Chinese hedge scope detection based on phrase semantic representation

Chinese hedge scope detection is dependent on syntactic and semantic information. Most previous methods typically use lexical and syntactic information as a basic unit of classification, which make these methods lose part of the effective structure information. In order to enhance detection performance, we take the phrase, which are extracted from the parse tree by some heuristic rules, as the classification unit (candidate phrase). Furthermore, a novel hierarchical neural network is proposed to learn the semantic representation of the phrase and its context. Experiments on the Chinese Biomedical Hedge Information (CBHI) corpus show that our system could achieve state-of-the-art performance without using any complicated feature engineering.