Fine-grained Entity Typing via Label Reasoning

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 4611–4622 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 4611 Fine-grained Entity Typing via Label Reasoning Qing Liu1,3, Hongyu Lin1∗, Xinyan Xiao4, Xianpei Han1,2∗, Le Sun1,2, Hua Wu4 1Chinese Information Processing Laboratory 2State Key Laboratory of Computer Science Institute of Software, Chinese Academy of Sciences, Beijing, China 3University of Chinese Academy of Sciences, Beijing, China 4Baidu Inc., Beijing, China {liuqing2020,hongyu,xianpei,sunle}@iscas.ac.cn {xiaoxinyan,wu_hua}@baidu.com Abstract Conventional entity typing approaches are based on independent classification paradigms, which make them difficult to recognize inter- dependent, long-tailed and fine-grained entity types. In this paper, we argue that the im- plicitly entailed extrinsic and intrinsic depen- dencies between labels can provide critical knowledge to tackle the above challenges. To this end, we propose Label Reasoning Net- work(LRN), which sequentially reasons fine- grained entity labels by discovering and ex- ploiting label dependencies knowledge en- tailed in the data. Specifically, LRN utilizes an auto-regressive network to conduct deduc- tive reasoning and a bipartite attribute graph to conduct inductive reasoning between labels, which can effectively model, learn and reason complex label dependencies in a sequence-to- set, end-to-end manner. Experiments show that LRN achieves the state-of-the-art perfor- mance on standard ultra fine-grained entity typing benchmarks, and can also resolve the long tail label problem effectively. 1 Introduction Fine-grained entity typing (FET) aims to classify entity mentions to a fine-grained semantic label set, e.g., classify “FBI agents" in “They were ar- rested by FBI agents." as {organization, adminis- tration, force, agent, police}. By providing fine- grained semantic labels, FET is critical for entity recognition (Lin et al., 2019a,b, 2020; Zhang et al., 2021b,a) and can benefit many NLP tasks, such as relation extraction (Yaghoobzadeh et al., 2017; Zhang et al., 2019), entity linking (Onoe and Dur- rett, 2020) and question answering (Yavuz et al., 2016). The fundamental challenge of FET comes from its large-scale and fine-grained entity label set, which leads to significant difference between FET and conventional entity typing. First, due to the ∗Corresponding authors. person theorist musician commander guitarist composer singer (a) Extrinsic dependency military scholartheoretical expert commander scientist theorist label attribute (b) Intrinsic dependency Only …, [they] theorize, could collapse some of the war ship … “they” person “they theorize” person “war ship” person theorist commander expert scholar scientist ++ + Deductive Reasoning Inductive Reasoning Context (c) Label reasoning process Figure 1: Examples of deductive reasoning based on the extrinsic dependency and inductive reasoning based on the intrinsic dependency, where the labels per- son, theorist and commander are deducted respectively and the label scientist is inducted from the attributes {expert, scholar}. massive label set, it is impossible to independently recognize each entity label without considering their dependencies. For this, existing approaches use the predefined label hierarchies (Ren et al., 2016a; Shimaoka et al., 2017; Abhishek et al., 2017; Karn et al., 2017; Xu and Barbosa, 2018; Wu et al., 2019; Chen et al., 2020; Ren, 2020) or label co-occurrence statistics from training data (Rabi- novich and …

Content truncated to 500 words