Named Entity Recognition in Multi-level Contexts

Proceedings of the 1st Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics and the 10th International Joint Conference on Natural Language Processing, pages 181–190 December 4 - 7, 2020. c⃝2020 Association for Computational Linguistics 181 Named Entity Recognition in Multi-level Contexts Yubo Chen Chuhan Wu Tao Qi Zhigang Yuan Yongfeng Huang Department of Electronic Engineering & BNRist, Tsinghua University, Beijing 100084, China {ybch14, wuchuhan15, taoqi.qt}@gmail.com {zgyuan, yfhuang}@tsinghua.edu.cn Abstract Named entity recognition is a critical task in the natural language processing field. Most existing methods for this task can only ex- ploit contextual information within a sentence. However, their performance on recognizing en- tities in limited or ambiguous sentence-level contexts is usually unsatisfactory. Fortunately, other sentences in the same document can pro- vide supplementary document-level contexts to help recognize these entities. In addition, words themselves contain word-level contex- tual information since they usually have dif- ferent preferences of entity type and relative position from named entities. In this paper, we propose a unified framework to incorporate multi-level contexts for named entity recogni- tion. We use TagLM as our basic model to capture sentence-level contexts. To incorpo- rate document-level contexts, we propose to capture interactions between sentences via a multi-head self attention network. To mine word-level contexts, we propose an auxiliary task to predict the type of each word to capture its type preference. We jointly train our model in entity recognition and the auxiliary classifi- cation task via multi-task learning. The exper- imental results on several benchmark datasets validate the effectiveness of our method. 1 Introduction Named Entity Recognition (NER) is defined as automatically identifying and classifying named entities into specific categories (e.g., person, loca- tion, organization) in text. It is a critical task in Natural Language Processing (NLP) and a prereq- uisite for many downstream tasks, such as entity linking (Luo et al., 2015), relation extraction (Feld- man and Rosenfeld, 2006) and question answering (Lee et al., 2006). NER is usually modeled as a sentence-level se- quence labeling task in previous work. For exam- ple, Lample et al. (2016) used long-short term Sentence 1 When Fred was still in High School he set up a business with his mother called Elizabeth Trump (PER) and Son. × When Fred was still in High School he set up a business with his mother called Elizabeth Trump and Son (ORG). � Russ Berrie and Co Inc (ORG) said on Friday that A. Curts Cooke (PER) will retire as chief operating officer. � Russ Berrie and Co Inc (PER) said on Friday that A. Curts Cooke (PER) will retire as chief operating officer. × Sentence 2 When Fred was still in high school he set up a business with his mother called Elizabeth Trump and Son. While in college, Donald Trump (PER) began his first real estate career at his father’s company, Elizabeth Trump and Son (ORG). Action Performance Cos Inc (ORG) said Friday it has agreed to acquire Motorsport Traditions Ltd (ORG) and Creative Marketing & Promotions Inc (ORG) for …

Content truncated to 500 words