Enumerations: Data and Literary Study by Andrew Piper (review)

Soon after publishing Enumerations: Data and Literary Study, Andrew Piper launched a website called The Fish and The Painting that guides readers in methods of "data-driven research in the humanities." The website's enigmatic title refers to a critique leveled against computational literary criticism of the sort Piper practices throughout Enumerations. "A critic once quipped," Piper explains, "that using numbers to study literature, culture, and history was like throwing a fish at a painting. You could, but why would you?"

Enumerations seeks to answer that question by using computational models to study literary history. Each chapter models a literary feature by applying computational models to texts that have been selected to shed light on the feature in question. Computationally analyzing these texts yields the data promised in the book's subtitle. The data ultimately becomes both an object of critical attention and an occasion for Piper to turn to more familiar methods like close reading to place macro patterns in micro contexts. Enumerations thus reads literature at several scales: one so large that it is accessible only through computation, while the others operate at more familiar scales like a chapter or a stanza. Piper's theory of computational modeling may disarm critics who would describe computational textual analysis as invested in an "objectivity" that remains epistemologically inappropriate for literary criticism. Instead, Piper embraces a theory of literary modeling best summarized by an aphorism from statistics: All models are wrong, but some models are useful.

Each of Piper's six chapters studies a literary feature with a new computational model. As the page numbers increase, so too does the complexity of the models. This is by design. In its first and simplest model, Enumerations begins with a question of quantity: How many punctuation marks are there in poems over time? Uncovering what Piper describes as the "general economy of punctuation" (24) allows the critic to not only generalize well about the history of punctuation in poetry, but also identify moments where that trend has been interrupted. Between 1800 and 2000, Piper's model shows a steady decline in the amount of punctuation in poetry in the aggregate, though some poets like Amiri Baraka buck that trend. From here, the models get more complex, and so do the questions. Subsequent chapters analyze plot by comparing vocabulary distributions over narrative time, interpret a single topic about life and death from a topic model, assess computationally salient distinctions between fictional and nonfictional writing, measure the semantic space that surrounds specific characters in fiction, and attempt to measure the (in)stability of an author's oeuvre.

These computational models have two properties that Piper especially values, properties that he suggests can be useful for other critics to think with regardless of whether they use computational approaches in their own work—explicitness and representativeness. The computer code structuring the models makes explicit a set of repeatable and unambiguous instructions: if painting, throw fish. To this end, Piper has released the code underpinning the models in Enumerations for us to test. However—and this is less a weakness of Piper's book than a challenge for the field of computational literary studies today—the number of people with both the technical and literary expertise required to evaluate the fit between Piper's 7,000 lines of code and what they seek to measure is quite small. Indeed, for many in literary studies, computer code is hardly less obscure than the black box of critical charisma that Piper argues good modeling can unseat. This recalls a question asked by Matthew Kirschenbaum nearly a decade ago, but no less relevant today: "What is digital humanities, and what's it doing in English departments?" Piper's admirable openness about his models does not necessarily answer Kirschenbaum's question, particularly when we ask for whom the code reads as explicit.

For Piper, the second aspect that models foreground is the representativeness of their evidence. "In traditional critical practices, we only ever hear about the passages and works...