Recently, text miners have analyzed gendered discourse based on a binary opposition, male/female (M/F), trying to determine distinctively ‘female writing style,’ ‘female keywords,’ or ‘female themes’ (Rybicki 2015; Jockers 2011, 2013). The terms ‘male’ and ‘female’ suggest biology and hence were abandoned by literary critics during the big feminist recovery projects of ‘women writers’: “women’ was used in preference to ‘female’ despite the fact that the former is a noun, not an adjective, indicating that gender was a cultural formation rather than a biological one. More theoretically enlightened text miners have used the tools of data analytics to trace changes through time in those attributes assigned to women and those assigned to men, examining how notions of gender change over time (Garg, Scheibinger, Jurafksy, Zou 2017; Underwood, Bamman, and Lee 2018; Olsen 1992). There is another way to analyze gender historically using digital tools without assuming a biological basis for differences between men and women that involves searching for gender categories beyond the binary opposition M/F. In her book
Gender Trouble, Judith Butler encourages undermining the M/F binary opposition by ‘proliferating’ identity categories (Butler 1990, pp. 17, 146). Butler says that ‘the very notion of the subject, intelligible only through its appearance as gendered, admits of possibilities that have been forcibly foreclosed by . . . various reifications of gender’ into the M/F binary (Butler 1990, p. 33).

The result of Butler’s call to multiply gender categories has been the creation of what Bowker and Star call “boundary objects” (1999): “cisgender” and “transgender” have expanded the binary while still relying on the underlying classification of m/f. But to apply these categories on historical documents is anachronistic: there are historically accurate gender categories that have been identified by others for eighteenth-century such as “molly” (Alan Bray 1988; Randolph Trumbach 1991) and “sapphist” (Lisa Lynn Moore 1997; Yopie Prins 1999). But what about others that have not yet been identified by readers? The Feminist Controversy in England project tries to find foreclosed identity categories, to uncover historically specific gender designations in novels, pamphlets, and essays written by women between 1788 and 1810 in England.

In 1974, Garland Publishing (now no longer in existence) published a collection of 44 treatises by women authors published on topics related to emerging feminism, edited by Gina Luria Walker (
https://books.google.com/books/about/The_Feminist_Controversy_in_England_1788.html?id=j1pqMwEACAAJ ). Mary Wollstonecraft’s ground-breaking
Vindication of the Rights of Women (1792) was among them. They were facsimile editions. We have used Optical Character Recognition software (Tesseract 3, trained for 18
^th-century typefaces by the Early Modern OCR Project,
http://emop.tamu.edu), corrected the OCR using TypeWright (
http://www.18thconnect.org/typewright/documents), run through Named Entity Recognition software to identify character names, and uploaded into the Catma.de interface where they have been tagged by three different teams: undergraduate students, graduate students, and the Professor who is the Principal Investigator on the project. Each team used its own taxonomy explicitly defined in Catma.de except the PI who derives a set of tags from the texts themselves.

The first, most basic taxonomy according to which the texts were tagged by undergraduat students identifies personality traits and activities of characters. The second more interpretive set of tags, encoded by graduate students, involves formal features of novels and essays–protagonists, narrators, and other character types. After these two procedures, each text’s tags are clustered by character in a graph, an interactive d3.js interface that allows a third round of tagging by the PI: the personality traits and activities (character attributes) are tagged either as ‘gender-normative’ or ‘different,’ and the different categories are given what Johnny Saldaña calls ‘In Vivo’ codes, short phrases that come from the language of the text itself (Saldaña 2009, 2016). Afterwards, these In Vivo codes are regularized across the whole set of documents. A second visualization interface provides a network view of all the characters grouped by their connections to tagged attributes, both gender normative and different. The goal has been to discover characters clustering around a set of non-normative character attributes–that is, to find personality traits and activities that are both different and shared, which is to say not merely a matter of any specific character’s personality. We argue that such clusters present alternative gender categories, based of course upon m/f norms (as are ‘cisgender’ and ‘transgender’) but contesting those norms nonetheless.

At DH2018, we present preliminary findings using our prototype visualization interfaces. As we have discovered so far, many characters share attributes in common with Harriet Freke, a character in Maria Edgeworth’s
Belinda. Thus we argue that ‘freke’ represents a specific gender category found in many of the transcribed texts. The goal is to postulate non-binary gender terms that have been derived from the texts themselves, and to demonstrate how this procedure offers an alternative method for historicizing gender.