Hugging Face's logo

Datasets:
ratimics
/
syntax-gossip

Tasks:
Token Classification
Feature Extraction
Modalities:
Text
Formats:
json
Languages:
English
Size:
1K - 10K
Tags:
syntax
dependency-parsing
universal-dependencies
syntax-ir
candidate
Libraries:
Datasets
pandas
Polars
License:
Dataset card Data Studio Files
xet
 Community
1
syntax-gossip / README.md
ratimics's picture
ratimics
syntax-gossip v0.1.0-candidate.1: 5000 validated Syntax IR records (train 4000/val 500/test 500)
c971fa3 verified
preview code
|
Raw History Blame Contribute Delete
1.34 kB
metadata 
license: cc-by-nc-sa-4.0
task_categories:
  - token-classification
  - feature-extraction
language:
  - en
tags:
  - syntax
  - dependency-parsing
  - universal-dependencies
  - syntax-ir
  - candidate

Syntax Gossip (candidate)

Syntax IR records over Crownless event-to-speech gossip: exact source bytes, UTF-8 byte spans, UDPipe UD dependencies, NP/VP chunks, clause spans with hearsay attribution links, and audit lineage per record.

  • Status: candidate (braid-syntax-gossip-v0.1.0-candidate.1). Not frozen, authorizes no training. Cloud verification and gold review are outstanding.
  • Splits: train.jsonl (4,000) / validation.jsonl (500) / test.jsonl (500).
  • Backend: udpipe-1.4.0+english-ewt-2.5, 138-name reviewed gazetteer. Full run: 5,000 parsed, 0 validation failures.
  • Verify: sha256sum -c checksums.sha256, then re-run record validation (source reconstruction, single root, acyclicity, clause integrity, attribution links) per the Syntax IR contract in manifest.json.
  • Rights: derived from Crownless Carriage gossip corpora, published by the rights holder. Model-facing text remains game fiction, not factual claims.

Built with the Braid syntactic adapter (crownless-syntax catalog candidate). See the Braid repository for the adapter source, white paper, and reproduction runbook.