Abstract:

English poetry has served as a consistent flashpoint in assessments LLM aesthetic and human-like performance. Poetry’s structures (rhyme, meter, and recognizable poetic forms such as sonnets and sestinas) as well as some its themes (emotional expression) have become a stand-in for non-literary scholars to argue that LLMs possess surprising artistic capabilities. At the same time, many of these studies under-theorize and universalize poems and poetry, and even the many ways that readers interact with art. This paper explores a dataset of poems in the Princeton Prosody Archive (“Found Poems”) that were used as examples for a variety of teaching tasks and explores how these examples might provide future AI researchers with both the requisite historical understanding of how English poetry has always been variously measured as well as the conceptual nuance of how using “poetry” as a stand in for universal human aesthetic experience narrows our understanding of both poetry and history. 

Bio:

Meredith Martin is Professor of English and Faculty Director of the Center for Digital Humanities at Princeton, which started under her leadership in 2014. Her book, The Rise and Fall of Meter, Poetry and English National Culture, 1860-1930 (Princeton UP, 2012), was the winner of the MLA Prize for a First Book, the Warren Brooks Prize for Literary Criticism, and co-winner of the Sonya Rudikoff Prize for the Best First Book in Victorian Studies. Her second book Poetry’s Data: Digital Humanities and the History of Prosody (Princeton UP, 2025) was published in April 2025 and argues that poetry can teach us how to think critically about data, and that critical data studies can teach us something about how we read poems. With several collaborators, she has been building and directing, since 2007, the Princeton Prosody Archive, which contains writing on poetics, prosody, rhetoric, grammar, speech, and literary history published between 1570-1923.