining-from-more-anime title: 'Project status: Japanese immersion learning with Anki and selective glossing: 5. Mining from more anime' report: https://ai.hiina.space/reports/jp-learning-and-anki-tooling.md section: 5 version: 1 updated: '2026-10-27' author: Hiina --- Index of everything on this site: https://ai.hiina.space/index.md # Project status: Japanese immersion learning with Anki and selective glossing: 5. Mining from more anime - Section 5 of 12 of the report "Project status: Japanese immersion learning with Anki and selective glossing" by Hiina (v1, updated 2026-10-27) - Report overview and contents: https://ai.hiina.space/reports/jp-learning-and-anki-tooling.md - Previous: https://ai.hiina.space/reports/jp-learning-and-anki-tooling/04-deck-hygiene-leeches-and-sentence-cued-cards.md - Next: https://ai.hiina.space/reports/jp-learning-and-anki-tooling/06-planned-tools-deck-hygiene-agent-and-concept.md ## 5. Mining from more anime ### Rebuilding the premine pipeline on Jiten.moe The open question from the *Katanagatari* premine was whether its filter thresholds were right. Research reframed it: much of that pipeline now exists upstream, and a better threshold is personal coverage, not a frequency cut-off. Jiten.moe is open source (Apache-2.0, Sudachi.rs plus JMdict). It has per-title vocabulary lists with Anki/CSV export, and a public API with `/api/media-deck/{id}/vocabulary` (sort by deck or global frequency, 200 per page) and `/api/vocabulary/parse`. Rate limits are 120 a minute anonymously and 300 with a key, and the endpoints are unversioned ([Sirush/Jiten](https://github.com/Sirush/Jiten); [Jiten API](https://jiten.moe/guides/using-the-api)). Jiten also has Word Sets that address the project's two main premine failure modes, function words and proper nouns: Particles & Common Grammar, Kaishi 1.5k, and Names (300k+), People, Place names, Other names. Each can be marked Mastered or Blacklisted without creating cards. Jiten's recommended setup for experienced learners is to master particles plus Kaishi and blacklist the names sets ([Jiten word sets](https://jiten.moe/guides/word-sets)). Known words can be imported from Anki into Jiten through AnkiConnect (add `https://jiten.moe` to `webCorsOriginList`), and Jiten then reports coverage of any show before it is started ([Jiten import](https://jiten.moe/guides/importing-known-words-from-anki)). Its rule of thumb: 80% coverage is challenging (about 2 lookups per sentence), 90% is a good immersion level (about 1), and 98% or more is comfortable ([Jiten: choosing media](https://jiten.moe/guides/choosing-media-at-your-level)). The project will use this to choose the next anime. Planned changes to the premine pipeline: - **Keep:** the Lapis field mapping, the offline known-word subtraction, and the JMdict definitions. - **Replace:** frequency ranking and tokenisation, with Jiten's per-title list or parse endpoint. - **Add:** a cheap LLM pass with two jobs. Given the line and the numbered JMdict senses, it picks a sense index (which would have fixed the 一寸 gloss). It also decides whether a token is a character name in that context. No public tool or evaluation was found for either job. The ecosystem offers only list-based name handling, such as Jiten's Names word sets, and Sudachi's Full dictionary with its named-entity split mode ([Sudachi](https://github.com/WorksApplications/Sudachi)). The *Katanagatari* failures serve as a regression set. - **Output route:** whether Jiten's Anki export is Lapis-compatible is unverified, so the Jiten API plus the project's own Lapis writer is the safer route. Two alternatives were noted: - **Anki Miner.** GPL-3.0, v3.10.1 on 22 Oct 2026. Turns video plus subs into cards, with i+1, frequency-range, blacklist and wordset filters, a Word Curator step, and Whisper subtitle generation. Single source, fast-moving, and it does not document how it decides a word is known ([PyPI](https://pypi.org/project/anki-miner/)). - **AnkiMorphs.** Reorders new cards so those with the fewest unknown morphs come first, which suits a 200-card premine dump ([AnkiMorphs docs](https://mortii.github.io/anki-morphs/)). ### Reactive mining while watching (Linux) - **asbplayer in Firefox:** v1.22.0, October 2026, with Jimaku auto-locate for subtitles. Its new Anki add-on (v1.0.0, needs Anki 25.02+, installed from a `.ankiaddon` file rather than AnkiWeb) lets Yomitan create the Lapis card while asbplayer patches in the screenshot, audio and subtitle sentence ([asbplayer releases](https://github.com/killergerbah/asbplayer/releases); [one-click mining](https://docs.asbplayer.dev/docs/guides/one-click-mining)). asbplayer can also colour subtitle words by known status, using a word database rebuilt from Anki fields ([annotation guide](https://docs.asbplayer.dev/docs/guides/annotation)). That is a free first cut of the selective-gloss idea. - **mpv with mpvacious:** v26.7.28.0, Linux-first, needs mpv 0.41+. Its README favours GoldenDict-ng or Rikaitan over Yomitan ([mpvacious](https://github.com/Ajatt-Tools/mpvacious)). Subtitles come from Jimaku and are retimed with alass, which can align to the video or to another subtitle track ([alass](https://github.com/kaegi/alass)). Kitsunekko's 2026 status could not be checked. Infrastructure note: AnkiConnect's GitHub repo was archived on 4 Nov 2025 and moved to sr.ht ([FooSoft/anki-connect](https://github.com/FooSoft/anki-connect)). Yomitan, asbplayer and Jiten import all depend on it, so the project pins its AnkiConnect version rather than assuming upstream updates. --- Text licensed CC BY 4.0 (attribution: Hiina, ai.hiina.space). Source: https://github.com/hiinaspace/ai-hiina-space