Detecting Poetic Metaphors by LDA-based Topic Distribution

Abstract

It is difficult to automatically extract a metaphor from Chinese poetry. In Chinese poetry, a metaphor appears when a word has a different, implicit connotation from its original, explicit significance. The meaning of a word in a non-literary text is its original, explicit sense. Thereby, we assume the metaphorical word, which has different nuances in a poem and non-literary texts (which form a semantically inconsistent pair). Depending on the text, a word is semantically inconsistent. For example, a “moon” is a satellite of the Earth in a non-literary setting, while in the poem “Quiet Night Thoughts,” the term “moon” means homesickness. Hence, the “moon” is an SIP in “Quiet Night Thoughts” and non-literary texts. This paper aims to detect SIPs in Chinese poems and non-literary texts. In particular, we discern SIP based on latent Dirichlet allocation (LDA) topic modeling. Subsequently, the proposed method has been evaluated by discovering SIP in Chinese poetry and non-literary texts.