AdvPicker: Effectively Leveraging Unlabeled Data via Adversarial Discriminator for Cross-Lingual NER

Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, pages 743–753 August 1–6, 2021. ©2021 Association for Computational Linguistics 743 AdvPicker: Effectively Leveraging Unlabeled Data via Adversarial Discriminator for Cross-Lingual NER Weile Chen1, Huiqiang Jiang2, Qianhui Wu3, B¨orje F. Karlsson4, and Yi Guan1 1Harbin Institute of Technology, Harbin, 15000, China 2Peking University, Beijing, 100871, China 3Department of Automation, Tsinghua University, Beijing, 100084, China 4Microsoft Research, Beijing, 100080, China chen.weile7@gmail.com, jhq@pku.edu.cn, guanyi@hit.edu.cn, wuqianhui@tsinghua.org.cn, borje.karlsson@microsoft.com Abstract Neural methods have been shown to achieve high performance in Named Entity Recogni- tion (NER), but rely on costly high-quality la- beled data for training, which is not always available across languages. While previous works have shown that unlabeled data in a tar- get language can be used to improve cross- lingual model performance, we propose a novel adversarial approach (AdvPicker) to bet- ter leverage such data and further improve re- sults. We design an adversarial learning frame- work in which an encoder learns entity domain knowledge from labeled source-language data and better shared features are captured via ad- versarial training - where a discriminator se- lects less language-dependent target-language data via similarity to the source language. Experimental results on standard benchmark datasets well demonstrate that the proposed method benefits strongly from this data selec- tion process and outperforms existing state-of- the-art methods; without requiring any addi- tional external resources (e.g., gazetteers or via machine translation). 1 1 Introduction Named entity recognition (NER) is a fundamental information extraction task, which seeks to identify named entities in text and classify them into pre- defined entity types (such as person, organization, location, etc.) and it is key in various downstream tasks, e.g., question answering (Moll´a et al., 2006). Neural NER models are highly successful for lan- guages with a large amount of quality annotated data. However, most languages don’t have enough labeled data to train a fully supervised model. This motivates research on cross-lingual transfer, which leverages labeled data from a source language (e.g., 1Code is publicly available at https://aka.ms/ AdvPicker English) to address the lack of training data prob- lem in a target language. In this paper, following Wu and Dredze (2019) and Wu et al. (2020a), we fo- cus on zero-shot cross-lingual NER, where labeled data is not available in the target language. The state-of-the-art methods for zero-shot cross-lingual NER are mainly divided into three categories: i) feature-based methods (Wu and Dredze, 2019; Wu et al., 2020b; Pfeiffer et al., 2020), which train a NER model to capture language-independent features of the labeled source-language data and then apply it to the target language; ii) translation-based methods (Mayhew et al., 2017; Xie et al., 2018), which build pseudo target-language dataset via translating from labeled source-language data and mapping entity labels; and iii) pseudo-labeling methods, which generate pseudo-labeled data for training a target-language NER model via a source-language model (Wu et al., 2020a) or annotation projection (Ni et al., 2017). However, each method has its …

Content truncated to 500 words