Domain-Adapted Dependency Parsing for Cross-Domain Named Entity Recognition

In recent years, many researchers have leveraged structural information from dependency trees to improve Named Entity Recognition (NER). Most of their methods take dependency-tree labels as input features for NER model training. However, such is not inherently provided in most corpora, making the with low usability practice. To effectively exploit potential word-dependency knowledge, motivated by success Multi-Task Learning on cross-domain NER, we investigate a novel learning method incorporating Dependency Parsing (DP) its auxiliary task. Then, considering high consistency relations across domains, present an unsupervised domain-adapted transfer knowledge high-resource domains low-resource ones. With help DP bridge different both useful and cross-task can be learned our considerably benefit NER. make better use between DP, unify tasks shared network architecture joint learning, using Maximum Mean Discrepancy(MMD). Finally, through extensive experiments, show proposed only advantage but also significantly outperform other Our code open-source available at https://github.com/xianghuisun/DADP.