Semantic Roles for Nominal Predicates: Building a Lexical Resource

Abstract: The linguistic annotation of noun-verb complex predicates (also termed as light verb constructions) is challenging as these predicates are highly productive in Hindi. For semantic role labelling, each argument of the noun-verb complex predicate must be given a role label. For complex predicates, frame files need to be created specifying the role labels for each noun-verb complex predicate. The creation of frame files is usually done manually, but we propose an automatic method to expedite this process. We use two resources for this method: Hindi PropBank frame files for simple verbs and the annotated Hindi Treebank. Our method perfectly predicts 65% of the roles in 3015 unique noun-verb combinations, with an additional 22% partial predictions, giving us 87% useful predictions to build our annotation resource.