Locate and Label: A Two-stage Identifier for Nested Named Entity Recognition

Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, pages 2782–2794 August 1–6, 2021. ©2021 Association for Computational Linguistics 2782 Locate and Label: A Two-stage Identifier for Nested Named Entity Recognition Yongliang Shen1, Xinyin Ma1, Zeqi Tan1, Shuai Zhang1, Wen Wang2, Weiming Lu1∗ 1College of Computer Science and Technology, Zhejiang University 2University of Science and Technology of China {syl, luwm}@zju.edu.cn Abstract Named entity recognition (NER) is a well- studied task in natural language processing. Traditional NER research only deals with flat entities and ignores nested entities. The span-based methods treat entity recognition as a span classification task. Although these methods have the innate ability to handle nested NER, they suffer from high computa- tional cost, ignorance of boundary informa- tion, under-utilization of the spans that par- tially match with entities, and difficulties in long entity recognition. To tackle these issues, we propose a two-stage entity identifier. First we generate span proposals by filtering and boundary regression on the seed spans to lo- cate the entities, and then label the boundary- adjusted span proposals with the correspond- ing categories. Our method effectively utilizes the boundary information of entities and par- tially matched spans during training. Through boundary regression, entities of any length can be covered theoretically, which improves the ability to recognize long entities. In addition, many low-quality seed spans are filtered out in the first stage, which reduces the time com- plexity of inference. Experiments on nested NER datasets demonstrate that our proposed method outperforms previous state-of-the-art models. 1 Introduction Named entity recognition (NER) is a fundamental task in natural language processing, focusing on identifying the spans of text that refer to entities. NER is widely used in downstream tasks, such as entity linking (Ganea and Hofmann, 2017; Le and Titov, 2018) and relation extraction (Li and Ji, 2014; Miwa and Bansal, 2016). Previous works usually treat NER as a sequence labeling task, assigning a single tag to each to- ∗ Corresponding author Object Detection Named Entity Recognition Flat Nested cow cow tenis racket person A U . S . tourist was detained and accused of spying after photographing a riot in the province of Irian Jaya . GPE PER GPE GPE FAC He said that 47,000 cars drive by the site daily and because the players have to drive by it every day it could become a hangout for them . VEH PER PER FAC PER FAC Figure 1: A Comparison of Named Entity Recognition and Object Detection. Examples of flat and nesetd en- tities or objects sampled from the COCO 2017 dataset and the ACE04 dataset, respectively. ken in a sentence. Such models lack the abil- ity to identify nested named entities. Various ap- proaches for nested NER have been proposed in recent years. Some works revised sequence models to support nested entities using different strategies (Alex et al., 2007; Ju et al., 2018; Strakov´a et al., 2019; Wang et al., 2020a) and some works …

Content truncated to 500 words