Improving Multilingual Named Entity Recognition with Wikipedia Entity Type Mapping
This work addresses the challenge of improving NER accuracy for multilingual applications, particularly in low-resource or domain-shift scenarios, but it is incremental as it builds on existing statistical models with a knowledge base enhancement.
The paper tackled the problem of multilingual named entity recognition (NER) systems making mistakes due to incorrect entity type features by utilizing Wikipedia as an open knowledge base to construct entity type mappings, resulting in up to an 18.3 F1 score improvement on unseen entities, especially in new domains or with limited training data.
The state-of-the-art named entity recognition (NER) systems are statistical machine learning models that have strong generalization capability (i.e., can recognize unseen entities that do not appear in training data) based on lexical and contextual information. However, such a model could still make mistakes if its features favor a wrong entity type. In this paper, we utilize Wikipedia as an open knowledge base to improve multilingual NER systems. Central to our approach is the construction of high-accuracy, high-coverage multilingual Wikipedia entity type mappings. These mappings are built from weakly annotated data and can be extended to new languages with no human annotation or language-dependent knowledge involved. Based on these mappings, we develop several approaches to improve an NER system. We evaluate the performance of the approaches via experiments on NER systems trained for 6 languages. Experimental results show that the proposed approaches are effective in improving the accuracy of such systems on unseen entities, especially when a system is applied to a new domain or it is trained with little training data (up to 18.3 F1 score improvement).