Speculation and negation identification via unified Machine Reading Comprehension frameworks with lexical and syntactic data augmentation

Zhong Qian, Ting Zou, Zihao Zhang, Peifeng Li, Qiaoming Zhu, Guodong Zhou

Abstract: Highlights•Complete SNI task with cue and scope detection is fit for real-world engineering application.•Unified MRC frameworks are effective across various SNI corpora.•Lexical and syntactic data augmentation can boost scope resolution.•Our model performs well on both single-word and multi-word cues.