nostr-summary on Nostr: [ koteitan/zundamstr ] v1.1.0/raise measured accuracy on unseen data from 84% to ...
[ koteitan/zundamstr ] v1.1.0/raise measured accuracy on unseen data from 84% to 89.3% over 9 rounds
- Route nd/yr, でしょう/だろう, ですけど/ですから through a conjugation-table
yogen test (isYogenEnd) instead of the YOGEN character class
- Open four classes previously left unconverted by design: request て-form
(→てほしいのだ), volitional, 様態のそう, and dialect や/にゃ/じゃ
- Generalize the polite-form reduction to relay any trailing particle instead
of enumerating 丁寧語+語尾 pairs one by one
- Recognise negation variants (ねえ, かもしれん, 小せえ, 〜くて) and 連用形名詞
as verbs by widening the stem tables and the E/A/U/O-row mappings
https://github.com/koteitan/zundamstr/commit/e9b5b300866c49af0fdb5af3373105d5524e19c6Published at
2026-08-01 21:01:08 CESTEvent JSON
{
"id": "a6e451565eacba158641d70ec6d492d2873f0aa737523626bf184512640dea1c",
"pubkey": "7febe2a59aa826f8f6337b4101939eaac82fd4e6ca5f3c7b8f13ca03ad26f2bf",
"created_at": 1785610868,
"kind": 1,
"tags": [
[
"proxy",
"https://github.com/koteitan/zundamstr/commit/e9b5b300866c49af0fdb5af3373105d5524e19c6",
"web"
]
],
"content": "[ koteitan/zundamstr ] v1.1.0/raise measured accuracy on unseen data from 84% to 89.3% over 9 rounds\n\n- Route nd/yr, でしょう/だろう, ですけど/ですから through a conjugation-table\n yogen test (isYogenEnd) instead of the YOGEN character class\n- Open four classes previously left unconverted by design: request て-form\n (→てほしいのだ), volitional, 様態のそう, and dialect や/にゃ/じゃ\n- Generalize the polite-form reduction to relay any trailing particle instead\n of enumerating 丁寧語+語尾 pairs one by one\n- Recognise negation variants (ねえ, かもしれん, 小せえ, 〜くて) and 連用形名詞\n as verbs by widening the stem tables and the E/A/U/O-row mappings\nhttps://github.com/koteitan/zundamstr/commit/e9b5b300866c49af0fdb5af3373105d5524e19c6",
"sig": "1a946aa33653674e6c72e1076c5d1e3c16d4bc22a83e821565420eb189cd6bbf8b16a57904758120c40faccc32077835f9e99748c3991593681e3d08ef228765"
}
