Experiments in Computational Criticism #4: Visualizing VLC Keywords

Here are some data visualizations I’ve wanted to make since @VLCjournal released its issue on “Keywords.”

Editors Danny Hack and Rachel Ablow emphasized that these keywords were not meant to capture the precise contours of the field. Nonetheless, I wondered what the field would look like if one did visualize the articles in this issue.

 I also wondered in what sense these terms were “keywords.” Were they “keywords” in the sense that they were explanatory words of particular significance, or key in the sense that they indicated or represented the content of a larger text or set of texts: in this case, critical work on Victorian literature and culture (www.oed.com/view/Entry/312961)?

 It struck me that a good measure of this issue might be the co-occurrence of these terms throughout the journal articles (using Ken Benoit’s quanteda package (https://quanteda.io/)). If one suggests that “empire” is key to contemporary work on Victorian literature and culture, then one would expect to find that articles on another keyword, such as “science,” might also mention “empire.” Visualizing a network of these co-occurrences would depict in some fashion the contours of the field as viewed through the Keywords issue, and allow one to measure, in some form, whether certain words are currently more “key” than others.

 If one uses the quanteda package (https://quanteda.io/) to look at the data knowing what the listed keywords might be, the contours of the field might look like this. In the network below I have, for clarity, only plotted the keywords which co-occur more than 10 times. In other words, each edge signals that the two vertices it connects appear together within at least 10 of the entries in the Keywords issue. As we can see, there is a somewhat dense network on the right of keywords which did frequently co-occur, such as “literature,” “work,” “reading,” and “politics.” On the left, however, we have those terms which may be key, but which seem to be less central to the field, given that they are referenced less frequently in other articles: terms like “Anthropocene” and “child.” Much of this seems to align with what I intuit about the shape of critical work on Victorian literature and culture. “Form,” “reading,” “literature,” and “politics” certainly seem to be central concerns.

However, this is only what the network looks like if we only look at the words authors said were key. If we use quanteda to determine the top features which co-occur in all the articles (editing the data somewhat to remove common terms pronouns, articles, etc.), the result looks like the network below. I have, for clarity, only plotted the top features which co-occur more than 55 times. As we can see, this is a much more conventional view of work on Victorian literature and culture, centered on words like “Victorian,” “Nineteenth-Century,” “history,” “novel,” etc. Very few of the Keywords provided by the journal appear, even though each of the articles is centered on these keywords.

 The data I have visualized here is not meant to be a definitive analysis of the Keywords issue or of modern scholarship on Victorian literature and culture. I have done relatively little work cleaning these text files, for instance, meaning that there certainly could be textual features biasing these visual representations. Nonetheless, there are two lessons I think we can draw from these visualizations.

1.       The editors are right to insist that the keywords they have provided are not meant to provide an image of the contours of the field. Indeed, we see from a comparison of the two images that even the contours of this Keywords issue can take different forms when approached in different ways.

2.       The differences between these visualizations may be a sign that the keywords are functioning exactly as intended, providing a key to work in the field that allows us to view the field in new ways. We all know that academic work on Victorian literature and culture prioritizes literature and history. The importance of work on less obvious issues such as the Victorian child or on ecology can easily be overshadowed. The two figures above demonstrate that we still need scholars (such as the contributors to the “Keywords” issue) to reflect on the topics they find especially generative, in order to draw attention to new aspects of the field which might later become central.