In recent years, great success has been achieved in the field of natural language processing (NLP), thanks in part to the considerable amount of annotated resources. For named entity recognition (NER), most languages do not have such an abundance of labeled data as English, so the performances of those languages are relatively lower. To improve the performance, we propose a general approach called Back Attention Network (BAN). BAN uses a translation system to translate other language sentences into English and then applies a new mechanism named back attention knowledge transfer to obtain task-specific information from pre-trained high-resource languages NER model. This strategy can transfer high-layer features of well-trained model and enrich the semantic representations of the original language. Experiments on three different language datasets indicate that the proposed approach outperforms other state-of-the-art methods.
Download