Metadata harmonisation: align spoken-data fields with UniDive naming conventions
Back to Ligurian GLT · Back to index
Repo: https://github.com/UniversalDependencies/UD_Ligurian-GLT
Cross-posting from the UniDive WG1 T1.5 (spoken language guidelines) metadata harmonisation review. We compared UD_Ligurian-GLT’s current CoNLL-U metadata against the proposed naming conventions (see also the full treebank status table). This is a suggestion for maintainers to review - feel free to push back on anything that doesn’t fit the corpus. The comparison was carried out semi-automatically with the help of Claude (Anthropic); errors or misunderstandings are possible, so please double-check anything unclear.
1. Is the spoken portion identifiable?
This treebank mixes spoken and written material but its .conllu files don’t explicitly mark which sentences are spoken. We looked for a pattern in the data (a weak signal):
Finding: Only 12 documents total, with mixed short prefixes - too few for an automatic pattern, but small enough to check by hand. The README mentions a radio broadcast among the sources, but we couldn’t work out which document_id(s) or sentences it corresponds to.
Evidence: document_id values: bdl-c00, cairo, esl-c01, wp-arba, wp-tintin, and others (12 total, no dominant separator/prefix pattern).
Suggestion: Since there are only 12 documents, could you confirm per-document whether each is spoken or written (e.g. do wp-* mean Wikipedia/written, is cairo a transcribed story), and specifically which document(s) correspond to the radio broadcast mentioned in the README?
2. Sentence-level (naming conventions)
| Field | Suggestion |
|---|---|
parallel_id |
corpus-specific (sentence-level) - verify against metadata.html |
Implementation notes
Needs manual input from maintainers
- Modality: only 12 documents total, no dominant naming pattern - needs a per-document call from maintainers (spoken vs. written for each of
bdl-c00,cairo,esl-c01,wp-arba,wp-tintin, and the rest), plus identification of which document(s) correspond to the radio broadcast. With only 12 documents this is faster to answer directly than to script; once the per-document list exists, tagging viaharmonize_metadata.py tag-modality --spoken-if '<confirmed ids>'(or a plain per-documentrename-comment/manual edit, since 12 is small) is trivial. parallel_id: corpus-specific field, needs a decision on whether/how it maps to the naming conventions before any action.