The Spanish UP is built on top of the universal dependency treebank for Spanish (AnCora corpus), version 1.4: https://github.com/UniversalDependencies/UD_Spanish-AnCora It inherites their licence and file structure.
There are three files in CoNLL-U format: A train, a test and a dev split of annotated data. Like all UPs, the Spanish UP adds a layer of semantic role labeling annotations based on the English Proposition Bank to the treebank. These are added as additional columns to the CoNLL format, as in earlier practice with the CoNLL-X format.
In addition, there are a list of verb overview files, one for each Spanish verb. They indicate what frames a German verb can take and give examples. They are meant to be viewed in a browser.
We are currently making no changes to the underlying UD for Spanish.
- We believe the recall for raised arguments to be low for Spanish.
- For Spanish, we did not yet execute the curation approach of our COLING 2016 paper and used heuristic frame aliasing instead.