Project status: Japanese immersion learning with Anki and selective glossing: 1. Summary

Section 1 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
Next: https://ai.hiina.space/reports/jp-learning-and-anki-tooling/02-goal-and-current-state.md

1. Summary

What the project is. A staged plan to make Anki serve anime, game, visual novel (VN) and reading immersion in Japanese, and then to build one shared learner model with selective-gloss tools on top of it.
Current state (25 Oct 2026). Plan proposed; no settings changed and no tools built. Kaishi 1.5k is finished. The collection, last inventoried 30 Jul 2026, holds 269 radical notes, 432 hand-mined Lapis notes and 202 premined cards for the anime Katanagatari.
Next gate. Four weeks after a settings, leech and sentence-rotation pass, recall on a word-only probe of 50 mature Lapis cards should be within 5 points of the deck's normal true retention. That measures whether cards are being recalled from the word or from their example sentence.
Build order. (1) Anki settings plus a manual leech and Click-card pass; (2) a thin read-only learner model; (3) an agent-maintained deck-hygiene loop and a premine pipeline rebuilt on Jiten.moe; (4) a hint-first selective-gloss overlay for games, VNs and anime; (5) a Japanese-page gloss mode; (6) a revived concept-level grammar tutor; (7) an English-to-Japanese hybrid-reading rewriter, last.
Strongest evidence. Varying the sentence cue across retrievals beats a constant cue for foreign vocabulary (PNAS 2024, six experiments). Retrieved corpus sentences are preferred over LLM-generated ones for Japanese learners (ACL SRW 2024). Glossed reading beats unglossed reading. Hint-first AI tutoring avoids the learning loss seen with unrestricted AI help in a field RCT.
Limits. No study tests any of these tools on Japanese learners. The supporting evidence (glossing, varied retrieval cues, hint-first tutoring) comes from other materials and settings. Many community guides failed to load during research, so community-consensus claims rest on vnclub, TheMoeWay, Donkuri and tool documentation.
Clearest gap found. No existing tool glosses only the words a learner probably does not know, or offers graded hints instead of translation, for games, VNs or anime. No tool puts immersion exposures and lookups into the same table as SRS reviews.

Text licensed CC BY 4.0 (attribution: Hiina, ai.hiina.space). Source: https://github.com/hiinaspace/ai-hiina-space