De-biasing Distantly Supervised Named Entity Recognition via Causal Intervention

Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, pages 4803–4813 August 1–6, 2021. ©2021 Association for Computational Linguistics 4803 De-biasing Distantly Supervised Named Entity Recognition via Causal Intervention Wenkai Zhang1,3, Hongyu Lin1,∗, Xianpei Han1,2,∗, Le Sun 1,2 1Chinese Information Processing Laboratory 2State Key Laboratory of Computer Science Institute of Software, Chinese Academy of Sciences, Beijing, China 3 University of Chinese Academy of Sciences, Beijing, China {wenkai2019,hongyu,xianpei,sunle}@iscas.ac.cn Abstract Distant supervision tackles the data bottleneck in NER by automatically generating training instances via dictionary matching. Unfortu- nately, the learning of DS-NER is severely dictionary-biased, which suffers from spu- rious correlations and therefore undermines the effectiveness and the robustness of the learned models. In this paper, we fundamen- tally explain the dictionary bias via a Struc- tural Causal Model (SCM), categorize the bias into intra-dictionary and inter-dictionary bi- ases, and identify their causes. Based on the SCM, we learn de-biased DS-NER via causal interventions. For intra-dictionary bias, we conduct backdoor adjustment to remove the spurious correlations introduced by the dictio- nary confounder. For inter-dictionary bias, we propose a causal invariance regularizer which will make DS-NER models more robust to the perturbation of dictionaries. Experiments on four datasets and three DS-NER models show that our method can significantly improve the performance of DS-NER. 1 Introduction Named entity recognition (NER) aims to identify text spans pertaining to specific semantic types, which is a fundamental task of information extrac- tion, and enables various downstream applications such as Relation Extraction (Lin et al., 2016) and Question Answering (Bordes et al., 2015). The past several years have witnessed the remarkable success of supervised NER methods using neural networks (Lample et al., 2016; Ma and Hovy, 2016; Lin et al., 2020), which can automatically extract effective features from data and conduct NER in an end-to-end manner. Unfortunately, supervised methods rely on high-quality labeled data, which is very labor-intensive, and thus severely restricts ∗Corresponding authors 0.3 0.4 0.5 0.6 D1 D2 D3 D4 All mentions (b) 0.0 0.2 0.4 0.6 0.8 1.0 Twitter CoNLL Webpage Wikigold In dictionary Not in dictionary (a) Figure 1: Dictionary bias in DS-NER happens both at intra-dictionary and inter-dictionary aspects: (a) Aver- aged likehoods of mentions in/not in the dictionary sig- nificantly diverge. (b) Mention likehoods of the models using different dictionaries significantly diverge. the application of current NER models. To resolve the data bottleneck, a promising approach is dis- tant supervision based NER (DS-NER). DS-NER automatically generates training data by matching entities in easily-obtained dictionaries with plain texts. Then this distantly-labeled data is used to train NER models, commonly be accompanied by a denoising step. DS-NER significantly reduces the annotation cost for building an effective NER model, and therefore has attracted great attention in recent years (Yang et al., 2018; Shang et al., 2018; Peng et al., 2019; Cao et al., 2019; Liang et al., 2020; Zhang et al., 2021). However, the learning of DS-NER is dictionary- biased, which severely …

Content truncated to 500 words