Download measured/hin.json from bcv-commons/gram-struct: direct link, hf CLI and curl.
- Browser
- Download file 2.09 kB
-
https://huggingface.co/datasets/bcv-commons/gram-struct/resolve/main/measured/hin.json
- Command line
-
hf download hf://datasets/bcv-commons/gram-struct/measured/hin.json
-
curl -L -o hin.json https://huggingface.co/datasets/bcv-commons/gram-struct/resolve/main/measured/hin.json
2.09 kB
| { | |
| "conventions_md": "config/llm_conventions/hin.md", | |
| "mechanisms": { | |
| "fertility_priors": { | |
| "date": "2026-09-24", | |
| "enabled": true, | |
| "gold": "clear/IRVHin", | |
| "lambda": 2.0, | |
| "note": "measured 2026-09-24, whole Bible vs Clear gold. lambda swept {0.5,1,2} on hin per the plan: all three clear the noise floor (2x baseline-replicate spread; eflomal F1 spread 0.0003, +gloss 0.0002) with exact_span up. lambda=2 gives the largest gain: eflomal F1 .5421->.5487 (Delta+0.0066), exact_span 36714.5->36862 (+147.5); +gloss F1 .5423->.5489 (Delta+0.0066), exact 36746.5->36911 (+164.5). VALIDATED against both plan controls at the same evidence scale (~7125 anchors, lambda=2): a typology-BLIND null prior gives a BIGGER F1 gain (+0.0098) but exact_span DOWN (-285.5) -- pure recall inflation that fails the plan's own exact_span-up bar; an inverted placebo (same weight, wrong anchors) gives only +0.0013 (~1/5 the real prior's gain). Confirms the typology-conditioned PLACEMENT is doing the work, not generic FERF regularization.", | |
| "source": "measured" | |
| }, | |
| "spanext.definite_trigger": { | |
| "date": "2026-09-24", | |
| "enabled": false, | |
| "gold": "clear/IRVHin", | |
| "note": "not applicable to hin — Hindi has no article_order direction at all (GB022/023 both 0), so definite_trigger has nothing to fire on regardless of this setting.", | |
| "source": "measured" | |
| }, | |
| "spanext.relation_trigger": { | |
| "date": "2026-10-08", | |
| "enabled": true, | |
| "gold": "clear/IRVHin", | |
| "note": "RE-MEASURED 2026-10-08 after porting the rectum test to MACULA construct_role (span_extension rela_of; BHSA rela no longer exists on the default spine, so the trigger had been inert since 2026-10-04): hinirv OT, word grain, with the audit ceiling lifted so the base path runs: F1 .5651 -> .5653 (+0.0002), exact +21 -> wash. Kept true (no harm); inert in practice while span extension does not run for hin (see _audit_note).", | |
| "source": "measured" | |
| }, | |
| "spanext.typology_fallback": { | |
| "date": "2026-09-24", | |
| "enabled": false, | |
| "gold": "clear/IRVHin", | |
| "source": "measured" | |
| } | |
| } | |
| } | |