Woolery - Extending Frame Semantics to Structured Documents.

This paper presents Woolery, a system for semantic annotation and mapping of structured documents (such as JSON key-value pairs) to FrameNet. Implemented as a graphical interface, Woolery provides an annotator with a guided means to map keys in a JSON document to FrameNet elements, without the need for extensive knowledge of FrameNet's semantic structures. Candidate frame elements are identified via a search across FrameNet's internal representations, or via mapping keys to their potential WordNet synsets. Final element selection is automated via a pretrained language model. Initial results are promising, with the model giving an overall accuracy of 77.8% when labeling frames across a diverse corpus of JSON document schemas.