46

Why I Dig: Feminist Approaches to Text Analysis

Lisa Marie Rhody

In an oft-quoted interview for Denver Quarterly in 1997, Jorie Graham, who had just received a Pulitzer Prize, offered a blistering critique of the state of contemporary American poetry. Lamenting the field’s lack of ambition, she proposed her own poetic experiments as a model for how contemporary poetry could pursue “the big hunger.” Graham employs postmodern and feminist strategies—syntactic complexity, narrative gaps, and vertigo-inducing alterations in perspective—to pose expansive, metaphysical questions that demand equal parts patience, participation, and restraint from her readers in the midst of uncomfortable textual circumstances.

It is Graham’s “big hunger”—the desire to ask sweeping metaphysical questions that speak directly to a contemporary moment—that has inspired my own work, where I have similarly sought to pose ambitious large-scale questions about literary tradition and genre convention. My “big hunger” compels me to consider what literary studies can bring to bear on the text mining of big data, a practice similarly steeped in the masculinized rhetoric of scale and ambition. Does the rhetoric of text analysis or its assumed empiricism dissuade feminist scholars from using it to pose questions about difference, erasure, and absence? How might the feminist literary critic approach text analysis without succumbing to the positivistic claims of objectivity that such methods so often encourage? What happens when we introduce diverse practitioners into text analysis, as Audre Lorde suggests, “as a fund of necessary polarities between which our creativity can spark like a dialectic?” (Lorde, 112).

In 2012, Bethany Nowviskie responded to discussions about a “Digging into Data” grant competition by asking “what do girls dig?” and wondering whether data mining had become “a gentleman’s sport.” Three years later, the question seems to have changed to “should women dig?” or, perhaps more fundamentally, should anyone dig? My own sense is that not only should women dig, but that it is necessary that we do so. Just as Lorde’s dialectic and Graham’s poetics activate feminist strategies of discomfort to unsettle deep-seated ideologies, so can feminist approaches to text analysis in literary studies. Feminist theory and methodology has already addressed research practices in the natural and physical sciences and social sciences; similar engagement is necessary in literary studies with respect to text analysis. “Big data” and computational text analysis, in fact, give feminist literary critics opportunities to embrace grand technical and social challenges through theorization and praxis, incorporating moments of productive discomfort into four stages of the research process: the assembling of text corpora; the translation of methodologies across disciplines; the gauging of interpretive value; and the selection of appropriate modes of scholarly expression. The language of feminist analysis, which challenges traditional modes of knowledge production, is well aligned with the practice of defamiliarizing a textual corpora in order to ask “new” questions at different scales.

While there is no single “feminist epistemology” or “feminist method,” feminist approaches to text mining might begin by exposing implicit and explicit choices that influence the construction of textual corpora, articulating the rationale for their selection, and carefully scoping the claims they make in deference to the representative limitations of their datasets. For example, by calling attention to androcentric biases, feminist empiricist approaches to text mining might deconstruct perceived errors in datasets or correct biases through activist interventions with the corpora that make gender a category of analysis (Scott). When topic modeling poetic corpora in my own research, for example, I have included detailed descriptions of my dataset, which was scraped from the American Academy of Poets (poets.org) website in 2012 and includes a disappointing ratio of three poems by male poets to every one by a woman. Productive discomfort rendered through the assembly of my dataset means balancing recuperative strategies for including much higher representations of poetry by women with a competing acknowledgment that the representation of work by women in the collection remains unsatisfactory.

Since most text-mining methods used in literary study have been forged in linguistics or computer sciences, the algorithms we use for literary text analysis projects, such as topic modeling or sentiment analysis, potentially demonstrate an intrinsic bias toward the source texts used during development. Though the difference may be negligible, feminist approaches might lay bare differences in disciplinary values embedded within an algorithm’s formal logic to expose moments of productive discomfort among transdisciplinary engagements, an approach akin to feminist calls for the integration of quantitative and qualitative research (Sprague and Zimmerman). For example, when beginning to work with topic modeling, my training as a scholar of poetry contrasted with the method’s tacit expectation that “small” words, including pronouns, articles, and prepositions, should be removed from the text. For literary critics of poetry, the position and selection of articles and pronouns signals an interpretive significance that is disregarded in the usual preparation of texts for computational analysis. To account for this difference in disciplinary practice, I performed extensive preliminary studies to measure the degree to which the extraction of small words interfered with values in literary criticism, and in doing so sought to assure my literary studies audiences that I had met their disciplinary, methodological expectations.1

Algorithms do precisely as they are told, enacting a carefully articulated and rigid logic; in text analysis, that logic is based on assumptions about how language operates. Algorithms do not interpret their own failures, but their errors generate moments of rupture for the feminist literary scholar, in whose hands error and marginality expose the fault lines encoded in predictive methods such as classifying and organizing text. Consequently, my approach to text-mining methods such as topic modeling differs from others. I use large poetic corpora to interrogate the assumptions of topic modeling—that documents sharing similar words likewise share thematic coherence. Where most topic-modeling results demonstrate thematic coherence, mine represent discourse coherence; subsequently, the model points to ekphrastic poems by women who share similar discourses as their male counterparts, but do so ironically. By challenging accepted research practices and testing the assumed logic of algorithms rather than using it to classify texts into settled categories, feminist interventions into topic modeling effectively resist the positivistic paradigms.

New forms of knowledge sharing that are familiar to practitioners in text analysis may elicit frustration or even confusion from colleagues in literary studies, exposing a fourth site of productive discomfort: scholarly communication. Unfamiliar forms of social scholarly exchange, such as R and Python vignettes in GitHub,2 blog posts (Swafford), or datasets on Figshare (Rand, Kraft-Todd, and Gruber), can unsettle deeply held expectations for sharing academic work. Here, feminist epistemologies of knowledge production afford scholarly debate in text analysis a language to begin engaging and responding to the collective unease of readers and authors. The formation of democratic communities of knowledge production requires mutual respect for the epistemic contributions of participants and care that their work is given credibility regardless of their academic status, identity, or mode of discourse (Grasswick).

So why do I dig? I dig as an act of re-vision. I dig to raise more questions than I can answer. I dig because—much like Jorie Graham’s poetic “big hunger”—it affords one more way “to wrest beauty and meaning—however tentative and qualified—from the abyss of language and randomness of experience” (Costello). I dig because I believe that feminist literary studies can offer a necessary corrective to the androcentric tendencies of “big data,” one that, in its unsettling, directly addresses the technological and epistemological challenges of our age.

Notes

Bibliography