Document-Level Event Factuality Identification Using Negation and Speculation Scope

Document-level Event Factuality Identification (DEFI) is to identify the factuality of an event in document. DEFI is an important foundation for many Natural Language Understanding (NLU) tasks, such as information extraction and text understanding. The negation and speculation scope refers to the continuous segment of text, which is controlled by the words with negative or speculative semantics. In this paper, we explore the importance of negation and speculation scope for DEFI, and propose two methods to use the scope features. The model of detecting scope is trained from cross-domain corpus, and applied to the Document-Level Event Factuality (DLEF) corpus. Experimental results show that our DEFI model is superior to several baselines.