This dissertation is about cross-document coreference between events in the news. The first two parts focus on data used to study event coreference and the last two parts contribute to modelling the event coreference phenomenon. Firstly, I investigate the available data sets to determine their representativeness with regard to the referential and lexical diversity of event coreference in the news. Thereafter, I explore how one can make a data set more representative of event coreference in the news by creating the ECB+ corpus. Next, I research the best ways to model the phenomenon of gradable coreference and to consider partial coreference in event coreference resolution. Finally, I deliberate about the role of event times and entities in event coreference resolution. The last part of the work results in developing the Bag of Events approach to event coreference resolution which makes use of partial coreference between mentions of event components from a unit of discourse. This dissertation provides a rigorous account of how the diversity of event coreference in the news can be sampled and modelled to perform event coreference resolution. The outcome of this research lays the foundation for highly accurate coreference resolvers.
|Award date||15 Apr 2021|
|Place of Publication||s.l.|
|Publication status||Published - 15 Apr 2021|
- event coreference, corpus annotation, cross-document coreference, event coreference resolution