Tags: yongsk0066/mce
Tags
SCiL 2026 Paper-3 — Camera Ready Results Snapshot This tag marks the exact commit corresponding to the measurements reported in: Jang, Yongseok (2026). Comonadic Morphophonology: A Compositional Framework for Context-Dependent Morphological Rules in Finnish. Proceedings of the Society for Computation in Linguistics (SCiL). Reproducing the reported numbers from a fresh clone: git clone https://github.com/yongsk0066/mce.git cd mce git checkout scil-2026-camera-ready-results git submodule update --init vendor/ud-finnish-tdt export MCE_DICT_PATH="$(pwd)/data" just eval # UPOS 94.66% dev, Lemma 93.09% dev just eval-test # UPOS 94.58% test just bench-throughput # ~88k tok/s just wasm-size # ~380 KB See README "Reproducing the SCiL 2026 Results" for details.
feat: v0.3.1 — generation fixes, parallel loading, Cache API, demo en… …hancements Verb generation accuracy 47.7% → 69.9% (6 bugs fixed). Parallel fetch + progressive loading + Cache API for instant revisits. Verb paradigm demo tile, enriched descriptions, tag UX fix. Code dedup, serde-wasm-bindgen removal, ignored test fixes.
feat: v0.3.0 — suggest() SpellChecker integration, plural generation,… … research docs suggest() single backend: - Route suggest()/spell_check() through SpellChecker pipeline (trie + cache + morph) - spell_check() vs is_valid_word() architecturally differentiated - mce-speller promoted to explicit dependency Plural noun generation: - generate_form(base, case, "plural") now active (was ignored) - generate_paradigm() returns 22 forms (11 sg + 11 pl) - 6 new functions: plural_stem, genitive_plural, partitive_plural, etc. - 31 new tests Research documentation (v0.4.0+ planning): - docs/research/compound-improvement-plan.md (779 lines) - docs/research/kotus-integration-plan.md (610 lines) - docs/research/long-term-roadmap.md (635 lines) Tests: 1,496 -> 1,532 (+36)
docs: comprehensive documentation + Carbon docs site + npm publish
- README.md (195 lines): features, quick start, mermaid architecture, benchmarks
- CLAUDE.md: English rewrite with current metrics
- LEARNING.md (557 lines): 9-section domain knowledge guide
- ARCHITECTURE.md (397 lines): 5 mermaid + 3 ASCII diagrams, MonoSketch eval
- docs/index.html (1477 lines): Carbon Design System site with 8 interactive demos
- docs/carbon-type.{css,scss}: Carbon typography from corevoikko
- 14 crate/directory READMEs
- npm: @yongsk0066/mce@0.1.0 published (332KB WASM)