Assignment 1

Gender identification

How much can text analysis unearth about the gender identity of the author of the text?

by Arshiya Khattak and Sarah Al-Towaity

The cultural theory of gender and the distinction between what is biologically-assigned and what is culturally constructed are part of an exploration that bore a lot of complexity to the question: what exactly constitutes gender? So far, we have been mainly using quantitative tools to explore text genres (science fiction, fan-fiction, and academia). In this analysis, we explore whether the same tools could be used to learn more about those who produce text: the authors. We engage in the question of gender identification – more specifically, whether computational text mining can reveal a writer’s gender primarily from their work. Through our experiment with four science fiction works, written by four science fiction authors, two of which, in binary sex terms, are male, and the other two are female, we conclude that gender identification through surface-level text analysis, even between corpora containing texts with the same themes and genres, inherently lies on gender stereotypes. We not only ascribe this conclusion to the complexity of gender identity in general, but also, to the neglect of other confounding factors such as age, historical period, and societal status that can murk our analysis.

Pre-analysis

Going into the analysis, we knew that with text-based gender identification, corpus selection is critical. In an attempt to constrain possible confounding variables that variation in topic and theme can produce, we chose four science fiction works, each pair of which shares a common central theme. In each pair, one is written by a “female” author, while the other is written by a “male” author. The theme of the first pair was resurrection and giving life to dead corpses. In this category, we picked Mary Shelley’s “Frankenstein” and H. G. Wells’ “The Island of Dr Moreau”. In “Frankenstein”, a scientist imparts sentience into a non-living corpse in a scientific experiment. Analogously, “The Island of Dr Moreau” tells the story of a mad scientist who creates humanoid-like creatures from animal parts. For the second half of the corpus, we wanted to experiment with gender-based themes with “The Ruins of Isis” by Marion Zimmer Bradley that narrates the travels of an anthropologist through a female-dominated world and “Flatland: A Romance of Many Dimensions” by Edwin Abbott Abbot in which a two-dimensional world is inhabited by men who are complete polygons and women who are reduced to being simple lines. In our analysis, we retrieved all texts from Project Gutenberg, save for “The Ruins of Isis”, and we use RStudio, AntConc, and Voyant tools to explore indications of gender in the computational exploration of these texts.

Frankenstein and The Island of Doctor Moreau

"Frankenstein" and "The Island of Doctor Moreau" are both tales of scientists playing God ⁠— trying their hand at creation. While Victor Frankenstein fiddles with science to breathe life into a creature strewn together from pieces of corpses, Dr. Moreau attempts to create a new kind of life, Beast folk, creatures who are both man and animal. At first glance, both novels seem eerily similar, dealing with themes of life and death, and what it truly means to be human. But to ensure our research is honest, it is important to note the biggest contextual difference between the two works: "Frankenstein" was published in 1816, much earlier than "The Island Of Dr. Moreau", which was published in 1896.

Image retrieved from: https://m.imdb.com/title/tt0116654/movieconnections/?ref_=tt_trv_cnn

Image retrieved from: https://www.amazon.com/Frankenstein-Mary-Shelley/dp/1512308056

With this difference in mind, we continued our study of gender identification using the two pieces of literature. Our first comparison was done through RStudio, using the software to generate a visual representation of the differences between the books. The most glaring difference that we noticed in the scatterplot was the sort of language used in each book. Shelley tended to use softer language, with words such as “affection”, “gentle”, “family” and “child” being more present in "Frankenstein". In contrast, words like “brute”, “beast”, “afraid” and “bad” were more common in "The Island of Doctor Moreau". This aligns with conventional beliefs about gender norms ⁠— that women tend to be more soft spoken, while men are far harsher in their language. However, this finding remains inherently flawed for multiple reasons, the fundamental reason being the categorization of certain words as male and others as female. Larura Mandell, digital humanities researcher and professor, ascribed the failure behind Jan Rybicki’s stylometric gender identification experiment to the “inability, in this study, to produce a stable ‘canon’ of male and female keywords that would survive a change of corpus or shifts in literary evolution, or both”. It is evident, then, that the writing of female and male authors varies drastically across both genres and historical periods. In our conclusion, the words that we perceive as being feminine or masculine do not arrive from an in-depth understanding of the genre and the time of the texts, but rather, gender stereotypes as a whole.

Scatterplot demonstrating shared terms between Shelley's "Frankenstein" and Wells' "The Island of Doctor Moreau" and their frequencies relative to both works, generated using RStudio

Moreover, even though both books revolve around similar themes, they grapple with them in different ways. Shelley focuses more on the abstract themes of the novel, with words such as “life”, “death”, “hope”, “horror” and “despair” being comparatively more common in her work. In comparison, it is evident that Wells spends far more time on graphic descriptions, with words like “blood”, “body” and “black” appearing very frequently. In addition, words such as “hand”, “eyes” and “arm” tend to appear more often as well. Knowing the book’s subject matter, it is not unreasonable to deduce that these words are mostly used to describe dismembered body parts of various creatures. This difference in the texts could also hint towards the gender of each author, as media created and consumed by men tends to be more gore than media created and consumed by women. However, once again, this analysis appears feeble at best and entirely nonsensical at worst. According to outdated gender stereotypes, women are less likely to create content that veers towards science fiction and horror in the first place. This doesn’t hold, as Shelley’s work on "Frankenstein" preceded — and even in many cases inspired — other novels. In the words of Laura Mandell again, differences between male and female writings arise from “individual social context than with gender as a social category”. The differences in the narration of Shelley’s book and Wells' book can be explained through this as well. Many believe that Shelley’s work was a consequence of her own grim personal life, while it would seem that Wells’ book was more of a response to discussions surrounding animal vivisection, which were gaining traction in Europe at the time. Since Shelley’s writing came from her own morbid experiences with death, it would make sense for her to emphasize those aspects in her work, rather than the process of mutilation and creation.

The Ruins of Isis and Flatland

Image retrieved from: https://en.wikipedia.org/wiki/The_Ruins_of_Isis

Moving onto “The Ruins of Isis” and “Flatland”, we primarily rely on Voyant tools and AntConc to produce our analysis as the former is not archived by Project Gutenberg. Since both works involve gender-based world-building, we ran a direct reference to target corpus comparison using AntConc’s Keyword List tool. From that alone, we see that female-gendered words such as “women”, “pro-matriarch”, and “mother” and more stereotypically feminine value-centered terms such as “unity” and “love” are more common in “The Ruins of Isis” than in “Flatland”. Reversing the target and reference corpora, we obtain terms such as “line”, “circle”, “sphere” that are more frequent in “Flatland” than in its counterpart – terms that relate more to the specific taxonomy of the story itself, since characters are figure, and are otherwise nonsensical without performing some form of close reading. However, using Voyant’s Term Berry yielded some interesting results. The word “men” in “Flatland” is connected to “law” and “king”, both of which are power terms that are generally associated with masculinity, whereas “woman” is related to “half” and “man”. Superficially, the implication is that in “Flatland”, men are empowered, whereas a woman is relegated to being only half as worth as a man is, and since the assumption is that men in the 19th century were at the top of the societal hierarchy, such associations could be assumed to be produced by a male author. On the other hand, the focus of terms in “The Ruins of Isis” on matriarchal power, coupled with its release in the late 1900s at a time when the feminist movement experienced a lot of traction, hints at a female author.

AntCon's Keyword List with "The Ruins of `Isis" as the target corpus and "Flatland" as the reference.

AntCon's Keyword List with "Flatland" as the target corpus and "The Ruins of `Isis" as the reference.

It is imperative to note, however, that the favored gender in each created reality is aligned with each author’s supposed gender identity and we have chosen our corpus strategically for it to be so. Moreover, we knew the gender of the author before conducting any sort of analysis. It seems like in this case, as Mandell concluded perfectly, gender classification did not work without help, which began as early as when we selected our texts. Another problem that arises out of deductions of surface-level data is that certain nuanced themes can go undetected. For instance, with close reading, “Flatland” turns out to be a satirical science fiction work, intended to condemn society, not commend it. With this post-analysis knowledge in mind, one could just as easily wager that it might as well have been written by an indignant woman, aiming to expose the flaws of 19th century society. Thus, the reproducibility of our gender identification process is, quite frankly, dubious.

Image retrieved from: https://en.wikipedia.org/wiki/Flatland

Conclusion

Concluding, it seems we were able to classify authors based on gender because we relied on stereotypes for identification and chose pieces where such stereotypes were found and interpreted as gender differences. However, does this conclusion mean that we can never reliably use quantitative textual analysis for gender classification? Would we ever be able to conduct more nuanced computational analysis to match the complexity of gender identity? Or is it necessary to introduce qualitative methods to increase reliability? The answer to any of these questions is the topic of many discussions, but for now, we can settle with the conclusion that surface-level lexical-based gender identification inflates stereotypical gender differences that, in reality, may not necessarily be statistically significant at all.

Ready for grading!

Date: 26th October 2021

Page updated

Report abuse