Abstract

Research has shown that sequence-to-sequence neural models, particularlythose with the attention mechanism, can successfully generate classical Chinesepoems. However, neural models are not capable of generating poems that matchspecific styles, such as the impulsive style of Li Bai, a famous poet in theTang Dynasty. This work proposes a memory-augmented neural model to enable thegeneration of style-specific poetry. The key idea is a memory structure thatstores how poems with a desired style were generated by humans, and usessimilar fragments to adjust the generation. We demonstrate that the proposedalgorithm generates poems with flexible styles, including styles of aparticular era and an individual poet.

 

Quick Read (beta)