Chinese Poetry Generation with Recurrent Neural Networks
34636Authors
nan
Abstract
We propose a model for Chinese poem generation based on recurrent neural net- works which we argue is ideally suited to capturing poetic content and form. Our generator jointly performs content selec- tion (“what to say”) and surface realization (“how to say”) by learning representations of individual characters, and their com- binations into one or more lines as well as how these mutually reinforce and con- strain each other. Poem lines are gener- ated incrementally by taking into account the entire history of what has been gen- erated so far rather than the limited hori- zon imposed by the previous line or lexical n-grams. Experimental results show that our model outperforms competitive Chi- nese poetry generation systems using both automatic and manual evaluation methods. 1
Keywords
Full Content
Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 670–680, October 25-29, 2014, Doha, Qatar. c⃝2014 Association for Computational Linguistics Chinese Poetry Generation with Recurrent Neural Networks Xingxing Zhang and Mirella Lapata Institute for Language, Cognition and Computation School of Informatics, University of Edinburgh 10 Crichton Street, Edinburgh EH8 9AB x.zhang@ed.ac.uk, mlap@inf.ed.ac.uk Abstract We propose a model for Chinese poem generation based on recurrent neural net- works which we argue is ideally suited to capturing poetic content and form. Our generator jointly performs content selec- tion (“what to say”) and surface realization (“how to say”) by learning representations of individual characters, and their com- binations into one or more lines as well as how these mutually reinforce and con- strain each other. Poem lines are gener- ated incrementally by taking into account the entire history of what has been gen- erated so far rather than the limited hori- zon imposed by the previous line or lexical n-grams. Experimental results show that our model outperforms competitive Chi- nese poetry generation systems using both automatic and manual evaluation methods. 1 Introduction Classical poems are a significant part of China’s cultural heritage. Their popularity manifests itself in many aspects of everyday life, e.g., as a means of expressing personal emotion, political views, or communicating messages at festive occasions as well as funerals. Amongst the many differ- ent types of classical Chinese poetry, quatrain and regulated verse are perhaps the best-known ones. Both types of poem must meet a set of structural, phonological, and semantic requirements, render- ing their composition a formidable task left to the very best scholars. An example of a quatrain is shown in Table 1. Quatrains have four lines, each five or seven char- acters long. Characters in turn follow specific phonological patterns, within each line and across lines. For instance, the final characters in the sec- ond, fourth and (optionally) first line must rhyme, 相 思 Missing You 红 豆 生 南 国, (* Z P P Z) Red berries born in the warm southland. 春 来 发 几 枝 枝 枝? (P P Z Z P) How many branches flush in the spring? 愿 君 多 采 撷, (* P P Z Z) Take home an armful, for my sake, 此 物 最 相 思 思 思。 (* Z Z P P) As a symbol of our love. Table 1: An example of a 5-char quatrain ex- hibiting one of the most popular tonal patterns. The tone of each character is shown at the end of each line (within parentheses); P and Z are short- hands for Ping and Ze tones, respectively; * indi- cates that the tone is not fixed and can be either. Rhyming characters are shown in boldface. whereas there are no rhyming constraints for the third line. Moreover, poems must follow a pre- scribed tonal pattern. In traditional Chinese, ev- ery character has one tone, Ping (level tone) or Ze (downward tone). The poem in Table 1 exempli- fies …
Content truncated to 500 words
Labeled Text
We propose a model for Chinese poem generation based on recurrent neural networks which we argue is ideally suited to capturing poetic content and form. Our generator jointly performs content selection (“what to say”) and surface realization (“how to say”) by learning representations of individual characters, and their combinations into one or more lines as well as how these mutually reinforce and constrain each other. Poem lines are generated incrementally by taking into account the entire history of what has been generated so far rather than the limited horizon imposed by the previous line or lexical n-grams. Experimental results show that our model outperforms competitive Chinese poetry generation systems using both automatic and manual evaluation methods. 1
Publication Details
- Paper ID:
- 34636
- Year:
- 2014
- Publication:
- TIPSTER
- Publication Type:
- conf
- DOI:
- nan