Coping With Implicit Arguments And Events Coreference

In this paper we present ongoing work for the creation of a linguistically-based system for event coreference. We assume that this task requires deep understanding of text and that statistically-based methods, both supervised and unsupervised are inadequate. The reason for this choice is due to the fact that event coreference can only take place whenever argumenthood is properly computed. It is a fact that in many cases, arguments of predicates are implicit and thus linguistically unexpressed. This prevents training to produce sensible results. We also assume that spatiotemporal locations need to be taken into account and this is also very often left implicit. We used GETARUNS system to develop the coreference system which works on the basis of the discourse model and the automatically annotated markables. We present data from the analysis, both on unexpressed implicit arguments and the description of the coreference algorithm.