Chinese Poetry Generation with a Salient-Clue Mechanism

41919

Authors

nan

Abstract

AbstractAs a precious part of the human cultural heritage, Chinese poetry has influenced people for generations. Automatic poetry composition is a challenge for AI. In recent years, significant progress has been made in this area benefiting from the development of neural networks. However, the coherence in meaning, theme or even artistic conception for a generated poem as a whole still remains a big problem. In this paper, we propose a novel Salient-Clue mechanism for Chinese poetry generation. Different from previous work which tried to exploit all the context information, our model selects the most salient characters automatically from each so-far generated line to gradually form a salient clue, which is utilized to guide successive poem generation process so as to eliminate interruptions and improve coherence. Besides, our model can be flexibly extended to control the generated poem in different aspects, for example, poetry style, which further enhances the coherence. Experimental results show that our model is very effective, outperforming three strong baselines.

Keywords

nan

Full Content

Proceedings of the 22nd Conference on Computational Natural Language Learning (CoNLL 2018), pages 241–250 Brussels, Belgium, October 31 - November 1, 2018. c⃝2018 Association for Computational Linguistics 241 Chinese Poetry Generation with a Salient-Clue Mechanism Xiaoyuan Yi1,2,3, Ruoyu Li5, Maosong Sun1,2,4∗ 1Department of Computer Science and Technology, Tsinghua University 2Institute for Artificial Intelligence, Tsinghua University 3State Key Lab on Intelligent Technology and Systems, Tsinghua University 4Beijing Advanced Innovation Center for Imaging Technology, Capital Normal University 56ESTATES PTE LTD, Singapore yi-xy16@mails.tsinghua.edu.cn, liruoyu@6estates.com, sms@mail.tsinghua.edu.cn Abstract As a precious part of the human cultural her- itage, Chinese poetry has influenced people for generations. Automatic poetry composition is a challenge for AI. In recent years, significant progress has been made in this area benefit- ing from the development of neural networks. However, the coherence in meaning, theme or even artistic conception for a generated poem as a whole still remains a big problem. In this paper, we propose a novel Salient-Clue mech- anism for Chinese poetry generation. Differ- ent from previous work which tried to exploit all the context information, our model selects the most salient characters automatically from each so-far generated line to gradually form a salient clue, which is utilized to guide succes- sive poem generation process so as to elimi- nate interruptions and improve coherence. Be- sides, our model can be flexibly extended to control the generated poem in different as- pects, for example, poetry style, which further enhances the coherence. Experimental results show that our model is very effective, outper- forming three strong baselines. 1 Introduction As a fascinating literary form starting from the Pre-Qin Period, Chinese poetry has influenced people for generations and thus influenced Chi- nese culture and history in thousands of years. Po- ets often write poems to record interesting events and express their feelings. In fact, the ability to create high-quality poetry has become an indicator of knowledge, wisdom and elegance of a person in China. Generally, a Chinese poem should meet two kinds of requirements. One is from the perspec- tive of form: it must obey some structural and phonological rules strictly. For example (as shown in Figure 1), quatrain (Jueju in Chinese), one of the most popular types of Chinese poetry, contains ∗Corresponding author: sms@mail.tsinghua.edu.cn. Figure 1: A Wujue generated by our model. The tone of each character is given in parentheses, where P and Z represent Ping (level tone) and Ze (oblique tone) re- spectively. Rhyming characters are underlined. The left part is an artistic illustration of the salient clue. four lines with each consisting of five or seven characters (called Wujue and Qijue respectively); characters with particular tone must be in partic- ular positions to make the poem cadenced and full of rhythmic beauty; and, the last character of the first (optional), second and fourth lines must rhyme. The other one is from the perspective of content, concerning: (1) if each line of the poem is adequate syntactically and semantically; (2) if the association between two adjacent lines is rea- sonable; and …

Content truncated to 500 words

Labeled Text

AbstractAs a precious part of the human cultural heritage, Chinese poetry has influenced people for generations. Automatic poetry composition is a challenge for AI. In recent years, significant progress has been made in this area benefiting from the development of neural networks. However, the coherence in meaning, theme or even artistic conception for a generated poem as a whole still remains a big problem. In this paper, we propose a novel Salient-Clue mechanism for Chinese poetry generation. Different from previous work which tried to exploit all the context information, our model selects the most salient characters automatically from each so-far generated line to gradually form a salient clue, which is utilized to guide successive poem generation process so as to eliminate interruptions and improve coherence. Besides, our model can be flexibly extended to control the generated poem in different aspects, for example, poetry style, which further enhances the coherence. Experimental results show that our model is very effective, outperforming three strong baselines.

Publication Details

Paper ID:
41919
Year:
2018
Publication:
TIPSTER
Publication Type:
conf
DOI:
nan