NCLS: Neural Cross-Lingual Summarization

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 3054–3064, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 3054 NCLS: Neural Cross-Lingual Summarization Junnan Zhu1,2, Qian Wang1,2, Yining Wang1,2, Yu Zhou1,2∗, Jiajun Zhang1,2, Shaonan Wang1,2, and Chengqing Zong1,2,3 1 National Laboratory of Pattern Recognition, Institute of Automation, CAS, Beijing, China 2 University of Chinese Academy of Sciences, Beijing, China 3 CAS Center for Excellence in Brain Science and Intelligence Technology, Beijing, China {junnan.zhu, yzhou, jjzhang, cqzong}@nlpr.ia.ac.cn Abstract Cross-lingual summarization (CLS) is the task to produce a summary in one particular lan- guage for a source document in a differ- ent language. Existing methods simply di- vide this task into two steps: summarization and translation, leading to the problem of er- ror propagation. To handle that, we present an end-to-end CLS framework, which we re- fer to as Neural Cross-Lingual Summarization (NCLS), for the first time. Moreover, we pro- pose to further improve NCLS by incorporat- ing two related tasks, monolingual summariza- tion and machine translation, into the train- ing process of CLS under multi-task learn- ing. Due to the lack of supervised CLS data, we propose a round-trip translation strategy to acquire two high-quality large-scale CLS datasets based on existing monolingual sum- marization datasets. Experimental results have shown that our NCLS achieves remarkable im- provement over traditional pipeline methods on both English-to-Chinese and Chinese-to- English CLS human-corrected test sets. In ad- dition, NCLS with multi-task learning can fur- ther significantly improve the quality of gen- erated summaries. We make our dataset and code publicly available here: http://www. nlpr.ia.ac.cn/cip/dataset.htm. 1 Introduction Given a document in one source language, cross- lingual summarization aims to produce a summary in a different target language, which can help peo- ple efficiently acquire the gist of an article in a foreign language. Traditional approaches to CLS are based on the pipeline paradigm, which ei- ther first translates the original document into tar- get language and then summarizes the translated document (Leuski et al., 2003) or first summa- rizes the original document and then translates the ∗Corresponding author. summary into target language (Lim et al., 2004; Orasan and Chiorean, 2008; Wan et al., 2010). However, the current machine translation (MT) is not perfect, which results in the error propagation problem. Although end-to-end deep learning has made great progress in natural language process- ing, no one has yet applied it to CLS due to the lack of large-scale supervised dataset. The input and output of CLS are in two dif- ferent languages, which makes the data acquisi- tion much more difficult than monolingual sum- marization (MS). To the best of our knowledge, no one has studied how to automatically build a high-quality large-scale CLS dataset. Therefore, in this work, we introduce a novel approach to di- rectly address the lack of data. Specifically, we propose a simple yet effective round-trip trans- lation strategy to obtain cross-lingual document- summary pairs from existing …

Content truncated to 500 words