Generating Classical Chinese Poems from Vernacular Chinese

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 6155–6164, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 6155 Generating Classical Chinese Poems from Vernacular Chinese Zhichao Yang1∗, Pengshan Cai1∗, Yansong Feng2, Fei Li1, Weijiang Feng3, Elena Suet-Ying Chiu1, Hong Yu1 1 University of Massachusetts, MA, USA {zhichaoyang, pengshancai}@umass.edu foxlf823@gmail.com chiu@llc.umass.edu hong yu@uml.edu 2 Institute of Computer Science and Technology, Peking University, China fengyansong@pku.edu.cn 3 College of Computer, National University of Defense Technology, China fengweijiang14@nudt.edu.cn Abstract Classical Chinese poetry is a jewel in the trea- sure house of Chinese culture. Previous poem generation models only allow users to employ keywords to interfere the meaning of gener- ated poems, leaving the dominion of gener- ation to the model. In this paper, we pro- pose a novel task of generating classical Chi- nese poems from vernacular, which allows users to have more control over the semantic of generated poems. We adapt the approach of unsupervised machine translation (UMT) to our task. We use segmentation-based padding and reinforcement learning to address under- translation and over-translation respectively. According to experiments, our approach sig- nificantly improve the perplexity and BLEU compared with typical UMT models. Further- more, we explored guidelines on how to write the input vernacular to generate better poems. Human evaluation showed our approach can generate high-quality poems which are com- parable to amateur poems. 1 Introduction During thousands of years, millions of classical Chinese poems have been written. They contain ancient poets’ emotions such as their apprecia- tion for nature, desiring for freedom and concerns for their countries. Among various types of clas- sical poetry, quatrain poems stand out. On the one hand, their aestheticism and terseness exhibit unique elegance. On the other hand, composing such poems is extremely challenging due to their phonological, tonal and structural restrictions. Most previous models for generating classical Chinese poems (He et al., 2012; Zhang and Lap- ata, 2014) are based on limited keywords or char- acters at fixed positions (e.g., acrostic poems). ∗Equal contribution Since users could only interfere with the semantic of generated poems using a few input words, mod- els control the procedure of poem generation. In this paper, we proposed a novel model for classical Chinese poem generation. As illustrated in Figure 1, our model generates a classical Chinese poem based on a vernacular Chinese paragraph. Our ob- jective is not only to make the model generate aes- thetic and terse poems, but also keep rich semantic of the original vernacular paragraph. Therefore, our model gives users more control power over the semantic of generated poems by carefully writing the vernacular paragraph. Although a great number of classical poems and vernacular paragraphs are easily available, there exist only limited human-annotated pairs of poems and their corresponding vernacular translations. Thus, it is unlikely to train such poem generation model using supervised approaches. Inspired by unsupervised machine translation (UMT) (Lam- ple et al., 2018b), we treated our …

Content truncated to 500 words