Because of the limitation of the paper, we will show more examples in this supplementary material. 

We will show that the results of two configurations of the memory-augmented neural network, one can generate the highest level of poetry that is considered by most experts through the analysis of the experimental results; the other can generate different styles of poems with different style memory. The former is one-iteration model augmented with memory， The latter is one-iteration model augmented with style memory.

This document contains a total of 4 files，each consists of randomly selected 50 poems, explained as follows:

poems_by__C1_with_memory.txt : the poems are generated by one-iteration model augmented with memory;
Battlefield_style_poems.txt : the poems are generated by one-iteration model augmented with battlefield style memory;
Pastoral_style_poems.txt : the poems are generated by one-iteration model augmented with pastoral style memory;
Romantic_style_poems.txt : the poems are generated by one-iteration model augmented with romantic style memory.

