Machine Masquerades a Poet: Using Unsupervised T5 Transformer for Semantic Style Transformation in Poetry Generation

This paper presents a novel approach to automatically capturing the unique style of various poets and using it to convert given poems into the styles of those poets. The method combines the power of web scraping, and T5 transformer (11 Billion Parameters). A dataset of poems was collected by web scraping popular online libraries, such as Project Gutenberg and Open Library. These poems were then pre-processed to remove using HTML tags and meta data. The pre-trained T5 Transformer-11b was fine tuned on this corpus of text. The results of this study were highly promising. The proposed method accurately captured the styles of various poets, effectively capturing their overall tone, ideologies, and poetic style. By providing a starting poem, the model generated new poems in the style of a specific poet, successfully mimicking their unique writing characteristics. These findings highlight the potential of machine learning algorithms in understanding and reproducing the intricate nuances of poetic styles. This work opens avenues for automated poem generation, enabling individuals to experience the styles and voices of renowned poets in a novel way.