Semantic role labelling with minimal resources: Experiments with French

Abstract

This paper describes a series of French semantic role labelling experiments which show that a small set of manually annotated training data is superior to a much larger set containing semantic role labels which have been projected from a source language via word alignment. Using universal part-of-speech tags and dependencies makes little difference over the original fine-grained tagset and dependency scheme. Moreover, there seems to be no improvement gained from projecting semantic roles between direct translations than between indirect translations.