A Deep Bidirectional Highway Long Short-Term Memory Network Approach to Chinese Semantic Role Labeling

摘要: Existing approaches to Chinese semantic role labeling (SRL) mainly adopt deep long short-term memory (LSTM) neural networks address the long-term dependencies problem. However, LSTM cannot vanishing gradient problem properly. In addition, complexity of language, as a hieroglyphic decreases performance traditional SRL SRL. To these problems, this paper proposes new approach with bidirectional highway network. The proposed is further improved by introducing conditional random fields (CRFs) constraints and part-of-speech (POS) feature since POS tags are classes formal equivalents words in linguistics. experimental results on commonly used Proposition Bank dataset show that outperforms existing approaches. With an easily acquired reliable for practical applications, substantially improves