Parallel Instance Query Network for Named Entity Recognition

Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 947 - 961 May 22-27, 2022 c⃝2022 Association for Computational Linguistics Parallel Instance Query Network for Named Entity Recognition Yongliang Shen1∗, Xiaobin Wang2, Zeqi Tan1, Guangwei Xu2, Pengjun Xie2, Fei Huang2, Weiming Lu1†, Yueting Zhuang1 1College of Computer Science and Technology, Zhejiang University 2DAMO Academy, Alibaba Group {syl, luwm}@zju.edu.cn xuanjie.wxb@alibaba-inc.com Abstract Named entity recognition (NER) is a funda- mental task in natural language processing. Re- cent works treat named entity recognition as a reading comprehension task, constructing type- specific queries manually to extract entities. This paradigm suffers from three issues. First, type-specific queries can only extract one type of entities per inference, which is inefficient. Second, the extraction for different types of entities is isolated, ignoring the dependencies between them. Third, query construction relies on external knowledge and is difficult to apply to realistic scenarios with hundreds of entity types. To deal with them, we propose Parallel Instance Query Network (PIQN), which sets up global and learnable instance queries to extract entities from a sentence in a parallel manner. Each instance query predicts one entity, and by feeding all instance queries simultaneously, we can query all entities in parallel. Instead of being constructed from external knowledge, in- stance queries can learn their different query se- mantics during training. For training the model, we treat label assignment as a one-to-many Lin- ear Assignment Problem (LAP) and dynami- cally assign gold entities to instance queries with minimal assignment cost. Experiments on both nested and flat NER datasets demonstrate that our proposed method outperforms previous state-of-the-art models1. 1 Introduction Named Entity Recognition (NER) aims to identify text spans to specific entity types such as Person, Location, Organization. It has been widely used in many downstream applications such as entity linking (Ganea and Hofmann, 2017; Le and Titov, 2018) and relation extraction (Li and Ji, 2014; * This work was conducted when Yongliang Shen was interning at Alibaba DAMO Academy. † Corresponding author. 1 Our code is available at https://github.com/ tricktreat/piqn. (a) Type-specific Query (b) Instance Query Model Context All Entities Model Context PER Entities PER Query Instance Queries PER LOC FAC Figure 1: (a) For a sentence, type-specific queries can only extract entities of one type per inference, so the model needs to be run multiple times. (b) In contrast, instance-based queries can be input into the model si- multaneously, and all entities can be extracted in parallel. Furthermore, the parallel manner can model the interac- tions between entities of different types. Miwa and Bansal, 2016; Shen et al., 2021b). Tradi- tional approaches for NER are based on sequence labeling, assigning a single tag to each word in a sentence. However, the words of nested entities have more than one tag, thus these methods lack the ability to identify nested entities. Recently, Ju et al. (2018); Straková et al. (2019); Wang et al. (2020a) redesign sequence labeling models to support nested structures using different strategies. Instead …

Content truncated to 500 words