Speculation and Negation Scope Resolution via Machine Reading Comprehension Formulation with Data Augmentation

Zhong Qian, Tiening Sun, Ting Zou, Peifeng Li, Qiaoming Zhu, Guodong Zhou

Abstract: As a key sub-task in the field of speculation and negation extraction, Speculation and Negation Scope Resolution (SpNeSR) focuses on extracting speculative and negative texts within sentences, i.e., distinguishing between factual and non-factual information, which means it is an important and fundamental task in Natural Language Processing (NLP) community. Previous work utilized various methods for SpNeSR that are quite domain-specific, and failed to build a unified framework with good generalization. In addition, they were limited by the sizes of datasets and ignored producing more samples for training, since SpNeSR is a data-hungry task. With consideration of the above problems, we not only propose a unified Machine Reading Comprehension (MRC) formulation for SpNeSR, but also design a Data Augmentation (DA) method fit for scopes. Experimental results on several English and Chinese corpora manifest that both MRC and DA mechanism are effective, and our MRC model with DA is superior to several state-of-the-arts.