Nested Named Entity Recognition as Corpus Aware Holistic Structure Parsing

Proceedings of the 29th International Conference on Computational Linguistics, pages 2472–2482 October 12–17, 2022. 2472 Nested Named Entity Recognition as Corpus Aware Holistic Structure Parsing Yifei Yang1,2,†, Zuchao Li1,2,†, Hai Zhao1,2,∗ 1Department of Computer Science and Engineering, Shanghai Jiao Tong University 2MoE Key Lab of Artificial Intelligence, AI Institute, Shanghai Jiao Tong University {yifeiyang, charlee}@sjtu.edu.cn, zhaohai@cs.sjtu.edu.cn Abstract As a fundamental natural language process- ing task and one of core knowledge extraction techniques, named entity recognition (NER) is widely used to extract information from texts for downstream tasks. Nested NER is a branch of NER in which the named entities (NEs) are nested with each other. However, most of the previous studies on nested NER usually apply linear structure to model the nested NEs which are actually accommodated in a hierarchical structure. Thus in order to address this mis- match, this work models the full nested NEs in a sentence as a holistic structure, then we propose a holistic structure parsing algorithm to disclose the entire NEs once for all. Be- sides, there is no research on applying corpus- level information to NER currently. To make up for the loss of this information, we intro- duce Point-wise Mutual Information (PMI) and other frequency features from corpus-aware statistics for even better performance by holis- tic modeling from sentence-level to corpus- level. Experiments show that our model yields promising results on widely-used benchmarks which approach or even achieve state-of-the- art. Further empirical studies show that our proposed corpus-aware features can substan- tially improve NER domain adaptation, which demonstrates the surprising advantage of our proposed corpus-level holistic structure model- ing.1 1 Introduction Named Entity Recognition is to find predefined named entities such as locations, organizations or people in text, which usually serves as an upstream natural language processing (NLP) task (Huang et al., 2015; Lample et al., 2016; Ma and Hovy, ∗ Corresponding author. † These authors made equal contribution to this paper. This work was supported by Key Projects of National Natural Science Foundation of China (U1836222 and 61733011). 1https://github.com/yangyifei729/ NerAsParsing … San Francisco of France GPE GPE GPE … Figure 1: An example of nested named entities. The solid lines or brackets represents the start and end po- sitions of a named entity. ‘GPE’ is a named entity category, representing geographic or political entities. France and San Francisco are nested by the San Fran- cisco of France which forms a hierarchy structure. 2016) and one of key knowledge extraction tech- niques in knowledge engineering. Early works on NER (Liu et al., 2011; Ritter et al., 2011) mostly cope with only flat NEs that have no overlapped relationship with each other. However, as a common language phenomenon, nested NER appears universally in many corpora such as the field of biology or news events. An instance of overlapped NEs is shown in Figure 1. Since solving the Nested NER task will bring more contextual information than the Flat NER task and promotes downstream tasks, it recently has aroused great research interest (Luo and Zhao, …

Content truncated to 500 words