Abstract
Versification has not been central to studies of AI poetry and poetic style. If we build tools to study the prosodic and phonological capacities of AI, we learn not just about AI verse but about hidden histories of reception and genre. The computational humanities can unpack these histories and explain how LLMs reify a limited kind of poetry and language. We can also compile data to train models that behave differently. I share a >100K line data set of scanned 18th century iambic pentameter, compare its versification with generative AI verse, and train BERT and T5 models with this data to scan meter and classify complexity. This tool lays the groundwork for exposing text-based LLMs to phonology and poetic prosody.