A Multitask Learning Approach for Named Entity Recognition by Exploiting Sentence-Level Semantics Globally

Named entity recognition (NER) is one fundamental task in natural language processing, which usually viewed as a sequence labeling problem and typically addressed by neural conditional random field (CRF) models, such BiLSTM-CRF. Intuitively, the types contain rich semantic information type sentence can globally reflect sentence-level semantics. However, most previous works recognize named entities based on feature representation of each token input sentence, token-level features cannot capture global-entity-type-related sentence. In this paper, we propose joint model to exploit global-type-related for NER. Concretely, introduce new auxiliary task, namely prediction (TSP), supervise constrain global learning process. Furthermore, multitask method used integrate into NER model. Experiments four datasets different languages domains show that our final highly effective, consistently outperforming BiLSTM-CRF baseline leading competitive results all datasets.