Sources of Evidence for Implicit Argument Resolution

Traditionally, semantic role labelling systems have focused on searching the fillers of those explicit roles appearing within sentence boundaries. However, when the participants of a predicate are implicit and can not be found inside sentence boundaries, this approach obtains incomplete predicative structures with null arguments. Previous research facing this task have coincided in identifying the implicit argument filling as a special case of anaphora or coreference resolution. In this work, we review a number of theories that model the behaviour of discourse coreference and propose some adaptations to capture evidence for the implicit argument resolution task. We empirically demonstrate that exploiting such evidence our system outperforms previous approaches evaluated on the SemEval-2010 task 10 dataset. We complete our study identifying those cases that traditional coreference theories can not cover.