PART II ][ Chapter 17

The Elusive Digital / Critical Synthesis

Seth Long and James Baker

In the first edition of Debates in the Digital Humanities, Alan Liu argued that digital humanists risk losing a seat at the “table of debate” if they continue to emphasize tools and databases to the exclusion of cultural criticism. Digital humanists, Liu wrote, must learn to “move seamlessly between text analysis and cultural analysis” if they are not to become a service industry to the humanities, providing tools and data for other scholars but not contributing to key debates. Not long afterward, the expression “hacking and yacking” emerged as another iteration of the difference between cultural critique and tool- and data-centric research. In 2014, Bethany Nowviskie tried to put the expression to rest, noting that the digital humanities has plenty of room for both hacking and yacking (“On the Origin”). Roopika Risam has likewise expressed optimism about “transcending” the “simplistic hack/yack binary.” The field continues to seem optimistic about the compatibility of digital methods and cultural criticism. Liu’s challenge is being met.

Taking a contrary stance, we want to argue that hacking and yacking are not as easily compatible as some have claimed. Digital and critical work are not incommensurable, but they do denote two different sets of practices. In our view, many critical DH projects give a perfunctory nod to one method of work while devoting the majority of intellectual labor to the other. They struggle toward a synthesis of these different modes of research and inquiry, though ideally a critical DH project will meet the requirements of the digital and the critical in equal measure. We want to remind the field how difficult it is to achieve the digital/critical synthesis. Liu’s challenge is a more wicked problem than we want to admit, particularly in the context of computational work.

To pinpoint where we think the problem lies, consider the typical workflow of a computational DH project: data collection, tagging, and organization; computational analysis of data; interpretation of results. We suggest that the second and third steps, computation and interpretation, are the ones not easily synthesized with a critical approach but that the first step—data collection and management—is more amenable to cultural criticism. At the end of this chapter, we explain our optimistic take on this initial stage of research, but first we want to explore the difficulty of maintaining a critical stance when one enters the computation and interpretation stages.

One obstacle in these stages is a tension between the critique of structure and the discovery or use of structure. As Liu notes, cultural criticism accepts the structural nature of human existence along lines of class and identity, but it does so as a means of questioning the utility and/or ontological status of those structuring categories. Cultural criticism recognizes structure as a pretext to argue that our structuring categories result not from natural but contingent social conditions that may be altered to produce better social structures. Categories, structures, patterns: for the cultural critic, these are to be questioned, critiqued, speculatively reimagined. In contrast, when working at a command line, categories, structures, and patterns are to be used or discovered. For example, ethnicity and gender may be socially conditioned and thus fluid, but the researcher studying their effects on cultural production must treat both as finite and discrete. Even an unsupervised algorithm such as a topic model needs to be told how many discrete topics to look for. To some extent, math forces one to assume discreteness over fluidity.

Another obstacle—one related to analysis—involves how digital humanists versus cultural critics treat gaps in data. The former recognize the reality of gaps—as, for example, Lauren Klein recognizes the gaps in Thomas Jefferson’s letters—but proceed to compute and construct knowledge despite those gaps. The latter, in contrast, take as their core project a discussion of the problematic ethics of constructing knowledge from data with racialized or gendered gaps. Indeed, for some critics, creating knowledge from data in which marginal identities are absent will reproduce unequal power dynamics, no matter what are the intentions of the researcher (see, for example, Wyse’s critique of white sociology, 15–33).

A third problem and perhaps the largest one facing a digital/critical synthesis is that, in computational work, one’s results are necessarily open to more than one explanation. To clarify what we mean, consider the traditional style of cultural critique. Grounded in a writer’s ethos, it undertakes an inquiry that so deftly combines evidence (often close readings) and interpretation that it is not always clear where evidence ends and interpretation begins. Cultural criticism is a rhetorical as much as a “factual” endeavor. Computational work, in contrast, clearly separates data, methods, and interpretation. Data are analyzed, and then the results are interpreted. Due to this strict separation of evidence and argument, of analysis and interpretation, computational work facilitates a culture of multiple explanations and methodological fine-tuning; this is why data-driven debates often devolve into debates about formulas and parameters. As the social and natural sciences demonstrate, computational results rarely foreclose all but a single explanation. As an example, consider the humorous coordinate axis whose points have been plotted to look like a duck in one direction and a rabbit in another. The data are the data; their interpretation is up to the viewer. Thus any project that separates categorical results from their interpretation will struggle to spotlight only and necessarily a progressive social critique. Critique occurs in interpretation, so when working with computational results, a critical interpretation will rarely be the only possible one.

Despite this tension between critical and digital work, many scholars have offered examples of their successful syntheses. For example, Ted Underwood’s work on gender in fiction has been highlighted as an example of a digital/critical synthesis (Spahr et al., “Beyond Resistance”). Using a logistic regression, Underwood finds that between 1800 and 1989, the words associated with male versus female characters in fiction became more volatile over time: the more contemporary the textual data, the more difficult it is for a regression model to predict whether a set of words is being applied to a male or a female. “Gender,” Underwood concludes, “is not at all the same thing in 1980 that it was in 1840.”

The social critic could certainly use Underwood’s data to make an argument about gendered power dynamics or the social construction of gender. But Underwood makes neither move. First, he notes that despite the trend away from sharply gendered terms, there remain important countertrends; for example, physical descriptions emerge as salient aspects of gender distinction in the twentieth century. Second, he frames “fluidity” not as a critical lens but as a problem of statistical gradation that computational methods can help solve: “There’s nothing very novel about the discovery that gender is fluid. But of course, we like to say everything is fluid: genres, roles, geographies. The advantage of a comparative method is that it lets us say specifically what we mean. Fluid compared to what?”

Is this methodological framing the same thing as cultural criticism? Just because a DH project utilizes gender as a category, is it by default a critical DH project? In line with our preceding discussion, there seem to be a few reasons why it is not.

First, though one could use Underwood’s data in a project about the fluidity of gender, such arguments would necessarily be subsequent to the logistic regression, which treats male and female not as social categories but as a 0 and a 1 with equal weight. Because the critical reading of the results would follow the computation of the results, it is not clear in what way the computational work itself is “critical,” given that its results do not support only and necessarily a critical reading. The results could be used to support gender essentialism as easily as gender fluidity (e.g., gender differentiation in fiction does not disappear, but just surfaces along a new terminological axis). As we noted, it is not easy to fuse a social critique with a methodology that separates results from interpretation. Indeed, the messiness of interpretation and algorithmic parameter tweaking are the price one pays when adopting structural methods, and it is unclear how one balances this messy uncertainty with the conviction of cultural (especially activist) critique. (In their essay on the whiteness of MFA programs, “The Program Era,” Juliana Spahr and Stephanie Young attempt that balance, but tellingly voice their frustration with it in their concluding paragraph.)

Second, in terms of knowledge construction, Underwood is not asking the questions about power dynamics that Adeline Koh and others involved in the #TransformDH collective posit as key for critical DH work. For these scholars, technology, its uses, and data collection and methodology selection should be submitted to a critique informed by the lenses of race, gender, and sexuality, so that when we sit down to the command line, we do so with activist as well as academic goals in mind (Lothian). A researcher may study the stylistics of gender with a logistic regression, but unless she incorporates issues of gendered power into the project, she is not involved in a critical DH project. It is not clear if Underwood’s research—a history of the stylistics of gender—is critical in the sense deployed by these scholars. Indeed, a strike against assuming so is that Underwood adopts a binary notion of gender as an explanatory set. In line with the activist goals championed by Koh and Lothian, a critic might ask why he used a binary logistic regression rather than a multinomial logistic regression, leaving a third option for “other,” which might have moved his project in a critical direction (even here, however, one would be subsuming the notion of gender nondiscreteness under a single discrete category).

In our view, Underwood’s project demonstrates not the ease but rather the elusiveness of a digital/critical synthesis. Alan Liu points to Franco Moretti’s work as another example of digital/critical synthesis, but both Moretti’s detractors (Prendergast; Allington, Brouillette, and Golumbia) and Moretti himself (155–58) would question that claim.

The digital humanities qua criticism is difficult. In our view, the best strategy is to admit that hacking and yacking are different activities. One relies on structures and categories, and the other critiques them. But there is nothing wrong with difference. Simply let the two modes of work exist, not in opposition, but in dialectical tension. Allow computational and critical practices to remain separate, but throw them together in projects in which neither takes precedence. Perhaps more than digital/critical synthesis, what the digital humanities needs is a new style of research writing that revels in the play of competing methodological and epistemological emphases.

Despite our skepticism about the ease with which computational and critical work might be merged, we nevertheless want to conclude on an optimistic note. Recall the DH workflow with which we began the chapter: data collection, tagging, and organization; computational analysis of data; interpretation of results. Although we have argued that computation and interpretation pose problems for a digital/critical synthesis, the collection, labeling, and organization of data at the beginning of a project are much more amenable to critical insight. It is here, in the initial stages of a project, that a researcher might return absent identities to the data, as well as label or structure data in a manner informed by previously marginalized knowledge bases—perhaps by inviting members of marginalized communities to participate in the structuring of the data before its analysis. Doing so may or may not affect one’s analysis or the range of interpretations it allows, but it will at least partially address the critical points regarding marginalization and power. Indeed, Koh’s critical questions seem mainly targeted to this initial stage of research: “Which agents do we give agency to in a project and why? Who are the voices that are allowed to speak, and who are heard?” Perhaps then it is in deciding what data to use and how to structure them that the epistemological and the activist concerns of critique might be enacted.

Bibliography