🌸 kotonoha v0.0.83

🌸 言の葉 (kotonoha) v0.0.83

📝 Changes in this release

• feat: adopt DeepSeek as primary AI with Gemini 3.1 flash-lite secondary fallback

🌸 kotonoha v0.0.82

🌸 言の葉 (kotonoha) v0.0.82

📝 Changes in this release

• style: format dialogue context and bump to v0.0.82

🌸 kotonoha v0.0.81

🌸 言の葉 (kotonoha) v0.0.81 ✨

Context-enriched AI disambiguation pipeline with surrounding dialogue windows, temporal cutoff guards, ASS speaker preservation, and clean show/episode metadata extraction.

🌟 Key Highlights

• • 💬 Surrounding Dialogue Context Window: Passes up to $\pm 2$ surrounding dialogue lines to Google Gemini for enhanced word sense disambiguation, register recognition, and accurate contextual definitions.
• • ⏱️ Temporal Proximity Guard: Implements a 10-second temporal cutoff to prevent bleeding across scene transitions, commercial breaks, or OP/ED music themes.
• • 🎬 Clean Show & Episode Metadata: Automatically parses and sanitizes media filenames (stripping bracketed release group tags, video codecs, and hashes) to ground the AI in the appropriate anime/drama domain.
• • 🎭 ASS Speaker Name Preservation: Captures and prepends character/actor names from Advanced SubStation Alpha subtitles (e.g. フェルン: ...) into the dialogue context.
• • 🔍 Interactive Dialogue Inspector (v key): Reviewers can now press v during card review to inspect the surrounding conversation snippet with relative timeline offsets directly in the terminal.
• • ✒️ Strict Codebase Quality: 113 unit & integration tests passing (100% green), zero compiler warnings, zero clippy warnings (all = "deny"), and full architecture compliance.

🚀 Direct Download & Run

# Download and run standalone Linux x86_64 binary
curl -fsSL https://github.com/Praveensenpai/kotonoha/releases/download/v0.0.81/kotonoha-x86_64-unknown-linux-gnu.tar.gz | tar -xz
./kotonoha --help

🌸 kotonoha v0.0.80

🌸 言の葉 (kotonoha) v0.0.80

📝 Changes in this release

• feat(nlp): resolve digit-to-kanji counter compounds, filter audio screams, and normalize formal 時 (v0.0.80)

🌸 kotonoha v0.0.79

🌸 言の葉 (kotonoha) v0.0.79

📝 Changes in this release

• fix(dict): resolve canonical readings, homophone disambiguation, and nickname tagging (v0.0.79)

🌸 kotonoha v0.0.78

🌸 言の葉 (kotonoha) v0.0.78

📝 Changes in this release

• feat(nlp): enhance real-world subtitle tokenizer and full inflected predicate highlighting (v0.0.78)

🌸 kotonoha v0.0.77 ✨

🌸 言の葉 (kotonoha) v0.0.77 ✨

Synchronized AI candidate indexing, exact-match offline dictionary priority, and contextual reading preservation!

🌟 Key Highlights

• • 🧠 Exact-Match Offline Dictionary Priority: query_offline_terms now queries exact expression and reading matches before falling back to prefix lookups. High-frequency compound words (e.g. 先々 read as さきざき, 先に, 先安) no longer crowd out or evict single-kanji and base words (先 read as さき).
• • 🎯 AI & Local Dictionary Index Parity: Connected prepare_ai_batch directly to local SQLite dictionaries (lookup_all_candidates_cached) with consistent contextual sorting (sort_candidates_for_context). Candidate indices recommended by Gemini now match local dictionary entries with 100% precision.
• • 📖 Contextual Reading Preservation: align_contextual_reading runs after AI analysis and guarantees that Sudachi's grammatical sentence reading (and pitch accent) is preserved on the card rather than overwritten by alternative pronunciations.
• • 🤖 AI Contextual Reading Extraction: Added recommended_reading schema support to Gemini analysis so AI can directly provide or verify furigana in colloquial dialogue.
• • 📦 Precompiled Linux Binary: Standalone x86_64 release archive with SHA256 checksums attached.

🚀 Direct Install

cargo install --git https://github.com/Praveensenpai/kotonoha --tag v0.0.77

🌸 kotonoha v0.0.76 ✨

🌸 言の葉 (kotonoha) v0.0.76 ✨

Linguistic UniDic verb resolution, contextual auxiliary verb handling, and rigorous candidate accuracy across full anime series!

🌟 Key Highlights

• • 🧠 Grammatical Subsidiary Verb Resolution: UniDic 非自立可能 is now contextually evaluated based on connective て/で attachment. Over 1,000+ legitimate independent verbs and adjectives (間違う, 行く, 見る, やる, 出す, 来る, 置く, いい, よい) are correctly preserved as content words rather than discarded as junk.
• • 🎯 Precise $i+1$ Candidate Filtering: Prevents false single-unknown misclassification by ensuring sentences containing multiple distinct unknown words are strictly categorized under multi-unknown tiers.
• • 🧹 Clean Modular Architecture: Refactored miner test suite into modern src/miner/tests.rs (<400 lines compliance), 0 compiler/clippy warnings, and 86 passing tests.
• • 📦 Precompiled Linux Binary: Standalone x86_64 release archive with SHA256 checksums attached.

🚀 Direct Install

cargo install --git https://github.com/Praveensenpai/kotonoha --tag v0.0.76

🌸 kotonoha v0.0.75 ✨

🌸 kotonoha v0.0.75 ✨

Official verified release of the Naturalness & i+1 Quality Scorer with real-world anime dialogue regression tests!

🌟 Key Highlights

• • 🧠 Naturalness & i+1 Quality Scorer: Replaced naive shortest-character heuristics with on-device Japanese syntactic evaluation (QualityScorer), completely eliminating broken dialogue fragments.
• • 🛡️ Defect Gating: Penalizes dangling connectives (て…, けど…, たら…) and hanging case particles (彼が…, 本当は…), ensuring only grammatically self-contained sentences win.
• • 🎯 Case Marker Relational Bonus: Rewards sentences with explicit case particles (を, に, が, で) that establish clear grammatical context for the target word.
• • 🎬 Anime Dataset Regression Tests: Verified across the complete Yuru Camp episode dataset, confirming superior sentence selection for all target vocabulary.
• • 📦 Precompiled Linux Binary: Standalone x86_64 release archive with SHA256 checksums attached.

🚀 Direct Install

cargo install --git https://github.com/Praveensenpai/kotonoha --tag v0.0.75

🌸 kotonoha v0.0.74 ✨

🌸 kotonoha v0.0.74 ✨

Aesthetic README overhaul, visual architecture pipeline, and official release of the on-device Naturalness & i+1 Quality Scorer!

🌟 Key Highlights

• • 🎨 Aesthetic Architecture Overhaul: Complete visual documentation overhaul featuring a comprehensive ASCII pipeline diagram, bilingual thematic headers, and Catppuccin badge palette.
• • 🧠 Naturalness & i+1 Quality Scorer: Fully documented and active on-device sentence scoring engine (QualityScorer) that eliminates broken fragments and dangling particles (て…, けど…, 彼が…).
• • 📦 .koto Learning Bundles Documentation: Comprehensive documentation for ultra-compact .koto archives (~14 MB, >98% space saved), hot-swap subtitle replacement (kotonoha bundle replace), and duplicate fingerprint guards.
• • 🚀 Standalone x86_64 Release: Precompiled, optimized Linux x86_64 release archive with SHA256 checksums attached.

🚀 Direct Install

cargo install --git https://github.com/Praveensenpai/kotonoha --tag v0.0.74

Or run the quick one-liner installer:

curl -sSL https://raw.githubusercontent.com/Praveensenpai/kotonoha/main/install.sh | bash