The publication of Mark Twain’s autobiography 100 years after his death presents a difficult proposition for the reading public with an unwieldy size of over 2000 pages. The work also presents an opportunity for computational text analysis to provide insights where more traditional close readings of it might prove difficult. My dissertation, “The Art of Literary Modeling”, demonstrates new scholarly activities that provide context sensitive, evidentiary bases for making interpretive claims from computational models of literature. Its first chapter answers the question, “What does it mean to read literature as data?” That chapter’s discussion of “literary” data quality looks to versions of Emily Dickinson’s poetry amid her multiple, posthumous publications. It adapts data quality metrics from information science to provide an overall assessment of those digital versions, and produces a new collection of poems that reflect their data quality with respect to her oeuvre. This next chapter will answer the followup question: “What does it mean to model literature as data?” Applying standards for model selection and quality analysis from statistics known as “information criterion” reveals differences between various models of a work. That comparison made possible by information criteria presents a new, qualitative decision point for the study of literature as data. By looking to passages of Twain’s autobiography noted by high quality models of it and then further assessing them with scholarly understandings of those passages, it becomes possible to produce new findings on the autobiography and sound methodology for model quality generalizable to humanities research beyond this project.

---