A Chinese Corpus for Fine-grained Entity Typing

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 4451–4457 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 4451 A Chinese Corpus for Fine-grained Entity Typing Chin Lee1, Hongliang Dai1, Yangqiu Song1, Xin Li2 1HKUST 2Tencent Technology (SZ) Co., Ltd. cleeag@ust.hk, {hdai,yqsong}@cse.ust, alonsoli@tencent.com Abstract Fine-grained entity typing is a challenging task with wide applications. However, most existing datasets for this task are in English. In this paper, we introduce a corpus for Chinese fine-grained entity typing that contains 4,800 mentions manually labeled through crowdsourcing. Each mention is annotated with free-form entity types. To make our dataset useful in more possible scenarios, we also categorize all the fine-grained types into 10 general types. Finally, we conduct experiments with some neural models whose structures are typical in fine-grained entity typing and show how well they perform on our dataset. We also show the possibility of improving Chinese fine-grained entity typing through cross-lingual transfer learning. Keywords: Fine-grained entity typing, Entity typing 1. Introduction The task of fine-grained entity typing (Ling and Weld, 2012; Gillick et al., 2014) assigns fine-grained types such as /person/politician, /organization/company to entity mentions in texts. It provides additional details to entity mentions compared with the typing in tradi- tional named entity recognition tasks (Chinchor, 1998; Finkel et al., 2005), which typically categorize entity mentions into very general types such as person, loca- tion, or organization. Ultra-fine Entity Typing (Choi et al., 2018) introduces a new fine-grained entity typing task that requires to predict an open set of types for entity mentions. The dataset constructed for this task uses a very large tag set that contains around 10k free-form type phrases, while previous fine-grained entity typing datasets usu- ally use tag sets with no greater than 200 types. This task presents a much closer view for each entity men- tion. Consider the sentence: “Tim Cook announced the new iPhone this morning.” With the dataset con- structed by (Gillick et al., 2014), the mention “Tim Cook” can only be identified as /person/business. But with ultra-fine entity typing, “Tim Cook” can be cat- egorized under types such as businessman, executive, public figure, etc. These free-form type phrases provide a more comprehensive and detailed description on the entity mention. Unfortunately, most corpora (Ling and Weld, 2012; Weischedel and Brunstein, 2005; Gillick et al., 2014; Choi et al., 2018) of fine-grained entity typing are in English. To our knowledge, there doesn’t exist a large- scale fine-grained entity typing dataset exclusively in Chinese. In view of the growth of the research in Chi- nese NLP, a dataset for Chinese fine-grained entity typing will provide great value. Thus, in this paper, we present a Chinese corpus of extremely fine-grained en- tity typing containing over 7,100 unique entity types. We adopt a similar policy as the Ultra-fine Entity Typ- Sentence with Mention Label Types 高尔基大街（现易名为特维尔大 街）是莫斯科一条最主要的大街 Gorky Street (now as known as Tverskaya Street) is one of the main streets in Moscow. 街 道/street, 旅 游 景 点/ …

Content truncated to 500 words