These Python scripts empirically determine the types of frame roles based on the knowledge graphs given in input and create new domain-specific axioms in OWLStar.
The pipeline can be described as a multi-step approach:

1. Roles identification: obtain <frame_type, role, object_type> triples, where frame_type and object_type are the types of the frame and the class covering role respectively.
2. Type estimation: empirically estimate the probability of having a certain class as the type of the argument of frame roles
3. Type generalization: obtain general types through WordNet mappings​
4. Ontology creation: mapping of triples to OWLStar

This module has been developed specifically for dealing with MUSICBO ontologies for the Polifonia project.

The scripts have been developed and tested with the following packages and their respective versions:

• python3: 3.9.7
• rdflib: 6.2.0
• SPARQLWrapper: 2.0.0

Run the following command

python3 selrestr_maker.py -d [KG_DIR]

where [KG_DIR] is the directory where the .nq files are stored.

It is also possible to run a OWLStar to OWL parser:

python3 owlstar2owl.py -i [INPUT_TTL] -o [OUTPUT_TTL]

As of now, the tool is only able to produce .ttl files

• Stanford slides (beware that this link will automatically download a pdf file)
• VerbNet selectional restrictions