Metadata harmonisation: align spoken-data fields with UniDive naming conventions
Back to South_Levantine_Arabic MADAR · Back to index
Repo: https://github.com/UniversalDependencies/UD_South_Levantine_Arabic-MADAR
Cross-posting from the UniDive WG1 T1.5 (spoken language guidelines) metadata harmonisation review. We compared UD_South_Levantine_Arabic-MADAR’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?
genre lists spoken social, but the README indicates the 100 sentences are actually written: they’re manually translated (not transcribed) short conversational tourism-related texts from the MADAR Parallel Corpus, itself derived from the written Basic Traveling Expression Corpus (BTEC). We couldn’t find any indication that a sentence is a transcription of actual speech.
Suggestion: Could you confirm whether any sentences are actually spoken, or should spoken be dropped from genre?
Implementation notes
- Needs manual input from maintainers: whether
spokenshould be dropped fromgenre(or whether any of the 100 sentences really are transcribed speech) is a factual question about the corpus, not something inferable from the data itself. No script applies until this is answered - if the answer is “drop it”, it’s then a trivialsed -i '' 's/, spoken//;s/spoken, //;s/^spoken$/prose/' *.conllu-style fix on the# genreline (exact command depends on the confirmed wording).
This issue was prepared as part of the UniDive WG1 T1.5 spoken language guidelines effort. Happy to help implement these changes ourselves if that’s easier than doing it on your end - just let us know.