From Shakespeare to Li-Bai: Adapting a Sonnet Model to Chinese Poetry
28461Authors
nan
Abstract
AbstractIn this paper, we adapt Deep-speare, a joint neural network model for English sonnets, to Chinese poetry. We illustrate characteristics of Chinese quatrain and explain our architecture as well as training and generation procedure, which differs from Shakespeare sonnets in several aspects. We analyse the generated poetry and find that model works well for Chinese poetry, as it can: (1) generate coherent 4-line quatrains of different topics; and (2) capture rhyme automatically (to a certain extent).
Keywords
Full Content
From Shakespeare to Li-Bai: Adapting a Sonnet Model to Chinese Poetry Zhuohan Xie Jey Han Lau Trevor Cohn The University of Melbourne zhuohanx@student.unimelb.edu.au jeyhan.lau@gmail.com t.cohn@unimelb.edu.au Abstract In this paper, we adapt Deep-speare, a joint neural network model for English sonnets, to Chinese poetry. We illustrate the characteris- tics of Chinese quatrain and explain our archi- tecture as well as training and generation pro- cedure, which differs from Shakespeare son- nets in several aspects. We analyse the gen- erated poetry and find that the adapted model works well for Chinese poetry, as it can: (1) generate coherent 4-line quatrains of different topics; and (2) capture rhyme automatically to a certain extent. 1 Introduction Classical poetry is an important Chinese cultural heritage and holds great value. The most popular poetry type is quatrain. Table 1 shows a seven- character quatrain written by Li Bai, one of the most famous poets in Chinese history, which we will use to explain characteristics of Chinese qua- train. First, a Chinese quatrain must contain four sen- tences where each sentence has the same number (five or seven) of characters. Usually, these sen- tences follow the “introduction, follow-up, transi- tion and conclusion” mode, which means the first sentence starts a topic, the second continues it, then the third sentence extends it to a new realm and the last provides a summarisation (Wang, 2002). The rhythmic scheme in Chinese quatrain is simple: the last characters of the second and fourth sentences should rhyme. Optionally, the last character of the first sentence can also rhyme with them, as the qua- train shows in Table 1. As for tone, each character contains one single syllable and there are two tones for middle Chinese (Pulleyblank, 2011): “Ping” (level tone) or “Ze” (downward tone). Generally, tones at the same positions of two adjacent sen- tences are opposite and four common tonal pat- terns are illustrated in Wang (2002). Character 望庐山瀑布 【唐】李白 日照香炉生紫烟, * Z P P Z Z P 遥看瀑布挂前川。 * P * Z Z P P 飞流直下三千尺, * P * Z P P Z 疑是银河落九天。 * Z P P Z Z P Waterfall on Mount Lu Li Bai The Sunlit Censer peak exhales a wreath of cloud. Like an unpended stream the waterfall sounds loud. Its torrent dashes down three thousand feet from high. As if the Silver River fell from azure sky. Table 1: A 7-character Tang Quatrain by Li Bai. tone is indicated in Table 1 where ‘P’ indicates “Ping”, ‘Z’ refers to “Ze” and ‘*’ means this char- acter can be either tone. Following all these constraints, a well-written Chinese quatrain are full of rhythm while express- ing abundant emotions. 1.1 Motivation Inspired by Deep-speare (Lau et al., 2018), a joint neural network model that is trained on English sonnets to generate quatrains in iambic pentame- ter (Halle and Keyser, 1971), we seek to adapt this model to poetry in other languages. The abun- dance of Chinese quatrain and its important cul- tural status makes it a natural …
Content truncated to 500 words
Labeled Text
AbstractIn this paper, we adapt Deep-speare, a joint neural network model for English sonnets, to Chinese poetry. We illustrate characteristics of Chinese quatrain and explain our architecture as well as training and generation procedure, which differs from Shakespeare sonnets in several aspects. We analyse the generated poetry and find that model works well for Chinese poetry, as it can: (1) generate coherent 4-line quatrains of different topics; and (2) capture rhyme automatically (to a certain extent).
Publication Details
- Paper ID:
- 28461
- Year:
- 2019
- Publication:
- TIPSTER
- Publication Type:
- conf
- DOI:
- nan