Coreference resolution for Italian: Assessing the impact of linguistic components

This paper presents a systematic evaluation of two linguistic components required to build a coreference resolution system: mention detection and mention description. We compare gold standard annotations against the output of the mod- ules based on the state-of-the-art NLP for Italian. Our experiments suggest the most promising direction for future work on coreference in Italian: we show that, while automatic mention description affects the performance only mildly, the mention de- tection module plays a crucial role for the end-to-end coreference performance. We also show that, while a considerable number of mentions in Italian are zero pronouns, their omission doesn’t affect a general-purpose coreference resolver, sug- gesting that more specialized algorithms are needed for this subtask.