Abstract
Vision is a common source of inspiration for poetry. The objects and thesentimental imprints that one perceives from an image may lead to variousfeelings depending on the reader. In this paper, we present a system of poetrygeneration from images to mimic the process. Given an image, we first extract afew keywords representing objects and sentiments perceived from the image.These keywords are then expanded to related ones based on their associations inhuman written poems. Finally, verses are generated gradually from the keywordsusing recurrent neural networks trained on existing poems. Our approach isevaluated by human assessors and compared to other generation baselines. Theresults show that our method can generate poems that are more artistic than thebaseline methods. This is one of the few attempts to generate poetry fromimages. By deploying our proposed approach, XiaoIce has already generated morethan 12 million poems for users since its release in July 2017. A book of itspoems has been published by Cheers Publishing, which claimed that the book isthe first-ever poetry collection written by an AI in human history.