VerbAtlas (Di Fabio et al., 2019) offers, with respect to other inventories such as FrameNet (Baker, 2014) and PropBank (Palmer et al., 2005):
- cross-domain roles
- semantically consistent frames
- informative role labels
One main problem concerning VerbAtlas is that it’s only verb-specific. The goal is to create a nominal dataset that can be used with VerbAtlas.
The aim of this repo is to show how (and to which extent) we can use different strategies in order to delegate all (or at least most of) the difficulties, that can be found in generating such dataset, from manual to automatic annotation.
The report.pdf file describes the solution adopted, compared with alternative approaches to said problem.