Traditional sequence tagging methods for named entity recognition (NER) face challenges when handling nested entities, where an entity is nested in another. Most previous methods for nested NER ignore the effect of entity boundary information or type information. Considering that entity boundary information and type information can be utilized to improve the performance of boundary detection, we propose a nested NER model with a multi-agent communication module. The type tagger and boundary tagger in the multi-agent communication module iteratively utilize the information from each other, which improves the boundary detection and the final performance of nested NER. Empirical experiments conducted on two nested NER datasets show the effectiveness of our model.

Named entity recognition (NER) is a basic task in natural language processing. Traditionally, sequence labeling methods are applied to named entity recognition and achieve good performance. However, sequence labeling methods can not be straightly applied to recognize nested named entities where an entity is included in another entity. Recently, some new methods are proposed for nested named entity recognition. Most of them ignore that entity type information can help recognize entity boundaries or ignore that entity boundary information can help recognize entity type, which limits the performance of nested NER. Considering the effect of entity type information and entity boundary information, in this paper, we propose a multi-agent communication module to utilize these two kinds of information. Our multi-agent communication module contains a type labeling agent and a boundary labeling agent. The type labeling agent can utilize boundary information from boundary labeling agent to recognize entity type. And the boundary labeling agent can utilize type information from type labeling agent to recognize entity boundaries. They communicate and collaborate iteratively to finish the entity boundary recognition. Compared with previous methods, with the assist of entity type information and entity boundary information, the performance of boundary recognition improves. The improvement of boundary recognition is beneficial to recognizing nested named entities, which improves the performance of nested named entity recognition. Empirical experiments are conducted on three nested NER datasets. And the experimental results show the effectiveness of our model.