2020 Virtual undergraduate Research symposium
Robot, Found: The Collaborative Human-Machine Poetry Writing Project
PROJECT NUMBER: 45
AUTHOR: Alexandria Leto, Computer Science
MENTOR: Thomas Williams, Computer Science | MENTOR: Toni Lefton, Humanities and Social Sciences
ABSTRACT
—
Found poetry is the practice of taking words or phrases from outside sources and rearranging them to create an original poem. One form of found poetry is blackout poetry, in which a section of text is selected, and the poet omits words such that remaining text forms a poem. This study aims to explore the use of computationally generated text as a source for found poetry, specifically in the style of blackout poetry.
Markov chains are an effective way to generate grammatically and thematically plausible text. A model is trained by inputting a large amount of text. The Markov algorithm then breaks the text into a series of words. An N-sized window slides across the words in their original order. The first N words are stored to generate a set of plausible N+1 words. The N+1 word is randomly selected from this set.
In creating several Markov models trained on various texts, a text-generation model that is functional for the purpose of writing poetry was developed. This model included text from news sources, poems, and artistic literature. A set of 15 poems was created as evidence of the merit of writing poetry in this way. During this process, it was found that the generated text is particularly useful in inspiring the poet with combinations of words and phrases that the human mind may not normally formulate, proving useful in developing fresh metaphors and themes. The blackout poetry method will be further utilized to explore public perceptions about computationally assisted art.
VISUAL PRESENTATION
—
AUTHOR BIOGRAPHY
—
Alexandria Leto is a senior graduating with her bachelor’s degree in Computer Science in May. After graduation, she plans to attend graduate school to pursue her doctoral degree in Computational Mathematics. She aims to eventually become a teaching professor. Her passions include all things art and math.
Hello Alexandria, very interesting work blending engineering and art!
Are there methods of quantifying your work to analyze how it compares to human generated poems? Such as opinion polls? Do you think this would be useful information to further refining your code?
Hello Alexandria, this is a very interesting project. It will be interesting to do a follow-study that looks at how humans especially poets evaluate these ai-generated poems. Good work!