Dynamic Propbanking with Deep Linguistic Grammars

In the development of annotated corpora with deep linguistic representations, the category to be assigned to each markable (i.e. the fully fledged grammatical representation of each sentence) is so complex that it cannot be safely constructed manually and the annotation cannot be performed without some supporting application, viz. a computational grammar. This paper discusses and present solutions for the adaptation of deep linguistic grammars as supporting tools in the construction of dynamic propbanks. It reports on the results of an experimental study with a pilot application of this approach to propbanking. Estimated scores for inter-annotator agreement and for development effort are presented.